]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - Intel(R) Matrix Storage Manager Support | |
3 | * | |
4 | * Copyright (C) 2002-2008 Intel Corporation | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms and conditions of the GNU General Public License, | |
8 | * version 2, as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | ||
20 | #define HAVE_STDINT_H 1 | |
21 | #include "mdadm.h" | |
22 | #include "mdmon.h" | |
23 | #include "sha1.h" | |
24 | #include "platform-intel.h" | |
25 | #include <values.h> | |
26 | #include <scsi/sg.h> | |
27 | #include <ctype.h> | |
28 | #include <dirent.h> | |
29 | ||
30 | /* MPB == Metadata Parameter Block */ | |
31 | #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. " | |
32 | #define MPB_SIG_LEN (strlen(MPB_SIGNATURE)) | |
33 | #define MPB_VERSION_RAID0 "1.0.00" | |
34 | #define MPB_VERSION_RAID1 "1.1.00" | |
35 | #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00" | |
36 | #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01" | |
37 | #define MPB_VERSION_RAID5 "1.2.02" | |
38 | #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04" | |
39 | #define MPB_VERSION_CNG "1.2.06" | |
40 | #define MPB_VERSION_ATTRIBS "1.3.00" | |
41 | #define MAX_SIGNATURE_LENGTH 32 | |
42 | #define MAX_RAID_SERIAL_LEN 16 | |
43 | ||
44 | /* supports RAID0 */ | |
45 | #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001) | |
46 | /* supports RAID1 */ | |
47 | #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002) | |
48 | /* supports RAID10 */ | |
49 | #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004) | |
50 | /* supports RAID1E */ | |
51 | #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008) | |
52 | /* supports RAID5 */ | |
53 | #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010) | |
54 | /* supports RAID CNG */ | |
55 | #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020) | |
56 | /* supports expanded stripe sizes of 256K, 512K and 1MB */ | |
57 | #define MPB_ATTRIB_EXP_STRIPE_SIZE __cpu_to_le32(0x00000040) | |
58 | ||
59 | /* The OROM Support RST Caching of Volumes */ | |
60 | #define MPB_ATTRIB_NVM __cpu_to_le32(0x02000000) | |
61 | /* The OROM supports creating disks greater than 2TB */ | |
62 | #define MPB_ATTRIB_2TB_DISK __cpu_to_le32(0x04000000) | |
63 | /* The OROM supports Bad Block Management */ | |
64 | #define MPB_ATTRIB_BBM __cpu_to_le32(0x08000000) | |
65 | ||
66 | /* THe OROM Supports NVM Caching of Volumes */ | |
67 | #define MPB_ATTRIB_NEVER_USE2 __cpu_to_le32(0x10000000) | |
68 | /* The OROM supports creating volumes greater than 2TB */ | |
69 | #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000) | |
70 | /* originally for PMP, now it's wasted b/c. Never use this bit! */ | |
71 | #define MPB_ATTRIB_NEVER_USE __cpu_to_le32(0x40000000) | |
72 | /* Verify MPB contents against checksum after reading MPB */ | |
73 | #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000) | |
74 | ||
75 | /* Define all supported attributes that have to be accepted by mdadm | |
76 | */ | |
77 | #define MPB_ATTRIB_SUPPORTED (MPB_ATTRIB_CHECKSUM_VERIFY | \ | |
78 | MPB_ATTRIB_2TB | \ | |
79 | MPB_ATTRIB_2TB_DISK | \ | |
80 | MPB_ATTRIB_RAID0 | \ | |
81 | MPB_ATTRIB_RAID1 | \ | |
82 | MPB_ATTRIB_RAID10 | \ | |
83 | MPB_ATTRIB_RAID5 | \ | |
84 | MPB_ATTRIB_EXP_STRIPE_SIZE) | |
85 | ||
86 | /* Define attributes that are unused but not harmful */ | |
87 | #define MPB_ATTRIB_IGNORED (MPB_ATTRIB_NEVER_USE) | |
88 | ||
89 | #define MPB_SECTOR_CNT 2210 | |
90 | #define IMSM_RESERVED_SECTORS 4096 | |
91 | #define NUM_BLOCKS_DIRTY_STRIPE_REGION 2056 | |
92 | #define SECT_PER_MB_SHIFT 11 | |
93 | ||
94 | /* Disk configuration info. */ | |
95 | #define IMSM_MAX_DEVICES 255 | |
96 | struct imsm_disk { | |
97 | __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */ | |
98 | __u32 total_blocks; /* 0xE8 - 0xEB total blocks */ | |
99 | __u32 scsi_id; /* 0xEC - 0xEF scsi ID */ | |
100 | #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */ | |
101 | #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */ | |
102 | #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */ | |
103 | __u32 status; /* 0xF0 - 0xF3 */ | |
104 | __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */ | |
105 | #define IMSM_DISK_FILLERS 4 | |
106 | __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */ | |
107 | }; | |
108 | ||
109 | /* RAID map configuration infos. */ | |
110 | struct imsm_map { | |
111 | __u32 pba_of_lba0; /* start address of partition */ | |
112 | __u32 blocks_per_member;/* blocks per member */ | |
113 | __u32 num_data_stripes; /* number of data stripes */ | |
114 | __u16 blocks_per_strip; | |
115 | __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */ | |
116 | #define IMSM_T_STATE_NORMAL 0 | |
117 | #define IMSM_T_STATE_UNINITIALIZED 1 | |
118 | #define IMSM_T_STATE_DEGRADED 2 | |
119 | #define IMSM_T_STATE_FAILED 3 | |
120 | __u8 raid_level; | |
121 | #define IMSM_T_RAID0 0 | |
122 | #define IMSM_T_RAID1 1 | |
123 | #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */ | |
124 | __u8 num_members; /* number of member disks */ | |
125 | __u8 num_domains; /* number of parity domains */ | |
126 | __u8 failed_disk_num; /* valid only when state is degraded */ | |
127 | __u8 ddf; | |
128 | __u32 filler[7]; /* expansion area */ | |
129 | #define IMSM_ORD_REBUILD (1 << 24) | |
130 | __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members], | |
131 | * top byte contains some flags | |
132 | */ | |
133 | } __attribute__ ((packed)); | |
134 | ||
135 | struct imsm_vol { | |
136 | __u32 curr_migr_unit; | |
137 | __u32 checkpoint_id; /* id to access curr_migr_unit */ | |
138 | __u8 migr_state; /* Normal or Migrating */ | |
139 | #define MIGR_INIT 0 | |
140 | #define MIGR_REBUILD 1 | |
141 | #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */ | |
142 | #define MIGR_GEN_MIGR 3 | |
143 | #define MIGR_STATE_CHANGE 4 | |
144 | #define MIGR_REPAIR 5 | |
145 | __u8 migr_type; /* Initializing, Rebuilding, ... */ | |
146 | __u8 dirty; | |
147 | __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */ | |
148 | __u16 verify_errors; /* number of mismatches */ | |
149 | __u16 bad_blocks; /* number of bad blocks during verify */ | |
150 | __u32 filler[4]; | |
151 | struct imsm_map map[1]; | |
152 | /* here comes another one if migr_state */ | |
153 | } __attribute__ ((packed)); | |
154 | ||
155 | struct imsm_dev { | |
156 | __u8 volume[MAX_RAID_SERIAL_LEN]; | |
157 | __u32 size_low; | |
158 | __u32 size_high; | |
159 | #define DEV_BOOTABLE __cpu_to_le32(0x01) | |
160 | #define DEV_BOOT_DEVICE __cpu_to_le32(0x02) | |
161 | #define DEV_READ_COALESCING __cpu_to_le32(0x04) | |
162 | #define DEV_WRITE_COALESCING __cpu_to_le32(0x08) | |
163 | #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10) | |
164 | #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20) | |
165 | #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40) | |
166 | #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80) | |
167 | #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100) | |
168 | #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200) | |
169 | #define DEV_CLONE_N_GO __cpu_to_le32(0x400) | |
170 | #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800) | |
171 | #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000) | |
172 | __u32 status; /* Persistent RaidDev status */ | |
173 | __u32 reserved_blocks; /* Reserved blocks at beginning of volume */ | |
174 | __u8 migr_priority; | |
175 | __u8 num_sub_vols; | |
176 | __u8 tid; | |
177 | __u8 cng_master_disk; | |
178 | __u16 cache_policy; | |
179 | __u8 cng_state; | |
180 | __u8 cng_sub_state; | |
181 | #define IMSM_DEV_FILLERS 10 | |
182 | __u32 filler[IMSM_DEV_FILLERS]; | |
183 | struct imsm_vol vol; | |
184 | } __attribute__ ((packed)); | |
185 | ||
186 | struct imsm_super { | |
187 | __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */ | |
188 | __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */ | |
189 | __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */ | |
190 | __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */ | |
191 | __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */ | |
192 | __u32 error_log_size; /* 0x30 - 0x33 in bytes */ | |
193 | __u32 attributes; /* 0x34 - 0x37 */ | |
194 | __u8 num_disks; /* 0x38 Number of configured disks */ | |
195 | __u8 num_raid_devs; /* 0x39 Number of configured volumes */ | |
196 | __u8 error_log_pos; /* 0x3A */ | |
197 | __u8 fill[1]; /* 0x3B */ | |
198 | __u32 cache_size; /* 0x3c - 0x40 in mb */ | |
199 | __u32 orig_family_num; /* 0x40 - 0x43 original family num */ | |
200 | __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */ | |
201 | __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */ | |
202 | #define IMSM_FILLERS 35 | |
203 | __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */ | |
204 | struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */ | |
205 | /* here comes imsm_dev[num_raid_devs] */ | |
206 | /* here comes BBM logs */ | |
207 | } __attribute__ ((packed)); | |
208 | ||
209 | #define BBM_LOG_MAX_ENTRIES 254 | |
210 | ||
211 | struct bbm_log_entry { | |
212 | __u64 defective_block_start; | |
213 | #define UNREADABLE 0xFFFFFFFF | |
214 | __u32 spare_block_offset; | |
215 | __u16 remapped_marked_count; | |
216 | __u16 disk_ordinal; | |
217 | } __attribute__ ((__packed__)); | |
218 | ||
219 | struct bbm_log { | |
220 | __u32 signature; /* 0xABADB10C */ | |
221 | __u32 entry_count; | |
222 | __u32 reserved_spare_block_count; /* 0 */ | |
223 | __u32 reserved; /* 0xFFFF */ | |
224 | __u64 first_spare_lba; | |
225 | struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES]; | |
226 | } __attribute__ ((__packed__)); | |
227 | ||
228 | ||
229 | #ifndef MDASSEMBLE | |
230 | static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" }; | |
231 | #endif | |
232 | ||
233 | #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209 | |
234 | ||
235 | #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */ | |
236 | ||
237 | #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must | |
238 | * be recovered using srcMap */ | |
239 | #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has | |
240 | * already been migrated and must | |
241 | * be recovered from checkpoint area */ | |
242 | struct migr_record { | |
243 | __u32 rec_status; /* Status used to determine how to restart | |
244 | * migration in case it aborts | |
245 | * in some fashion */ | |
246 | __u32 curr_migr_unit; /* 0..numMigrUnits-1 */ | |
247 | __u32 family_num; /* Family number of MPB | |
248 | * containing the RaidDev | |
249 | * that is migrating */ | |
250 | __u32 ascending_migr; /* True if migrating in increasing | |
251 | * order of lbas */ | |
252 | __u32 blocks_per_unit; /* Num disk blocks per unit of operation */ | |
253 | __u32 dest_depth_per_unit; /* Num member blocks each destMap | |
254 | * member disk | |
255 | * advances per unit-of-operation */ | |
256 | __u32 ckpt_area_pba; /* Pba of first block of ckpt copy area */ | |
257 | __u32 dest_1st_member_lba; /* First member lba on first | |
258 | * stripe of destination */ | |
259 | __u32 num_migr_units; /* Total num migration units-of-op */ | |
260 | __u32 post_migr_vol_cap; /* Size of volume after | |
261 | * migration completes */ | |
262 | __u32 post_migr_vol_cap_hi; /* Expansion space for LBA64 */ | |
263 | __u32 ckpt_read_disk_num; /* Which member disk in destSubMap[0] the | |
264 | * migration ckpt record was read from | |
265 | * (for recovered migrations) */ | |
266 | } __attribute__ ((__packed__)); | |
267 | ||
268 | static __u8 migr_type(struct imsm_dev *dev) | |
269 | { | |
270 | if (dev->vol.migr_type == MIGR_VERIFY && | |
271 | dev->status & DEV_VERIFY_AND_FIX) | |
272 | return MIGR_REPAIR; | |
273 | else | |
274 | return dev->vol.migr_type; | |
275 | } | |
276 | ||
277 | static void set_migr_type(struct imsm_dev *dev, __u8 migr_type) | |
278 | { | |
279 | /* for compatibility with older oroms convert MIGR_REPAIR, into | |
280 | * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status | |
281 | */ | |
282 | if (migr_type == MIGR_REPAIR) { | |
283 | dev->vol.migr_type = MIGR_VERIFY; | |
284 | dev->status |= DEV_VERIFY_AND_FIX; | |
285 | } else { | |
286 | dev->vol.migr_type = migr_type; | |
287 | dev->status &= ~DEV_VERIFY_AND_FIX; | |
288 | } | |
289 | } | |
290 | ||
291 | static unsigned int sector_count(__u32 bytes) | |
292 | { | |
293 | return ((bytes + (512-1)) & (~(512-1))) / 512; | |
294 | } | |
295 | ||
296 | static unsigned int mpb_sectors(struct imsm_super *mpb) | |
297 | { | |
298 | return sector_count(__le32_to_cpu(mpb->mpb_size)); | |
299 | } | |
300 | ||
301 | struct intel_dev { | |
302 | struct imsm_dev *dev; | |
303 | struct intel_dev *next; | |
304 | unsigned index; | |
305 | }; | |
306 | ||
307 | struct intel_hba { | |
308 | enum sys_dev_type type; | |
309 | char *path; | |
310 | char *pci_id; | |
311 | struct intel_hba *next; | |
312 | }; | |
313 | ||
314 | enum action { | |
315 | DISK_REMOVE = 1, | |
316 | DISK_ADD | |
317 | }; | |
318 | /* internal representation of IMSM metadata */ | |
319 | struct intel_super { | |
320 | union { | |
321 | void *buf; /* O_DIRECT buffer for reading/writing metadata */ | |
322 | struct imsm_super *anchor; /* immovable parameters */ | |
323 | }; | |
324 | union { | |
325 | void *migr_rec_buf; /* buffer for I/O operations */ | |
326 | struct migr_record *migr_rec; /* migration record */ | |
327 | }; | |
328 | size_t len; /* size of the 'buf' allocation */ | |
329 | void *next_buf; /* for realloc'ing buf from the manager */ | |
330 | size_t next_len; | |
331 | int updates_pending; /* count of pending updates for mdmon */ | |
332 | int current_vol; /* index of raid device undergoing creation */ | |
333 | __u32 create_offset; /* common start for 'current_vol' */ | |
334 | __u32 random; /* random data for seeding new family numbers */ | |
335 | struct intel_dev *devlist; | |
336 | struct dl { | |
337 | struct dl *next; | |
338 | int index; | |
339 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
340 | int major, minor; | |
341 | char *devname; | |
342 | struct imsm_disk disk; | |
343 | int fd; | |
344 | int extent_cnt; | |
345 | struct extent *e; /* for determining freespace @ create */ | |
346 | int raiddisk; /* slot to fill in autolayout */ | |
347 | enum action action; | |
348 | } *disks, *current_disk; | |
349 | struct dl *disk_mgmt_list; /* list of disks to add/remove while mdmon | |
350 | active */ | |
351 | struct dl *missing; /* disks removed while we weren't looking */ | |
352 | struct bbm_log *bbm_log; | |
353 | struct intel_hba *hba; /* device path of the raid controller for this metadata */ | |
354 | const struct imsm_orom *orom; /* platform firmware support */ | |
355 | struct intel_super *next; /* (temp) list for disambiguating family_num */ | |
356 | }; | |
357 | ||
358 | struct intel_disk { | |
359 | struct imsm_disk disk; | |
360 | #define IMSM_UNKNOWN_OWNER (-1) | |
361 | int owner; | |
362 | struct intel_disk *next; | |
363 | }; | |
364 | ||
365 | struct extent { | |
366 | unsigned long long start, size; | |
367 | }; | |
368 | ||
369 | /* definitions of reshape process types */ | |
370 | enum imsm_reshape_type { | |
371 | CH_TAKEOVER, | |
372 | CH_MIGRATION, | |
373 | }; | |
374 | ||
375 | /* definition of messages passed to imsm_process_update */ | |
376 | enum imsm_update_type { | |
377 | update_activate_spare, | |
378 | update_create_array, | |
379 | update_kill_array, | |
380 | update_rename_array, | |
381 | update_add_remove_disk, | |
382 | update_reshape_container_disks, | |
383 | update_reshape_migration, | |
384 | update_takeover, | |
385 | update_general_migration_checkpoint, | |
386 | }; | |
387 | ||
388 | struct imsm_update_activate_spare { | |
389 | enum imsm_update_type type; | |
390 | struct dl *dl; | |
391 | int slot; | |
392 | int array; | |
393 | struct imsm_update_activate_spare *next; | |
394 | }; | |
395 | ||
396 | struct geo_params { | |
397 | int dev_id; | |
398 | char *dev_name; | |
399 | long long size; | |
400 | int level; | |
401 | int layout; | |
402 | int chunksize; | |
403 | int raid_disks; | |
404 | }; | |
405 | ||
406 | enum takeover_direction { | |
407 | R10_TO_R0, | |
408 | R0_TO_R10 | |
409 | }; | |
410 | struct imsm_update_takeover { | |
411 | enum imsm_update_type type; | |
412 | int subarray; | |
413 | enum takeover_direction direction; | |
414 | }; | |
415 | ||
416 | struct imsm_update_reshape { | |
417 | enum imsm_update_type type; | |
418 | int old_raid_disks; | |
419 | int new_raid_disks; | |
420 | ||
421 | int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */ | |
422 | }; | |
423 | ||
424 | struct imsm_update_reshape_migration { | |
425 | enum imsm_update_type type; | |
426 | int old_raid_disks; | |
427 | int new_raid_disks; | |
428 | /* fields for array migration changes | |
429 | */ | |
430 | int subdev; | |
431 | int new_level; | |
432 | int new_layout; | |
433 | int new_chunksize; | |
434 | ||
435 | int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */ | |
436 | }; | |
437 | ||
438 | struct imsm_update_general_migration_checkpoint { | |
439 | enum imsm_update_type type; | |
440 | __u32 curr_migr_unit; | |
441 | }; | |
442 | ||
443 | struct disk_info { | |
444 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
445 | }; | |
446 | ||
447 | struct imsm_update_create_array { | |
448 | enum imsm_update_type type; | |
449 | int dev_idx; | |
450 | struct imsm_dev dev; | |
451 | }; | |
452 | ||
453 | struct imsm_update_kill_array { | |
454 | enum imsm_update_type type; | |
455 | int dev_idx; | |
456 | }; | |
457 | ||
458 | struct imsm_update_rename_array { | |
459 | enum imsm_update_type type; | |
460 | __u8 name[MAX_RAID_SERIAL_LEN]; | |
461 | int dev_idx; | |
462 | }; | |
463 | ||
464 | struct imsm_update_add_remove_disk { | |
465 | enum imsm_update_type type; | |
466 | }; | |
467 | ||
468 | ||
469 | static const char *_sys_dev_type[] = { | |
470 | [SYS_DEV_UNKNOWN] = "Unknown", | |
471 | [SYS_DEV_SAS] = "SAS", | |
472 | [SYS_DEV_SATA] = "SATA" | |
473 | }; | |
474 | ||
475 | const char *get_sys_dev_type(enum sys_dev_type type) | |
476 | { | |
477 | if (type >= SYS_DEV_MAX) | |
478 | type = SYS_DEV_UNKNOWN; | |
479 | ||
480 | return _sys_dev_type[type]; | |
481 | } | |
482 | ||
483 | static struct intel_hba * alloc_intel_hba(struct sys_dev *device) | |
484 | { | |
485 | struct intel_hba *result = malloc(sizeof(*result)); | |
486 | if (result) { | |
487 | result->type = device->type; | |
488 | result->path = strdup(device->path); | |
489 | result->next = NULL; | |
490 | if (result->path && (result->pci_id = strrchr(result->path, '/')) != NULL) | |
491 | result->pci_id++; | |
492 | } | |
493 | return result; | |
494 | } | |
495 | ||
496 | static struct intel_hba * find_intel_hba(struct intel_hba *hba, struct sys_dev *device) | |
497 | { | |
498 | struct intel_hba *result=NULL; | |
499 | for (result = hba; result; result = result->next) { | |
500 | if (result->type == device->type && strcmp(result->path, device->path) == 0) | |
501 | break; | |
502 | } | |
503 | return result; | |
504 | } | |
505 | ||
506 | static int attach_hba_to_super(struct intel_super *super, struct sys_dev *device) | |
507 | { | |
508 | struct intel_hba *hba; | |
509 | ||
510 | /* check if disk attached to Intel HBA */ | |
511 | hba = find_intel_hba(super->hba, device); | |
512 | if (hba != NULL) | |
513 | return 1; | |
514 | /* Check if HBA is already attached to super */ | |
515 | if (super->hba == NULL) { | |
516 | super->hba = alloc_intel_hba(device); | |
517 | return 1; | |
518 | } | |
519 | ||
520 | hba = super->hba; | |
521 | /* Intel metadata allows for all disks attached to the same type HBA. | |
522 | * Do not sypport odf HBA types mixing | |
523 | */ | |
524 | if (device->type != hba->type) | |
525 | return 2; | |
526 | ||
527 | while (hba->next) | |
528 | hba = hba->next; | |
529 | ||
530 | hba->next = alloc_intel_hba(device); | |
531 | return 1; | |
532 | } | |
533 | ||
534 | static struct sys_dev* find_disk_attached_hba(int fd, const char *devname) | |
535 | { | |
536 | struct sys_dev *list, *elem, *prev; | |
537 | char *disk_path; | |
538 | ||
539 | if ((list = find_intel_devices()) == NULL) | |
540 | return 0; | |
541 | ||
542 | if (fd < 0) | |
543 | disk_path = (char *) devname; | |
544 | else | |
545 | disk_path = diskfd_to_devpath(fd); | |
546 | ||
547 | if (!disk_path) { | |
548 | free_sys_dev(&list); | |
549 | return 0; | |
550 | } | |
551 | ||
552 | for (prev = NULL, elem = list; elem; prev = elem, elem = elem->next) { | |
553 | if (path_attached_to_hba(disk_path, elem->path)) { | |
554 | if (prev == NULL) | |
555 | list = list->next; | |
556 | else | |
557 | prev->next = elem->next; | |
558 | elem->next = NULL; | |
559 | if (disk_path != devname) | |
560 | free(disk_path); | |
561 | free_sys_dev(&list); | |
562 | return elem; | |
563 | } | |
564 | } | |
565 | if (disk_path != devname) | |
566 | free(disk_path); | |
567 | free_sys_dev(&list); | |
568 | ||
569 | return NULL; | |
570 | } | |
571 | ||
572 | ||
573 | static int find_intel_hba_capability(int fd, struct intel_super *super, | |
574 | char *devname); | |
575 | ||
576 | static struct supertype *match_metadata_desc_imsm(char *arg) | |
577 | { | |
578 | struct supertype *st; | |
579 | ||
580 | if (strcmp(arg, "imsm") != 0 && | |
581 | strcmp(arg, "default") != 0 | |
582 | ) | |
583 | return NULL; | |
584 | ||
585 | st = malloc(sizeof(*st)); | |
586 | if (!st) | |
587 | return NULL; | |
588 | memset(st, 0, sizeof(*st)); | |
589 | st->container_dev = NoMdDev; | |
590 | st->ss = &super_imsm; | |
591 | st->max_devs = IMSM_MAX_DEVICES; | |
592 | st->minor_version = 0; | |
593 | st->sb = NULL; | |
594 | return st; | |
595 | } | |
596 | ||
597 | #ifndef MDASSEMBLE | |
598 | static __u8 *get_imsm_version(struct imsm_super *mpb) | |
599 | { | |
600 | return &mpb->sig[MPB_SIG_LEN]; | |
601 | } | |
602 | #endif | |
603 | ||
604 | /* retrieve a disk directly from the anchor when the anchor is known to be | |
605 | * up-to-date, currently only at load time | |
606 | */ | |
607 | static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index) | |
608 | { | |
609 | if (index >= mpb->num_disks) | |
610 | return NULL; | |
611 | return &mpb->disk[index]; | |
612 | } | |
613 | ||
614 | /* retrieve the disk description based on a index of the disk | |
615 | * in the sub-array | |
616 | */ | |
617 | static struct dl *get_imsm_dl_disk(struct intel_super *super, __u8 index) | |
618 | { | |
619 | struct dl *d; | |
620 | ||
621 | for (d = super->disks; d; d = d->next) | |
622 | if (d->index == index) | |
623 | return d; | |
624 | ||
625 | return NULL; | |
626 | } | |
627 | /* retrieve a disk from the parsed metadata */ | |
628 | static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index) | |
629 | { | |
630 | struct dl *dl; | |
631 | ||
632 | dl = get_imsm_dl_disk(super, index); | |
633 | if (dl) | |
634 | return &dl->disk; | |
635 | ||
636 | return NULL; | |
637 | } | |
638 | ||
639 | /* generate a checksum directly from the anchor when the anchor is known to be | |
640 | * up-to-date, currently only at load or write_super after coalescing | |
641 | */ | |
642 | static __u32 __gen_imsm_checksum(struct imsm_super *mpb) | |
643 | { | |
644 | __u32 end = mpb->mpb_size / sizeof(end); | |
645 | __u32 *p = (__u32 *) mpb; | |
646 | __u32 sum = 0; | |
647 | ||
648 | while (end--) { | |
649 | sum += __le32_to_cpu(*p); | |
650 | p++; | |
651 | } | |
652 | ||
653 | return sum - __le32_to_cpu(mpb->check_sum); | |
654 | } | |
655 | ||
656 | static size_t sizeof_imsm_map(struct imsm_map *map) | |
657 | { | |
658 | return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1); | |
659 | } | |
660 | ||
661 | struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map) | |
662 | { | |
663 | /* A device can have 2 maps if it is in the middle of a migration. | |
664 | * If second_map is: | |
665 | * 0 - we return the first map | |
666 | * 1 - we return the second map if it exists, else NULL | |
667 | * -1 - we return the second map if it exists, else the first | |
668 | */ | |
669 | struct imsm_map *map = &dev->vol.map[0]; | |
670 | ||
671 | if (second_map == 1 && !dev->vol.migr_state) | |
672 | return NULL; | |
673 | else if (second_map == 1 || | |
674 | (second_map < 0 && dev->vol.migr_state)) { | |
675 | void *ptr = map; | |
676 | ||
677 | return ptr + sizeof_imsm_map(map); | |
678 | } else | |
679 | return map; | |
680 | ||
681 | } | |
682 | ||
683 | /* return the size of the device. | |
684 | * migr_state increases the returned size if map[0] were to be duplicated | |
685 | */ | |
686 | static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state) | |
687 | { | |
688 | size_t size = sizeof(*dev) - sizeof(struct imsm_map) + | |
689 | sizeof_imsm_map(get_imsm_map(dev, 0)); | |
690 | ||
691 | /* migrating means an additional map */ | |
692 | if (dev->vol.migr_state) | |
693 | size += sizeof_imsm_map(get_imsm_map(dev, 1)); | |
694 | else if (migr_state) | |
695 | size += sizeof_imsm_map(get_imsm_map(dev, 0)); | |
696 | ||
697 | return size; | |
698 | } | |
699 | ||
700 | #ifndef MDASSEMBLE | |
701 | /* retrieve disk serial number list from a metadata update */ | |
702 | static struct disk_info *get_disk_info(struct imsm_update_create_array *update) | |
703 | { | |
704 | void *u = update; | |
705 | struct disk_info *inf; | |
706 | ||
707 | inf = u + sizeof(*update) - sizeof(struct imsm_dev) + | |
708 | sizeof_imsm_dev(&update->dev, 0); | |
709 | ||
710 | return inf; | |
711 | } | |
712 | #endif | |
713 | ||
714 | static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index) | |
715 | { | |
716 | int offset; | |
717 | int i; | |
718 | void *_mpb = mpb; | |
719 | ||
720 | if (index >= mpb->num_raid_devs) | |
721 | return NULL; | |
722 | ||
723 | /* devices start after all disks */ | |
724 | offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb; | |
725 | ||
726 | for (i = 0; i <= index; i++) | |
727 | if (i == index) | |
728 | return _mpb + offset; | |
729 | else | |
730 | offset += sizeof_imsm_dev(_mpb + offset, 0); | |
731 | ||
732 | return NULL; | |
733 | } | |
734 | ||
735 | static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index) | |
736 | { | |
737 | struct intel_dev *dv; | |
738 | ||
739 | if (index >= super->anchor->num_raid_devs) | |
740 | return NULL; | |
741 | for (dv = super->devlist; dv; dv = dv->next) | |
742 | if (dv->index == index) | |
743 | return dv->dev; | |
744 | return NULL; | |
745 | } | |
746 | ||
747 | /* | |
748 | * for second_map: | |
749 | * == 0 get first map | |
750 | * == 1 get second map | |
751 | * == -1 than get map according to the current migr_state | |
752 | */ | |
753 | static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, | |
754 | int slot, | |
755 | int second_map) | |
756 | { | |
757 | struct imsm_map *map; | |
758 | ||
759 | map = get_imsm_map(dev, second_map); | |
760 | ||
761 | /* top byte identifies disk under rebuild */ | |
762 | return __le32_to_cpu(map->disk_ord_tbl[slot]); | |
763 | } | |
764 | ||
765 | #define ord_to_idx(ord) (((ord) << 8) >> 8) | |
766 | static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot, int second_map) | |
767 | { | |
768 | __u32 ord = get_imsm_ord_tbl_ent(dev, slot, second_map); | |
769 | ||
770 | return ord_to_idx(ord); | |
771 | } | |
772 | ||
773 | static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord) | |
774 | { | |
775 | map->disk_ord_tbl[slot] = __cpu_to_le32(ord); | |
776 | } | |
777 | ||
778 | static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx) | |
779 | { | |
780 | int slot; | |
781 | __u32 ord; | |
782 | ||
783 | for (slot = 0; slot < map->num_members; slot++) { | |
784 | ord = __le32_to_cpu(map->disk_ord_tbl[slot]); | |
785 | if (ord_to_idx(ord) == idx) | |
786 | return slot; | |
787 | } | |
788 | ||
789 | return -1; | |
790 | } | |
791 | ||
792 | static int get_imsm_raid_level(struct imsm_map *map) | |
793 | { | |
794 | if (map->raid_level == 1) { | |
795 | if (map->num_members == 2) | |
796 | return 1; | |
797 | else | |
798 | return 10; | |
799 | } | |
800 | ||
801 | return map->raid_level; | |
802 | } | |
803 | ||
804 | static int cmp_extent(const void *av, const void *bv) | |
805 | { | |
806 | const struct extent *a = av; | |
807 | const struct extent *b = bv; | |
808 | if (a->start < b->start) | |
809 | return -1; | |
810 | if (a->start > b->start) | |
811 | return 1; | |
812 | return 0; | |
813 | } | |
814 | ||
815 | static int count_memberships(struct dl *dl, struct intel_super *super) | |
816 | { | |
817 | int memberships = 0; | |
818 | int i; | |
819 | ||
820 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
821 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
822 | struct imsm_map *map = get_imsm_map(dev, 0); | |
823 | ||
824 | if (get_imsm_disk_slot(map, dl->index) >= 0) | |
825 | memberships++; | |
826 | } | |
827 | ||
828 | return memberships; | |
829 | } | |
830 | ||
831 | static __u32 imsm_min_reserved_sectors(struct intel_super *super); | |
832 | ||
833 | static struct extent *get_extents(struct intel_super *super, struct dl *dl) | |
834 | { | |
835 | /* find a list of used extents on the given physical device */ | |
836 | struct extent *rv, *e; | |
837 | int i; | |
838 | int memberships = count_memberships(dl, super); | |
839 | __u32 reservation; | |
840 | ||
841 | /* trim the reserved area for spares, so they can join any array | |
842 | * regardless of whether the OROM has assigned sectors from the | |
843 | * IMSM_RESERVED_SECTORS region | |
844 | */ | |
845 | if (dl->index == -1) | |
846 | reservation = imsm_min_reserved_sectors(super); | |
847 | else | |
848 | reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
849 | ||
850 | rv = malloc(sizeof(struct extent) * (memberships + 1)); | |
851 | if (!rv) | |
852 | return NULL; | |
853 | e = rv; | |
854 | ||
855 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
856 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
857 | struct imsm_map *map = get_imsm_map(dev, 0); | |
858 | ||
859 | if (get_imsm_disk_slot(map, dl->index) >= 0) { | |
860 | e->start = __le32_to_cpu(map->pba_of_lba0); | |
861 | e->size = __le32_to_cpu(map->blocks_per_member); | |
862 | e++; | |
863 | } | |
864 | } | |
865 | qsort(rv, memberships, sizeof(*rv), cmp_extent); | |
866 | ||
867 | /* determine the start of the metadata | |
868 | * when no raid devices are defined use the default | |
869 | * ...otherwise allow the metadata to truncate the value | |
870 | * as is the case with older versions of imsm | |
871 | */ | |
872 | if (memberships) { | |
873 | struct extent *last = &rv[memberships - 1]; | |
874 | __u32 remainder; | |
875 | ||
876 | remainder = __le32_to_cpu(dl->disk.total_blocks) - | |
877 | (last->start + last->size); | |
878 | /* round down to 1k block to satisfy precision of the kernel | |
879 | * 'size' interface | |
880 | */ | |
881 | remainder &= ~1UL; | |
882 | /* make sure remainder is still sane */ | |
883 | if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9) | |
884 | remainder = ROUND_UP(super->len, 512) >> 9; | |
885 | if (reservation > remainder) | |
886 | reservation = remainder; | |
887 | } | |
888 | e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation; | |
889 | e->size = 0; | |
890 | return rv; | |
891 | } | |
892 | ||
893 | /* try to determine how much space is reserved for metadata from | |
894 | * the last get_extents() entry, otherwise fallback to the | |
895 | * default | |
896 | */ | |
897 | static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl) | |
898 | { | |
899 | struct extent *e; | |
900 | int i; | |
901 | __u32 rv; | |
902 | ||
903 | /* for spares just return a minimal reservation which will grow | |
904 | * once the spare is picked up by an array | |
905 | */ | |
906 | if (dl->index == -1) | |
907 | return MPB_SECTOR_CNT; | |
908 | ||
909 | e = get_extents(super, dl); | |
910 | if (!e) | |
911 | return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
912 | ||
913 | /* scroll to last entry */ | |
914 | for (i = 0; e[i].size; i++) | |
915 | continue; | |
916 | ||
917 | rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start; | |
918 | ||
919 | free(e); | |
920 | ||
921 | return rv; | |
922 | } | |
923 | ||
924 | static int is_spare(struct imsm_disk *disk) | |
925 | { | |
926 | return (disk->status & SPARE_DISK) == SPARE_DISK; | |
927 | } | |
928 | ||
929 | static int is_configured(struct imsm_disk *disk) | |
930 | { | |
931 | return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK; | |
932 | } | |
933 | ||
934 | static int is_failed(struct imsm_disk *disk) | |
935 | { | |
936 | return (disk->status & FAILED_DISK) == FAILED_DISK; | |
937 | } | |
938 | ||
939 | /* try to determine how much space is reserved for metadata from | |
940 | * the last get_extents() entry on the smallest active disk, | |
941 | * otherwise fallback to the default | |
942 | */ | |
943 | static __u32 imsm_min_reserved_sectors(struct intel_super *super) | |
944 | { | |
945 | struct extent *e; | |
946 | int i; | |
947 | __u32 min_active, remainder; | |
948 | __u32 rv = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
949 | struct dl *dl, *dl_min = NULL; | |
950 | ||
951 | if (!super) | |
952 | return rv; | |
953 | ||
954 | min_active = 0; | |
955 | for (dl = super->disks; dl; dl = dl->next) { | |
956 | if (dl->index < 0) | |
957 | continue; | |
958 | if (dl->disk.total_blocks < min_active || min_active == 0) { | |
959 | dl_min = dl; | |
960 | min_active = dl->disk.total_blocks; | |
961 | } | |
962 | } | |
963 | if (!dl_min) | |
964 | return rv; | |
965 | ||
966 | /* find last lba used by subarrays on the smallest active disk */ | |
967 | e = get_extents(super, dl_min); | |
968 | if (!e) | |
969 | return rv; | |
970 | for (i = 0; e[i].size; i++) | |
971 | continue; | |
972 | ||
973 | remainder = min_active - e[i].start; | |
974 | free(e); | |
975 | ||
976 | /* to give priority to recovery we should not require full | |
977 | IMSM_RESERVED_SECTORS from the spare */ | |
978 | rv = MPB_SECTOR_CNT + NUM_BLOCKS_DIRTY_STRIPE_REGION; | |
979 | ||
980 | /* if real reservation is smaller use that value */ | |
981 | return (remainder < rv) ? remainder : rv; | |
982 | } | |
983 | ||
984 | /* Return minimum size of a spare that can be used in this array*/ | |
985 | static unsigned long long min_acceptable_spare_size_imsm(struct supertype *st) | |
986 | { | |
987 | struct intel_super *super = st->sb; | |
988 | struct dl *dl; | |
989 | struct extent *e; | |
990 | int i; | |
991 | unsigned long long rv = 0; | |
992 | ||
993 | if (!super) | |
994 | return rv; | |
995 | /* find first active disk in array */ | |
996 | dl = super->disks; | |
997 | while (dl && (is_failed(&dl->disk) || dl->index == -1)) | |
998 | dl = dl->next; | |
999 | if (!dl) | |
1000 | return rv; | |
1001 | /* find last lba used by subarrays */ | |
1002 | e = get_extents(super, dl); | |
1003 | if (!e) | |
1004 | return rv; | |
1005 | for (i = 0; e[i].size; i++) | |
1006 | continue; | |
1007 | if (i > 0) | |
1008 | rv = e[i-1].start + e[i-1].size; | |
1009 | free(e); | |
1010 | ||
1011 | /* add the amount of space needed for metadata */ | |
1012 | rv = rv + imsm_min_reserved_sectors(super); | |
1013 | ||
1014 | return rv * 512; | |
1015 | } | |
1016 | ||
1017 | #ifndef MDASSEMBLE | |
1018 | static __u64 blocks_per_migr_unit(struct intel_super *super, | |
1019 | struct imsm_dev *dev); | |
1020 | ||
1021 | static void print_imsm_dev(struct intel_super *super, | |
1022 | struct imsm_dev *dev, | |
1023 | char *uuid, | |
1024 | int disk_idx) | |
1025 | { | |
1026 | __u64 sz; | |
1027 | int slot, i; | |
1028 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1029 | struct imsm_map *map2 = get_imsm_map(dev, 1); | |
1030 | __u32 ord; | |
1031 | ||
1032 | printf("\n"); | |
1033 | printf("[%.16s]:\n", dev->volume); | |
1034 | printf(" UUID : %s\n", uuid); | |
1035 | printf(" RAID Level : %d", get_imsm_raid_level(map)); | |
1036 | if (map2) | |
1037 | printf(" <-- %d", get_imsm_raid_level(map2)); | |
1038 | printf("\n"); | |
1039 | printf(" Members : %d", map->num_members); | |
1040 | if (map2) | |
1041 | printf(" <-- %d", map2->num_members); | |
1042 | printf("\n"); | |
1043 | printf(" Slots : ["); | |
1044 | for (i = 0; i < map->num_members; i++) { | |
1045 | ord = get_imsm_ord_tbl_ent(dev, i, 0); | |
1046 | printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U"); | |
1047 | } | |
1048 | printf("]"); | |
1049 | if (map2) { | |
1050 | printf(" <-- ["); | |
1051 | for (i = 0; i < map2->num_members; i++) { | |
1052 | ord = get_imsm_ord_tbl_ent(dev, i, 1); | |
1053 | printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U"); | |
1054 | } | |
1055 | printf("]"); | |
1056 | } | |
1057 | printf("\n"); | |
1058 | printf(" Failed disk : "); | |
1059 | if (map->failed_disk_num == 0xff) | |
1060 | printf("none"); | |
1061 | else | |
1062 | printf("%i", map->failed_disk_num); | |
1063 | printf("\n"); | |
1064 | slot = get_imsm_disk_slot(map, disk_idx); | |
1065 | if (slot >= 0) { | |
1066 | ord = get_imsm_ord_tbl_ent(dev, slot, -1); | |
1067 | printf(" This Slot : %d%s\n", slot, | |
1068 | ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : ""); | |
1069 | } else | |
1070 | printf(" This Slot : ?\n"); | |
1071 | sz = __le32_to_cpu(dev->size_high); | |
1072 | sz <<= 32; | |
1073 | sz += __le32_to_cpu(dev->size_low); | |
1074 | printf(" Array Size : %llu%s\n", (unsigned long long)sz, | |
1075 | human_size(sz * 512)); | |
1076 | sz = __le32_to_cpu(map->blocks_per_member); | |
1077 | printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz, | |
1078 | human_size(sz * 512)); | |
1079 | printf(" Sector Offset : %u\n", | |
1080 | __le32_to_cpu(map->pba_of_lba0)); | |
1081 | printf(" Num Stripes : %u\n", | |
1082 | __le32_to_cpu(map->num_data_stripes)); | |
1083 | printf(" Chunk Size : %u KiB", | |
1084 | __le16_to_cpu(map->blocks_per_strip) / 2); | |
1085 | if (map2) | |
1086 | printf(" <-- %u KiB", | |
1087 | __le16_to_cpu(map2->blocks_per_strip) / 2); | |
1088 | printf("\n"); | |
1089 | printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks)); | |
1090 | printf(" Migrate State : "); | |
1091 | if (dev->vol.migr_state) { | |
1092 | if (migr_type(dev) == MIGR_INIT) | |
1093 | printf("initialize\n"); | |
1094 | else if (migr_type(dev) == MIGR_REBUILD) | |
1095 | printf("rebuild\n"); | |
1096 | else if (migr_type(dev) == MIGR_VERIFY) | |
1097 | printf("check\n"); | |
1098 | else if (migr_type(dev) == MIGR_GEN_MIGR) | |
1099 | printf("general migration\n"); | |
1100 | else if (migr_type(dev) == MIGR_STATE_CHANGE) | |
1101 | printf("state change\n"); | |
1102 | else if (migr_type(dev) == MIGR_REPAIR) | |
1103 | printf("repair\n"); | |
1104 | else | |
1105 | printf("<unknown:%d>\n", migr_type(dev)); | |
1106 | } else | |
1107 | printf("idle\n"); | |
1108 | printf(" Map State : %s", map_state_str[map->map_state]); | |
1109 | if (dev->vol.migr_state) { | |
1110 | struct imsm_map *map = get_imsm_map(dev, 1); | |
1111 | ||
1112 | printf(" <-- %s", map_state_str[map->map_state]); | |
1113 | printf("\n Checkpoint : %u (%llu)", | |
1114 | __le32_to_cpu(dev->vol.curr_migr_unit), | |
1115 | (unsigned long long)blocks_per_migr_unit(super, dev)); | |
1116 | } | |
1117 | printf("\n"); | |
1118 | printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean"); | |
1119 | } | |
1120 | ||
1121 | static void print_imsm_disk(struct imsm_disk *disk, int index, __u32 reserved) | |
1122 | { | |
1123 | char str[MAX_RAID_SERIAL_LEN + 1]; | |
1124 | __u64 sz; | |
1125 | ||
1126 | if (index < -1 || !disk) | |
1127 | return; | |
1128 | ||
1129 | printf("\n"); | |
1130 | snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial); | |
1131 | if (index >= 0) | |
1132 | printf(" Disk%02d Serial : %s\n", index, str); | |
1133 | else | |
1134 | printf(" Disk Serial : %s\n", str); | |
1135 | printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "", | |
1136 | is_configured(disk) ? " active" : "", | |
1137 | is_failed(disk) ? " failed" : ""); | |
1138 | printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id)); | |
1139 | sz = __le32_to_cpu(disk->total_blocks) - reserved; | |
1140 | printf(" Usable Size : %llu%s\n", (unsigned long long)sz, | |
1141 | human_size(sz * 512)); | |
1142 | } | |
1143 | ||
1144 | static int is_gen_migration(struct imsm_dev *dev); | |
1145 | ||
1146 | void examine_migr_rec_imsm(struct intel_super *super) | |
1147 | { | |
1148 | struct migr_record *migr_rec = super->migr_rec; | |
1149 | struct imsm_super *mpb = super->anchor; | |
1150 | int i; | |
1151 | ||
1152 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
1153 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
1154 | if (is_gen_migration(dev) == 0) | |
1155 | continue; | |
1156 | ||
1157 | printf("\nMigration Record Information:"); | |
1158 | if (super->disks->index > 1) { | |
1159 | printf(" Empty\n "); | |
1160 | printf("Examine one of first two disks in array\n"); | |
1161 | break; | |
1162 | } | |
1163 | printf("\n Status : "); | |
1164 | if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL) | |
1165 | printf("Normal\n"); | |
1166 | else | |
1167 | printf("Contains Data\n"); | |
1168 | printf(" Current Unit : %u\n", | |
1169 | __le32_to_cpu(migr_rec->curr_migr_unit)); | |
1170 | printf(" Family : %u\n", | |
1171 | __le32_to_cpu(migr_rec->family_num)); | |
1172 | printf(" Ascending : %u\n", | |
1173 | __le32_to_cpu(migr_rec->ascending_migr)); | |
1174 | printf(" Blocks Per Unit : %u\n", | |
1175 | __le32_to_cpu(migr_rec->blocks_per_unit)); | |
1176 | printf(" Dest. Depth Per Unit : %u\n", | |
1177 | __le32_to_cpu(migr_rec->dest_depth_per_unit)); | |
1178 | printf(" Checkpoint Area pba : %u\n", | |
1179 | __le32_to_cpu(migr_rec->ckpt_area_pba)); | |
1180 | printf(" First member lba : %u\n", | |
1181 | __le32_to_cpu(migr_rec->dest_1st_member_lba)); | |
1182 | printf(" Total Number of Units : %u\n", | |
1183 | __le32_to_cpu(migr_rec->num_migr_units)); | |
1184 | printf(" Size of volume : %u\n", | |
1185 | __le32_to_cpu(migr_rec->post_migr_vol_cap)); | |
1186 | printf(" Expansion space for LBA64 : %u\n", | |
1187 | __le32_to_cpu(migr_rec->post_migr_vol_cap_hi)); | |
1188 | printf(" Record was read from : %u\n", | |
1189 | __le32_to_cpu(migr_rec->ckpt_read_disk_num)); | |
1190 | ||
1191 | break; | |
1192 | } | |
1193 | } | |
1194 | #endif /* MDASSEMBLE */ | |
1195 | /******************************************************************************* | |
1196 | * function: imsm_check_attributes | |
1197 | * Description: Function checks if features represented by attributes flags | |
1198 | * are supported by mdadm. | |
1199 | * Parameters: | |
1200 | * attributes - Attributes read from metadata | |
1201 | * Returns: | |
1202 | * 0 - passed attributes contains unsupported features flags | |
1203 | * 1 - all features are supported | |
1204 | ******************************************************************************/ | |
1205 | static int imsm_check_attributes(__u32 attributes) | |
1206 | { | |
1207 | int ret_val = 1; | |
1208 | __u32 not_supported = MPB_ATTRIB_SUPPORTED^0xffffffff; | |
1209 | ||
1210 | not_supported &= ~MPB_ATTRIB_IGNORED; | |
1211 | ||
1212 | not_supported &= attributes; | |
1213 | if (not_supported) { | |
1214 | fprintf(stderr, Name "(IMSM): Unsupported attributes : %x\n", | |
1215 | (unsigned)__le32_to_cpu(not_supported)); | |
1216 | if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) { | |
1217 | dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY \n"); | |
1218 | not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY; | |
1219 | } | |
1220 | if (not_supported & MPB_ATTRIB_2TB) { | |
1221 | dprintf("\t\tMPB_ATTRIB_2TB\n"); | |
1222 | not_supported ^= MPB_ATTRIB_2TB; | |
1223 | } | |
1224 | if (not_supported & MPB_ATTRIB_RAID0) { | |
1225 | dprintf("\t\tMPB_ATTRIB_RAID0\n"); | |
1226 | not_supported ^= MPB_ATTRIB_RAID0; | |
1227 | } | |
1228 | if (not_supported & MPB_ATTRIB_RAID1) { | |
1229 | dprintf("\t\tMPB_ATTRIB_RAID1\n"); | |
1230 | not_supported ^= MPB_ATTRIB_RAID1; | |
1231 | } | |
1232 | if (not_supported & MPB_ATTRIB_RAID10) { | |
1233 | dprintf("\t\tMPB_ATTRIB_RAID10\n"); | |
1234 | not_supported ^= MPB_ATTRIB_RAID10; | |
1235 | } | |
1236 | if (not_supported & MPB_ATTRIB_RAID1E) { | |
1237 | dprintf("\t\tMPB_ATTRIB_RAID1E\n"); | |
1238 | not_supported ^= MPB_ATTRIB_RAID1E; | |
1239 | } | |
1240 | if (not_supported & MPB_ATTRIB_RAID5) { | |
1241 | dprintf("\t\tMPB_ATTRIB_RAID5\n"); | |
1242 | not_supported ^= MPB_ATTRIB_RAID5; | |
1243 | } | |
1244 | if (not_supported & MPB_ATTRIB_RAIDCNG) { | |
1245 | dprintf("\t\tMPB_ATTRIB_RAIDCNG\n"); | |
1246 | not_supported ^= MPB_ATTRIB_RAIDCNG; | |
1247 | } | |
1248 | if (not_supported & MPB_ATTRIB_BBM) { | |
1249 | dprintf("\t\tMPB_ATTRIB_BBM\n"); | |
1250 | not_supported ^= MPB_ATTRIB_BBM; | |
1251 | } | |
1252 | if (not_supported & MPB_ATTRIB_CHECKSUM_VERIFY) { | |
1253 | dprintf("\t\tMPB_ATTRIB_CHECKSUM_VERIFY (== MPB_ATTRIB_LEGACY)\n"); | |
1254 | not_supported ^= MPB_ATTRIB_CHECKSUM_VERIFY; | |
1255 | } | |
1256 | if (not_supported & MPB_ATTRIB_EXP_STRIPE_SIZE) { | |
1257 | dprintf("\t\tMPB_ATTRIB_EXP_STRIP_SIZE\n"); | |
1258 | not_supported ^= MPB_ATTRIB_EXP_STRIPE_SIZE; | |
1259 | } | |
1260 | if (not_supported & MPB_ATTRIB_2TB_DISK) { | |
1261 | dprintf("\t\tMPB_ATTRIB_2TB_DISK\n"); | |
1262 | not_supported ^= MPB_ATTRIB_2TB_DISK; | |
1263 | } | |
1264 | if (not_supported & MPB_ATTRIB_NEVER_USE2) { | |
1265 | dprintf("\t\tMPB_ATTRIB_NEVER_USE2\n"); | |
1266 | not_supported ^= MPB_ATTRIB_NEVER_USE2; | |
1267 | } | |
1268 | if (not_supported & MPB_ATTRIB_NEVER_USE) { | |
1269 | dprintf("\t\tMPB_ATTRIB_NEVER_USE\n"); | |
1270 | not_supported ^= MPB_ATTRIB_NEVER_USE; | |
1271 | } | |
1272 | ||
1273 | if (not_supported) | |
1274 | dprintf(Name "(IMSM): Unknown attributes : %x\n", not_supported); | |
1275 | ||
1276 | ret_val = 0; | |
1277 | } | |
1278 | ||
1279 | return ret_val; | |
1280 | } | |
1281 | ||
1282 | #ifndef MDASSEMBLE | |
1283 | static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map); | |
1284 | ||
1285 | static void examine_super_imsm(struct supertype *st, char *homehost) | |
1286 | { | |
1287 | struct intel_super *super = st->sb; | |
1288 | struct imsm_super *mpb = super->anchor; | |
1289 | char str[MAX_SIGNATURE_LENGTH]; | |
1290 | int i; | |
1291 | struct mdinfo info; | |
1292 | char nbuf[64]; | |
1293 | __u32 sum; | |
1294 | __u32 reserved = imsm_reserved_sectors(super, super->disks); | |
1295 | struct dl *dl; | |
1296 | ||
1297 | snprintf(str, MPB_SIG_LEN, "%s", mpb->sig); | |
1298 | printf(" Magic : %s\n", str); | |
1299 | snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb)); | |
1300 | printf(" Version : %s\n", get_imsm_version(mpb)); | |
1301 | printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num)); | |
1302 | printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num)); | |
1303 | printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num)); | |
1304 | printf(" Attributes : "); | |
1305 | if (imsm_check_attributes(mpb->attributes)) | |
1306 | printf("All supported\n"); | |
1307 | else | |
1308 | printf("not supported\n"); | |
1309 | getinfo_super_imsm(st, &info, NULL); | |
1310 | fname_from_uuid(st, &info, nbuf, ':'); | |
1311 | printf(" UUID : %s\n", nbuf + 5); | |
1312 | sum = __le32_to_cpu(mpb->check_sum); | |
1313 | printf(" Checksum : %08x %s\n", sum, | |
1314 | __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect"); | |
1315 | printf(" MPB Sectors : %d\n", mpb_sectors(mpb)); | |
1316 | printf(" Disks : %d\n", mpb->num_disks); | |
1317 | printf(" RAID Devices : %d\n", mpb->num_raid_devs); | |
1318 | print_imsm_disk(__get_imsm_disk(mpb, super->disks->index), super->disks->index, reserved); | |
1319 | if (super->bbm_log) { | |
1320 | struct bbm_log *log = super->bbm_log; | |
1321 | ||
1322 | printf("\n"); | |
1323 | printf("Bad Block Management Log:\n"); | |
1324 | printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size)); | |
1325 | printf(" Signature : %x\n", __le32_to_cpu(log->signature)); | |
1326 | printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count)); | |
1327 | printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count)); | |
1328 | printf(" First Spare : %llx\n", | |
1329 | (unsigned long long) __le64_to_cpu(log->first_spare_lba)); | |
1330 | } | |
1331 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
1332 | struct mdinfo info; | |
1333 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
1334 | ||
1335 | super->current_vol = i; | |
1336 | getinfo_super_imsm(st, &info, NULL); | |
1337 | fname_from_uuid(st, &info, nbuf, ':'); | |
1338 | print_imsm_dev(super, dev, nbuf + 5, super->disks->index); | |
1339 | } | |
1340 | for (i = 0; i < mpb->num_disks; i++) { | |
1341 | if (i == super->disks->index) | |
1342 | continue; | |
1343 | print_imsm_disk(__get_imsm_disk(mpb, i), i, reserved); | |
1344 | } | |
1345 | ||
1346 | for (dl = super->disks; dl; dl = dl->next) | |
1347 | if (dl->index == -1) | |
1348 | print_imsm_disk(&dl->disk, -1, reserved); | |
1349 | ||
1350 | examine_migr_rec_imsm(super); | |
1351 | } | |
1352 | ||
1353 | static void brief_examine_super_imsm(struct supertype *st, int verbose) | |
1354 | { | |
1355 | /* We just write a generic IMSM ARRAY entry */ | |
1356 | struct mdinfo info; | |
1357 | char nbuf[64]; | |
1358 | struct intel_super *super = st->sb; | |
1359 | ||
1360 | if (!super->anchor->num_raid_devs) { | |
1361 | printf("ARRAY metadata=imsm\n"); | |
1362 | return; | |
1363 | } | |
1364 | ||
1365 | getinfo_super_imsm(st, &info, NULL); | |
1366 | fname_from_uuid(st, &info, nbuf, ':'); | |
1367 | printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5); | |
1368 | } | |
1369 | ||
1370 | static void brief_examine_subarrays_imsm(struct supertype *st, int verbose) | |
1371 | { | |
1372 | /* We just write a generic IMSM ARRAY entry */ | |
1373 | struct mdinfo info; | |
1374 | char nbuf[64]; | |
1375 | char nbuf1[64]; | |
1376 | struct intel_super *super = st->sb; | |
1377 | int i; | |
1378 | ||
1379 | if (!super->anchor->num_raid_devs) | |
1380 | return; | |
1381 | ||
1382 | getinfo_super_imsm(st, &info, NULL); | |
1383 | fname_from_uuid(st, &info, nbuf, ':'); | |
1384 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
1385 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
1386 | ||
1387 | super->current_vol = i; | |
1388 | getinfo_super_imsm(st, &info, NULL); | |
1389 | fname_from_uuid(st, &info, nbuf1, ':'); | |
1390 | printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n", | |
1391 | dev->volume, nbuf + 5, i, nbuf1 + 5); | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | static void export_examine_super_imsm(struct supertype *st) | |
1396 | { | |
1397 | struct intel_super *super = st->sb; | |
1398 | struct imsm_super *mpb = super->anchor; | |
1399 | struct mdinfo info; | |
1400 | char nbuf[64]; | |
1401 | ||
1402 | getinfo_super_imsm(st, &info, NULL); | |
1403 | fname_from_uuid(st, &info, nbuf, ':'); | |
1404 | printf("MD_METADATA=imsm\n"); | |
1405 | printf("MD_LEVEL=container\n"); | |
1406 | printf("MD_UUID=%s\n", nbuf+5); | |
1407 | printf("MD_DEVICES=%u\n", mpb->num_disks); | |
1408 | } | |
1409 | ||
1410 | static void detail_super_imsm(struct supertype *st, char *homehost) | |
1411 | { | |
1412 | struct mdinfo info; | |
1413 | char nbuf[64]; | |
1414 | ||
1415 | getinfo_super_imsm(st, &info, NULL); | |
1416 | fname_from_uuid(st, &info, nbuf, ':'); | |
1417 | printf("\n UUID : %s\n", nbuf + 5); | |
1418 | } | |
1419 | ||
1420 | static void brief_detail_super_imsm(struct supertype *st) | |
1421 | { | |
1422 | struct mdinfo info; | |
1423 | char nbuf[64]; | |
1424 | getinfo_super_imsm(st, &info, NULL); | |
1425 | fname_from_uuid(st, &info, nbuf, ':'); | |
1426 | printf(" UUID=%s", nbuf + 5); | |
1427 | } | |
1428 | ||
1429 | static int imsm_read_serial(int fd, char *devname, __u8 *serial); | |
1430 | static void fd2devname(int fd, char *name); | |
1431 | ||
1432 | static int ahci_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose) | |
1433 | { | |
1434 | /* dump an unsorted list of devices attached to AHCI Intel storage | |
1435 | * controller, as well as non-connected ports | |
1436 | */ | |
1437 | int hba_len = strlen(hba_path) + 1; | |
1438 | struct dirent *ent; | |
1439 | DIR *dir; | |
1440 | char *path = NULL; | |
1441 | int err = 0; | |
1442 | unsigned long port_mask = (1 << port_count) - 1; | |
1443 | ||
1444 | if (port_count > (int)sizeof(port_mask) * 8) { | |
1445 | if (verbose) | |
1446 | fprintf(stderr, Name ": port_count %d out of range\n", port_count); | |
1447 | return 2; | |
1448 | } | |
1449 | ||
1450 | /* scroll through /sys/dev/block looking for devices attached to | |
1451 | * this hba | |
1452 | */ | |
1453 | dir = opendir("/sys/dev/block"); | |
1454 | for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) { | |
1455 | int fd; | |
1456 | char model[64]; | |
1457 | char vendor[64]; | |
1458 | char buf[1024]; | |
1459 | int major, minor; | |
1460 | char *device; | |
1461 | char *c; | |
1462 | int port; | |
1463 | int type; | |
1464 | ||
1465 | if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2) | |
1466 | continue; | |
1467 | path = devt_to_devpath(makedev(major, minor)); | |
1468 | if (!path) | |
1469 | continue; | |
1470 | if (!path_attached_to_hba(path, hba_path)) { | |
1471 | free(path); | |
1472 | path = NULL; | |
1473 | continue; | |
1474 | } | |
1475 | ||
1476 | /* retrieve the scsi device type */ | |
1477 | if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) { | |
1478 | if (verbose) | |
1479 | fprintf(stderr, Name ": failed to allocate 'device'\n"); | |
1480 | err = 2; | |
1481 | break; | |
1482 | } | |
1483 | sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor); | |
1484 | if (load_sys(device, buf) != 0) { | |
1485 | if (verbose) | |
1486 | fprintf(stderr, Name ": failed to read device type for %s\n", | |
1487 | path); | |
1488 | err = 2; | |
1489 | free(device); | |
1490 | break; | |
1491 | } | |
1492 | type = strtoul(buf, NULL, 10); | |
1493 | ||
1494 | /* if it's not a disk print the vendor and model */ | |
1495 | if (!(type == 0 || type == 7 || type == 14)) { | |
1496 | vendor[0] = '\0'; | |
1497 | model[0] = '\0'; | |
1498 | sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor); | |
1499 | if (load_sys(device, buf) == 0) { | |
1500 | strncpy(vendor, buf, sizeof(vendor)); | |
1501 | vendor[sizeof(vendor) - 1] = '\0'; | |
1502 | c = (char *) &vendor[sizeof(vendor) - 1]; | |
1503 | while (isspace(*c) || *c == '\0') | |
1504 | *c-- = '\0'; | |
1505 | ||
1506 | } | |
1507 | sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor); | |
1508 | if (load_sys(device, buf) == 0) { | |
1509 | strncpy(model, buf, sizeof(model)); | |
1510 | model[sizeof(model) - 1] = '\0'; | |
1511 | c = (char *) &model[sizeof(model) - 1]; | |
1512 | while (isspace(*c) || *c == '\0') | |
1513 | *c-- = '\0'; | |
1514 | } | |
1515 | ||
1516 | if (vendor[0] && model[0]) | |
1517 | sprintf(buf, "%.64s %.64s", vendor, model); | |
1518 | else | |
1519 | switch (type) { /* numbers from hald/linux/device.c */ | |
1520 | case 1: sprintf(buf, "tape"); break; | |
1521 | case 2: sprintf(buf, "printer"); break; | |
1522 | case 3: sprintf(buf, "processor"); break; | |
1523 | case 4: | |
1524 | case 5: sprintf(buf, "cdrom"); break; | |
1525 | case 6: sprintf(buf, "scanner"); break; | |
1526 | case 8: sprintf(buf, "media_changer"); break; | |
1527 | case 9: sprintf(buf, "comm"); break; | |
1528 | case 12: sprintf(buf, "raid"); break; | |
1529 | default: sprintf(buf, "unknown"); | |
1530 | } | |
1531 | } else | |
1532 | buf[0] = '\0'; | |
1533 | free(device); | |
1534 | ||
1535 | /* chop device path to 'host%d' and calculate the port number */ | |
1536 | c = strchr(&path[hba_len], '/'); | |
1537 | if (!c) { | |
1538 | if (verbose) | |
1539 | fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len); | |
1540 | err = 2; | |
1541 | break; | |
1542 | } | |
1543 | *c = '\0'; | |
1544 | if (sscanf(&path[hba_len], "host%d", &port) == 1) | |
1545 | port -= host_base; | |
1546 | else { | |
1547 | if (verbose) { | |
1548 | *c = '/'; /* repair the full string */ | |
1549 | fprintf(stderr, Name ": failed to determine port number for %s\n", | |
1550 | path); | |
1551 | } | |
1552 | err = 2; | |
1553 | break; | |
1554 | } | |
1555 | ||
1556 | /* mark this port as used */ | |
1557 | port_mask &= ~(1 << port); | |
1558 | ||
1559 | /* print out the device information */ | |
1560 | if (buf[0]) { | |
1561 | printf(" Port%d : - non-disk device (%s) -\n", port, buf); | |
1562 | continue; | |
1563 | } | |
1564 | ||
1565 | fd = dev_open(ent->d_name, O_RDONLY); | |
1566 | if (fd < 0) | |
1567 | printf(" Port%d : - disk info unavailable -\n", port); | |
1568 | else { | |
1569 | fd2devname(fd, buf); | |
1570 | printf(" Port%d : %s", port, buf); | |
1571 | if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0) | |
1572 | printf(" (%.*s)\n", MAX_RAID_SERIAL_LEN, buf); | |
1573 | else | |
1574 | printf(" ()\n"); | |
1575 | close(fd); | |
1576 | } | |
1577 | free(path); | |
1578 | path = NULL; | |
1579 | } | |
1580 | if (path) | |
1581 | free(path); | |
1582 | if (dir) | |
1583 | closedir(dir); | |
1584 | if (err == 0) { | |
1585 | int i; | |
1586 | ||
1587 | for (i = 0; i < port_count; i++) | |
1588 | if (port_mask & (1 << i)) | |
1589 | printf(" Port%d : - no device attached -\n", i); | |
1590 | } | |
1591 | ||
1592 | return err; | |
1593 | } | |
1594 | ||
1595 | static void print_found_intel_controllers(struct sys_dev *elem) | |
1596 | { | |
1597 | for (; elem; elem = elem->next) { | |
1598 | fprintf(stderr, Name ": found Intel(R) "); | |
1599 | if (elem->type == SYS_DEV_SATA) | |
1600 | fprintf(stderr, "SATA "); | |
1601 | else if (elem->type == SYS_DEV_SAS) | |
1602 | fprintf(stderr, "SAS "); | |
1603 | fprintf(stderr, "RAID controller"); | |
1604 | if (elem->pci_id) | |
1605 | fprintf(stderr, " at %s", elem->pci_id); | |
1606 | fprintf(stderr, ".\n"); | |
1607 | } | |
1608 | fflush(stderr); | |
1609 | } | |
1610 | ||
1611 | static int ahci_get_port_count(const char *hba_path, int *port_count) | |
1612 | { | |
1613 | struct dirent *ent; | |
1614 | DIR *dir; | |
1615 | int host_base = -1; | |
1616 | ||
1617 | *port_count = 0; | |
1618 | if ((dir = opendir(hba_path)) == NULL) | |
1619 | return -1; | |
1620 | ||
1621 | for (ent = readdir(dir); ent; ent = readdir(dir)) { | |
1622 | int host; | |
1623 | ||
1624 | if (sscanf(ent->d_name, "host%d", &host) != 1) | |
1625 | continue; | |
1626 | if (*port_count == 0) | |
1627 | host_base = host; | |
1628 | else if (host < host_base) | |
1629 | host_base = host; | |
1630 | ||
1631 | if (host + 1 > *port_count + host_base) | |
1632 | *port_count = host + 1 - host_base; | |
1633 | } | |
1634 | closedir(dir); | |
1635 | return host_base; | |
1636 | } | |
1637 | ||
1638 | static void print_imsm_capability(const struct imsm_orom *orom) | |
1639 | { | |
1640 | printf(" Platform : Intel(R) Matrix Storage Manager\n"); | |
1641 | printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver, | |
1642 | orom->hotfix_ver, orom->build); | |
1643 | printf(" RAID Levels :%s%s%s%s%s\n", | |
1644 | imsm_orom_has_raid0(orom) ? " raid0" : "", | |
1645 | imsm_orom_has_raid1(orom) ? " raid1" : "", | |
1646 | imsm_orom_has_raid1e(orom) ? " raid1e" : "", | |
1647 | imsm_orom_has_raid10(orom) ? " raid10" : "", | |
1648 | imsm_orom_has_raid5(orom) ? " raid5" : ""); | |
1649 | printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", | |
1650 | imsm_orom_has_chunk(orom, 2) ? " 2k" : "", | |
1651 | imsm_orom_has_chunk(orom, 4) ? " 4k" : "", | |
1652 | imsm_orom_has_chunk(orom, 8) ? " 8k" : "", | |
1653 | imsm_orom_has_chunk(orom, 16) ? " 16k" : "", | |
1654 | imsm_orom_has_chunk(orom, 32) ? " 32k" : "", | |
1655 | imsm_orom_has_chunk(orom, 64) ? " 64k" : "", | |
1656 | imsm_orom_has_chunk(orom, 128) ? " 128k" : "", | |
1657 | imsm_orom_has_chunk(orom, 256) ? " 256k" : "", | |
1658 | imsm_orom_has_chunk(orom, 512) ? " 512k" : "", | |
1659 | imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "", | |
1660 | imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "", | |
1661 | imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "", | |
1662 | imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "", | |
1663 | imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "", | |
1664 | imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "", | |
1665 | imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : ""); | |
1666 | printf(" Max Disks : %d\n", orom->tds); | |
1667 | printf(" Max Volumes : %d\n", orom->vpa); | |
1668 | return; | |
1669 | } | |
1670 | ||
1671 | static int detail_platform_imsm(int verbose, int enumerate_only) | |
1672 | { | |
1673 | /* There are two components to imsm platform support, the ahci SATA | |
1674 | * controller and the option-rom. To find the SATA controller we | |
1675 | * simply look in /sys/bus/pci/drivers/ahci to see if an ahci | |
1676 | * controller with the Intel vendor id is present. This approach | |
1677 | * allows mdadm to leverage the kernel's ahci detection logic, with the | |
1678 | * caveat that if ahci.ko is not loaded mdadm will not be able to | |
1679 | * detect platform raid capabilities. The option-rom resides in a | |
1680 | * platform "Adapter ROM". We scan for its signature to retrieve the | |
1681 | * platform capabilities. If raid support is disabled in the BIOS the | |
1682 | * option-rom capability structure will not be available. | |
1683 | */ | |
1684 | const struct imsm_orom *orom; | |
1685 | struct sys_dev *list, *hba; | |
1686 | int host_base = 0; | |
1687 | int port_count = 0; | |
1688 | int result=0; | |
1689 | ||
1690 | if (enumerate_only) { | |
1691 | if (check_env("IMSM_NO_PLATFORM")) | |
1692 | return 0; | |
1693 | list = find_intel_devices(); | |
1694 | if (!list) | |
1695 | return 2; | |
1696 | for (hba = list; hba; hba = hba->next) { | |
1697 | orom = find_imsm_capability(hba->type); | |
1698 | if (!orom) { | |
1699 | result = 2; | |
1700 | break; | |
1701 | } | |
1702 | } | |
1703 | free_sys_dev(&list); | |
1704 | return result; | |
1705 | } | |
1706 | ||
1707 | list = find_intel_devices(); | |
1708 | if (!list) { | |
1709 | if (verbose) | |
1710 | fprintf(stderr, Name ": no active Intel(R) RAID " | |
1711 | "controller found.\n"); | |
1712 | free_sys_dev(&list); | |
1713 | return 2; | |
1714 | } else if (verbose) | |
1715 | print_found_intel_controllers(list); | |
1716 | ||
1717 | for (hba = list; hba; hba = hba->next) { | |
1718 | orom = find_imsm_capability(hba->type); | |
1719 | if (!orom) | |
1720 | fprintf(stderr, Name ": imsm capabilities not found for controller: %s (type %s)\n", | |
1721 | hba->path, get_sys_dev_type(hba->type)); | |
1722 | else | |
1723 | print_imsm_capability(orom); | |
1724 | } | |
1725 | ||
1726 | for (hba = list; hba; hba = hba->next) { | |
1727 | printf(" I/O Controller : %s (%s)\n", | |
1728 | hba->path, get_sys_dev_type(hba->type)); | |
1729 | ||
1730 | if (hba->type == SYS_DEV_SATA) { | |
1731 | host_base = ahci_get_port_count(hba->path, &port_count); | |
1732 | if (ahci_enumerate_ports(hba->path, port_count, host_base, verbose)) { | |
1733 | if (verbose) | |
1734 | fprintf(stderr, Name ": failed to enumerate " | |
1735 | "ports on SATA controller at %s.", hba->pci_id); | |
1736 | result |= 2; | |
1737 | } | |
1738 | } | |
1739 | } | |
1740 | ||
1741 | free_sys_dev(&list); | |
1742 | return result; | |
1743 | } | |
1744 | #endif | |
1745 | ||
1746 | static int match_home_imsm(struct supertype *st, char *homehost) | |
1747 | { | |
1748 | /* the imsm metadata format does not specify any host | |
1749 | * identification information. We return -1 since we can never | |
1750 | * confirm nor deny whether a given array is "meant" for this | |
1751 | * host. We rely on compare_super and the 'family_num' fields to | |
1752 | * exclude member disks that do not belong, and we rely on | |
1753 | * mdadm.conf to specify the arrays that should be assembled. | |
1754 | * Auto-assembly may still pick up "foreign" arrays. | |
1755 | */ | |
1756 | ||
1757 | return -1; | |
1758 | } | |
1759 | ||
1760 | static void uuid_from_super_imsm(struct supertype *st, int uuid[4]) | |
1761 | { | |
1762 | /* The uuid returned here is used for: | |
1763 | * uuid to put into bitmap file (Create, Grow) | |
1764 | * uuid for backup header when saving critical section (Grow) | |
1765 | * comparing uuids when re-adding a device into an array | |
1766 | * In these cases the uuid required is that of the data-array, | |
1767 | * not the device-set. | |
1768 | * uuid to recognise same set when adding a missing device back | |
1769 | * to an array. This is a uuid for the device-set. | |
1770 | * | |
1771 | * For each of these we can make do with a truncated | |
1772 | * or hashed uuid rather than the original, as long as | |
1773 | * everyone agrees. | |
1774 | * In each case the uuid required is that of the data-array, | |
1775 | * not the device-set. | |
1776 | */ | |
1777 | /* imsm does not track uuid's so we synthesis one using sha1 on | |
1778 | * - The signature (Which is constant for all imsm array, but no matter) | |
1779 | * - the orig_family_num of the container | |
1780 | * - the index number of the volume | |
1781 | * - the 'serial' number of the volume. | |
1782 | * Hopefully these are all constant. | |
1783 | */ | |
1784 | struct intel_super *super = st->sb; | |
1785 | ||
1786 | char buf[20]; | |
1787 | struct sha1_ctx ctx; | |
1788 | struct imsm_dev *dev = NULL; | |
1789 | __u32 family_num; | |
1790 | ||
1791 | /* some mdadm versions failed to set ->orig_family_num, in which | |
1792 | * case fall back to ->family_num. orig_family_num will be | |
1793 | * fixed up with the first metadata update. | |
1794 | */ | |
1795 | family_num = super->anchor->orig_family_num; | |
1796 | if (family_num == 0) | |
1797 | family_num = super->anchor->family_num; | |
1798 | sha1_init_ctx(&ctx); | |
1799 | sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx); | |
1800 | sha1_process_bytes(&family_num, sizeof(__u32), &ctx); | |
1801 | if (super->current_vol >= 0) | |
1802 | dev = get_imsm_dev(super, super->current_vol); | |
1803 | if (dev) { | |
1804 | __u32 vol = super->current_vol; | |
1805 | sha1_process_bytes(&vol, sizeof(vol), &ctx); | |
1806 | sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx); | |
1807 | } | |
1808 | sha1_finish_ctx(&ctx, buf); | |
1809 | memcpy(uuid, buf, 4*4); | |
1810 | } | |
1811 | ||
1812 | #if 0 | |
1813 | static void | |
1814 | get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p) | |
1815 | { | |
1816 | __u8 *v = get_imsm_version(mpb); | |
1817 | __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH; | |
1818 | char major[] = { 0, 0, 0 }; | |
1819 | char minor[] = { 0 ,0, 0 }; | |
1820 | char patch[] = { 0, 0, 0 }; | |
1821 | char *ver_parse[] = { major, minor, patch }; | |
1822 | int i, j; | |
1823 | ||
1824 | i = j = 0; | |
1825 | while (*v != '\0' && v < end) { | |
1826 | if (*v != '.' && j < 2) | |
1827 | ver_parse[i][j++] = *v; | |
1828 | else { | |
1829 | i++; | |
1830 | j = 0; | |
1831 | } | |
1832 | v++; | |
1833 | } | |
1834 | ||
1835 | *m = strtol(minor, NULL, 0); | |
1836 | *p = strtol(patch, NULL, 0); | |
1837 | } | |
1838 | #endif | |
1839 | ||
1840 | static __u32 migr_strip_blocks_resync(struct imsm_dev *dev) | |
1841 | { | |
1842 | /* migr_strip_size when repairing or initializing parity */ | |
1843 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1844 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1845 | ||
1846 | switch (get_imsm_raid_level(map)) { | |
1847 | case 5: | |
1848 | case 10: | |
1849 | return chunk; | |
1850 | default: | |
1851 | return 128*1024 >> 9; | |
1852 | } | |
1853 | } | |
1854 | ||
1855 | static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev) | |
1856 | { | |
1857 | /* migr_strip_size when rebuilding a degraded disk, no idea why | |
1858 | * this is different than migr_strip_size_resync(), but it's good | |
1859 | * to be compatible | |
1860 | */ | |
1861 | struct imsm_map *map = get_imsm_map(dev, 1); | |
1862 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1863 | ||
1864 | switch (get_imsm_raid_level(map)) { | |
1865 | case 1: | |
1866 | case 10: | |
1867 | if (map->num_members % map->num_domains == 0) | |
1868 | return 128*1024 >> 9; | |
1869 | else | |
1870 | return chunk; | |
1871 | case 5: | |
1872 | return max((__u32) 64*1024 >> 9, chunk); | |
1873 | default: | |
1874 | return 128*1024 >> 9; | |
1875 | } | |
1876 | } | |
1877 | ||
1878 | static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev) | |
1879 | { | |
1880 | struct imsm_map *lo = get_imsm_map(dev, 0); | |
1881 | struct imsm_map *hi = get_imsm_map(dev, 1); | |
1882 | __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip); | |
1883 | __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip); | |
1884 | ||
1885 | return max((__u32) 1, hi_chunk / lo_chunk); | |
1886 | } | |
1887 | ||
1888 | static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev) | |
1889 | { | |
1890 | struct imsm_map *lo = get_imsm_map(dev, 0); | |
1891 | int level = get_imsm_raid_level(lo); | |
1892 | ||
1893 | if (level == 1 || level == 10) { | |
1894 | struct imsm_map *hi = get_imsm_map(dev, 1); | |
1895 | ||
1896 | return hi->num_domains; | |
1897 | } else | |
1898 | return num_stripes_per_unit_resync(dev); | |
1899 | } | |
1900 | ||
1901 | static __u8 imsm_num_data_members(struct imsm_dev *dev, int second_map) | |
1902 | { | |
1903 | /* named 'imsm_' because raid0, raid1 and raid10 | |
1904 | * counter-intuitively have the same number of data disks | |
1905 | */ | |
1906 | struct imsm_map *map = get_imsm_map(dev, second_map); | |
1907 | ||
1908 | switch (get_imsm_raid_level(map)) { | |
1909 | case 0: | |
1910 | case 1: | |
1911 | case 10: | |
1912 | return map->num_members; | |
1913 | case 5: | |
1914 | return map->num_members - 1; | |
1915 | default: | |
1916 | dprintf("%s: unsupported raid level\n", __func__); | |
1917 | return 0; | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | static __u32 parity_segment_depth(struct imsm_dev *dev) | |
1922 | { | |
1923 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1924 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1925 | ||
1926 | switch(get_imsm_raid_level(map)) { | |
1927 | case 1: | |
1928 | case 10: | |
1929 | return chunk * map->num_domains; | |
1930 | case 5: | |
1931 | return chunk * map->num_members; | |
1932 | default: | |
1933 | return chunk; | |
1934 | } | |
1935 | } | |
1936 | ||
1937 | static __u32 map_migr_block(struct imsm_dev *dev, __u32 block) | |
1938 | { | |
1939 | struct imsm_map *map = get_imsm_map(dev, 1); | |
1940 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1941 | __u32 strip = block / chunk; | |
1942 | ||
1943 | switch (get_imsm_raid_level(map)) { | |
1944 | case 1: | |
1945 | case 10: { | |
1946 | __u32 vol_strip = (strip * map->num_domains) + 1; | |
1947 | __u32 vol_stripe = vol_strip / map->num_members; | |
1948 | ||
1949 | return vol_stripe * chunk + block % chunk; | |
1950 | } case 5: { | |
1951 | __u32 stripe = strip / (map->num_members - 1); | |
1952 | ||
1953 | return stripe * chunk + block % chunk; | |
1954 | } | |
1955 | default: | |
1956 | return 0; | |
1957 | } | |
1958 | } | |
1959 | ||
1960 | static __u64 blocks_per_migr_unit(struct intel_super *super, | |
1961 | struct imsm_dev *dev) | |
1962 | { | |
1963 | /* calculate the conversion factor between per member 'blocks' | |
1964 | * (md/{resync,rebuild}_start) and imsm migration units, return | |
1965 | * 0 for the 'not migrating' and 'unsupported migration' cases | |
1966 | */ | |
1967 | if (!dev->vol.migr_state) | |
1968 | return 0; | |
1969 | ||
1970 | switch (migr_type(dev)) { | |
1971 | case MIGR_GEN_MIGR: { | |
1972 | struct migr_record *migr_rec = super->migr_rec; | |
1973 | return __le32_to_cpu(migr_rec->blocks_per_unit); | |
1974 | } | |
1975 | case MIGR_VERIFY: | |
1976 | case MIGR_REPAIR: | |
1977 | case MIGR_INIT: { | |
1978 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1979 | __u32 stripes_per_unit; | |
1980 | __u32 blocks_per_unit; | |
1981 | __u32 parity_depth; | |
1982 | __u32 migr_chunk; | |
1983 | __u32 block_map; | |
1984 | __u32 block_rel; | |
1985 | __u32 segment; | |
1986 | __u32 stripe; | |
1987 | __u8 disks; | |
1988 | ||
1989 | /* yes, this is really the translation of migr_units to | |
1990 | * per-member blocks in the 'resync' case | |
1991 | */ | |
1992 | stripes_per_unit = num_stripes_per_unit_resync(dev); | |
1993 | migr_chunk = migr_strip_blocks_resync(dev); | |
1994 | disks = imsm_num_data_members(dev, 0); | |
1995 | blocks_per_unit = stripes_per_unit * migr_chunk * disks; | |
1996 | stripe = __le16_to_cpu(map->blocks_per_strip) * disks; | |
1997 | segment = blocks_per_unit / stripe; | |
1998 | block_rel = blocks_per_unit - segment * stripe; | |
1999 | parity_depth = parity_segment_depth(dev); | |
2000 | block_map = map_migr_block(dev, block_rel); | |
2001 | return block_map + parity_depth * segment; | |
2002 | } | |
2003 | case MIGR_REBUILD: { | |
2004 | __u32 stripes_per_unit; | |
2005 | __u32 migr_chunk; | |
2006 | ||
2007 | stripes_per_unit = num_stripes_per_unit_rebuild(dev); | |
2008 | migr_chunk = migr_strip_blocks_rebuild(dev); | |
2009 | return migr_chunk * stripes_per_unit; | |
2010 | } | |
2011 | case MIGR_STATE_CHANGE: | |
2012 | default: | |
2013 | return 0; | |
2014 | } | |
2015 | } | |
2016 | ||
2017 | static int imsm_level_to_layout(int level) | |
2018 | { | |
2019 | switch (level) { | |
2020 | case 0: | |
2021 | case 1: | |
2022 | return 0; | |
2023 | case 5: | |
2024 | case 6: | |
2025 | return ALGORITHM_LEFT_ASYMMETRIC; | |
2026 | case 10: | |
2027 | return 0x102; | |
2028 | } | |
2029 | return UnSet; | |
2030 | } | |
2031 | ||
2032 | /******************************************************************************* | |
2033 | * Function: read_imsm_migr_rec | |
2034 | * Description: Function reads imsm migration record from last sector of disk | |
2035 | * Parameters: | |
2036 | * fd : disk descriptor | |
2037 | * super : metadata info | |
2038 | * Returns: | |
2039 | * 0 : success, | |
2040 | * -1 : fail | |
2041 | ******************************************************************************/ | |
2042 | static int read_imsm_migr_rec(int fd, struct intel_super *super) | |
2043 | { | |
2044 | int ret_val = -1; | |
2045 | unsigned long long dsize; | |
2046 | ||
2047 | get_dev_size(fd, NULL, &dsize); | |
2048 | if (lseek64(fd, dsize - 512, SEEK_SET) < 0) { | |
2049 | fprintf(stderr, | |
2050 | Name ": Cannot seek to anchor block: %s\n", | |
2051 | strerror(errno)); | |
2052 | goto out; | |
2053 | } | |
2054 | if (read(fd, super->migr_rec_buf, 512) != 512) { | |
2055 | fprintf(stderr, | |
2056 | Name ": Cannot read migr record block: %s\n", | |
2057 | strerror(errno)); | |
2058 | goto out; | |
2059 | } | |
2060 | ret_val = 0; | |
2061 | ||
2062 | out: | |
2063 | return ret_val; | |
2064 | } | |
2065 | ||
2066 | /******************************************************************************* | |
2067 | * Function: load_imsm_migr_rec | |
2068 | * Description: Function reads imsm migration record (it is stored at the last | |
2069 | * sector of disk) | |
2070 | * Parameters: | |
2071 | * super : imsm internal array info | |
2072 | * info : general array info | |
2073 | * Returns: | |
2074 | * 0 : success | |
2075 | * -1 : fail | |
2076 | ******************************************************************************/ | |
2077 | static int load_imsm_migr_rec(struct intel_super *super, struct mdinfo *info) | |
2078 | { | |
2079 | struct mdinfo *sd; | |
2080 | struct dl *dl = NULL; | |
2081 | char nm[30]; | |
2082 | int retval = -1; | |
2083 | int fd = -1; | |
2084 | ||
2085 | if (info) { | |
2086 | for (sd = info->devs ; sd ; sd = sd->next) { | |
2087 | /* read only from one of the first two slots */ | |
2088 | if ((sd->disk.raid_disk > 1) || | |
2089 | (sd->disk.raid_disk < 0)) | |
2090 | continue; | |
2091 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
2092 | fd = dev_open(nm, O_RDONLY); | |
2093 | if (fd >= 0) | |
2094 | break; | |
2095 | } | |
2096 | } | |
2097 | if (fd < 0) { | |
2098 | for (dl = super->disks; dl; dl = dl->next) { | |
2099 | /* read only from one of the first two slots */ | |
2100 | if (dl->index > 1) | |
2101 | continue; | |
2102 | sprintf(nm, "%d:%d", dl->major, dl->minor); | |
2103 | fd = dev_open(nm, O_RDONLY); | |
2104 | if (fd >= 0) | |
2105 | break; | |
2106 | } | |
2107 | } | |
2108 | if (fd < 0) | |
2109 | goto out; | |
2110 | retval = read_imsm_migr_rec(fd, super); | |
2111 | ||
2112 | out: | |
2113 | if (fd >= 0) | |
2114 | close(fd); | |
2115 | return retval; | |
2116 | } | |
2117 | ||
2118 | #ifndef MDASSEMBLE | |
2119 | /******************************************************************************* | |
2120 | * function: imsm_create_metadata_checkpoint_update | |
2121 | * Description: It creates update for checkpoint change. | |
2122 | * Parameters: | |
2123 | * super : imsm internal array info | |
2124 | * u : pointer to prepared update | |
2125 | * Returns: | |
2126 | * Uptate length. | |
2127 | * If length is equal to 0, input pointer u contains no update | |
2128 | ******************************************************************************/ | |
2129 | static int imsm_create_metadata_checkpoint_update( | |
2130 | struct intel_super *super, | |
2131 | struct imsm_update_general_migration_checkpoint **u) | |
2132 | { | |
2133 | ||
2134 | int update_memory_size = 0; | |
2135 | ||
2136 | dprintf("imsm_create_metadata_checkpoint_update(enter)\n"); | |
2137 | ||
2138 | if (u == NULL) | |
2139 | return 0; | |
2140 | *u = NULL; | |
2141 | ||
2142 | /* size of all update data without anchor */ | |
2143 | update_memory_size = | |
2144 | sizeof(struct imsm_update_general_migration_checkpoint); | |
2145 | ||
2146 | *u = calloc(1, update_memory_size); | |
2147 | if (*u == NULL) { | |
2148 | dprintf("error: cannot get memory for " | |
2149 | "imsm_create_metadata_checkpoint_update update\n"); | |
2150 | return 0; | |
2151 | } | |
2152 | (*u)->type = update_general_migration_checkpoint; | |
2153 | (*u)->curr_migr_unit = __le32_to_cpu(super->migr_rec->curr_migr_unit); | |
2154 | dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n", | |
2155 | (*u)->curr_migr_unit); | |
2156 | ||
2157 | return update_memory_size; | |
2158 | } | |
2159 | ||
2160 | ||
2161 | static void imsm_update_metadata_locally(struct supertype *st, | |
2162 | void *buf, int len); | |
2163 | ||
2164 | /******************************************************************************* | |
2165 | * Function: write_imsm_migr_rec | |
2166 | * Description: Function writes imsm migration record | |
2167 | * (at the last sector of disk) | |
2168 | * Parameters: | |
2169 | * super : imsm internal array info | |
2170 | * Returns: | |
2171 | * 0 : success | |
2172 | * -1 : if fail | |
2173 | ******************************************************************************/ | |
2174 | static int write_imsm_migr_rec(struct supertype *st) | |
2175 | { | |
2176 | struct intel_super *super = st->sb; | |
2177 | unsigned long long dsize; | |
2178 | char nm[30]; | |
2179 | int fd = -1; | |
2180 | int retval = -1; | |
2181 | struct dl *sd; | |
2182 | int len; | |
2183 | struct imsm_update_general_migration_checkpoint *u; | |
2184 | ||
2185 | for (sd = super->disks ; sd ; sd = sd->next) { | |
2186 | /* write to 2 first slots only */ | |
2187 | if ((sd->index < 0) || (sd->index > 1)) | |
2188 | continue; | |
2189 | sprintf(nm, "%d:%d", sd->major, sd->minor); | |
2190 | fd = dev_open(nm, O_RDWR); | |
2191 | if (fd < 0) | |
2192 | continue; | |
2193 | get_dev_size(fd, NULL, &dsize); | |
2194 | if (lseek64(fd, dsize - 512, SEEK_SET) < 0) { | |
2195 | fprintf(stderr, | |
2196 | Name ": Cannot seek to anchor block: %s\n", | |
2197 | strerror(errno)); | |
2198 | goto out; | |
2199 | } | |
2200 | if (write(fd, super->migr_rec_buf, 512) != 512) { | |
2201 | fprintf(stderr, | |
2202 | Name ": Cannot write migr record block: %s\n", | |
2203 | strerror(errno)); | |
2204 | goto out; | |
2205 | } | |
2206 | close(fd); | |
2207 | fd = -1; | |
2208 | } | |
2209 | /* update checkpoint information in metadata */ | |
2210 | len = imsm_create_metadata_checkpoint_update(super, &u); | |
2211 | ||
2212 | if (len <= 0) { | |
2213 | dprintf("imsm: Cannot prepare update\n"); | |
2214 | goto out; | |
2215 | } | |
2216 | /* update metadata locally */ | |
2217 | imsm_update_metadata_locally(st, u, len); | |
2218 | /* and possibly remotely */ | |
2219 | if (st->update_tail) { | |
2220 | append_metadata_update(st, u, len); | |
2221 | /* during reshape we do all work inside metadata handler | |
2222 | * manage_reshape(), so metadata update has to be triggered | |
2223 | * insida it | |
2224 | */ | |
2225 | flush_metadata_updates(st); | |
2226 | st->update_tail = &st->updates; | |
2227 | } else | |
2228 | free(u); | |
2229 | ||
2230 | retval = 0; | |
2231 | out: | |
2232 | if (fd >= 0) | |
2233 | close(fd); | |
2234 | return retval; | |
2235 | } | |
2236 | #endif /* MDASSEMBLE */ | |
2237 | ||
2238 | /* spare/missing disks activations are not allowe when | |
2239 | * array/container performs reshape operation, because | |
2240 | * all arrays in container works on the same disks set | |
2241 | */ | |
2242 | int imsm_reshape_blocks_arrays_changes(struct intel_super *super) | |
2243 | { | |
2244 | int rv = 0; | |
2245 | struct intel_dev *i_dev; | |
2246 | struct imsm_dev *dev; | |
2247 | ||
2248 | /* check whole container | |
2249 | */ | |
2250 | for (i_dev = super->devlist; i_dev; i_dev = i_dev->next) { | |
2251 | dev = i_dev->dev; | |
2252 | if (is_gen_migration(dev)) { | |
2253 | /* No repair during any migration in container | |
2254 | */ | |
2255 | rv = 1; | |
2256 | break; | |
2257 | } | |
2258 | } | |
2259 | return rv; | |
2260 | } | |
2261 | ||
2262 | static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap) | |
2263 | { | |
2264 | struct intel_super *super = st->sb; | |
2265 | struct migr_record *migr_rec = super->migr_rec; | |
2266 | struct imsm_dev *dev = get_imsm_dev(super, super->current_vol); | |
2267 | struct imsm_map *map = get_imsm_map(dev, 0); | |
2268 | struct imsm_map *prev_map = get_imsm_map(dev, 1); | |
2269 | struct imsm_map *map_to_analyse = map; | |
2270 | struct dl *dl; | |
2271 | char *devname; | |
2272 | unsigned int component_size_alligment; | |
2273 | int map_disks = info->array.raid_disks; | |
2274 | ||
2275 | memset(info, 0, sizeof(*info)); | |
2276 | if (prev_map) | |
2277 | map_to_analyse = prev_map; | |
2278 | ||
2279 | dl = super->current_disk; | |
2280 | ||
2281 | info->container_member = super->current_vol; | |
2282 | info->array.raid_disks = map->num_members; | |
2283 | info->array.level = get_imsm_raid_level(map_to_analyse); | |
2284 | info->array.layout = imsm_level_to_layout(info->array.level); | |
2285 | info->array.md_minor = -1; | |
2286 | info->array.ctime = 0; | |
2287 | info->array.utime = 0; | |
2288 | info->array.chunk_size = | |
2289 | __le16_to_cpu(map_to_analyse->blocks_per_strip) << 9; | |
2290 | info->array.state = !dev->vol.dirty; | |
2291 | info->custom_array_size = __le32_to_cpu(dev->size_high); | |
2292 | info->custom_array_size <<= 32; | |
2293 | info->custom_array_size |= __le32_to_cpu(dev->size_low); | |
2294 | info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb); | |
2295 | ||
2296 | if (prev_map && map->map_state == prev_map->map_state && | |
2297 | (migr_type(dev) == MIGR_GEN_MIGR)) { | |
2298 | info->reshape_active = 1; | |
2299 | info->new_level = get_imsm_raid_level(map); | |
2300 | info->new_layout = imsm_level_to_layout(info->new_level); | |
2301 | info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9; | |
2302 | info->delta_disks = map->num_members - prev_map->num_members; | |
2303 | if (info->delta_disks) { | |
2304 | /* this needs to be applied to every array | |
2305 | * in the container. | |
2306 | */ | |
2307 | info->reshape_active = CONTAINER_RESHAPE; | |
2308 | } | |
2309 | /* We shape information that we give to md might have to be | |
2310 | * modify to cope with md's requirement for reshaping arrays. | |
2311 | * For example, when reshaping a RAID0, md requires it to be | |
2312 | * presented as a degraded RAID4. | |
2313 | * Also if a RAID0 is migrating to a RAID5 we need to specify | |
2314 | * the array as already being RAID5, but the 'before' layout | |
2315 | * is a RAID4-like layout. | |
2316 | */ | |
2317 | switch (info->array.level) { | |
2318 | case 0: | |
2319 | switch(info->new_level) { | |
2320 | case 0: | |
2321 | /* conversion is happening as RAID4 */ | |
2322 | info->array.level = 4; | |
2323 | info->array.raid_disks += 1; | |
2324 | break; | |
2325 | case 5: | |
2326 | /* conversion is happening as RAID5 */ | |
2327 | info->array.level = 5; | |
2328 | info->array.layout = ALGORITHM_PARITY_N; | |
2329 | info->delta_disks -= 1; | |
2330 | break; | |
2331 | default: | |
2332 | /* FIXME error message */ | |
2333 | info->array.level = UnSet; | |
2334 | break; | |
2335 | } | |
2336 | break; | |
2337 | } | |
2338 | } else { | |
2339 | info->new_level = UnSet; | |
2340 | info->new_layout = UnSet; | |
2341 | info->new_chunk = info->array.chunk_size; | |
2342 | info->delta_disks = 0; | |
2343 | } | |
2344 | ||
2345 | if (dl) { | |
2346 | info->disk.major = dl->major; | |
2347 | info->disk.minor = dl->minor; | |
2348 | info->disk.number = dl->index; | |
2349 | info->disk.raid_disk = get_imsm_disk_slot(map_to_analyse, | |
2350 | dl->index); | |
2351 | } | |
2352 | ||
2353 | info->data_offset = __le32_to_cpu(map_to_analyse->pba_of_lba0); | |
2354 | info->component_size = | |
2355 | __le32_to_cpu(map_to_analyse->blocks_per_member); | |
2356 | ||
2357 | /* check component size aligment | |
2358 | */ | |
2359 | component_size_alligment = | |
2360 | info->component_size % (info->array.chunk_size/512); | |
2361 | ||
2362 | if (component_size_alligment && | |
2363 | (info->array.level != 1) && (info->array.level != UnSet)) { | |
2364 | dprintf("imsm: reported component size alligned from %llu ", | |
2365 | info->component_size); | |
2366 | info->component_size -= component_size_alligment; | |
2367 | dprintf("to %llu (%i).\n", | |
2368 | info->component_size, component_size_alligment); | |
2369 | } | |
2370 | ||
2371 | memset(info->uuid, 0, sizeof(info->uuid)); | |
2372 | info->recovery_start = MaxSector; | |
2373 | ||
2374 | info->reshape_progress = 0; | |
2375 | info->resync_start = MaxSector; | |
2376 | if ((map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED || | |
2377 | dev->vol.dirty) && | |
2378 | imsm_reshape_blocks_arrays_changes(super) == 0) { | |
2379 | info->resync_start = 0; | |
2380 | } | |
2381 | if (dev->vol.migr_state) { | |
2382 | switch (migr_type(dev)) { | |
2383 | case MIGR_REPAIR: | |
2384 | case MIGR_INIT: { | |
2385 | __u64 blocks_per_unit = blocks_per_migr_unit(super, | |
2386 | dev); | |
2387 | __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit); | |
2388 | ||
2389 | info->resync_start = blocks_per_unit * units; | |
2390 | break; | |
2391 | } | |
2392 | case MIGR_GEN_MIGR: { | |
2393 | __u64 blocks_per_unit = blocks_per_migr_unit(super, | |
2394 | dev); | |
2395 | __u64 units = __le32_to_cpu(migr_rec->curr_migr_unit); | |
2396 | unsigned long long array_blocks; | |
2397 | int used_disks; | |
2398 | ||
2399 | if (__le32_to_cpu(migr_rec->ascending_migr) && | |
2400 | (units < | |
2401 | (__le32_to_cpu(migr_rec->num_migr_units)-1)) && | |
2402 | (super->migr_rec->rec_status == | |
2403 | __cpu_to_le32(UNIT_SRC_IN_CP_AREA))) | |
2404 | units++; | |
2405 | ||
2406 | info->reshape_progress = blocks_per_unit * units; | |
2407 | ||
2408 | dprintf("IMSM: General Migration checkpoint : %llu " | |
2409 | "(%llu) -> read reshape progress : %llu\n", | |
2410 | (unsigned long long)units, | |
2411 | (unsigned long long)blocks_per_unit, | |
2412 | info->reshape_progress); | |
2413 | ||
2414 | used_disks = imsm_num_data_members(dev, 1); | |
2415 | if (used_disks > 0) { | |
2416 | array_blocks = map->blocks_per_member * | |
2417 | used_disks; | |
2418 | /* round array size down to closest MB | |
2419 | */ | |
2420 | info->custom_array_size = (array_blocks | |
2421 | >> SECT_PER_MB_SHIFT) | |
2422 | << SECT_PER_MB_SHIFT; | |
2423 | } | |
2424 | } | |
2425 | case MIGR_VERIFY: | |
2426 | /* we could emulate the checkpointing of | |
2427 | * 'sync_action=check' migrations, but for now | |
2428 | * we just immediately complete them | |
2429 | */ | |
2430 | case MIGR_REBUILD: | |
2431 | /* this is handled by container_content_imsm() */ | |
2432 | case MIGR_STATE_CHANGE: | |
2433 | /* FIXME handle other migrations */ | |
2434 | default: | |
2435 | /* we are not dirty, so... */ | |
2436 | info->resync_start = MaxSector; | |
2437 | } | |
2438 | } | |
2439 | ||
2440 | strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN); | |
2441 | info->name[MAX_RAID_SERIAL_LEN] = 0; | |
2442 | ||
2443 | info->array.major_version = -1; | |
2444 | info->array.minor_version = -2; | |
2445 | devname = devnum2devname(st->container_dev); | |
2446 | *info->text_version = '\0'; | |
2447 | if (devname) | |
2448 | sprintf(info->text_version, "/%s/%d", devname, info->container_member); | |
2449 | free(devname); | |
2450 | info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */ | |
2451 | uuid_from_super_imsm(st, info->uuid); | |
2452 | ||
2453 | if (dmap) { | |
2454 | int i, j; | |
2455 | for (i=0; i<map_disks; i++) { | |
2456 | dmap[i] = 0; | |
2457 | if (i < info->array.raid_disks) { | |
2458 | struct imsm_disk *dsk; | |
2459 | j = get_imsm_disk_idx(dev, i, -1); | |
2460 | dsk = get_imsm_disk(super, j); | |
2461 | if (dsk && (dsk->status & CONFIGURED_DISK)) | |
2462 | dmap[i] = 1; | |
2463 | } | |
2464 | } | |
2465 | } | |
2466 | } | |
2467 | ||
2468 | static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed); | |
2469 | static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev); | |
2470 | ||
2471 | static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index) | |
2472 | { | |
2473 | struct dl *d; | |
2474 | ||
2475 | for (d = super->missing; d; d = d->next) | |
2476 | if (d->index == index) | |
2477 | return &d->disk; | |
2478 | return NULL; | |
2479 | } | |
2480 | ||
2481 | static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map) | |
2482 | { | |
2483 | struct intel_super *super = st->sb; | |
2484 | struct imsm_disk *disk; | |
2485 | int map_disks = info->array.raid_disks; | |
2486 | int max_enough = -1; | |
2487 | int i; | |
2488 | struct imsm_super *mpb; | |
2489 | ||
2490 | if (super->current_vol >= 0) { | |
2491 | getinfo_super_imsm_volume(st, info, map); | |
2492 | return; | |
2493 | } | |
2494 | memset(info, 0, sizeof(*info)); | |
2495 | ||
2496 | /* Set raid_disks to zero so that Assemble will always pull in valid | |
2497 | * spares | |
2498 | */ | |
2499 | info->array.raid_disks = 0; | |
2500 | info->array.level = LEVEL_CONTAINER; | |
2501 | info->array.layout = 0; | |
2502 | info->array.md_minor = -1; | |
2503 | info->array.ctime = 0; /* N/A for imsm */ | |
2504 | info->array.utime = 0; | |
2505 | info->array.chunk_size = 0; | |
2506 | ||
2507 | info->disk.major = 0; | |
2508 | info->disk.minor = 0; | |
2509 | info->disk.raid_disk = -1; | |
2510 | info->reshape_active = 0; | |
2511 | info->array.major_version = -1; | |
2512 | info->array.minor_version = -2; | |
2513 | strcpy(info->text_version, "imsm"); | |
2514 | info->safe_mode_delay = 0; | |
2515 | info->disk.number = -1; | |
2516 | info->disk.state = 0; | |
2517 | info->name[0] = 0; | |
2518 | info->recovery_start = MaxSector; | |
2519 | info->recovery_blocked = imsm_reshape_blocks_arrays_changes(st->sb); | |
2520 | ||
2521 | /* do we have the all the insync disks that we expect? */ | |
2522 | mpb = super->anchor; | |
2523 | ||
2524 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
2525 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
2526 | int failed, enough, j, missing = 0; | |
2527 | struct imsm_map *map; | |
2528 | __u8 state; | |
2529 | ||
2530 | failed = imsm_count_failed(super, dev); | |
2531 | state = imsm_check_degraded(super, dev, failed); | |
2532 | map = get_imsm_map(dev, dev->vol.migr_state); | |
2533 | ||
2534 | /* any newly missing disks? | |
2535 | * (catches single-degraded vs double-degraded) | |
2536 | */ | |
2537 | for (j = 0; j < map->num_members; j++) { | |
2538 | __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1); | |
2539 | __u32 idx = ord_to_idx(ord); | |
2540 | ||
2541 | if (!(ord & IMSM_ORD_REBUILD) && | |
2542 | get_imsm_missing(super, idx)) { | |
2543 | missing = 1; | |
2544 | break; | |
2545 | } | |
2546 | } | |
2547 | ||
2548 | if (state == IMSM_T_STATE_FAILED) | |
2549 | enough = -1; | |
2550 | else if (state == IMSM_T_STATE_DEGRADED && | |
2551 | (state != map->map_state || missing)) | |
2552 | enough = 0; | |
2553 | else /* we're normal, or already degraded */ | |
2554 | enough = 1; | |
2555 | ||
2556 | /* in the missing/failed disk case check to see | |
2557 | * if at least one array is runnable | |
2558 | */ | |
2559 | max_enough = max(max_enough, enough); | |
2560 | } | |
2561 | dprintf("%s: enough: %d\n", __func__, max_enough); | |
2562 | info->container_enough = max_enough; | |
2563 | ||
2564 | if (super->disks) { | |
2565 | __u32 reserved = imsm_reserved_sectors(super, super->disks); | |
2566 | ||
2567 | disk = &super->disks->disk; | |
2568 | info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved; | |
2569 | info->component_size = reserved; | |
2570 | info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0; | |
2571 | /* we don't change info->disk.raid_disk here because | |
2572 | * this state will be finalized in mdmon after we have | |
2573 | * found the 'most fresh' version of the metadata | |
2574 | */ | |
2575 | info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0; | |
2576 | info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC); | |
2577 | } | |
2578 | ||
2579 | /* only call uuid_from_super_imsm when this disk is part of a populated container, | |
2580 | * ->compare_super may have updated the 'num_raid_devs' field for spares | |
2581 | */ | |
2582 | if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs) | |
2583 | uuid_from_super_imsm(st, info->uuid); | |
2584 | else | |
2585 | memcpy(info->uuid, uuid_zero, sizeof(uuid_zero)); | |
2586 | ||
2587 | /* I don't know how to compute 'map' on imsm, so use safe default */ | |
2588 | if (map) { | |
2589 | int i; | |
2590 | for (i = 0; i < map_disks; i++) | |
2591 | map[i] = 1; | |
2592 | } | |
2593 | ||
2594 | } | |
2595 | ||
2596 | /* allocates memory and fills disk in mdinfo structure | |
2597 | * for each disk in array */ | |
2598 | struct mdinfo *getinfo_super_disks_imsm(struct supertype *st) | |
2599 | { | |
2600 | struct mdinfo *mddev = NULL; | |
2601 | struct intel_super *super = st->sb; | |
2602 | struct imsm_disk *disk; | |
2603 | int count = 0; | |
2604 | struct dl *dl; | |
2605 | if (!super || !super->disks) | |
2606 | return NULL; | |
2607 | dl = super->disks; | |
2608 | mddev = malloc(sizeof(*mddev)); | |
2609 | if (!mddev) { | |
2610 | fprintf(stderr, Name ": Failed to allocate memory.\n"); | |
2611 | return NULL; | |
2612 | } | |
2613 | memset(mddev, 0, sizeof(*mddev)); | |
2614 | while (dl) { | |
2615 | struct mdinfo *tmp; | |
2616 | disk = &dl->disk; | |
2617 | tmp = malloc(sizeof(*tmp)); | |
2618 | if (!tmp) { | |
2619 | fprintf(stderr, Name ": Failed to allocate memory.\n"); | |
2620 | if (mddev) | |
2621 | sysfs_free(mddev); | |
2622 | return NULL; | |
2623 | } | |
2624 | memset(tmp, 0, sizeof(*tmp)); | |
2625 | if (mddev->devs) | |
2626 | tmp->next = mddev->devs; | |
2627 | mddev->devs = tmp; | |
2628 | tmp->disk.number = count++; | |
2629 | tmp->disk.major = dl->major; | |
2630 | tmp->disk.minor = dl->minor; | |
2631 | tmp->disk.state = is_configured(disk) ? | |
2632 | (1 << MD_DISK_ACTIVE) : 0; | |
2633 | tmp->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0; | |
2634 | tmp->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC); | |
2635 | tmp->disk.raid_disk = -1; | |
2636 | dl = dl->next; | |
2637 | } | |
2638 | return mddev; | |
2639 | } | |
2640 | ||
2641 | static int update_super_imsm(struct supertype *st, struct mdinfo *info, | |
2642 | char *update, char *devname, int verbose, | |
2643 | int uuid_set, char *homehost) | |
2644 | { | |
2645 | /* For 'assemble' and 'force' we need to return non-zero if any | |
2646 | * change was made. For others, the return value is ignored. | |
2647 | * Update options are: | |
2648 | * force-one : This device looks a bit old but needs to be included, | |
2649 | * update age info appropriately. | |
2650 | * assemble: clear any 'faulty' flag to allow this device to | |
2651 | * be assembled. | |
2652 | * force-array: Array is degraded but being forced, mark it clean | |
2653 | * if that will be needed to assemble it. | |
2654 | * | |
2655 | * newdev: not used ???? | |
2656 | * grow: Array has gained a new device - this is currently for | |
2657 | * linear only | |
2658 | * resync: mark as dirty so a resync will happen. | |
2659 | * name: update the name - preserving the homehost | |
2660 | * uuid: Change the uuid of the array to match watch is given | |
2661 | * | |
2662 | * Following are not relevant for this imsm: | |
2663 | * sparc2.2 : update from old dodgey metadata | |
2664 | * super-minor: change the preferred_minor number | |
2665 | * summaries: update redundant counters. | |
2666 | * homehost: update the recorded homehost | |
2667 | * _reshape_progress: record new reshape_progress position. | |
2668 | */ | |
2669 | int rv = 1; | |
2670 | struct intel_super *super = st->sb; | |
2671 | struct imsm_super *mpb; | |
2672 | ||
2673 | /* we can only update container info */ | |
2674 | if (!super || super->current_vol >= 0 || !super->anchor) | |
2675 | return 1; | |
2676 | ||
2677 | mpb = super->anchor; | |
2678 | ||
2679 | if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private) | |
2680 | rv = -1; | |
2681 | else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) { | |
2682 | mpb->orig_family_num = *((__u32 *) info->update_private); | |
2683 | rv = 0; | |
2684 | } else if (strcmp(update, "uuid") == 0) { | |
2685 | __u32 *new_family = malloc(sizeof(*new_family)); | |
2686 | ||
2687 | /* update orig_family_number with the incoming random | |
2688 | * data, report the new effective uuid, and store the | |
2689 | * new orig_family_num for future updates. | |
2690 | */ | |
2691 | if (new_family) { | |
2692 | memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32)); | |
2693 | uuid_from_super_imsm(st, info->uuid); | |
2694 | *new_family = mpb->orig_family_num; | |
2695 | info->update_private = new_family; | |
2696 | rv = 0; | |
2697 | } | |
2698 | } else if (strcmp(update, "assemble") == 0) | |
2699 | rv = 0; | |
2700 | else | |
2701 | rv = -1; | |
2702 | ||
2703 | /* successful update? recompute checksum */ | |
2704 | if (rv == 0) | |
2705 | mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb)); | |
2706 | ||
2707 | return rv; | |
2708 | } | |
2709 | ||
2710 | static size_t disks_to_mpb_size(int disks) | |
2711 | { | |
2712 | size_t size; | |
2713 | ||
2714 | size = sizeof(struct imsm_super); | |
2715 | size += (disks - 1) * sizeof(struct imsm_disk); | |
2716 | size += 2 * sizeof(struct imsm_dev); | |
2717 | /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */ | |
2718 | size += (4 - 2) * sizeof(struct imsm_map); | |
2719 | /* 4 possible disk_ord_tbl's */ | |
2720 | size += 4 * (disks - 1) * sizeof(__u32); | |
2721 | ||
2722 | return size; | |
2723 | } | |
2724 | ||
2725 | static __u64 avail_size_imsm(struct supertype *st, __u64 devsize) | |
2726 | { | |
2727 | if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS)) | |
2728 | return 0; | |
2729 | ||
2730 | return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS); | |
2731 | } | |
2732 | ||
2733 | static void free_devlist(struct intel_super *super) | |
2734 | { | |
2735 | struct intel_dev *dv; | |
2736 | ||
2737 | while (super->devlist) { | |
2738 | dv = super->devlist->next; | |
2739 | free(super->devlist->dev); | |
2740 | free(super->devlist); | |
2741 | super->devlist = dv; | |
2742 | } | |
2743 | } | |
2744 | ||
2745 | static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src) | |
2746 | { | |
2747 | memcpy(dest, src, sizeof_imsm_dev(src, 0)); | |
2748 | } | |
2749 | ||
2750 | static int compare_super_imsm(struct supertype *st, struct supertype *tst) | |
2751 | { | |
2752 | /* | |
2753 | * return: | |
2754 | * 0 same, or first was empty, and second was copied | |
2755 | * 1 second had wrong number | |
2756 | * 2 wrong uuid | |
2757 | * 3 wrong other info | |
2758 | */ | |
2759 | struct intel_super *first = st->sb; | |
2760 | struct intel_super *sec = tst->sb; | |
2761 | ||
2762 | if (!first) { | |
2763 | st->sb = tst->sb; | |
2764 | tst->sb = NULL; | |
2765 | return 0; | |
2766 | } | |
2767 | /* in platform dependent environment test if the disks | |
2768 | * use the same Intel hba | |
2769 | */ | |
2770 | if (!check_env("IMSM_NO_PLATFORM")) { | |
2771 | if (!first->hba || !sec->hba || | |
2772 | (first->hba->type != sec->hba->type)) { | |
2773 | fprintf(stderr, | |
2774 | "HBAs of devices does not match %s != %s\n", | |
2775 | first->hba ? get_sys_dev_type(first->hba->type) : NULL, | |
2776 | sec->hba ? get_sys_dev_type(sec->hba->type) : NULL); | |
2777 | return 3; | |
2778 | } | |
2779 | } | |
2780 | ||
2781 | /* if an anchor does not have num_raid_devs set then it is a free | |
2782 | * floating spare | |
2783 | */ | |
2784 | if (first->anchor->num_raid_devs > 0 && | |
2785 | sec->anchor->num_raid_devs > 0) { | |
2786 | /* Determine if these disks might ever have been | |
2787 | * related. Further disambiguation can only take place | |
2788 | * in load_super_imsm_all | |
2789 | */ | |
2790 | __u32 first_family = first->anchor->orig_family_num; | |
2791 | __u32 sec_family = sec->anchor->orig_family_num; | |
2792 | ||
2793 | if (memcmp(first->anchor->sig, sec->anchor->sig, | |
2794 | MAX_SIGNATURE_LENGTH) != 0) | |
2795 | return 3; | |
2796 | ||
2797 | if (first_family == 0) | |
2798 | first_family = first->anchor->family_num; | |
2799 | if (sec_family == 0) | |
2800 | sec_family = sec->anchor->family_num; | |
2801 | ||
2802 | if (first_family != sec_family) | |
2803 | return 3; | |
2804 | ||
2805 | } | |
2806 | ||
2807 | ||
2808 | /* if 'first' is a spare promote it to a populated mpb with sec's | |
2809 | * family number | |
2810 | */ | |
2811 | if (first->anchor->num_raid_devs == 0 && | |
2812 | sec->anchor->num_raid_devs > 0) { | |
2813 | int i; | |
2814 | struct intel_dev *dv; | |
2815 | struct imsm_dev *dev; | |
2816 | ||
2817 | /* we need to copy raid device info from sec if an allocation | |
2818 | * fails here we don't associate the spare | |
2819 | */ | |
2820 | for (i = 0; i < sec->anchor->num_raid_devs; i++) { | |
2821 | dv = malloc(sizeof(*dv)); | |
2822 | if (!dv) | |
2823 | break; | |
2824 | dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1)); | |
2825 | if (!dev) { | |
2826 | free(dv); | |
2827 | break; | |
2828 | } | |
2829 | dv->dev = dev; | |
2830 | dv->index = i; | |
2831 | dv->next = first->devlist; | |
2832 | first->devlist = dv; | |
2833 | } | |
2834 | if (i < sec->anchor->num_raid_devs) { | |
2835 | /* allocation failure */ | |
2836 | free_devlist(first); | |
2837 | fprintf(stderr, "imsm: failed to associate spare\n"); | |
2838 | return 3; | |
2839 | } | |
2840 | first->anchor->num_raid_devs = sec->anchor->num_raid_devs; | |
2841 | first->anchor->orig_family_num = sec->anchor->orig_family_num; | |
2842 | first->anchor->family_num = sec->anchor->family_num; | |
2843 | memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH); | |
2844 | for (i = 0; i < sec->anchor->num_raid_devs; i++) | |
2845 | imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i)); | |
2846 | } | |
2847 | ||
2848 | return 0; | |
2849 | } | |
2850 | ||
2851 | static void fd2devname(int fd, char *name) | |
2852 | { | |
2853 | struct stat st; | |
2854 | char path[256]; | |
2855 | char dname[PATH_MAX]; | |
2856 | char *nm; | |
2857 | int rv; | |
2858 | ||
2859 | name[0] = '\0'; | |
2860 | if (fstat(fd, &st) != 0) | |
2861 | return; | |
2862 | sprintf(path, "/sys/dev/block/%d:%d", | |
2863 | major(st.st_rdev), minor(st.st_rdev)); | |
2864 | ||
2865 | rv = readlink(path, dname, sizeof(dname)-1); | |
2866 | if (rv <= 0) | |
2867 | return; | |
2868 | ||
2869 | dname[rv] = '\0'; | |
2870 | nm = strrchr(dname, '/'); | |
2871 | if (nm) { | |
2872 | nm++; | |
2873 | snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm); | |
2874 | } | |
2875 | } | |
2876 | ||
2877 | extern int scsi_get_serial(int fd, void *buf, size_t buf_len); | |
2878 | ||
2879 | static int imsm_read_serial(int fd, char *devname, | |
2880 | __u8 serial[MAX_RAID_SERIAL_LEN]) | |
2881 | { | |
2882 | unsigned char scsi_serial[255]; | |
2883 | int rv; | |
2884 | int rsp_len; | |
2885 | int len; | |
2886 | char *dest; | |
2887 | char *src; | |
2888 | char *rsp_buf; | |
2889 | int i; | |
2890 | ||
2891 | memset(scsi_serial, 0, sizeof(scsi_serial)); | |
2892 | ||
2893 | rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial)); | |
2894 | ||
2895 | if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) { | |
2896 | memset(serial, 0, MAX_RAID_SERIAL_LEN); | |
2897 | fd2devname(fd, (char *) serial); | |
2898 | return 0; | |
2899 | } | |
2900 | ||
2901 | if (rv != 0) { | |
2902 | if (devname) | |
2903 | fprintf(stderr, | |
2904 | Name ": Failed to retrieve serial for %s\n", | |
2905 | devname); | |
2906 | return rv; | |
2907 | } | |
2908 | ||
2909 | rsp_len = scsi_serial[3]; | |
2910 | if (!rsp_len) { | |
2911 | if (devname) | |
2912 | fprintf(stderr, | |
2913 | Name ": Failed to retrieve serial for %s\n", | |
2914 | devname); | |
2915 | return 2; | |
2916 | } | |
2917 | rsp_buf = (char *) &scsi_serial[4]; | |
2918 | ||
2919 | /* trim all whitespace and non-printable characters and convert | |
2920 | * ':' to ';' | |
2921 | */ | |
2922 | for (i = 0, dest = rsp_buf; i < rsp_len; i++) { | |
2923 | src = &rsp_buf[i]; | |
2924 | if (*src > 0x20) { | |
2925 | /* ':' is reserved for use in placeholder serial | |
2926 | * numbers for missing disks | |
2927 | */ | |
2928 | if (*src == ':') | |
2929 | *dest++ = ';'; | |
2930 | else | |
2931 | *dest++ = *src; | |
2932 | } | |
2933 | } | |
2934 | len = dest - rsp_buf; | |
2935 | dest = rsp_buf; | |
2936 | ||
2937 | /* truncate leading characters */ | |
2938 | if (len > MAX_RAID_SERIAL_LEN) { | |
2939 | dest += len - MAX_RAID_SERIAL_LEN; | |
2940 | len = MAX_RAID_SERIAL_LEN; | |
2941 | } | |
2942 | ||
2943 | memset(serial, 0, MAX_RAID_SERIAL_LEN); | |
2944 | memcpy(serial, dest, len); | |
2945 | ||
2946 | return 0; | |
2947 | } | |
2948 | ||
2949 | static int serialcmp(__u8 *s1, __u8 *s2) | |
2950 | { | |
2951 | return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN); | |
2952 | } | |
2953 | ||
2954 | static void serialcpy(__u8 *dest, __u8 *src) | |
2955 | { | |
2956 | strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN); | |
2957 | } | |
2958 | ||
2959 | static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super) | |
2960 | { | |
2961 | struct dl *dl; | |
2962 | ||
2963 | for (dl = super->disks; dl; dl = dl->next) | |
2964 | if (serialcmp(dl->serial, serial) == 0) | |
2965 | break; | |
2966 | ||
2967 | return dl; | |
2968 | } | |
2969 | ||
2970 | static struct imsm_disk * | |
2971 | __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx) | |
2972 | { | |
2973 | int i; | |
2974 | ||
2975 | for (i = 0; i < mpb->num_disks; i++) { | |
2976 | struct imsm_disk *disk = __get_imsm_disk(mpb, i); | |
2977 | ||
2978 | if (serialcmp(disk->serial, serial) == 0) { | |
2979 | if (idx) | |
2980 | *idx = i; | |
2981 | return disk; | |
2982 | } | |
2983 | } | |
2984 | ||
2985 | return NULL; | |
2986 | } | |
2987 | ||
2988 | static int | |
2989 | load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd) | |
2990 | { | |
2991 | struct imsm_disk *disk; | |
2992 | struct dl *dl; | |
2993 | struct stat stb; | |
2994 | int rv; | |
2995 | char name[40]; | |
2996 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
2997 | ||
2998 | rv = imsm_read_serial(fd, devname, serial); | |
2999 | ||
3000 | if (rv != 0) | |
3001 | return 2; | |
3002 | ||
3003 | dl = calloc(1, sizeof(*dl)); | |
3004 | if (!dl) { | |
3005 | if (devname) | |
3006 | fprintf(stderr, | |
3007 | Name ": failed to allocate disk buffer for %s\n", | |
3008 | devname); | |
3009 | return 2; | |
3010 | } | |
3011 | ||
3012 | fstat(fd, &stb); | |
3013 | dl->major = major(stb.st_rdev); | |
3014 | dl->minor = minor(stb.st_rdev); | |
3015 | dl->next = super->disks; | |
3016 | dl->fd = keep_fd ? fd : -1; | |
3017 | assert(super->disks == NULL); | |
3018 | super->disks = dl; | |
3019 | serialcpy(dl->serial, serial); | |
3020 | dl->index = -2; | |
3021 | dl->e = NULL; | |
3022 | fd2devname(fd, name); | |
3023 | if (devname) | |
3024 | dl->devname = strdup(devname); | |
3025 | else | |
3026 | dl->devname = strdup(name); | |
3027 | ||
3028 | /* look up this disk's index in the current anchor */ | |
3029 | disk = __serial_to_disk(dl->serial, super->anchor, &dl->index); | |
3030 | if (disk) { | |
3031 | dl->disk = *disk; | |
3032 | /* only set index on disks that are a member of a | |
3033 | * populated contianer, i.e. one with raid_devs | |
3034 | */ | |
3035 | if (is_failed(&dl->disk)) | |
3036 | dl->index = -2; | |
3037 | else if (is_spare(&dl->disk)) | |
3038 | dl->index = -1; | |
3039 | } | |
3040 | ||
3041 | return 0; | |
3042 | } | |
3043 | ||
3044 | #ifndef MDASSEMBLE | |
3045 | /* When migrating map0 contains the 'destination' state while map1 | |
3046 | * contains the current state. When not migrating map0 contains the | |
3047 | * current state. This routine assumes that map[0].map_state is set to | |
3048 | * the current array state before being called. | |
3049 | * | |
3050 | * Migration is indicated by one of the following states | |
3051 | * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed) | |
3052 | * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal | |
3053 | * map1state=unitialized) | |
3054 | * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal | |
3055 | * map1state=normal) | |
3056 | * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal | |
3057 | * map1state=degraded) | |
3058 | * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal | |
3059 | * map1state=normal) | |
3060 | */ | |
3061 | static void migrate(struct imsm_dev *dev, struct intel_super *super, | |
3062 | __u8 to_state, int migr_type) | |
3063 | { | |
3064 | struct imsm_map *dest; | |
3065 | struct imsm_map *src = get_imsm_map(dev, 0); | |
3066 | ||
3067 | dev->vol.migr_state = 1; | |
3068 | set_migr_type(dev, migr_type); | |
3069 | dev->vol.curr_migr_unit = 0; | |
3070 | dest = get_imsm_map(dev, 1); | |
3071 | ||
3072 | /* duplicate and then set the target end state in map[0] */ | |
3073 | memcpy(dest, src, sizeof_imsm_map(src)); | |
3074 | if ((migr_type == MIGR_REBUILD) || | |
3075 | (migr_type == MIGR_GEN_MIGR)) { | |
3076 | __u32 ord; | |
3077 | int i; | |
3078 | ||
3079 | for (i = 0; i < src->num_members; i++) { | |
3080 | ord = __le32_to_cpu(src->disk_ord_tbl[i]); | |
3081 | set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord)); | |
3082 | } | |
3083 | } | |
3084 | ||
3085 | if (migr_type == MIGR_GEN_MIGR) | |
3086 | /* Clear migration record */ | |
3087 | memset(super->migr_rec, 0, sizeof(struct migr_record)); | |
3088 | ||
3089 | src->map_state = to_state; | |
3090 | } | |
3091 | ||
3092 | static void end_migration(struct imsm_dev *dev, __u8 map_state) | |
3093 | { | |
3094 | struct imsm_map *map = get_imsm_map(dev, 0); | |
3095 | struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state); | |
3096 | int i, j; | |
3097 | ||
3098 | /* merge any IMSM_ORD_REBUILD bits that were not successfully | |
3099 | * completed in the last migration. | |
3100 | * | |
3101 | * FIXME add support for raid-level-migration | |
3102 | */ | |
3103 | for (i = 0; i < prev->num_members; i++) | |
3104 | for (j = 0; j < map->num_members; j++) | |
3105 | /* during online capacity expansion | |
3106 | * disks position can be changed if takeover is used | |
3107 | */ | |
3108 | if (ord_to_idx(map->disk_ord_tbl[j]) == | |
3109 | ord_to_idx(prev->disk_ord_tbl[i])) { | |
3110 | map->disk_ord_tbl[j] |= prev->disk_ord_tbl[i]; | |
3111 | break; | |
3112 | } | |
3113 | ||
3114 | dev->vol.migr_state = 0; | |
3115 | set_migr_type(dev, 0); | |
3116 | dev->vol.curr_migr_unit = 0; | |
3117 | map->map_state = map_state; | |
3118 | } | |
3119 | #endif | |
3120 | ||
3121 | static int parse_raid_devices(struct intel_super *super) | |
3122 | { | |
3123 | int i; | |
3124 | struct imsm_dev *dev_new; | |
3125 | size_t len, len_migr; | |
3126 | size_t max_len = 0; | |
3127 | size_t space_needed = 0; | |
3128 | struct imsm_super *mpb = super->anchor; | |
3129 | ||
3130 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
3131 | struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i); | |
3132 | struct intel_dev *dv; | |
3133 | ||
3134 | len = sizeof_imsm_dev(dev_iter, 0); | |
3135 | len_migr = sizeof_imsm_dev(dev_iter, 1); | |
3136 | if (len_migr > len) | |
3137 | space_needed += len_migr - len; | |
3138 | ||
3139 | dv = malloc(sizeof(*dv)); | |
3140 | if (!dv) | |
3141 | return 1; | |
3142 | if (max_len < len_migr) | |
3143 | max_len = len_migr; | |
3144 | if (max_len > len_migr) | |
3145 | space_needed += max_len - len_migr; | |
3146 | dev_new = malloc(max_len); | |
3147 | if (!dev_new) { | |
3148 | free(dv); | |
3149 | return 1; | |
3150 | } | |
3151 | imsm_copy_dev(dev_new, dev_iter); | |
3152 | dv->dev = dev_new; | |
3153 | dv->index = i; | |
3154 | dv->next = super->devlist; | |
3155 | super->devlist = dv; | |
3156 | } | |
3157 | ||
3158 | /* ensure that super->buf is large enough when all raid devices | |
3159 | * are migrating | |
3160 | */ | |
3161 | if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) { | |
3162 | void *buf; | |
3163 | ||
3164 | len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512); | |
3165 | if (posix_memalign(&buf, 512, len) != 0) | |
3166 | return 1; | |
3167 | ||
3168 | memcpy(buf, super->buf, super->len); | |
3169 | memset(buf + super->len, 0, len - super->len); | |
3170 | free(super->buf); | |
3171 | super->buf = buf; | |
3172 | super->len = len; | |
3173 | } | |
3174 | ||
3175 | return 0; | |
3176 | } | |
3177 | ||
3178 | /* retrieve a pointer to the bbm log which starts after all raid devices */ | |
3179 | struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb) | |
3180 | { | |
3181 | void *ptr = NULL; | |
3182 | ||
3183 | if (__le32_to_cpu(mpb->bbm_log_size)) { | |
3184 | ptr = mpb; | |
3185 | ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size); | |
3186 | } | |
3187 | ||
3188 | return ptr; | |
3189 | } | |
3190 | ||
3191 | /******************************************************************************* | |
3192 | * Function: check_mpb_migr_compatibility | |
3193 | * Description: Function checks for unsupported migration features: | |
3194 | * - migration optimization area (pba_of_lba0) | |
3195 | * - descending reshape (ascending_migr) | |
3196 | * Parameters: | |
3197 | * super : imsm metadata information | |
3198 | * Returns: | |
3199 | * 0 : migration is compatible | |
3200 | * -1 : migration is not compatible | |
3201 | ******************************************************************************/ | |
3202 | int check_mpb_migr_compatibility(struct intel_super *super) | |
3203 | { | |
3204 | struct imsm_map *map0, *map1; | |
3205 | struct migr_record *migr_rec = super->migr_rec; | |
3206 | int i; | |
3207 | ||
3208 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
3209 | struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i); | |
3210 | ||
3211 | if (dev_iter && | |
3212 | dev_iter->vol.migr_state == 1 && | |
3213 | dev_iter->vol.migr_type == MIGR_GEN_MIGR) { | |
3214 | /* This device is migrating */ | |
3215 | map0 = get_imsm_map(dev_iter, 0); | |
3216 | map1 = get_imsm_map(dev_iter, 1); | |
3217 | if (map0->pba_of_lba0 != map1->pba_of_lba0) | |
3218 | /* migration optimization area was used */ | |
3219 | return -1; | |
3220 | if (migr_rec->ascending_migr == 0 | |
3221 | && migr_rec->dest_depth_per_unit > 0) | |
3222 | /* descending reshape not supported yet */ | |
3223 | return -1; | |
3224 | } | |
3225 | } | |
3226 | return 0; | |
3227 | } | |
3228 | ||
3229 | static void __free_imsm(struct intel_super *super, int free_disks); | |
3230 | ||
3231 | /* load_imsm_mpb - read matrix metadata | |
3232 | * allocates super->mpb to be freed by free_imsm | |
3233 | */ | |
3234 | static int load_imsm_mpb(int fd, struct intel_super *super, char *devname) | |
3235 | { | |
3236 | unsigned long long dsize; | |
3237 | unsigned long long sectors; | |
3238 | struct stat; | |
3239 | struct imsm_super *anchor; | |
3240 | __u32 check_sum; | |
3241 | ||
3242 | get_dev_size(fd, NULL, &dsize); | |
3243 | if (dsize < 1024) { | |
3244 | if (devname) | |
3245 | fprintf(stderr, | |
3246 | Name ": %s: device to small for imsm\n", | |
3247 | devname); | |
3248 | return 1; | |
3249 | } | |
3250 | ||
3251 | if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) { | |
3252 | if (devname) | |
3253 | fprintf(stderr, Name | |
3254 | ": Cannot seek to anchor block on %s: %s\n", | |
3255 | devname, strerror(errno)); | |
3256 | return 1; | |
3257 | } | |
3258 | ||
3259 | if (posix_memalign((void**)&anchor, 512, 512) != 0) { | |
3260 | if (devname) | |
3261 | fprintf(stderr, | |
3262 | Name ": Failed to allocate imsm anchor buffer" | |
3263 | " on %s\n", devname); | |
3264 | return 1; | |
3265 | } | |
3266 | if (read(fd, anchor, 512) != 512) { | |
3267 | if (devname) | |
3268 | fprintf(stderr, | |
3269 | Name ": Cannot read anchor block on %s: %s\n", | |
3270 | devname, strerror(errno)); | |
3271 | free(anchor); | |
3272 | return 1; | |
3273 | } | |
3274 | ||
3275 | if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) { | |
3276 | if (devname) | |
3277 | fprintf(stderr, | |
3278 | Name ": no IMSM anchor on %s\n", devname); | |
3279 | free(anchor); | |
3280 | return 2; | |
3281 | } | |
3282 | ||
3283 | __free_imsm(super, 0); | |
3284 | /* reload capability and hba */ | |
3285 | ||
3286 | /* capability and hba must be updated with new super allocation */ | |
3287 | find_intel_hba_capability(fd, super, devname); | |
3288 | super->len = ROUND_UP(anchor->mpb_size, 512); | |
3289 | if (posix_memalign(&super->buf, 512, super->len) != 0) { | |
3290 | if (devname) | |
3291 | fprintf(stderr, | |
3292 | Name ": unable to allocate %zu byte mpb buffer\n", | |
3293 | super->len); | |
3294 | free(anchor); | |
3295 | return 2; | |
3296 | } | |
3297 | memcpy(super->buf, anchor, 512); | |
3298 | ||
3299 | sectors = mpb_sectors(anchor) - 1; | |
3300 | free(anchor); | |
3301 | ||
3302 | if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) { | |
3303 | fprintf(stderr, Name | |
3304 | ": %s could not allocate migr_rec buffer\n", __func__); | |
3305 | free(super->buf); | |
3306 | return 2; | |
3307 | } | |
3308 | ||
3309 | if (!sectors) { | |
3310 | check_sum = __gen_imsm_checksum(super->anchor); | |
3311 | if (check_sum != __le32_to_cpu(super->anchor->check_sum)) { | |
3312 | if (devname) | |
3313 | fprintf(stderr, | |
3314 | Name ": IMSM checksum %x != %x on %s\n", | |
3315 | check_sum, | |
3316 | __le32_to_cpu(super->anchor->check_sum), | |
3317 | devname); | |
3318 | return 2; | |
3319 | } | |
3320 | ||
3321 | return 0; | |
3322 | } | |
3323 | ||
3324 | /* read the extended mpb */ | |
3325 | if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) { | |
3326 | if (devname) | |
3327 | fprintf(stderr, | |
3328 | Name ": Cannot seek to extended mpb on %s: %s\n", | |
3329 | devname, strerror(errno)); | |
3330 | return 1; | |
3331 | } | |
3332 | ||
3333 | if ((unsigned)read(fd, super->buf + 512, super->len - 512) != super->len - 512) { | |
3334 | if (devname) | |
3335 | fprintf(stderr, | |
3336 | Name ": Cannot read extended mpb on %s: %s\n", | |
3337 | devname, strerror(errno)); | |
3338 | return 2; | |
3339 | } | |
3340 | ||
3341 | check_sum = __gen_imsm_checksum(super->anchor); | |
3342 | if (check_sum != __le32_to_cpu(super->anchor->check_sum)) { | |
3343 | if (devname) | |
3344 | fprintf(stderr, | |
3345 | Name ": IMSM checksum %x != %x on %s\n", | |
3346 | check_sum, __le32_to_cpu(super->anchor->check_sum), | |
3347 | devname); | |
3348 | return 3; | |
3349 | } | |
3350 | ||
3351 | /* FIXME the BBM log is disk specific so we cannot use this global | |
3352 | * buffer for all disks. Ok for now since we only look at the global | |
3353 | * bbm_log_size parameter to gate assembly | |
3354 | */ | |
3355 | super->bbm_log = __get_imsm_bbm_log(super->anchor); | |
3356 | ||
3357 | return 0; | |
3358 | } | |
3359 | ||
3360 | static int read_imsm_migr_rec(int fd, struct intel_super *super); | |
3361 | ||
3362 | static int | |
3363 | load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd) | |
3364 | { | |
3365 | int err; | |
3366 | ||
3367 | err = load_imsm_mpb(fd, super, devname); | |
3368 | if (err) | |
3369 | return err; | |
3370 | err = load_imsm_disk(fd, super, devname, keep_fd); | |
3371 | if (err) | |
3372 | return err; | |
3373 | err = parse_raid_devices(super); | |
3374 | ||
3375 | return err; | |
3376 | } | |
3377 | ||
3378 | static void __free_imsm_disk(struct dl *d) | |
3379 | { | |
3380 | if (d->fd >= 0) | |
3381 | close(d->fd); | |
3382 | if (d->devname) | |
3383 | free(d->devname); | |
3384 | if (d->e) | |
3385 | free(d->e); | |
3386 | free(d); | |
3387 | ||
3388 | } | |
3389 | ||
3390 | static void free_imsm_disks(struct intel_super *super) | |
3391 | { | |
3392 | struct dl *d; | |
3393 | ||
3394 | while (super->disks) { | |
3395 | d = super->disks; | |
3396 | super->disks = d->next; | |
3397 | __free_imsm_disk(d); | |
3398 | } | |
3399 | while (super->disk_mgmt_list) { | |
3400 | d = super->disk_mgmt_list; | |
3401 | super->disk_mgmt_list = d->next; | |
3402 | __free_imsm_disk(d); | |
3403 | } | |
3404 | while (super->missing) { | |
3405 | d = super->missing; | |
3406 | super->missing = d->next; | |
3407 | __free_imsm_disk(d); | |
3408 | } | |
3409 | ||
3410 | } | |
3411 | ||
3412 | /* free all the pieces hanging off of a super pointer */ | |
3413 | static void __free_imsm(struct intel_super *super, int free_disks) | |
3414 | { | |
3415 | struct intel_hba *elem, *next; | |
3416 | ||
3417 | if (super->buf) { | |
3418 | free(super->buf); | |
3419 | super->buf = NULL; | |
3420 | } | |
3421 | /* unlink capability description */ | |
3422 | super->orom = NULL; | |
3423 | if (super->migr_rec_buf) { | |
3424 | free(super->migr_rec_buf); | |
3425 | super->migr_rec_buf = NULL; | |
3426 | } | |
3427 | if (free_disks) | |
3428 | free_imsm_disks(super); | |
3429 | free_devlist(super); | |
3430 | elem = super->hba; | |
3431 | while (elem) { | |
3432 | if (elem->path) | |
3433 | free((void *)elem->path); | |
3434 | next = elem->next; | |
3435 | free(elem); | |
3436 | elem = next; | |
3437 | } | |
3438 | super->hba = NULL; | |
3439 | } | |
3440 | ||
3441 | static void free_imsm(struct intel_super *super) | |
3442 | { | |
3443 | __free_imsm(super, 1); | |
3444 | free(super); | |
3445 | } | |
3446 | ||
3447 | static void free_super_imsm(struct supertype *st) | |
3448 | { | |
3449 | struct intel_super *super = st->sb; | |
3450 | ||
3451 | if (!super) | |
3452 | return; | |
3453 | ||
3454 | free_imsm(super); | |
3455 | st->sb = NULL; | |
3456 | } | |
3457 | ||
3458 | static struct intel_super *alloc_super(void) | |
3459 | { | |
3460 | struct intel_super *super = malloc(sizeof(*super)); | |
3461 | ||
3462 | if (super) { | |
3463 | memset(super, 0, sizeof(*super)); | |
3464 | super->current_vol = -1; | |
3465 | super->create_offset = ~((__u32 ) 0); | |
3466 | } | |
3467 | return super; | |
3468 | } | |
3469 | ||
3470 | /* | |
3471 | * find and allocate hba and OROM/EFI based on valid fd of RAID component device | |
3472 | */ | |
3473 | static int find_intel_hba_capability(int fd, struct intel_super *super, char *devname) | |
3474 | { | |
3475 | struct sys_dev *hba_name; | |
3476 | int rv = 0; | |
3477 | ||
3478 | if ((fd < 0) || check_env("IMSM_NO_PLATFORM")) { | |
3479 | super->orom = NULL; | |
3480 | super->hba = NULL; | |
3481 | return 0; | |
3482 | } | |
3483 | hba_name = find_disk_attached_hba(fd, NULL); | |
3484 | if (!hba_name) { | |
3485 | if (devname) | |
3486 | fprintf(stderr, | |
3487 | Name ": %s is not attached to Intel(R) RAID controller.\n", | |
3488 | devname); | |
3489 | return 1; | |
3490 | } | |
3491 | rv = attach_hba_to_super(super, hba_name); | |
3492 | if (rv == 2) { | |
3493 | if (devname) { | |
3494 | struct intel_hba *hba = super->hba; | |
3495 | ||
3496 | fprintf(stderr, Name ": %s is attached to Intel(R) %s RAID " | |
3497 | "controller (%s),\n" | |
3498 | " but the container is assigned to Intel(R) " | |
3499 | "%s RAID controller (", | |
3500 | devname, | |
3501 | hba_name->path, | |
3502 | hba_name->pci_id ? : "Err!", | |
3503 | get_sys_dev_type(hba_name->type)); | |
3504 | ||
3505 | while (hba) { | |
3506 | fprintf(stderr, "%s", hba->pci_id ? : "Err!"); | |
3507 | if (hba->next) | |
3508 | fprintf(stderr, ", "); | |
3509 | hba = hba->next; | |
3510 | } | |
3511 | ||
3512 | fprintf(stderr, ").\n" | |
3513 | " Mixing devices attached to different controllers " | |
3514 | "is not allowed.\n"); | |
3515 | } | |
3516 | free_sys_dev(&hba_name); | |
3517 | return 2; | |
3518 | } | |
3519 | super->orom = find_imsm_capability(hba_name->type); | |
3520 | free_sys_dev(&hba_name); | |
3521 | if (!super->orom) | |
3522 | return 3; | |
3523 | return 0; | |
3524 | } | |
3525 | ||
3526 | /* find_missing - helper routine for load_super_imsm_all that identifies | |
3527 | * disks that have disappeared from the system. This routine relies on | |
3528 | * the mpb being uptodate, which it is at load time. | |
3529 | */ | |
3530 | static int find_missing(struct intel_super *super) | |
3531 | { | |
3532 | int i; | |
3533 | struct imsm_super *mpb = super->anchor; | |
3534 | struct dl *dl; | |
3535 | struct imsm_disk *disk; | |
3536 | ||
3537 | for (i = 0; i < mpb->num_disks; i++) { | |
3538 | disk = __get_imsm_disk(mpb, i); | |
3539 | dl = serial_to_dl(disk->serial, super); | |
3540 | if (dl) | |
3541 | continue; | |
3542 | ||
3543 | dl = malloc(sizeof(*dl)); | |
3544 | if (!dl) | |
3545 | return 1; | |
3546 | dl->major = 0; | |
3547 | dl->minor = 0; | |
3548 | dl->fd = -1; | |
3549 | dl->devname = strdup("missing"); | |
3550 | dl->index = i; | |
3551 | serialcpy(dl->serial, disk->serial); | |
3552 | dl->disk = *disk; | |
3553 | dl->e = NULL; | |
3554 | dl->next = super->missing; | |
3555 | super->missing = dl; | |
3556 | } | |
3557 | ||
3558 | return 0; | |
3559 | } | |
3560 | ||
3561 | #ifndef MDASSEMBLE | |
3562 | static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list) | |
3563 | { | |
3564 | struct intel_disk *idisk = disk_list; | |
3565 | ||
3566 | while (idisk) { | |
3567 | if (serialcmp(idisk->disk.serial, serial) == 0) | |
3568 | break; | |
3569 | idisk = idisk->next; | |
3570 | } | |
3571 | ||
3572 | return idisk; | |
3573 | } | |
3574 | ||
3575 | static int __prep_thunderdome(struct intel_super **table, int tbl_size, | |
3576 | struct intel_super *super, | |
3577 | struct intel_disk **disk_list) | |
3578 | { | |
3579 | struct imsm_disk *d = &super->disks->disk; | |
3580 | struct imsm_super *mpb = super->anchor; | |
3581 | int i, j; | |
3582 | ||
3583 | for (i = 0; i < tbl_size; i++) { | |
3584 | struct imsm_super *tbl_mpb = table[i]->anchor; | |
3585 | struct imsm_disk *tbl_d = &table[i]->disks->disk; | |
3586 | ||
3587 | if (tbl_mpb->family_num == mpb->family_num) { | |
3588 | if (tbl_mpb->check_sum == mpb->check_sum) { | |
3589 | dprintf("%s: mpb from %d:%d matches %d:%d\n", | |
3590 | __func__, super->disks->major, | |
3591 | super->disks->minor, | |
3592 | table[i]->disks->major, | |
3593 | table[i]->disks->minor); | |
3594 | break; | |
3595 | } | |
3596 | ||
3597 | if (((is_configured(d) && !is_configured(tbl_d)) || | |
3598 | is_configured(d) == is_configured(tbl_d)) && | |
3599 | tbl_mpb->generation_num < mpb->generation_num) { | |
3600 | /* current version of the mpb is a | |
3601 | * better candidate than the one in | |
3602 | * super_table, but copy over "cross | |
3603 | * generational" status | |
3604 | */ | |
3605 | struct intel_disk *idisk; | |
3606 | ||
3607 | dprintf("%s: mpb from %d:%d replaces %d:%d\n", | |
3608 | __func__, super->disks->major, | |
3609 | super->disks->minor, | |
3610 | table[i]->disks->major, | |
3611 | table[i]->disks->minor); | |
3612 | ||
3613 | idisk = disk_list_get(tbl_d->serial, *disk_list); | |
3614 | if (idisk && is_failed(&idisk->disk)) | |
3615 | tbl_d->status |= FAILED_DISK; | |
3616 | break; | |
3617 | } else { | |
3618 | struct intel_disk *idisk; | |
3619 | struct imsm_disk *disk; | |
3620 | ||
3621 | /* tbl_mpb is more up to date, but copy | |
3622 | * over cross generational status before | |
3623 | * returning | |
3624 | */ | |
3625 | disk = __serial_to_disk(d->serial, mpb, NULL); | |
3626 | if (disk && is_failed(disk)) | |
3627 | d->status |= FAILED_DISK; | |
3628 | ||
3629 | idisk = disk_list_get(d->serial, *disk_list); | |
3630 | if (idisk) { | |
3631 | idisk->owner = i; | |
3632 | if (disk && is_configured(disk)) | |
3633 | idisk->disk.status |= CONFIGURED_DISK; | |
3634 | } | |
3635 | ||
3636 | dprintf("%s: mpb from %d:%d prefer %d:%d\n", | |
3637 | __func__, super->disks->major, | |
3638 | super->disks->minor, | |
3639 | table[i]->disks->major, | |
3640 | table[i]->disks->minor); | |
3641 | ||
3642 | return tbl_size; | |
3643 | } | |
3644 | } | |
3645 | } | |
3646 | ||
3647 | if (i >= tbl_size) | |
3648 | table[tbl_size++] = super; | |
3649 | else | |
3650 | table[i] = super; | |
3651 | ||
3652 | /* update/extend the merged list of imsm_disk records */ | |
3653 | for (j = 0; j < mpb->num_disks; j++) { | |
3654 | struct imsm_disk *disk = __get_imsm_disk(mpb, j); | |
3655 | struct intel_disk *idisk; | |
3656 | ||
3657 | idisk = disk_list_get(disk->serial, *disk_list); | |
3658 | if (idisk) { | |
3659 | idisk->disk.status |= disk->status; | |
3660 | if (is_configured(&idisk->disk) || | |
3661 | is_failed(&idisk->disk)) | |
3662 | idisk->disk.status &= ~(SPARE_DISK); | |
3663 | } else { | |
3664 | idisk = calloc(1, sizeof(*idisk)); | |
3665 | if (!idisk) | |
3666 | return -1; | |
3667 | idisk->owner = IMSM_UNKNOWN_OWNER; | |
3668 | idisk->disk = *disk; | |
3669 | idisk->next = *disk_list; | |
3670 | *disk_list = idisk; | |
3671 | } | |
3672 | ||
3673 | if (serialcmp(idisk->disk.serial, d->serial) == 0) | |
3674 | idisk->owner = i; | |
3675 | } | |
3676 | ||
3677 | return tbl_size; | |
3678 | } | |
3679 | ||
3680 | static struct intel_super * | |
3681 | validate_members(struct intel_super *super, struct intel_disk *disk_list, | |
3682 | const int owner) | |
3683 | { | |
3684 | struct imsm_super *mpb = super->anchor; | |
3685 | int ok_count = 0; | |
3686 | int i; | |
3687 | ||
3688 | for (i = 0; i < mpb->num_disks; i++) { | |
3689 | struct imsm_disk *disk = __get_imsm_disk(mpb, i); | |
3690 | struct intel_disk *idisk; | |
3691 | ||
3692 | idisk = disk_list_get(disk->serial, disk_list); | |
3693 | if (idisk) { | |
3694 | if (idisk->owner == owner || | |
3695 | idisk->owner == IMSM_UNKNOWN_OWNER) | |
3696 | ok_count++; | |
3697 | else | |
3698 | dprintf("%s: '%.16s' owner %d != %d\n", | |
3699 | __func__, disk->serial, idisk->owner, | |
3700 | owner); | |
3701 | } else { | |
3702 | dprintf("%s: unknown disk %x [%d]: %.16s\n", | |
3703 | __func__, __le32_to_cpu(mpb->family_num), i, | |
3704 | disk->serial); | |
3705 | break; | |
3706 | } | |
3707 | } | |
3708 | ||
3709 | if (ok_count == mpb->num_disks) | |
3710 | return super; | |
3711 | return NULL; | |
3712 | } | |
3713 | ||
3714 | static void show_conflicts(__u32 family_num, struct intel_super *super_list) | |
3715 | { | |
3716 | struct intel_super *s; | |
3717 | ||
3718 | for (s = super_list; s; s = s->next) { | |
3719 | if (family_num != s->anchor->family_num) | |
3720 | continue; | |
3721 | fprintf(stderr, "Conflict, offlining family %#x on '%s'\n", | |
3722 | __le32_to_cpu(family_num), s->disks->devname); | |
3723 | } | |
3724 | } | |
3725 | ||
3726 | static struct intel_super * | |
3727 | imsm_thunderdome(struct intel_super **super_list, int len) | |
3728 | { | |
3729 | struct intel_super *super_table[len]; | |
3730 | struct intel_disk *disk_list = NULL; | |
3731 | struct intel_super *champion, *spare; | |
3732 | struct intel_super *s, **del; | |
3733 | int tbl_size = 0; | |
3734 | int conflict; | |
3735 | int i; | |
3736 | ||
3737 | memset(super_table, 0, sizeof(super_table)); | |
3738 | for (s = *super_list; s; s = s->next) | |
3739 | tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list); | |
3740 | ||
3741 | for (i = 0; i < tbl_size; i++) { | |
3742 | struct imsm_disk *d; | |
3743 | struct intel_disk *idisk; | |
3744 | struct imsm_super *mpb = super_table[i]->anchor; | |
3745 | ||
3746 | s = super_table[i]; | |
3747 | d = &s->disks->disk; | |
3748 | ||
3749 | /* 'd' must appear in merged disk list for its | |
3750 | * configuration to be valid | |
3751 | */ | |
3752 | idisk = disk_list_get(d->serial, disk_list); | |
3753 | if (idisk && idisk->owner == i) | |
3754 | s = validate_members(s, disk_list, i); | |
3755 | else | |
3756 | s = NULL; | |
3757 | ||
3758 | if (!s) | |
3759 | dprintf("%s: marking family: %#x from %d:%d offline\n", | |
3760 | __func__, mpb->family_num, | |
3761 | super_table[i]->disks->major, | |
3762 | super_table[i]->disks->minor); | |
3763 | super_table[i] = s; | |
3764 | } | |
3765 | ||
3766 | /* This is where the mdadm implementation differs from the Windows | |
3767 | * driver which has no strict concept of a container. We can only | |
3768 | * assemble one family from a container, so when returning a prodigal | |
3769 | * array member to this system the code will not be able to disambiguate | |
3770 | * the container contents that should be assembled ("foreign" versus | |
3771 | * "local"). It requires user intervention to set the orig_family_num | |
3772 | * to a new value to establish a new container. The Windows driver in | |
3773 | * this situation fixes up the volume name in place and manages the | |
3774 | * foreign array as an independent entity. | |
3775 | */ | |
3776 | s = NULL; | |
3777 | spare = NULL; | |
3778 | conflict = 0; | |
3779 | for (i = 0; i < tbl_size; i++) { | |
3780 | struct intel_super *tbl_ent = super_table[i]; | |
3781 | int is_spare = 0; | |
3782 | ||
3783 | if (!tbl_ent) | |
3784 | continue; | |
3785 | ||
3786 | if (tbl_ent->anchor->num_raid_devs == 0) { | |
3787 | spare = tbl_ent; | |
3788 | is_spare = 1; | |
3789 | } | |
3790 | ||
3791 | if (s && !is_spare) { | |
3792 | show_conflicts(tbl_ent->anchor->family_num, *super_list); | |
3793 | conflict++; | |
3794 | } else if (!s && !is_spare) | |
3795 | s = tbl_ent; | |
3796 | } | |
3797 | ||
3798 | if (!s) | |
3799 | s = spare; | |
3800 | if (!s) { | |
3801 | champion = NULL; | |
3802 | goto out; | |
3803 | } | |
3804 | champion = s; | |
3805 | ||
3806 | if (conflict) | |
3807 | fprintf(stderr, "Chose family %#x on '%s', " | |
3808 | "assemble conflicts to new container with '--update=uuid'\n", | |
3809 | __le32_to_cpu(s->anchor->family_num), s->disks->devname); | |
3810 | ||
3811 | /* collect all dl's onto 'champion', and update them to | |
3812 | * champion's version of the status | |
3813 | */ | |
3814 | for (s = *super_list; s; s = s->next) { | |
3815 | struct imsm_super *mpb = champion->anchor; | |
3816 | struct dl *dl = s->disks; | |
3817 | ||
3818 | if (s == champion) | |
3819 | continue; | |
3820 | ||
3821 | for (i = 0; i < mpb->num_disks; i++) { | |
3822 | struct imsm_disk *disk; | |
3823 | ||
3824 | disk = __serial_to_disk(dl->serial, mpb, &dl->index); | |
3825 | if (disk) { | |
3826 | dl->disk = *disk; | |
3827 | /* only set index on disks that are a member of | |
3828 | * a populated contianer, i.e. one with | |
3829 | * raid_devs | |
3830 | */ | |
3831 | if (is_failed(&dl->disk)) | |
3832 | dl->index = -2; | |
3833 | else if (is_spare(&dl->disk)) | |
3834 | dl->index = -1; | |
3835 | break; | |
3836 | } | |
3837 | } | |
3838 | ||
3839 | if (i >= mpb->num_disks) { | |
3840 | struct intel_disk *idisk; | |
3841 | ||
3842 | idisk = disk_list_get(dl->serial, disk_list); | |
3843 | if (idisk && is_spare(&idisk->disk) && | |
3844 | !is_failed(&idisk->disk) && !is_configured(&idisk->disk)) | |
3845 | dl->index = -1; | |
3846 | else { | |
3847 | dl->index = -2; | |
3848 | continue; | |
3849 | } | |
3850 | } | |
3851 | ||
3852 | dl->next = champion->disks; | |
3853 | champion->disks = dl; | |
3854 | s->disks = NULL; | |
3855 | } | |
3856 | ||
3857 | /* delete 'champion' from super_list */ | |
3858 | for (del = super_list; *del; ) { | |
3859 | if (*del == champion) { | |
3860 | *del = (*del)->next; | |
3861 | break; | |
3862 | } else | |
3863 | del = &(*del)->next; | |
3864 | } | |
3865 | champion->next = NULL; | |
3866 | ||
3867 | out: | |
3868 | while (disk_list) { | |
3869 | struct intel_disk *idisk = disk_list; | |
3870 | ||
3871 | disk_list = disk_list->next; | |
3872 | free(idisk); | |
3873 | } | |
3874 | ||
3875 | return champion; | |
3876 | } | |
3877 | ||
3878 | static int load_super_imsm_all(struct supertype *st, int fd, void **sbp, | |
3879 | char *devname) | |
3880 | { | |
3881 | struct mdinfo *sra; | |
3882 | struct intel_super *super_list = NULL; | |
3883 | struct intel_super *super = NULL; | |
3884 | int devnum = fd2devnum(fd); | |
3885 | struct mdinfo *sd; | |
3886 | int retry; | |
3887 | int err = 0; | |
3888 | int i; | |
3889 | ||
3890 | /* check if 'fd' an opened container */ | |
3891 | sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE); | |
3892 | if (!sra) | |
3893 | return 1; | |
3894 | ||
3895 | if (sra->array.major_version != -1 || | |
3896 | sra->array.minor_version != -2 || | |
3897 | strcmp(sra->text_version, "imsm") != 0) { | |
3898 | err = 1; | |
3899 | goto error; | |
3900 | } | |
3901 | /* load all mpbs */ | |
3902 | for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) { | |
3903 | struct intel_super *s = alloc_super(); | |
3904 | char nm[32]; | |
3905 | int dfd; | |
3906 | int rv; | |
3907 | ||
3908 | err = 1; | |
3909 | if (!s) | |
3910 | goto error; | |
3911 | s->next = super_list; | |
3912 | super_list = s; | |
3913 | ||
3914 | err = 2; | |
3915 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
3916 | dfd = dev_open(nm, O_RDWR); | |
3917 | if (dfd < 0) | |
3918 | goto error; | |
3919 | ||
3920 | rv = find_intel_hba_capability(dfd, s, devname); | |
3921 | /* no orom/efi or non-intel hba of the disk */ | |
3922 | if (rv != 0) | |
3923 | goto error; | |
3924 | ||
3925 | err = load_and_parse_mpb(dfd, s, NULL, 1); | |
3926 | ||
3927 | /* retry the load if we might have raced against mdmon */ | |
3928 | if (err == 3 && mdmon_running(devnum)) | |
3929 | for (retry = 0; retry < 3; retry++) { | |
3930 | usleep(3000); | |
3931 | err = load_and_parse_mpb(dfd, s, NULL, 1); | |
3932 | if (err != 3) | |
3933 | break; | |
3934 | } | |
3935 | if (err) | |
3936 | goto error; | |
3937 | } | |
3938 | ||
3939 | /* all mpbs enter, maybe one leaves */ | |
3940 | super = imsm_thunderdome(&super_list, i); | |
3941 | if (!super) { | |
3942 | err = 1; | |
3943 | goto error; | |
3944 | } | |
3945 | ||
3946 | if (find_missing(super) != 0) { | |
3947 | free_imsm(super); | |
3948 | err = 2; | |
3949 | goto error; | |
3950 | } | |
3951 | ||
3952 | /* load migration record */ | |
3953 | err = load_imsm_migr_rec(super, NULL); | |
3954 | if (err) { | |
3955 | err = 4; | |
3956 | goto error; | |
3957 | } | |
3958 | ||
3959 | /* Check migration compatibility */ | |
3960 | if (check_mpb_migr_compatibility(super) != 0) { | |
3961 | fprintf(stderr, Name ": Unsupported migration detected"); | |
3962 | if (devname) | |
3963 | fprintf(stderr, " on %s\n", devname); | |
3964 | else | |
3965 | fprintf(stderr, " (IMSM).\n"); | |
3966 | ||
3967 | err = 5; | |
3968 | goto error; | |
3969 | } | |
3970 | ||
3971 | err = 0; | |
3972 | ||
3973 | error: | |
3974 | while (super_list) { | |
3975 | struct intel_super *s = super_list; | |
3976 | ||
3977 | super_list = super_list->next; | |
3978 | free_imsm(s); | |
3979 | } | |
3980 | sysfs_free(sra); | |
3981 | ||
3982 | if (err) | |
3983 | return err; | |
3984 | ||
3985 | *sbp = super; | |
3986 | st->container_dev = devnum; | |
3987 | if (err == 0 && st->ss == NULL) { | |
3988 | st->ss = &super_imsm; | |
3989 | st->minor_version = 0; | |
3990 | st->max_devs = IMSM_MAX_DEVICES; | |
3991 | } | |
3992 | return 0; | |
3993 | } | |
3994 | ||
3995 | static int load_container_imsm(struct supertype *st, int fd, char *devname) | |
3996 | { | |
3997 | return load_super_imsm_all(st, fd, &st->sb, devname); | |
3998 | } | |
3999 | #endif | |
4000 | ||
4001 | static int load_super_imsm(struct supertype *st, int fd, char *devname) | |
4002 | { | |
4003 | struct intel_super *super; | |
4004 | int rv; | |
4005 | ||
4006 | if (test_partition(fd)) | |
4007 | /* IMSM not allowed on partitions */ | |
4008 | return 1; | |
4009 | ||
4010 | free_super_imsm(st); | |
4011 | ||
4012 | super = alloc_super(); | |
4013 | if (!super) { | |
4014 | fprintf(stderr, | |
4015 | Name ": malloc of %zu failed.\n", | |
4016 | sizeof(*super)); | |
4017 | return 1; | |
4018 | } | |
4019 | /* Load hba and capabilities if they exist. | |
4020 | * But do not preclude loading metadata in case capabilities or hba are | |
4021 | * non-compliant and ignore_hw_compat is set. | |
4022 | */ | |
4023 | rv = find_intel_hba_capability(fd, super, devname); | |
4024 | /* no orom/efi or non-intel hba of the disk */ | |
4025 | if ((rv != 0) && (st->ignore_hw_compat == 0)) { | |
4026 | if (devname) | |
4027 | fprintf(stderr, | |
4028 | Name ": No OROM/EFI properties for %s\n", devname); | |
4029 | free_imsm(super); | |
4030 | return 2; | |
4031 | } | |
4032 | rv = load_and_parse_mpb(fd, super, devname, 0); | |
4033 | ||
4034 | if (rv) { | |
4035 | if (devname) | |
4036 | fprintf(stderr, | |
4037 | Name ": Failed to load all information " | |
4038 | "sections on %s\n", devname); | |
4039 | free_imsm(super); | |
4040 | return rv; | |
4041 | } | |
4042 | ||
4043 | st->sb = super; | |
4044 | if (st->ss == NULL) { | |
4045 | st->ss = &super_imsm; | |
4046 | st->minor_version = 0; | |
4047 | st->max_devs = IMSM_MAX_DEVICES; | |
4048 | } | |
4049 | ||
4050 | /* load migration record */ | |
4051 | if (load_imsm_migr_rec(super, NULL) == 0) { | |
4052 | /* Check for unsupported migration features */ | |
4053 | if (check_mpb_migr_compatibility(super) != 0) { | |
4054 | fprintf(stderr, | |
4055 | Name ": Unsupported migration detected"); | |
4056 | if (devname) | |
4057 | fprintf(stderr, " on %s\n", devname); | |
4058 | else | |
4059 | fprintf(stderr, " (IMSM).\n"); | |
4060 | return 3; | |
4061 | } | |
4062 | } | |
4063 | ||
4064 | return 0; | |
4065 | } | |
4066 | ||
4067 | static __u16 info_to_blocks_per_strip(mdu_array_info_t *info) | |
4068 | { | |
4069 | if (info->level == 1) | |
4070 | return 128; | |
4071 | return info->chunk_size >> 9; | |
4072 | } | |
4073 | ||
4074 | static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains) | |
4075 | { | |
4076 | __u32 num_stripes; | |
4077 | ||
4078 | num_stripes = (info->size * 2) / info_to_blocks_per_strip(info); | |
4079 | num_stripes /= num_domains; | |
4080 | ||
4081 | return num_stripes; | |
4082 | } | |
4083 | ||
4084 | static __u32 info_to_blocks_per_member(mdu_array_info_t *info) | |
4085 | { | |
4086 | if (info->level == 1) | |
4087 | return info->size * 2; | |
4088 | else | |
4089 | return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1); | |
4090 | } | |
4091 | ||
4092 | static void imsm_update_version_info(struct intel_super *super) | |
4093 | { | |
4094 | /* update the version and attributes */ | |
4095 | struct imsm_super *mpb = super->anchor; | |
4096 | char *version; | |
4097 | struct imsm_dev *dev; | |
4098 | struct imsm_map *map; | |
4099 | int i; | |
4100 | ||
4101 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4102 | dev = get_imsm_dev(super, i); | |
4103 | map = get_imsm_map(dev, 0); | |
4104 | if (__le32_to_cpu(dev->size_high) > 0) | |
4105 | mpb->attributes |= MPB_ATTRIB_2TB; | |
4106 | ||
4107 | /* FIXME detect when an array spans a port multiplier */ | |
4108 | #if 0 | |
4109 | mpb->attributes |= MPB_ATTRIB_PM; | |
4110 | #endif | |
4111 | ||
4112 | if (mpb->num_raid_devs > 1 || | |
4113 | mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) { | |
4114 | version = MPB_VERSION_ATTRIBS; | |
4115 | switch (get_imsm_raid_level(map)) { | |
4116 | case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break; | |
4117 | case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break; | |
4118 | case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break; | |
4119 | case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break; | |
4120 | } | |
4121 | } else { | |
4122 | if (map->num_members >= 5) | |
4123 | version = MPB_VERSION_5OR6_DISK_ARRAY; | |
4124 | else if (dev->status == DEV_CLONE_N_GO) | |
4125 | version = MPB_VERSION_CNG; | |
4126 | else if (get_imsm_raid_level(map) == 5) | |
4127 | version = MPB_VERSION_RAID5; | |
4128 | else if (map->num_members >= 3) | |
4129 | version = MPB_VERSION_3OR4_DISK_ARRAY; | |
4130 | else if (get_imsm_raid_level(map) == 1) | |
4131 | version = MPB_VERSION_RAID1; | |
4132 | else | |
4133 | version = MPB_VERSION_RAID0; | |
4134 | } | |
4135 | strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version); | |
4136 | } | |
4137 | } | |
4138 | ||
4139 | static int check_name(struct intel_super *super, char *name, int quiet) | |
4140 | { | |
4141 | struct imsm_super *mpb = super->anchor; | |
4142 | char *reason = NULL; | |
4143 | int i; | |
4144 | ||
4145 | if (strlen(name) > MAX_RAID_SERIAL_LEN) | |
4146 | reason = "must be 16 characters or less"; | |
4147 | ||
4148 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4149 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
4150 | ||
4151 | if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) { | |
4152 | reason = "already exists"; | |
4153 | break; | |
4154 | } | |
4155 | } | |
4156 | ||
4157 | if (reason && !quiet) | |
4158 | fprintf(stderr, Name ": imsm volume name %s\n", reason); | |
4159 | ||
4160 | return !reason; | |
4161 | } | |
4162 | ||
4163 | static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info, | |
4164 | unsigned long long size, char *name, | |
4165 | char *homehost, int *uuid) | |
4166 | { | |
4167 | /* We are creating a volume inside a pre-existing container. | |
4168 | * so st->sb is already set. | |
4169 | */ | |
4170 | struct intel_super *super = st->sb; | |
4171 | struct imsm_super *mpb = super->anchor; | |
4172 | struct intel_dev *dv; | |
4173 | struct imsm_dev *dev; | |
4174 | struct imsm_vol *vol; | |
4175 | struct imsm_map *map; | |
4176 | int idx = mpb->num_raid_devs; | |
4177 | int i; | |
4178 | unsigned long long array_blocks; | |
4179 | size_t size_old, size_new; | |
4180 | __u32 num_data_stripes; | |
4181 | ||
4182 | if (super->orom && mpb->num_raid_devs >= super->orom->vpa) { | |
4183 | fprintf(stderr, Name": This imsm-container already has the " | |
4184 | "maximum of %d volumes\n", super->orom->vpa); | |
4185 | return 0; | |
4186 | } | |
4187 | ||
4188 | /* ensure the mpb is large enough for the new data */ | |
4189 | size_old = __le32_to_cpu(mpb->mpb_size); | |
4190 | size_new = disks_to_mpb_size(info->nr_disks); | |
4191 | if (size_new > size_old) { | |
4192 | void *mpb_new; | |
4193 | size_t size_round = ROUND_UP(size_new, 512); | |
4194 | ||
4195 | if (posix_memalign(&mpb_new, 512, size_round) != 0) { | |
4196 | fprintf(stderr, Name": could not allocate new mpb\n"); | |
4197 | return 0; | |
4198 | } | |
4199 | if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) { | |
4200 | fprintf(stderr, Name | |
4201 | ": %s could not allocate migr_rec buffer\n", | |
4202 | __func__); | |
4203 | free(super->buf); | |
4204 | free(super); | |
4205 | free(mpb_new); | |
4206 | return 0; | |
4207 | } | |
4208 | memcpy(mpb_new, mpb, size_old); | |
4209 | free(mpb); | |
4210 | mpb = mpb_new; | |
4211 | super->anchor = mpb_new; | |
4212 | mpb->mpb_size = __cpu_to_le32(size_new); | |
4213 | memset(mpb_new + size_old, 0, size_round - size_old); | |
4214 | } | |
4215 | super->current_vol = idx; | |
4216 | ||
4217 | /* handle 'failed_disks' by either: | |
4218 | * a) create dummy disk entries in the table if this the first | |
4219 | * volume in the array. We add them here as this is the only | |
4220 | * opportunity to add them. add_to_super_imsm_volume() | |
4221 | * handles the non-failed disks and continues incrementing | |
4222 | * mpb->num_disks. | |
4223 | * b) validate that 'failed_disks' matches the current number | |
4224 | * of missing disks if the container is populated | |
4225 | */ | |
4226 | if (super->current_vol == 0) { | |
4227 | mpb->num_disks = 0; | |
4228 | for (i = 0; i < info->failed_disks; i++) { | |
4229 | struct imsm_disk *disk; | |
4230 | ||
4231 | mpb->num_disks++; | |
4232 | disk = __get_imsm_disk(mpb, i); | |
4233 | disk->status = CONFIGURED_DISK | FAILED_DISK; | |
4234 | disk->scsi_id = __cpu_to_le32(~(__u32)0); | |
4235 | snprintf((char *) disk->serial, MAX_RAID_SERIAL_LEN, | |
4236 | "missing:%d", i); | |
4237 | } | |
4238 | find_missing(super); | |
4239 | } else { | |
4240 | int missing = 0; | |
4241 | struct dl *d; | |
4242 | ||
4243 | for (d = super->missing; d; d = d->next) | |
4244 | missing++; | |
4245 | if (info->failed_disks > missing) { | |
4246 | fprintf(stderr, Name": unable to add 'missing' disk to container\n"); | |
4247 | return 0; | |
4248 | } | |
4249 | } | |
4250 | ||
4251 | if (!check_name(super, name, 0)) | |
4252 | return 0; | |
4253 | dv = malloc(sizeof(*dv)); | |
4254 | if (!dv) { | |
4255 | fprintf(stderr, Name ": failed to allocate device list entry\n"); | |
4256 | return 0; | |
4257 | } | |
4258 | dev = calloc(1, sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1)); | |
4259 | if (!dev) { | |
4260 | free(dv); | |
4261 | fprintf(stderr, Name": could not allocate raid device\n"); | |
4262 | return 0; | |
4263 | } | |
4264 | ||
4265 | strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN); | |
4266 | if (info->level == 1) | |
4267 | array_blocks = info_to_blocks_per_member(info); | |
4268 | else | |
4269 | array_blocks = calc_array_size(info->level, info->raid_disks, | |
4270 | info->layout, info->chunk_size, | |
4271 | info->size*2); | |
4272 | /* round array size down to closest MB */ | |
4273 | array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT; | |
4274 | ||
4275 | dev->size_low = __cpu_to_le32((__u32) array_blocks); | |
4276 | dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32)); | |
4277 | dev->status = (DEV_READ_COALESCING | DEV_WRITE_COALESCING); | |
4278 | vol = &dev->vol; | |
4279 | vol->migr_state = 0; | |
4280 | set_migr_type(dev, MIGR_INIT); | |
4281 | vol->dirty = !info->state; | |
4282 | vol->curr_migr_unit = 0; | |
4283 | map = get_imsm_map(dev, 0); | |
4284 | map->pba_of_lba0 = __cpu_to_le32(super->create_offset); | |
4285 | map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info)); | |
4286 | map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info)); | |
4287 | map->failed_disk_num = ~0; | |
4288 | map->map_state = info->failed_disks ? IMSM_T_STATE_DEGRADED : IMSM_T_STATE_NORMAL; | |
4289 | map->ddf = 1; | |
4290 | ||
4291 | if (info->level == 1 && info->raid_disks > 2) { | |
4292 | free(dev); | |
4293 | free(dv); | |
4294 | fprintf(stderr, Name": imsm does not support more than 2 disks" | |
4295 | "in a raid1 volume\n"); | |
4296 | return 0; | |
4297 | } | |
4298 | ||
4299 | map->raid_level = info->level; | |
4300 | if (info->level == 10) { | |
4301 | map->raid_level = 1; | |
4302 | map->num_domains = info->raid_disks / 2; | |
4303 | } else if (info->level == 1) | |
4304 | map->num_domains = info->raid_disks; | |
4305 | else | |
4306 | map->num_domains = 1; | |
4307 | ||
4308 | num_data_stripes = info_to_num_data_stripes(info, map->num_domains); | |
4309 | map->num_data_stripes = __cpu_to_le32(num_data_stripes); | |
4310 | ||
4311 | map->num_members = info->raid_disks; | |
4312 | for (i = 0; i < map->num_members; i++) { | |
4313 | /* initialized in add_to_super */ | |
4314 | set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD); | |
4315 | } | |
4316 | mpb->num_raid_devs++; | |
4317 | ||
4318 | dv->dev = dev; | |
4319 | dv->index = super->current_vol; | |
4320 | dv->next = super->devlist; | |
4321 | super->devlist = dv; | |
4322 | ||
4323 | imsm_update_version_info(super); | |
4324 | ||
4325 | return 1; | |
4326 | } | |
4327 | ||
4328 | static int init_super_imsm(struct supertype *st, mdu_array_info_t *info, | |
4329 | unsigned long long size, char *name, | |
4330 | char *homehost, int *uuid) | |
4331 | { | |
4332 | /* This is primarily called by Create when creating a new array. | |
4333 | * We will then get add_to_super called for each component, and then | |
4334 | * write_init_super called to write it out to each device. | |
4335 | * For IMSM, Create can create on fresh devices or on a pre-existing | |
4336 | * array. | |
4337 | * To create on a pre-existing array a different method will be called. | |
4338 | * This one is just for fresh drives. | |
4339 | */ | |
4340 | struct intel_super *super; | |
4341 | struct imsm_super *mpb; | |
4342 | size_t mpb_size; | |
4343 | char *version; | |
4344 | ||
4345 | if (st->sb) | |
4346 | return init_super_imsm_volume(st, info, size, name, homehost, uuid); | |
4347 | ||
4348 | if (info) | |
4349 | mpb_size = disks_to_mpb_size(info->nr_disks); | |
4350 | else | |
4351 | mpb_size = 512; | |
4352 | ||
4353 | super = alloc_super(); | |
4354 | if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) { | |
4355 | free(super); | |
4356 | super = NULL; | |
4357 | } | |
4358 | if (!super) { | |
4359 | fprintf(stderr, Name | |
4360 | ": %s could not allocate superblock\n", __func__); | |
4361 | return 0; | |
4362 | } | |
4363 | if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) { | |
4364 | fprintf(stderr, Name | |
4365 | ": %s could not allocate migr_rec buffer\n", __func__); | |
4366 | free(super->buf); | |
4367 | free(super); | |
4368 | return 0; | |
4369 | } | |
4370 | memset(super->buf, 0, mpb_size); | |
4371 | mpb = super->buf; | |
4372 | mpb->mpb_size = __cpu_to_le32(mpb_size); | |
4373 | st->sb = super; | |
4374 | ||
4375 | if (info == NULL) { | |
4376 | /* zeroing superblock */ | |
4377 | return 0; | |
4378 | } | |
4379 | ||
4380 | mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY; | |
4381 | ||
4382 | version = (char *) mpb->sig; | |
4383 | strcpy(version, MPB_SIGNATURE); | |
4384 | version += strlen(MPB_SIGNATURE); | |
4385 | strcpy(version, MPB_VERSION_RAID0); | |
4386 | ||
4387 | return 1; | |
4388 | } | |
4389 | ||
4390 | #ifndef MDASSEMBLE | |
4391 | static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk, | |
4392 | int fd, char *devname) | |
4393 | { | |
4394 | struct intel_super *super = st->sb; | |
4395 | struct imsm_super *mpb = super->anchor; | |
4396 | struct imsm_disk *_disk; | |
4397 | struct imsm_dev *dev; | |
4398 | struct imsm_map *map; | |
4399 | struct dl *dl, *df; | |
4400 | int slot; | |
4401 | ||
4402 | dev = get_imsm_dev(super, super->current_vol); | |
4403 | map = get_imsm_map(dev, 0); | |
4404 | ||
4405 | if (! (dk->state & (1<<MD_DISK_SYNC))) { | |
4406 | fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n", | |
4407 | devname); | |
4408 | return 1; | |
4409 | } | |
4410 | ||
4411 | if (fd == -1) { | |
4412 | /* we're doing autolayout so grab the pre-marked (in | |
4413 | * validate_geometry) raid_disk | |
4414 | */ | |
4415 | for (dl = super->disks; dl; dl = dl->next) | |
4416 | if (dl->raiddisk == dk->raid_disk) | |
4417 | break; | |
4418 | } else { | |
4419 | for (dl = super->disks; dl ; dl = dl->next) | |
4420 | if (dl->major == dk->major && | |
4421 | dl->minor == dk->minor) | |
4422 | break; | |
4423 | } | |
4424 | ||
4425 | if (!dl) { | |
4426 | fprintf(stderr, Name ": %s is not a member of the same container\n", devname); | |
4427 | return 1; | |
4428 | } | |
4429 | ||
4430 | /* add a pristine spare to the metadata */ | |
4431 | if (dl->index < 0) { | |
4432 | dl->index = super->anchor->num_disks; | |
4433 | super->anchor->num_disks++; | |
4434 | } | |
4435 | /* Check the device has not already been added */ | |
4436 | slot = get_imsm_disk_slot(map, dl->index); | |
4437 | if (slot >= 0 && | |
4438 | (get_imsm_ord_tbl_ent(dev, slot, -1) & IMSM_ORD_REBUILD) == 0) { | |
4439 | fprintf(stderr, Name ": %s has been included in this array twice\n", | |
4440 | devname); | |
4441 | return 1; | |
4442 | } | |
4443 | set_imsm_ord_tbl_ent(map, dk->raid_disk, dl->index); | |
4444 | dl->disk.status = CONFIGURED_DISK; | |
4445 | ||
4446 | /* update size of 'missing' disks to be at least as large as the | |
4447 | * largest acitve member (we only have dummy missing disks when | |
4448 | * creating the first volume) | |
4449 | */ | |
4450 | if (super->current_vol == 0) { | |
4451 | for (df = super->missing; df; df = df->next) { | |
4452 | if (dl->disk.total_blocks > df->disk.total_blocks) | |
4453 | df->disk.total_blocks = dl->disk.total_blocks; | |
4454 | _disk = __get_imsm_disk(mpb, df->index); | |
4455 | *_disk = df->disk; | |
4456 | } | |
4457 | } | |
4458 | ||
4459 | /* refresh unset/failed slots to point to valid 'missing' entries */ | |
4460 | for (df = super->missing; df; df = df->next) | |
4461 | for (slot = 0; slot < mpb->num_disks; slot++) { | |
4462 | __u32 ord = get_imsm_ord_tbl_ent(dev, slot, -1); | |
4463 | ||
4464 | if ((ord & IMSM_ORD_REBUILD) == 0) | |
4465 | continue; | |
4466 | set_imsm_ord_tbl_ent(map, slot, df->index | IMSM_ORD_REBUILD); | |
4467 | dprintf("set slot:%d to missing disk:%d\n", slot, df->index); | |
4468 | break; | |
4469 | } | |
4470 | ||
4471 | /* if we are creating the first raid device update the family number */ | |
4472 | if (super->current_vol == 0) { | |
4473 | __u32 sum; | |
4474 | struct imsm_dev *_dev = __get_imsm_dev(mpb, 0); | |
4475 | ||
4476 | _disk = __get_imsm_disk(mpb, dl->index); | |
4477 | if (!_dev || !_disk) { | |
4478 | fprintf(stderr, Name ": BUG mpb setup error\n"); | |
4479 | return 1; | |
4480 | } | |
4481 | *_dev = *dev; | |
4482 | *_disk = dl->disk; | |
4483 | sum = random32(); | |
4484 | sum += __gen_imsm_checksum(mpb); | |
4485 | mpb->family_num = __cpu_to_le32(sum); | |
4486 | mpb->orig_family_num = mpb->family_num; | |
4487 | } | |
4488 | super->current_disk = dl; | |
4489 | return 0; | |
4490 | } | |
4491 | ||
4492 | /* mark_spare() | |
4493 | * Function marks disk as spare and restores disk serial | |
4494 | * in case it was previously marked as failed by takeover operation | |
4495 | * reruns: | |
4496 | * -1 : critical error | |
4497 | * 0 : disk is marked as spare but serial is not set | |
4498 | * 1 : success | |
4499 | */ | |
4500 | int mark_spare(struct dl *disk) | |
4501 | { | |
4502 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
4503 | int ret_val = -1; | |
4504 | ||
4505 | if (!disk) | |
4506 | return ret_val; | |
4507 | ||
4508 | ret_val = 0; | |
4509 | if (!imsm_read_serial(disk->fd, NULL, serial)) { | |
4510 | /* Restore disk serial number, because takeover marks disk | |
4511 | * as failed and adds to serial ':0' before it becomes | |
4512 | * a spare disk. | |
4513 | */ | |
4514 | serialcpy(disk->serial, serial); | |
4515 | serialcpy(disk->disk.serial, serial); | |
4516 | ret_val = 1; | |
4517 | } | |
4518 | disk->disk.status = SPARE_DISK; | |
4519 | disk->index = -1; | |
4520 | ||
4521 | return ret_val; | |
4522 | } | |
4523 | ||
4524 | static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk, | |
4525 | int fd, char *devname) | |
4526 | { | |
4527 | struct intel_super *super = st->sb; | |
4528 | struct dl *dd; | |
4529 | unsigned long long size; | |
4530 | __u32 id; | |
4531 | int rv; | |
4532 | struct stat stb; | |
4533 | ||
4534 | /* If we are on an RAID enabled platform check that the disk is | |
4535 | * attached to the raid controller. | |
4536 | * We do not need to test disks attachment for container based additions, | |
4537 | * they shall be already tested when container was created/assembled. | |
4538 | */ | |
4539 | rv = find_intel_hba_capability(fd, super, devname); | |
4540 | /* no orom/efi or non-intel hba of the disk */ | |
4541 | if (rv != 0) { | |
4542 | dprintf("capability: %p fd: %d ret: %d\n", | |
4543 | super->orom, fd, rv); | |
4544 | return 1; | |
4545 | } | |
4546 | ||
4547 | if (super->current_vol >= 0) | |
4548 | return add_to_super_imsm_volume(st, dk, fd, devname); | |
4549 | ||
4550 | fstat(fd, &stb); | |
4551 | dd = malloc(sizeof(*dd)); | |
4552 | if (!dd) { | |
4553 | fprintf(stderr, | |
4554 | Name ": malloc failed %s:%d.\n", __func__, __LINE__); | |
4555 | return 1; | |
4556 | } | |
4557 | memset(dd, 0, sizeof(*dd)); | |
4558 | dd->major = major(stb.st_rdev); | |
4559 | dd->minor = minor(stb.st_rdev); | |
4560 | dd->devname = devname ? strdup(devname) : NULL; | |
4561 | dd->fd = fd; | |
4562 | dd->e = NULL; | |
4563 | dd->action = DISK_ADD; | |
4564 | rv = imsm_read_serial(fd, devname, dd->serial); | |
4565 | if (rv) { | |
4566 | fprintf(stderr, | |
4567 | Name ": failed to retrieve scsi serial, aborting\n"); | |
4568 | free(dd); | |
4569 | abort(); | |
4570 | } | |
4571 | ||
4572 | get_dev_size(fd, NULL, &size); | |
4573 | size /= 512; | |
4574 | serialcpy(dd->disk.serial, dd->serial); | |
4575 | dd->disk.total_blocks = __cpu_to_le32(size); | |
4576 | mark_spare(dd); | |
4577 | if (sysfs_disk_to_scsi_id(fd, &id) == 0) | |
4578 | dd->disk.scsi_id = __cpu_to_le32(id); | |
4579 | else | |
4580 | dd->disk.scsi_id = __cpu_to_le32(0); | |
4581 | ||
4582 | if (st->update_tail) { | |
4583 | dd->next = super->disk_mgmt_list; | |
4584 | super->disk_mgmt_list = dd; | |
4585 | } else { | |
4586 | dd->next = super->disks; | |
4587 | super->disks = dd; | |
4588 | super->updates_pending++; | |
4589 | } | |
4590 | ||
4591 | return 0; | |
4592 | } | |
4593 | ||
4594 | ||
4595 | static int remove_from_super_imsm(struct supertype *st, mdu_disk_info_t *dk) | |
4596 | { | |
4597 | struct intel_super *super = st->sb; | |
4598 | struct dl *dd; | |
4599 | ||
4600 | /* remove from super works only in mdmon - for communication | |
4601 | * manager - monitor. Check if communication memory buffer | |
4602 | * is prepared. | |
4603 | */ | |
4604 | if (!st->update_tail) { | |
4605 | fprintf(stderr, | |
4606 | Name ": %s shall be used in mdmon context only" | |
4607 | "(line %d).\n", __func__, __LINE__); | |
4608 | return 1; | |
4609 | } | |
4610 | dd = malloc(sizeof(*dd)); | |
4611 | if (!dd) { | |
4612 | fprintf(stderr, | |
4613 | Name ": malloc failed %s:%d.\n", __func__, __LINE__); | |
4614 | return 1; | |
4615 | } | |
4616 | memset(dd, 0, sizeof(*dd)); | |
4617 | dd->major = dk->major; | |
4618 | dd->minor = dk->minor; | |
4619 | dd->fd = -1; | |
4620 | mark_spare(dd); | |
4621 | dd->action = DISK_REMOVE; | |
4622 | ||
4623 | dd->next = super->disk_mgmt_list; | |
4624 | super->disk_mgmt_list = dd; | |
4625 | ||
4626 | ||
4627 | return 0; | |
4628 | } | |
4629 | ||
4630 | static int store_imsm_mpb(int fd, struct imsm_super *mpb); | |
4631 | ||
4632 | static union { | |
4633 | char buf[512]; | |
4634 | struct imsm_super anchor; | |
4635 | } spare_record __attribute__ ((aligned(512))); | |
4636 | ||
4637 | /* spare records have their own family number and do not have any defined raid | |
4638 | * devices | |
4639 | */ | |
4640 | static int write_super_imsm_spares(struct intel_super *super, int doclose) | |
4641 | { | |
4642 | struct imsm_super *mpb = super->anchor; | |
4643 | struct imsm_super *spare = &spare_record.anchor; | |
4644 | __u32 sum; | |
4645 | struct dl *d; | |
4646 | ||
4647 | spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)), | |
4648 | spare->generation_num = __cpu_to_le32(1UL), | |
4649 | spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY; | |
4650 | spare->num_disks = 1, | |
4651 | spare->num_raid_devs = 0, | |
4652 | spare->cache_size = mpb->cache_size, | |
4653 | spare->pwr_cycle_count = __cpu_to_le32(1), | |
4654 | ||
4655 | snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH, | |
4656 | MPB_SIGNATURE MPB_VERSION_RAID0); | |
4657 | ||
4658 | for (d = super->disks; d; d = d->next) { | |
4659 | if (d->index != -1) | |
4660 | continue; | |
4661 | ||
4662 | spare->disk[0] = d->disk; | |
4663 | sum = __gen_imsm_checksum(spare); | |
4664 | spare->family_num = __cpu_to_le32(sum); | |
4665 | spare->orig_family_num = 0; | |
4666 | sum = __gen_imsm_checksum(spare); | |
4667 | spare->check_sum = __cpu_to_le32(sum); | |
4668 | ||
4669 | if (store_imsm_mpb(d->fd, spare)) { | |
4670 | fprintf(stderr, "%s: failed for device %d:%d %s\n", | |
4671 | __func__, d->major, d->minor, strerror(errno)); | |
4672 | return 1; | |
4673 | } | |
4674 | if (doclose) { | |
4675 | close(d->fd); | |
4676 | d->fd = -1; | |
4677 | } | |
4678 | } | |
4679 | ||
4680 | return 0; | |
4681 | } | |
4682 | ||
4683 | static int write_super_imsm(struct supertype *st, int doclose) | |
4684 | { | |
4685 | struct intel_super *super = st->sb; | |
4686 | struct imsm_super *mpb = super->anchor; | |
4687 | struct dl *d; | |
4688 | __u32 generation; | |
4689 | __u32 sum; | |
4690 | int spares = 0; | |
4691 | int i; | |
4692 | __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk); | |
4693 | int num_disks = 0; | |
4694 | int clear_migration_record = 1; | |
4695 | ||
4696 | /* 'generation' is incremented everytime the metadata is written */ | |
4697 | generation = __le32_to_cpu(mpb->generation_num); | |
4698 | generation++; | |
4699 | mpb->generation_num = __cpu_to_le32(generation); | |
4700 | ||
4701 | /* fix up cases where previous mdadm releases failed to set | |
4702 | * orig_family_num | |
4703 | */ | |
4704 | if (mpb->orig_family_num == 0) | |
4705 | mpb->orig_family_num = mpb->family_num; | |
4706 | ||
4707 | for (d = super->disks; d; d = d->next) { | |
4708 | if (d->index == -1) | |
4709 | spares++; | |
4710 | else { | |
4711 | mpb->disk[d->index] = d->disk; | |
4712 | num_disks++; | |
4713 | } | |
4714 | } | |
4715 | for (d = super->missing; d; d = d->next) { | |
4716 | mpb->disk[d->index] = d->disk; | |
4717 | num_disks++; | |
4718 | } | |
4719 | mpb->num_disks = num_disks; | |
4720 | mpb_size += sizeof(struct imsm_disk) * mpb->num_disks; | |
4721 | ||
4722 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4723 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
4724 | struct imsm_dev *dev2 = get_imsm_dev(super, i); | |
4725 | if (dev && dev2) { | |
4726 | imsm_copy_dev(dev, dev2); | |
4727 | mpb_size += sizeof_imsm_dev(dev, 0); | |
4728 | } | |
4729 | if (is_gen_migration(dev2)) | |
4730 | clear_migration_record = 0; | |
4731 | } | |
4732 | mpb_size += __le32_to_cpu(mpb->bbm_log_size); | |
4733 | mpb->mpb_size = __cpu_to_le32(mpb_size); | |
4734 | ||
4735 | /* recalculate checksum */ | |
4736 | sum = __gen_imsm_checksum(mpb); | |
4737 | mpb->check_sum = __cpu_to_le32(sum); | |
4738 | ||
4739 | if (clear_migration_record) | |
4740 | memset(super->migr_rec_buf, 0, 512); | |
4741 | ||
4742 | /* write the mpb for disks that compose raid devices */ | |
4743 | for (d = super->disks; d ; d = d->next) { | |
4744 | if (d->index < 0 || is_failed(&d->disk)) | |
4745 | continue; | |
4746 | if (store_imsm_mpb(d->fd, mpb)) | |
4747 | fprintf(stderr, "%s: failed for device %d:%d %s\n", | |
4748 | __func__, d->major, d->minor, strerror(errno)); | |
4749 | if (clear_migration_record) { | |
4750 | unsigned long long dsize; | |
4751 | ||
4752 | get_dev_size(d->fd, NULL, &dsize); | |
4753 | if (lseek64(d->fd, dsize - 512, SEEK_SET) >= 0) { | |
4754 | if (write(d->fd, super->migr_rec_buf, 512) != 512) | |
4755 | perror("Write migr_rec failed"); | |
4756 | } | |
4757 | } | |
4758 | if (doclose) { | |
4759 | close(d->fd); | |
4760 | d->fd = -1; | |
4761 | } | |
4762 | } | |
4763 | ||
4764 | if (spares) | |
4765 | return write_super_imsm_spares(super, doclose); | |
4766 | ||
4767 | return 0; | |
4768 | } | |
4769 | ||
4770 | ||
4771 | static int create_array(struct supertype *st, int dev_idx) | |
4772 | { | |
4773 | size_t len; | |
4774 | struct imsm_update_create_array *u; | |
4775 | struct intel_super *super = st->sb; | |
4776 | struct imsm_dev *dev = get_imsm_dev(super, dev_idx); | |
4777 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4778 | struct disk_info *inf; | |
4779 | struct imsm_disk *disk; | |
4780 | int i; | |
4781 | ||
4782 | len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) + | |
4783 | sizeof(*inf) * map->num_members; | |
4784 | u = malloc(len); | |
4785 | if (!u) { | |
4786 | fprintf(stderr, "%s: failed to allocate update buffer\n", | |
4787 | __func__); | |
4788 | return 1; | |
4789 | } | |
4790 | ||
4791 | u->type = update_create_array; | |
4792 | u->dev_idx = dev_idx; | |
4793 | imsm_copy_dev(&u->dev, dev); | |
4794 | inf = get_disk_info(u); | |
4795 | for (i = 0; i < map->num_members; i++) { | |
4796 | int idx = get_imsm_disk_idx(dev, i, -1); | |
4797 | ||
4798 | disk = get_imsm_disk(super, idx); | |
4799 | serialcpy(inf[i].serial, disk->serial); | |
4800 | } | |
4801 | append_metadata_update(st, u, len); | |
4802 | ||
4803 | return 0; | |
4804 | } | |
4805 | ||
4806 | static int mgmt_disk(struct supertype *st) | |
4807 | { | |
4808 | struct intel_super *super = st->sb; | |
4809 | size_t len; | |
4810 | struct imsm_update_add_remove_disk *u; | |
4811 | ||
4812 | if (!super->disk_mgmt_list) | |
4813 | return 0; | |
4814 | ||
4815 | len = sizeof(*u); | |
4816 | u = malloc(len); | |
4817 | if (!u) { | |
4818 | fprintf(stderr, "%s: failed to allocate update buffer\n", | |
4819 | __func__); | |
4820 | return 1; | |
4821 | } | |
4822 | ||
4823 | u->type = update_add_remove_disk; | |
4824 | append_metadata_update(st, u, len); | |
4825 | ||
4826 | return 0; | |
4827 | } | |
4828 | ||
4829 | static int write_init_super_imsm(struct supertype *st) | |
4830 | { | |
4831 | struct intel_super *super = st->sb; | |
4832 | int current_vol = super->current_vol; | |
4833 | ||
4834 | /* we are done with current_vol reset it to point st at the container */ | |
4835 | super->current_vol = -1; | |
4836 | ||
4837 | if (st->update_tail) { | |
4838 | /* queue the recently created array / added disk | |
4839 | * as a metadata update */ | |
4840 | int rv; | |
4841 | ||
4842 | /* determine if we are creating a volume or adding a disk */ | |
4843 | if (current_vol < 0) { | |
4844 | /* in the mgmt (add/remove) disk case we are running | |
4845 | * in mdmon context, so don't close fd's | |
4846 | */ | |
4847 | return mgmt_disk(st); | |
4848 | } else | |
4849 | rv = create_array(st, current_vol); | |
4850 | ||
4851 | return rv; | |
4852 | } else { | |
4853 | struct dl *d; | |
4854 | for (d = super->disks; d; d = d->next) | |
4855 | Kill(d->devname, NULL, 0, 1, 1); | |
4856 | return write_super_imsm(st, 1); | |
4857 | } | |
4858 | } | |
4859 | #endif | |
4860 | ||
4861 | static int store_super_imsm(struct supertype *st, int fd) | |
4862 | { | |
4863 | struct intel_super *super = st->sb; | |
4864 | struct imsm_super *mpb = super ? super->anchor : NULL; | |
4865 | ||
4866 | if (!mpb) | |
4867 | return 1; | |
4868 | ||
4869 | #ifndef MDASSEMBLE | |
4870 | return store_imsm_mpb(fd, mpb); | |
4871 | #else | |
4872 | return 1; | |
4873 | #endif | |
4874 | } | |
4875 | ||
4876 | static int imsm_bbm_log_size(struct imsm_super *mpb) | |
4877 | { | |
4878 | return __le32_to_cpu(mpb->bbm_log_size); | |
4879 | } | |
4880 | ||
4881 | #ifndef MDASSEMBLE | |
4882 | static int validate_geometry_imsm_container(struct supertype *st, int level, | |
4883 | int layout, int raiddisks, int chunk, | |
4884 | unsigned long long size, char *dev, | |
4885 | unsigned long long *freesize, | |
4886 | int verbose) | |
4887 | { | |
4888 | int fd; | |
4889 | unsigned long long ldsize; | |
4890 | struct intel_super *super=NULL; | |
4891 | int rv = 0; | |
4892 | ||
4893 | if (level != LEVEL_CONTAINER) | |
4894 | return 0; | |
4895 | if (!dev) | |
4896 | return 1; | |
4897 | ||
4898 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
4899 | if (fd < 0) { | |
4900 | if (verbose) | |
4901 | fprintf(stderr, Name ": imsm: Cannot open %s: %s\n", | |
4902 | dev, strerror(errno)); | |
4903 | return 0; | |
4904 | } | |
4905 | if (!get_dev_size(fd, dev, &ldsize)) { | |
4906 | close(fd); | |
4907 | return 0; | |
4908 | } | |
4909 | ||
4910 | /* capabilities retrieve could be possible | |
4911 | * note that there is no fd for the disks in array. | |
4912 | */ | |
4913 | super = alloc_super(); | |
4914 | if (!super) { | |
4915 | fprintf(stderr, | |
4916 | Name ": malloc of %zu failed.\n", | |
4917 | sizeof(*super)); | |
4918 | close(fd); | |
4919 | return 0; | |
4920 | } | |
4921 | ||
4922 | rv = find_intel_hba_capability(fd, super, verbose ? dev : NULL); | |
4923 | if (rv != 0) { | |
4924 | #if DEBUG | |
4925 | char str[256]; | |
4926 | fd2devname(fd, str); | |
4927 | dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n", | |
4928 | fd, str, super->orom, rv, raiddisks); | |
4929 | #endif | |
4930 | /* no orom/efi or non-intel hba of the disk */ | |
4931 | close(fd); | |
4932 | free_imsm(super); | |
4933 | return 0; | |
4934 | } | |
4935 | close(fd); | |
4936 | if (super->orom && raiddisks > super->orom->tds) { | |
4937 | if (verbose) | |
4938 | fprintf(stderr, Name ": %d exceeds maximum number of" | |
4939 | " platform supported disks: %d\n", | |
4940 | raiddisks, super->orom->tds); | |
4941 | ||
4942 | free_imsm(super); | |
4943 | return 0; | |
4944 | } | |
4945 | ||
4946 | *freesize = avail_size_imsm(st, ldsize >> 9); | |
4947 | free_imsm(super); | |
4948 | ||
4949 | return 1; | |
4950 | } | |
4951 | ||
4952 | static unsigned long long find_size(struct extent *e, int *idx, int num_extents) | |
4953 | { | |
4954 | const unsigned long long base_start = e[*idx].start; | |
4955 | unsigned long long end = base_start + e[*idx].size; | |
4956 | int i; | |
4957 | ||
4958 | if (base_start == end) | |
4959 | return 0; | |
4960 | ||
4961 | *idx = *idx + 1; | |
4962 | for (i = *idx; i < num_extents; i++) { | |
4963 | /* extend overlapping extents */ | |
4964 | if (e[i].start >= base_start && | |
4965 | e[i].start <= end) { | |
4966 | if (e[i].size == 0) | |
4967 | return 0; | |
4968 | if (e[i].start + e[i].size > end) | |
4969 | end = e[i].start + e[i].size; | |
4970 | } else if (e[i].start > end) { | |
4971 | *idx = i; | |
4972 | break; | |
4973 | } | |
4974 | } | |
4975 | ||
4976 | return end - base_start; | |
4977 | } | |
4978 | ||
4979 | static unsigned long long merge_extents(struct intel_super *super, int sum_extents) | |
4980 | { | |
4981 | /* build a composite disk with all known extents and generate a new | |
4982 | * 'maxsize' given the "all disks in an array must share a common start | |
4983 | * offset" constraint | |
4984 | */ | |
4985 | struct extent *e = calloc(sum_extents, sizeof(*e)); | |
4986 | struct dl *dl; | |
4987 | int i, j; | |
4988 | int start_extent; | |
4989 | unsigned long long pos; | |
4990 | unsigned long long start = 0; | |
4991 | unsigned long long maxsize; | |
4992 | unsigned long reserve; | |
4993 | ||
4994 | if (!e) | |
4995 | return 0; | |
4996 | ||
4997 | /* coalesce and sort all extents. also, check to see if we need to | |
4998 | * reserve space between member arrays | |
4999 | */ | |
5000 | j = 0; | |
5001 | for (dl = super->disks; dl; dl = dl->next) { | |
5002 | if (!dl->e) | |
5003 | continue; | |
5004 | for (i = 0; i < dl->extent_cnt; i++) | |
5005 | e[j++] = dl->e[i]; | |
5006 | } | |
5007 | qsort(e, sum_extents, sizeof(*e), cmp_extent); | |
5008 | ||
5009 | /* merge extents */ | |
5010 | i = 0; | |
5011 | j = 0; | |
5012 | while (i < sum_extents) { | |
5013 | e[j].start = e[i].start; | |
5014 | e[j].size = find_size(e, &i, sum_extents); | |
5015 | j++; | |
5016 | if (e[j-1].size == 0) | |
5017 | break; | |
5018 | } | |
5019 | ||
5020 | pos = 0; | |
5021 | maxsize = 0; | |
5022 | start_extent = 0; | |
5023 | i = 0; | |
5024 | do { | |
5025 | unsigned long long esize; | |
5026 | ||
5027 | esize = e[i].start - pos; | |
5028 | if (esize >= maxsize) { | |
5029 | maxsize = esize; | |
5030 | start = pos; | |
5031 | start_extent = i; | |
5032 | } | |
5033 | pos = e[i].start + e[i].size; | |
5034 | i++; | |
5035 | } while (e[i-1].size); | |
5036 | free(e); | |
5037 | ||
5038 | if (maxsize == 0) | |
5039 | return 0; | |
5040 | ||
5041 | /* FIXME assumes volume at offset 0 is the first volume in a | |
5042 | * container | |
5043 | */ | |
5044 | if (start_extent > 0) | |
5045 | reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */ | |
5046 | else | |
5047 | reserve = 0; | |
5048 | ||
5049 | if (maxsize < reserve) | |
5050 | return 0; | |
5051 | ||
5052 | super->create_offset = ~((__u32) 0); | |
5053 | if (start + reserve > super->create_offset) | |
5054 | return 0; /* start overflows create_offset */ | |
5055 | super->create_offset = start + reserve; | |
5056 | ||
5057 | return maxsize - reserve; | |
5058 | } | |
5059 | ||
5060 | static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks) | |
5061 | { | |
5062 | if (level < 0 || level == 6 || level == 4) | |
5063 | return 0; | |
5064 | ||
5065 | /* if we have an orom prevent invalid raid levels */ | |
5066 | if (orom) | |
5067 | switch (level) { | |
5068 | case 0: return imsm_orom_has_raid0(orom); | |
5069 | case 1: | |
5070 | if (raiddisks > 2) | |
5071 | return imsm_orom_has_raid1e(orom); | |
5072 | return imsm_orom_has_raid1(orom) && raiddisks == 2; | |
5073 | case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4; | |
5074 | case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2; | |
5075 | } | |
5076 | else | |
5077 | return 1; /* not on an Intel RAID platform so anything goes */ | |
5078 | ||
5079 | return 0; | |
5080 | } | |
5081 | ||
5082 | static int imsm_default_chunk(const struct imsm_orom *orom) | |
5083 | { | |
5084 | /* up to 512 if the plaform supports it, otherwise the platform max. | |
5085 | * 128 if no platform detected | |
5086 | */ | |
5087 | int fs = max(7, orom ? fls(orom->sss) : 0); | |
5088 | ||
5089 | return min(512, (1 << fs)); | |
5090 | } | |
5091 | ||
5092 | #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg)) | |
5093 | static int | |
5094 | validate_geometry_imsm_orom(struct intel_super *super, int level, int layout, | |
5095 | int raiddisks, int *chunk, int verbose) | |
5096 | { | |
5097 | /* check/set platform and metadata limits/defaults */ | |
5098 | if (super->orom && raiddisks > super->orom->dpa) { | |
5099 | pr_vrb(": platform supports a maximum of %d disks per array\n", | |
5100 | super->orom->dpa); | |
5101 | return 0; | |
5102 | } | |
5103 | ||
5104 | /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */ | |
5105 | if (!is_raid_level_supported(super->orom, level, raiddisks)) { | |
5106 | pr_vrb(": platform does not support raid%d with %d disk%s\n", | |
5107 | level, raiddisks, raiddisks > 1 ? "s" : ""); | |
5108 | return 0; | |
5109 | } | |
5110 | ||
5111 | if (chunk && (*chunk == 0 || *chunk == UnSet)) | |
5112 | *chunk = imsm_default_chunk(super->orom); | |
5113 | ||
5114 | if (super->orom && chunk && !imsm_orom_has_chunk(super->orom, *chunk)) { | |
5115 | pr_vrb(": platform does not support a chunk size of: " | |
5116 | "%d\n", *chunk); | |
5117 | return 0; | |
5118 | } | |
5119 | ||
5120 | if (layout != imsm_level_to_layout(level)) { | |
5121 | if (level == 5) | |
5122 | pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n"); | |
5123 | else if (level == 10) | |
5124 | pr_vrb(": imsm raid 10 only supports the n2 layout\n"); | |
5125 | else | |
5126 | pr_vrb(": imsm unknown layout %#x for this raid level %d\n", | |
5127 | layout, level); | |
5128 | return 0; | |
5129 | } | |
5130 | return 1; | |
5131 | } | |
5132 | ||
5133 | /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd | |
5134 | * FIX ME add ahci details | |
5135 | */ | |
5136 | static int validate_geometry_imsm_volume(struct supertype *st, int level, | |
5137 | int layout, int raiddisks, int *chunk, | |
5138 | unsigned long long size, char *dev, | |
5139 | unsigned long long *freesize, | |
5140 | int verbose) | |
5141 | { | |
5142 | struct stat stb; | |
5143 | struct intel_super *super = st->sb; | |
5144 | struct imsm_super *mpb; | |
5145 | struct dl *dl; | |
5146 | unsigned long long pos = 0; | |
5147 | unsigned long long maxsize; | |
5148 | struct extent *e; | |
5149 | int i; | |
5150 | ||
5151 | /* We must have the container info already read in. */ | |
5152 | if (!super) | |
5153 | return 0; | |
5154 | ||
5155 | mpb = super->anchor; | |
5156 | ||
5157 | if (mpb->num_raid_devs > 0 && mpb->num_disks != raiddisks) { | |
5158 | fprintf(stderr, Name ": the option-rom requires all " | |
5159 | "member disks to be a member of all volumes.\n"); | |
5160 | return 0; | |
5161 | } | |
5162 | ||
5163 | if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose)) { | |
5164 | fprintf(stderr, Name ": RAID gemetry validation failed. " | |
5165 | "Cannot proceed with the action(s).\n"); | |
5166 | return 0; | |
5167 | } | |
5168 | if (!dev) { | |
5169 | /* General test: make sure there is space for | |
5170 | * 'raiddisks' device extents of size 'size' at a given | |
5171 | * offset | |
5172 | */ | |
5173 | unsigned long long minsize = size; | |
5174 | unsigned long long start_offset = MaxSector; | |
5175 | int dcnt = 0; | |
5176 | if (minsize == 0) | |
5177 | minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
5178 | for (dl = super->disks; dl ; dl = dl->next) { | |
5179 | int found = 0; | |
5180 | ||
5181 | pos = 0; | |
5182 | i = 0; | |
5183 | e = get_extents(super, dl); | |
5184 | if (!e) continue; | |
5185 | do { | |
5186 | unsigned long long esize; | |
5187 | esize = e[i].start - pos; | |
5188 | if (esize >= minsize) | |
5189 | found = 1; | |
5190 | if (found && start_offset == MaxSector) { | |
5191 | start_offset = pos; | |
5192 | break; | |
5193 | } else if (found && pos != start_offset) { | |
5194 | found = 0; | |
5195 | break; | |
5196 | } | |
5197 | pos = e[i].start + e[i].size; | |
5198 | i++; | |
5199 | } while (e[i-1].size); | |
5200 | if (found) | |
5201 | dcnt++; | |
5202 | free(e); | |
5203 | } | |
5204 | if (dcnt < raiddisks) { | |
5205 | if (verbose) | |
5206 | fprintf(stderr, Name ": imsm: Not enough " | |
5207 | "devices with space for this array " | |
5208 | "(%d < %d)\n", | |
5209 | dcnt, raiddisks); | |
5210 | return 0; | |
5211 | } | |
5212 | return 1; | |
5213 | } | |
5214 | ||
5215 | /* This device must be a member of the set */ | |
5216 | if (stat(dev, &stb) < 0) | |
5217 | return 0; | |
5218 | if ((S_IFMT & stb.st_mode) != S_IFBLK) | |
5219 | return 0; | |
5220 | for (dl = super->disks ; dl ; dl = dl->next) { | |
5221 | if (dl->major == (int)major(stb.st_rdev) && | |
5222 | dl->minor == (int)minor(stb.st_rdev)) | |
5223 | break; | |
5224 | } | |
5225 | if (!dl) { | |
5226 | if (verbose) | |
5227 | fprintf(stderr, Name ": %s is not in the " | |
5228 | "same imsm set\n", dev); | |
5229 | return 0; | |
5230 | } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) { | |
5231 | /* If a volume is present then the current creation attempt | |
5232 | * cannot incorporate new spares because the orom may not | |
5233 | * understand this configuration (all member disks must be | |
5234 | * members of each array in the container). | |
5235 | */ | |
5236 | fprintf(stderr, Name ": %s is a spare and a volume" | |
5237 | " is already defined for this container\n", dev); | |
5238 | fprintf(stderr, Name ": The option-rom requires all member" | |
5239 | " disks to be a member of all volumes\n"); | |
5240 | return 0; | |
5241 | } | |
5242 | ||
5243 | /* retrieve the largest free space block */ | |
5244 | e = get_extents(super, dl); | |
5245 | maxsize = 0; | |
5246 | i = 0; | |
5247 | if (e) { | |
5248 | do { | |
5249 | unsigned long long esize; | |
5250 | ||
5251 | esize = e[i].start - pos; | |
5252 | if (esize >= maxsize) | |
5253 | maxsize = esize; | |
5254 | pos = e[i].start + e[i].size; | |
5255 | i++; | |
5256 | } while (e[i-1].size); | |
5257 | dl->e = e; | |
5258 | dl->extent_cnt = i; | |
5259 | } else { | |
5260 | if (verbose) | |
5261 | fprintf(stderr, Name ": unable to determine free space for: %s\n", | |
5262 | dev); | |
5263 | return 0; | |
5264 | } | |
5265 | if (maxsize < size) { | |
5266 | if (verbose) | |
5267 | fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n", | |
5268 | dev, maxsize, size); | |
5269 | return 0; | |
5270 | } | |
5271 | ||
5272 | /* count total number of extents for merge */ | |
5273 | i = 0; | |
5274 | for (dl = super->disks; dl; dl = dl->next) | |
5275 | if (dl->e) | |
5276 | i += dl->extent_cnt; | |
5277 | ||
5278 | maxsize = merge_extents(super, i); | |
5279 | if (maxsize < size || maxsize == 0) { | |
5280 | if (verbose) | |
5281 | fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n", | |
5282 | maxsize, size); | |
5283 | return 0; | |
5284 | } | |
5285 | ||
5286 | *freesize = maxsize; | |
5287 | ||
5288 | return 1; | |
5289 | } | |
5290 | ||
5291 | static int reserve_space(struct supertype *st, int raiddisks, | |
5292 | unsigned long long size, int chunk, | |
5293 | unsigned long long *freesize) | |
5294 | { | |
5295 | struct intel_super *super = st->sb; | |
5296 | struct imsm_super *mpb = super->anchor; | |
5297 | struct dl *dl; | |
5298 | int i; | |
5299 | int extent_cnt; | |
5300 | struct extent *e; | |
5301 | unsigned long long maxsize; | |
5302 | unsigned long long minsize; | |
5303 | int cnt; | |
5304 | int used; | |
5305 | ||
5306 | /* find the largest common start free region of the possible disks */ | |
5307 | used = 0; | |
5308 | extent_cnt = 0; | |
5309 | cnt = 0; | |
5310 | for (dl = super->disks; dl; dl = dl->next) { | |
5311 | dl->raiddisk = -1; | |
5312 | ||
5313 | if (dl->index >= 0) | |
5314 | used++; | |
5315 | ||
5316 | /* don't activate new spares if we are orom constrained | |
5317 | * and there is already a volume active in the container | |
5318 | */ | |
5319 | if (super->orom && dl->index < 0 && mpb->num_raid_devs) | |
5320 | continue; | |
5321 | ||
5322 | e = get_extents(super, dl); | |
5323 | if (!e) | |
5324 | continue; | |
5325 | for (i = 1; e[i-1].size; i++) | |
5326 | ; | |
5327 | dl->e = e; | |
5328 | dl->extent_cnt = i; | |
5329 | extent_cnt += i; | |
5330 | cnt++; | |
5331 | } | |
5332 | ||
5333 | maxsize = merge_extents(super, extent_cnt); | |
5334 | minsize = size; | |
5335 | if (size == 0) | |
5336 | /* chunk is in K */ | |
5337 | minsize = chunk * 2; | |
5338 | ||
5339 | if (cnt < raiddisks || | |
5340 | (super->orom && used && used != raiddisks) || | |
5341 | maxsize < minsize || | |
5342 | maxsize == 0) { | |
5343 | fprintf(stderr, Name ": not enough devices with space to create array.\n"); | |
5344 | return 0; /* No enough free spaces large enough */ | |
5345 | } | |
5346 | ||
5347 | if (size == 0) { | |
5348 | size = maxsize; | |
5349 | if (chunk) { | |
5350 | size /= 2 * chunk; | |
5351 | size *= 2 * chunk; | |
5352 | } | |
5353 | } | |
5354 | ||
5355 | cnt = 0; | |
5356 | for (dl = super->disks; dl; dl = dl->next) | |
5357 | if (dl->e) | |
5358 | dl->raiddisk = cnt++; | |
5359 | ||
5360 | *freesize = size; | |
5361 | ||
5362 | return 1; | |
5363 | } | |
5364 | ||
5365 | static int validate_geometry_imsm(struct supertype *st, int level, int layout, | |
5366 | int raiddisks, int *chunk, unsigned long long size, | |
5367 | char *dev, unsigned long long *freesize, | |
5368 | int verbose) | |
5369 | { | |
5370 | int fd, cfd; | |
5371 | struct mdinfo *sra; | |
5372 | int is_member = 0; | |
5373 | ||
5374 | /* load capability | |
5375 | * if given unused devices create a container | |
5376 | * if given given devices in a container create a member volume | |
5377 | */ | |
5378 | if (level == LEVEL_CONTAINER) { | |
5379 | /* Must be a fresh device to add to a container */ | |
5380 | return validate_geometry_imsm_container(st, level, layout, | |
5381 | raiddisks, | |
5382 | chunk?*chunk:0, size, | |
5383 | dev, freesize, | |
5384 | verbose); | |
5385 | } | |
5386 | ||
5387 | if (!dev) { | |
5388 | if (st->sb && freesize) { | |
5389 | /* we are being asked to automatically layout a | |
5390 | * new volume based on the current contents of | |
5391 | * the container. If the the parameters can be | |
5392 | * satisfied reserve_space will record the disks, | |
5393 | * start offset, and size of the volume to be | |
5394 | * created. add_to_super and getinfo_super | |
5395 | * detect when autolayout is in progress. | |
5396 | */ | |
5397 | if (!validate_geometry_imsm_orom(st->sb, level, layout, | |
5398 | raiddisks, chunk, | |
5399 | verbose)) | |
5400 | return 0; | |
5401 | return reserve_space(st, raiddisks, size, | |
5402 | chunk?*chunk:0, freesize); | |
5403 | } | |
5404 | return 1; | |
5405 | } | |
5406 | if (st->sb) { | |
5407 | /* creating in a given container */ | |
5408 | return validate_geometry_imsm_volume(st, level, layout, | |
5409 | raiddisks, chunk, size, | |
5410 | dev, freesize, verbose); | |
5411 | } | |
5412 | ||
5413 | /* This device needs to be a device in an 'imsm' container */ | |
5414 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
5415 | if (fd >= 0) { | |
5416 | if (verbose) | |
5417 | fprintf(stderr, | |
5418 | Name ": Cannot create this array on device %s\n", | |
5419 | dev); | |
5420 | close(fd); | |
5421 | return 0; | |
5422 | } | |
5423 | if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) { | |
5424 | if (verbose) | |
5425 | fprintf(stderr, Name ": Cannot open %s: %s\n", | |
5426 | dev, strerror(errno)); | |
5427 | return 0; | |
5428 | } | |
5429 | /* Well, it is in use by someone, maybe an 'imsm' container. */ | |
5430 | cfd = open_container(fd); | |
5431 | close(fd); | |
5432 | if (cfd < 0) { | |
5433 | if (verbose) | |
5434 | fprintf(stderr, Name ": Cannot use %s: It is busy\n", | |
5435 | dev); | |
5436 | return 0; | |
5437 | } | |
5438 | sra = sysfs_read(cfd, 0, GET_VERSION); | |
5439 | if (sra && sra->array.major_version == -1 && | |
5440 | strcmp(sra->text_version, "imsm") == 0) | |
5441 | is_member = 1; | |
5442 | sysfs_free(sra); | |
5443 | if (is_member) { | |
5444 | /* This is a member of a imsm container. Load the container | |
5445 | * and try to create a volume | |
5446 | */ | |
5447 | struct intel_super *super; | |
5448 | ||
5449 | if (load_super_imsm_all(st, cfd, (void **) &super, NULL) == 0) { | |
5450 | st->sb = super; | |
5451 | st->container_dev = fd2devnum(cfd); | |
5452 | close(cfd); | |
5453 | return validate_geometry_imsm_volume(st, level, layout, | |
5454 | raiddisks, chunk, | |
5455 | size, dev, | |
5456 | freesize, 1) | |
5457 | ? 1 : -1; | |
5458 | } | |
5459 | } | |
5460 | ||
5461 | if (verbose) | |
5462 | fprintf(stderr, Name ": failed container membership check\n"); | |
5463 | ||
5464 | close(cfd); | |
5465 | return 0; | |
5466 | } | |
5467 | ||
5468 | static void default_geometry_imsm(struct supertype *st, int *level, int *layout, int *chunk) | |
5469 | { | |
5470 | struct intel_super *super = st->sb; | |
5471 | ||
5472 | if (level && *level == UnSet) | |
5473 | *level = LEVEL_CONTAINER; | |
5474 | ||
5475 | if (level && layout && *layout == UnSet) | |
5476 | *layout = imsm_level_to_layout(*level); | |
5477 | ||
5478 | if (chunk && (*chunk == UnSet || *chunk == 0)) | |
5479 | *chunk = imsm_default_chunk(super->orom); | |
5480 | } | |
5481 | ||
5482 | static void handle_missing(struct intel_super *super, struct imsm_dev *dev); | |
5483 | ||
5484 | static int kill_subarray_imsm(struct supertype *st) | |
5485 | { | |
5486 | /* remove the subarray currently referenced by ->current_vol */ | |
5487 | __u8 i; | |
5488 | struct intel_dev **dp; | |
5489 | struct intel_super *super = st->sb; | |
5490 | __u8 current_vol = super->current_vol; | |
5491 | struct imsm_super *mpb = super->anchor; | |
5492 | ||
5493 | if (super->current_vol < 0) | |
5494 | return 2; | |
5495 | super->current_vol = -1; /* invalidate subarray cursor */ | |
5496 | ||
5497 | /* block deletions that would change the uuid of active subarrays | |
5498 | * | |
5499 | * FIXME when immutable ids are available, but note that we'll | |
5500 | * also need to fixup the invalidated/active subarray indexes in | |
5501 | * mdstat | |
5502 | */ | |
5503 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5504 | char subarray[4]; | |
5505 | ||
5506 | if (i < current_vol) | |
5507 | continue; | |
5508 | sprintf(subarray, "%u", i); | |
5509 | if (is_subarray_active(subarray, st->devname)) { | |
5510 | fprintf(stderr, | |
5511 | Name ": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n", | |
5512 | current_vol, i); | |
5513 | ||
5514 | return 2; | |
5515 | } | |
5516 | } | |
5517 | ||
5518 | if (st->update_tail) { | |
5519 | struct imsm_update_kill_array *u = malloc(sizeof(*u)); | |
5520 | ||
5521 | if (!u) | |
5522 | return 2; | |
5523 | u->type = update_kill_array; | |
5524 | u->dev_idx = current_vol; | |
5525 | append_metadata_update(st, u, sizeof(*u)); | |
5526 | ||
5527 | return 0; | |
5528 | } | |
5529 | ||
5530 | for (dp = &super->devlist; *dp;) | |
5531 | if ((*dp)->index == current_vol) { | |
5532 | *dp = (*dp)->next; | |
5533 | } else { | |
5534 | handle_missing(super, (*dp)->dev); | |
5535 | if ((*dp)->index > current_vol) | |
5536 | (*dp)->index--; | |
5537 | dp = &(*dp)->next; | |
5538 | } | |
5539 | ||
5540 | /* no more raid devices, all active components are now spares, | |
5541 | * but of course failed are still failed | |
5542 | */ | |
5543 | if (--mpb->num_raid_devs == 0) { | |
5544 | struct dl *d; | |
5545 | ||
5546 | for (d = super->disks; d; d = d->next) | |
5547 | if (d->index > -2) | |
5548 | mark_spare(d); | |
5549 | } | |
5550 | ||
5551 | super->updates_pending++; | |
5552 | ||
5553 | return 0; | |
5554 | } | |
5555 | ||
5556 | static int update_subarray_imsm(struct supertype *st, char *subarray, | |
5557 | char *update, struct mddev_ident *ident) | |
5558 | { | |
5559 | /* update the subarray currently referenced by ->current_vol */ | |
5560 | struct intel_super *super = st->sb; | |
5561 | struct imsm_super *mpb = super->anchor; | |
5562 | ||
5563 | if (strcmp(update, "name") == 0) { | |
5564 | char *name = ident->name; | |
5565 | char *ep; | |
5566 | int vol; | |
5567 | ||
5568 | if (is_subarray_active(subarray, st->devname)) { | |
5569 | fprintf(stderr, | |
5570 | Name ": Unable to update name of active subarray\n"); | |
5571 | return 2; | |
5572 | } | |
5573 | ||
5574 | if (!check_name(super, name, 0)) | |
5575 | return 2; | |
5576 | ||
5577 | vol = strtoul(subarray, &ep, 10); | |
5578 | if (*ep != '\0' || vol >= super->anchor->num_raid_devs) | |
5579 | return 2; | |
5580 | ||
5581 | if (st->update_tail) { | |
5582 | struct imsm_update_rename_array *u = malloc(sizeof(*u)); | |
5583 | ||
5584 | if (!u) | |
5585 | return 2; | |
5586 | u->type = update_rename_array; | |
5587 | u->dev_idx = vol; | |
5588 | snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name); | |
5589 | append_metadata_update(st, u, sizeof(*u)); | |
5590 | } else { | |
5591 | struct imsm_dev *dev; | |
5592 | int i; | |
5593 | ||
5594 | dev = get_imsm_dev(super, vol); | |
5595 | snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name); | |
5596 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5597 | dev = get_imsm_dev(super, i); | |
5598 | handle_missing(super, dev); | |
5599 | } | |
5600 | super->updates_pending++; | |
5601 | } | |
5602 | } else | |
5603 | return 2; | |
5604 | ||
5605 | return 0; | |
5606 | } | |
5607 | ||
5608 | static int is_gen_migration(struct imsm_dev *dev) | |
5609 | { | |
5610 | if (dev == NULL) | |
5611 | return 0; | |
5612 | ||
5613 | if (!dev->vol.migr_state) | |
5614 | return 0; | |
5615 | ||
5616 | if (migr_type(dev) == MIGR_GEN_MIGR) | |
5617 | return 1; | |
5618 | ||
5619 | return 0; | |
5620 | } | |
5621 | #endif /* MDASSEMBLE */ | |
5622 | ||
5623 | static int is_rebuilding(struct imsm_dev *dev) | |
5624 | { | |
5625 | struct imsm_map *migr_map; | |
5626 | ||
5627 | if (!dev->vol.migr_state) | |
5628 | return 0; | |
5629 | ||
5630 | if (migr_type(dev) != MIGR_REBUILD) | |
5631 | return 0; | |
5632 | ||
5633 | migr_map = get_imsm_map(dev, 1); | |
5634 | ||
5635 | if (migr_map->map_state == IMSM_T_STATE_DEGRADED) | |
5636 | return 1; | |
5637 | else | |
5638 | return 0; | |
5639 | } | |
5640 | ||
5641 | static void update_recovery_start(struct intel_super *super, | |
5642 | struct imsm_dev *dev, | |
5643 | struct mdinfo *array) | |
5644 | { | |
5645 | struct mdinfo *rebuild = NULL; | |
5646 | struct mdinfo *d; | |
5647 | __u32 units; | |
5648 | ||
5649 | if (!is_rebuilding(dev)) | |
5650 | return; | |
5651 | ||
5652 | /* Find the rebuild target, but punt on the dual rebuild case */ | |
5653 | for (d = array->devs; d; d = d->next) | |
5654 | if (d->recovery_start == 0) { | |
5655 | if (rebuild) | |
5656 | return; | |
5657 | rebuild = d; | |
5658 | } | |
5659 | ||
5660 | if (!rebuild) { | |
5661 | /* (?) none of the disks are marked with | |
5662 | * IMSM_ORD_REBUILD, so assume they are missing and the | |
5663 | * disk_ord_tbl was not correctly updated | |
5664 | */ | |
5665 | dprintf("%s: failed to locate out-of-sync disk\n", __func__); | |
5666 | return; | |
5667 | } | |
5668 | ||
5669 | units = __le32_to_cpu(dev->vol.curr_migr_unit); | |
5670 | rebuild->recovery_start = units * blocks_per_migr_unit(super, dev); | |
5671 | } | |
5672 | ||
5673 | #ifndef MDASSEMBLE | |
5674 | static int recover_backup_imsm(struct supertype *st, struct mdinfo *info); | |
5675 | #endif | |
5676 | ||
5677 | static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray) | |
5678 | { | |
5679 | /* Given a container loaded by load_super_imsm_all, | |
5680 | * extract information about all the arrays into | |
5681 | * an mdinfo tree. | |
5682 | * If 'subarray' is given, just extract info about that array. | |
5683 | * | |
5684 | * For each imsm_dev create an mdinfo, fill it in, | |
5685 | * then look for matching devices in super->disks | |
5686 | * and create appropriate device mdinfo. | |
5687 | */ | |
5688 | struct intel_super *super = st->sb; | |
5689 | struct imsm_super *mpb = super->anchor; | |
5690 | struct mdinfo *rest = NULL; | |
5691 | unsigned int i; | |
5692 | int sb_errors = 0; | |
5693 | struct dl *d; | |
5694 | int spare_disks = 0; | |
5695 | ||
5696 | /* do not assemble arrays when not all attributes are supported */ | |
5697 | if (imsm_check_attributes(mpb->attributes) == 0) { | |
5698 | sb_errors = 1; | |
5699 | fprintf(stderr, Name ": Unsupported attributes in IMSM metadata." | |
5700 | "Arrays activation is blocked.\n"); | |
5701 | } | |
5702 | ||
5703 | /* check for bad blocks */ | |
5704 | if (imsm_bbm_log_size(super->anchor)) { | |
5705 | fprintf(stderr, Name ": BBM log found in IMSM metadata." | |
5706 | "Arrays activation is blocked.\n"); | |
5707 | sb_errors = 1; | |
5708 | } | |
5709 | ||
5710 | ||
5711 | /* count spare devices, not used in maps | |
5712 | */ | |
5713 | for (d = super->disks; d; d = d->next) | |
5714 | if (d->index == -1) | |
5715 | spare_disks++; | |
5716 | ||
5717 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5718 | struct imsm_dev *dev; | |
5719 | struct imsm_map *map; | |
5720 | struct imsm_map *map2; | |
5721 | struct mdinfo *this; | |
5722 | int slot, chunk; | |
5723 | char *ep; | |
5724 | ||
5725 | if (subarray && | |
5726 | (i != strtoul(subarray, &ep, 10) || *ep != '\0')) | |
5727 | continue; | |
5728 | ||
5729 | dev = get_imsm_dev(super, i); | |
5730 | map = get_imsm_map(dev, 0); | |
5731 | map2 = get_imsm_map(dev, 1); | |
5732 | ||
5733 | /* do not publish arrays that are in the middle of an | |
5734 | * unsupported migration | |
5735 | */ | |
5736 | if (dev->vol.migr_state && | |
5737 | (migr_type(dev) == MIGR_STATE_CHANGE)) { | |
5738 | fprintf(stderr, Name ": cannot assemble volume '%.16s':" | |
5739 | " unsupported migration in progress\n", | |
5740 | dev->volume); | |
5741 | continue; | |
5742 | } | |
5743 | /* do not publish arrays that are not support by controller's | |
5744 | * OROM/EFI | |
5745 | */ | |
5746 | ||
5747 | chunk = __le16_to_cpu(map->blocks_per_strip) >> 1; | |
5748 | this = malloc(sizeof(*this)); | |
5749 | if (!this) { | |
5750 | fprintf(stderr, Name ": failed to allocate %zu bytes\n", | |
5751 | sizeof(*this)); | |
5752 | break; | |
5753 | } | |
5754 | ||
5755 | super->current_vol = i; | |
5756 | getinfo_super_imsm_volume(st, this, NULL); | |
5757 | this->next = rest; | |
5758 | #ifndef MDASSEMBLE | |
5759 | /* mdadm does not support all metadata features- set the bit in all arrays state */ | |
5760 | if (!validate_geometry_imsm_orom(super, | |
5761 | get_imsm_raid_level(map), /* RAID level */ | |
5762 | imsm_level_to_layout(get_imsm_raid_level(map)), | |
5763 | map->num_members, /* raid disks */ | |
5764 | &chunk, | |
5765 | 1 /* verbose */)) { | |
5766 | fprintf(stderr, Name ": IMSM RAID geometry validation" | |
5767 | " failed. Array %s activation is blocked.\n", | |
5768 | dev->volume); | |
5769 | this->array.state |= | |
5770 | (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) | | |
5771 | (1<<MD_SB_BLOCK_VOLUME); | |
5772 | } | |
5773 | #endif | |
5774 | ||
5775 | /* if array has bad blocks, set suitable bit in all arrays state */ | |
5776 | if (sb_errors) | |
5777 | this->array.state |= | |
5778 | (1<<MD_SB_BLOCK_CONTAINER_RESHAPE) | | |
5779 | (1<<MD_SB_BLOCK_VOLUME); | |
5780 | ||
5781 | for (slot = 0 ; slot < map->num_members; slot++) { | |
5782 | unsigned long long recovery_start; | |
5783 | struct mdinfo *info_d; | |
5784 | struct dl *d; | |
5785 | int idx; | |
5786 | int skip; | |
5787 | __u32 ord; | |
5788 | ||
5789 | skip = 0; | |
5790 | idx = get_imsm_disk_idx(dev, slot, 0); | |
5791 | ord = get_imsm_ord_tbl_ent(dev, slot, -1); | |
5792 | for (d = super->disks; d ; d = d->next) | |
5793 | if (d->index == idx) | |
5794 | break; | |
5795 | ||
5796 | recovery_start = MaxSector; | |
5797 | if (d == NULL) | |
5798 | skip = 1; | |
5799 | if (d && is_failed(&d->disk)) | |
5800 | skip = 1; | |
5801 | if (ord & IMSM_ORD_REBUILD) | |
5802 | recovery_start = 0; | |
5803 | ||
5804 | /* | |
5805 | * if we skip some disks the array will be assmebled degraded; | |
5806 | * reset resync start to avoid a dirty-degraded | |
5807 | * situation when performing the intial sync | |
5808 | * | |
5809 | * FIXME handle dirty degraded | |
5810 | */ | |
5811 | if ((skip || recovery_start == 0) && !dev->vol.dirty) | |
5812 | this->resync_start = MaxSector; | |
5813 | if (skip) | |
5814 | continue; | |
5815 | ||
5816 | info_d = calloc(1, sizeof(*info_d)); | |
5817 | if (!info_d) { | |
5818 | fprintf(stderr, Name ": failed to allocate disk" | |
5819 | " for volume %.16s\n", dev->volume); | |
5820 | info_d = this->devs; | |
5821 | while (info_d) { | |
5822 | struct mdinfo *d = info_d->next; | |
5823 | ||
5824 | free(info_d); | |
5825 | info_d = d; | |
5826 | } | |
5827 | free(this); | |
5828 | this = rest; | |
5829 | break; | |
5830 | } | |
5831 | info_d->next = this->devs; | |
5832 | this->devs = info_d; | |
5833 | ||
5834 | info_d->disk.number = d->index; | |
5835 | info_d->disk.major = d->major; | |
5836 | info_d->disk.minor = d->minor; | |
5837 | info_d->disk.raid_disk = slot; | |
5838 | info_d->recovery_start = recovery_start; | |
5839 | if (map2) { | |
5840 | if (slot < map2->num_members) | |
5841 | info_d->disk.state = (1 << MD_DISK_ACTIVE); | |
5842 | else | |
5843 | this->array.spare_disks++; | |
5844 | } else { | |
5845 | if (slot < map->num_members) | |
5846 | info_d->disk.state = (1 << MD_DISK_ACTIVE); | |
5847 | else | |
5848 | this->array.spare_disks++; | |
5849 | } | |
5850 | if (info_d->recovery_start == MaxSector) | |
5851 | this->array.working_disks++; | |
5852 | ||
5853 | info_d->events = __le32_to_cpu(mpb->generation_num); | |
5854 | info_d->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
5855 | info_d->component_size = __le32_to_cpu(map->blocks_per_member); | |
5856 | } | |
5857 | /* now that the disk list is up-to-date fixup recovery_start */ | |
5858 | update_recovery_start(super, dev, this); | |
5859 | this->array.spare_disks += spare_disks; | |
5860 | ||
5861 | #ifndef MDASSEMBLE | |
5862 | /* check for reshape */ | |
5863 | if (this->reshape_active == 1) | |
5864 | recover_backup_imsm(st, this); | |
5865 | #endif | |
5866 | rest = this; | |
5867 | } | |
5868 | ||
5869 | return rest; | |
5870 | } | |
5871 | ||
5872 | ||
5873 | static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed) | |
5874 | { | |
5875 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5876 | ||
5877 | if (!failed) | |
5878 | return map->map_state == IMSM_T_STATE_UNINITIALIZED ? | |
5879 | IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL; | |
5880 | ||
5881 | switch (get_imsm_raid_level(map)) { | |
5882 | case 0: | |
5883 | return IMSM_T_STATE_FAILED; | |
5884 | break; | |
5885 | case 1: | |
5886 | if (failed < map->num_members) | |
5887 | return IMSM_T_STATE_DEGRADED; | |
5888 | else | |
5889 | return IMSM_T_STATE_FAILED; | |
5890 | break; | |
5891 | case 10: | |
5892 | { | |
5893 | /** | |
5894 | * check to see if any mirrors have failed, otherwise we | |
5895 | * are degraded. Even numbered slots are mirrored on | |
5896 | * slot+1 | |
5897 | */ | |
5898 | int i; | |
5899 | /* gcc -Os complains that this is unused */ | |
5900 | int insync = insync; | |
5901 | ||
5902 | for (i = 0; i < map->num_members; i++) { | |
5903 | __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1); | |
5904 | int idx = ord_to_idx(ord); | |
5905 | struct imsm_disk *disk; | |
5906 | ||
5907 | /* reset the potential in-sync count on even-numbered | |
5908 | * slots. num_copies is always 2 for imsm raid10 | |
5909 | */ | |
5910 | if ((i & 1) == 0) | |
5911 | insync = 2; | |
5912 | ||
5913 | disk = get_imsm_disk(super, idx); | |
5914 | if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD) | |
5915 | insync--; | |
5916 | ||
5917 | /* no in-sync disks left in this mirror the | |
5918 | * array has failed | |
5919 | */ | |
5920 | if (insync == 0) | |
5921 | return IMSM_T_STATE_FAILED; | |
5922 | } | |
5923 | ||
5924 | return IMSM_T_STATE_DEGRADED; | |
5925 | } | |
5926 | case 5: | |
5927 | if (failed < 2) | |
5928 | return IMSM_T_STATE_DEGRADED; | |
5929 | else | |
5930 | return IMSM_T_STATE_FAILED; | |
5931 | break; | |
5932 | default: | |
5933 | break; | |
5934 | } | |
5935 | ||
5936 | return map->map_state; | |
5937 | } | |
5938 | ||
5939 | static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev) | |
5940 | { | |
5941 | int i; | |
5942 | int failed = 0; | |
5943 | struct imsm_disk *disk; | |
5944 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5945 | struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state); | |
5946 | __u32 ord; | |
5947 | int idx; | |
5948 | ||
5949 | /* at the beginning of migration we set IMSM_ORD_REBUILD on | |
5950 | * disks that are being rebuilt. New failures are recorded to | |
5951 | * map[0]. So we look through all the disks we started with and | |
5952 | * see if any failures are still present, or if any new ones | |
5953 | * have arrived | |
5954 | * | |
5955 | * FIXME add support for online capacity expansion and | |
5956 | * raid-level-migration | |
5957 | */ | |
5958 | for (i = 0; i < prev->num_members; i++) { | |
5959 | ord = __le32_to_cpu(prev->disk_ord_tbl[i]); | |
5960 | ord |= __le32_to_cpu(map->disk_ord_tbl[i]); | |
5961 | idx = ord_to_idx(ord); | |
5962 | ||
5963 | disk = get_imsm_disk(super, idx); | |
5964 | if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD) | |
5965 | failed++; | |
5966 | } | |
5967 | ||
5968 | return failed; | |
5969 | } | |
5970 | ||
5971 | #ifndef MDASSEMBLE | |
5972 | static int imsm_open_new(struct supertype *c, struct active_array *a, | |
5973 | char *inst) | |
5974 | { | |
5975 | struct intel_super *super = c->sb; | |
5976 | struct imsm_super *mpb = super->anchor; | |
5977 | ||
5978 | if (atoi(inst) >= mpb->num_raid_devs) { | |
5979 | fprintf(stderr, "%s: subarry index %d, out of range\n", | |
5980 | __func__, atoi(inst)); | |
5981 | return -ENODEV; | |
5982 | } | |
5983 | ||
5984 | dprintf("imsm: open_new %s\n", inst); | |
5985 | a->info.container_member = atoi(inst); | |
5986 | return 0; | |
5987 | } | |
5988 | ||
5989 | static int is_resyncing(struct imsm_dev *dev) | |
5990 | { | |
5991 | struct imsm_map *migr_map; | |
5992 | ||
5993 | if (!dev->vol.migr_state) | |
5994 | return 0; | |
5995 | ||
5996 | if (migr_type(dev) == MIGR_INIT || | |
5997 | migr_type(dev) == MIGR_REPAIR) | |
5998 | return 1; | |
5999 | ||
6000 | if (migr_type(dev) == MIGR_GEN_MIGR) | |
6001 | return 0; | |
6002 | ||
6003 | migr_map = get_imsm_map(dev, 1); | |
6004 | ||
6005 | if ((migr_map->map_state == IMSM_T_STATE_NORMAL) && | |
6006 | (dev->vol.migr_type != MIGR_GEN_MIGR)) | |
6007 | return 1; | |
6008 | else | |
6009 | return 0; | |
6010 | } | |
6011 | ||
6012 | /* return true if we recorded new information */ | |
6013 | static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx) | |
6014 | { | |
6015 | __u32 ord; | |
6016 | int slot; | |
6017 | struct imsm_map *map; | |
6018 | char buf[MAX_RAID_SERIAL_LEN+3]; | |
6019 | unsigned int len, shift = 0; | |
6020 | ||
6021 | /* new failures are always set in map[0] */ | |
6022 | map = get_imsm_map(dev, 0); | |
6023 | ||
6024 | slot = get_imsm_disk_slot(map, idx); | |
6025 | if (slot < 0) | |
6026 | return 0; | |
6027 | ||
6028 | ord = __le32_to_cpu(map->disk_ord_tbl[slot]); | |
6029 | if (is_failed(disk) && (ord & IMSM_ORD_REBUILD)) | |
6030 | return 0; | |
6031 | ||
6032 | memcpy(buf, disk->serial, MAX_RAID_SERIAL_LEN); | |
6033 | buf[MAX_RAID_SERIAL_LEN] = '\000'; | |
6034 | strcat(buf, ":0"); | |
6035 | if ((len = strlen(buf)) >= MAX_RAID_SERIAL_LEN) | |
6036 | shift = len - MAX_RAID_SERIAL_LEN + 1; | |
6037 | strncpy((char *)disk->serial, &buf[shift], MAX_RAID_SERIAL_LEN); | |
6038 | ||
6039 | disk->status |= FAILED_DISK; | |
6040 | set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD); | |
6041 | if (map->failed_disk_num == 0xff) | |
6042 | map->failed_disk_num = slot; | |
6043 | return 1; | |
6044 | } | |
6045 | ||
6046 | static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx) | |
6047 | { | |
6048 | mark_failure(dev, disk, idx); | |
6049 | ||
6050 | if (disk->scsi_id == __cpu_to_le32(~(__u32)0)) | |
6051 | return; | |
6052 | ||
6053 | disk->scsi_id = __cpu_to_le32(~(__u32)0); | |
6054 | memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1); | |
6055 | } | |
6056 | ||
6057 | static void handle_missing(struct intel_super *super, struct imsm_dev *dev) | |
6058 | { | |
6059 | __u8 map_state; | |
6060 | struct dl *dl; | |
6061 | int failed; | |
6062 | ||
6063 | if (!super->missing) | |
6064 | return; | |
6065 | failed = imsm_count_failed(super, dev); | |
6066 | map_state = imsm_check_degraded(super, dev, failed); | |
6067 | ||
6068 | dprintf("imsm: mark missing\n"); | |
6069 | end_migration(dev, map_state); | |
6070 | for (dl = super->missing; dl; dl = dl->next) | |
6071 | mark_missing(dev, &dl->disk, dl->index); | |
6072 | super->updates_pending++; | |
6073 | } | |
6074 | ||
6075 | static unsigned long long imsm_set_array_size(struct imsm_dev *dev) | |
6076 | { | |
6077 | int used_disks = imsm_num_data_members(dev, 0); | |
6078 | unsigned long long array_blocks; | |
6079 | struct imsm_map *map; | |
6080 | ||
6081 | if (used_disks == 0) { | |
6082 | /* when problems occures | |
6083 | * return current array_blocks value | |
6084 | */ | |
6085 | array_blocks = __le32_to_cpu(dev->size_high); | |
6086 | array_blocks = array_blocks << 32; | |
6087 | array_blocks += __le32_to_cpu(dev->size_low); | |
6088 | ||
6089 | return array_blocks; | |
6090 | } | |
6091 | ||
6092 | /* set array size in metadata | |
6093 | */ | |
6094 | map = get_imsm_map(dev, 0); | |
6095 | array_blocks = map->blocks_per_member * used_disks; | |
6096 | ||
6097 | /* round array size down to closest MB | |
6098 | */ | |
6099 | array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT; | |
6100 | dev->size_low = __cpu_to_le32((__u32)array_blocks); | |
6101 | dev->size_high = __cpu_to_le32((__u32)(array_blocks >> 32)); | |
6102 | ||
6103 | return array_blocks; | |
6104 | } | |
6105 | ||
6106 | static void imsm_set_disk(struct active_array *a, int n, int state); | |
6107 | ||
6108 | static void imsm_progress_container_reshape(struct intel_super *super) | |
6109 | { | |
6110 | /* if no device has a migr_state, but some device has a | |
6111 | * different number of members than the previous device, start | |
6112 | * changing the number of devices in this device to match | |
6113 | * previous. | |
6114 | */ | |
6115 | struct imsm_super *mpb = super->anchor; | |
6116 | int prev_disks = -1; | |
6117 | int i; | |
6118 | int copy_map_size; | |
6119 | ||
6120 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
6121 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
6122 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6123 | struct imsm_map *map2; | |
6124 | int prev_num_members; | |
6125 | ||
6126 | if (dev->vol.migr_state) | |
6127 | return; | |
6128 | ||
6129 | if (prev_disks == -1) | |
6130 | prev_disks = map->num_members; | |
6131 | if (prev_disks == map->num_members) | |
6132 | continue; | |
6133 | ||
6134 | /* OK, this array needs to enter reshape mode. | |
6135 | * i.e it needs a migr_state | |
6136 | */ | |
6137 | ||
6138 | copy_map_size = sizeof_imsm_map(map); | |
6139 | prev_num_members = map->num_members; | |
6140 | map->num_members = prev_disks; | |
6141 | dev->vol.migr_state = 1; | |
6142 | dev->vol.curr_migr_unit = 0; | |
6143 | set_migr_type(dev, MIGR_GEN_MIGR); | |
6144 | for (i = prev_num_members; | |
6145 | i < map->num_members; i++) | |
6146 | set_imsm_ord_tbl_ent(map, i, i); | |
6147 | map2 = get_imsm_map(dev, 1); | |
6148 | /* Copy the current map */ | |
6149 | memcpy(map2, map, copy_map_size); | |
6150 | map2->num_members = prev_num_members; | |
6151 | ||
6152 | imsm_set_array_size(dev); | |
6153 | super->updates_pending++; | |
6154 | } | |
6155 | } | |
6156 | ||
6157 | /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild | |
6158 | * states are handled in imsm_set_disk() with one exception, when a | |
6159 | * resync is stopped due to a new failure this routine will set the | |
6160 | * 'degraded' state for the array. | |
6161 | */ | |
6162 | static int imsm_set_array_state(struct active_array *a, int consistent) | |
6163 | { | |
6164 | int inst = a->info.container_member; | |
6165 | struct intel_super *super = a->container->sb; | |
6166 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
6167 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6168 | int failed = imsm_count_failed(super, dev); | |
6169 | __u8 map_state = imsm_check_degraded(super, dev, failed); | |
6170 | __u32 blocks_per_unit; | |
6171 | ||
6172 | if (dev->vol.migr_state && | |
6173 | dev->vol.migr_type == MIGR_GEN_MIGR) { | |
6174 | /* array state change is blocked due to reshape action | |
6175 | * We might need to | |
6176 | * - abort the reshape (if last_checkpoint is 0 and action!= reshape) | |
6177 | * - finish the reshape (if last_checkpoint is big and action != reshape) | |
6178 | * - update curr_migr_unit | |
6179 | */ | |
6180 | if (a->curr_action == reshape) { | |
6181 | /* still reshaping, maybe update curr_migr_unit */ | |
6182 | goto mark_checkpoint; | |
6183 | } else { | |
6184 | if (a->last_checkpoint == 0 && a->prev_action == reshape) { | |
6185 | /* for some reason we aborted the reshape. | |
6186 | * | |
6187 | * disable automatic metadata rollback | |
6188 | * user action is required to recover process | |
6189 | */ | |
6190 | if (0) { | |
6191 | struct imsm_map *map2 = get_imsm_map(dev, 1); | |
6192 | dev->vol.migr_state = 0; | |
6193 | set_migr_type(dev, 0); | |
6194 | dev->vol.curr_migr_unit = 0; | |
6195 | memcpy(map, map2, sizeof_imsm_map(map2)); | |
6196 | super->updates_pending++; | |
6197 | } | |
6198 | } | |
6199 | if (a->last_checkpoint >= a->info.component_size) { | |
6200 | unsigned long long array_blocks; | |
6201 | int used_disks; | |
6202 | struct mdinfo *mdi; | |
6203 | ||
6204 | used_disks = imsm_num_data_members(dev, 0); | |
6205 | if (used_disks > 0) { | |
6206 | array_blocks = | |
6207 | map->blocks_per_member * | |
6208 | used_disks; | |
6209 | /* round array size down to closest MB | |
6210 | */ | |
6211 | array_blocks = (array_blocks | |
6212 | >> SECT_PER_MB_SHIFT) | |
6213 | << SECT_PER_MB_SHIFT; | |
6214 | a->info.custom_array_size = array_blocks; | |
6215 | /* encourage manager to update array | |
6216 | * size | |
6217 | */ | |
6218 | ||
6219 | a->check_reshape = 1; | |
6220 | } | |
6221 | /* finalize online capacity expansion/reshape */ | |
6222 | for (mdi = a->info.devs; mdi; mdi = mdi->next) | |
6223 | imsm_set_disk(a, | |
6224 | mdi->disk.raid_disk, | |
6225 | mdi->curr_state); | |
6226 | ||
6227 | imsm_progress_container_reshape(super); | |
6228 | } | |
6229 | } | |
6230 | } | |
6231 | ||
6232 | /* before we activate this array handle any missing disks */ | |
6233 | if (consistent == 2) | |
6234 | handle_missing(super, dev); | |
6235 | ||
6236 | if (consistent == 2 && | |
6237 | (!is_resync_complete(&a->info) || | |
6238 | map_state != IMSM_T_STATE_NORMAL || | |
6239 | dev->vol.migr_state)) | |
6240 | consistent = 0; | |
6241 | ||
6242 | if (is_resync_complete(&a->info)) { | |
6243 | /* complete intialization / resync, | |
6244 | * recovery and interrupted recovery is completed in | |
6245 | * ->set_disk | |
6246 | */ | |
6247 | if (is_resyncing(dev)) { | |
6248 | dprintf("imsm: mark resync done\n"); | |
6249 | end_migration(dev, map_state); | |
6250 | super->updates_pending++; | |
6251 | a->last_checkpoint = 0; | |
6252 | } | |
6253 | } else if ((!is_resyncing(dev) && !failed) && | |
6254 | (imsm_reshape_blocks_arrays_changes(super) == 0)) { | |
6255 | /* mark the start of the init process if nothing is failed */ | |
6256 | dprintf("imsm: mark resync start\n"); | |
6257 | if (map->map_state == IMSM_T_STATE_UNINITIALIZED) | |
6258 | migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_INIT); | |
6259 | else | |
6260 | migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_REPAIR); | |
6261 | super->updates_pending++; | |
6262 | } | |
6263 | ||
6264 | mark_checkpoint: | |
6265 | /* skip checkpointing for general migration, | |
6266 | * it is controlled in mdadm | |
6267 | */ | |
6268 | if (is_gen_migration(dev)) | |
6269 | goto skip_mark_checkpoint; | |
6270 | ||
6271 | /* check if we can update curr_migr_unit from resync_start, recovery_start */ | |
6272 | blocks_per_unit = blocks_per_migr_unit(super, dev); | |
6273 | if (blocks_per_unit) { | |
6274 | __u32 units32; | |
6275 | __u64 units; | |
6276 | ||
6277 | units = a->last_checkpoint / blocks_per_unit; | |
6278 | units32 = units; | |
6279 | ||
6280 | /* check that we did not overflow 32-bits, and that | |
6281 | * curr_migr_unit needs updating | |
6282 | */ | |
6283 | if (units32 == units && | |
6284 | units32 != 0 && | |
6285 | __le32_to_cpu(dev->vol.curr_migr_unit) != units32) { | |
6286 | dprintf("imsm: mark checkpoint (%u)\n", units32); | |
6287 | dev->vol.curr_migr_unit = __cpu_to_le32(units32); | |
6288 | super->updates_pending++; | |
6289 | } | |
6290 | } | |
6291 | ||
6292 | skip_mark_checkpoint: | |
6293 | /* mark dirty / clean */ | |
6294 | if (dev->vol.dirty != !consistent) { | |
6295 | dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty"); | |
6296 | if (consistent) | |
6297 | dev->vol.dirty = 0; | |
6298 | else | |
6299 | dev->vol.dirty = 1; | |
6300 | super->updates_pending++; | |
6301 | } | |
6302 | ||
6303 | return consistent; | |
6304 | } | |
6305 | ||
6306 | static void imsm_set_disk(struct active_array *a, int n, int state) | |
6307 | { | |
6308 | int inst = a->info.container_member; | |
6309 | struct intel_super *super = a->container->sb; | |
6310 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
6311 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6312 | struct imsm_disk *disk; | |
6313 | int failed; | |
6314 | __u32 ord; | |
6315 | __u8 map_state; | |
6316 | ||
6317 | if (n > map->num_members) | |
6318 | fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n", | |
6319 | n, map->num_members - 1); | |
6320 | ||
6321 | if (n < 0) | |
6322 | return; | |
6323 | ||
6324 | dprintf("imsm: set_disk %d:%x\n", n, state); | |
6325 | ||
6326 | ord = get_imsm_ord_tbl_ent(dev, n, -1); | |
6327 | disk = get_imsm_disk(super, ord_to_idx(ord)); | |
6328 | ||
6329 | /* check for new failures */ | |
6330 | if (state & DS_FAULTY) { | |
6331 | if (mark_failure(dev, disk, ord_to_idx(ord))) | |
6332 | super->updates_pending++; | |
6333 | } | |
6334 | ||
6335 | /* check if in_sync */ | |
6336 | if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) { | |
6337 | struct imsm_map *migr_map = get_imsm_map(dev, 1); | |
6338 | ||
6339 | set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord)); | |
6340 | super->updates_pending++; | |
6341 | } | |
6342 | ||
6343 | failed = imsm_count_failed(super, dev); | |
6344 | map_state = imsm_check_degraded(super, dev, failed); | |
6345 | ||
6346 | /* check if recovery complete, newly degraded, or failed */ | |
6347 | if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) { | |
6348 | end_migration(dev, map_state); | |
6349 | map = get_imsm_map(dev, 0); | |
6350 | map->failed_disk_num = ~0; | |
6351 | super->updates_pending++; | |
6352 | a->last_checkpoint = 0; | |
6353 | } else if (map_state == IMSM_T_STATE_DEGRADED && | |
6354 | map->map_state != map_state && | |
6355 | !dev->vol.migr_state) { | |
6356 | dprintf("imsm: mark degraded\n"); | |
6357 | map->map_state = map_state; | |
6358 | super->updates_pending++; | |
6359 | a->last_checkpoint = 0; | |
6360 | } else if (map_state == IMSM_T_STATE_FAILED && | |
6361 | map->map_state != map_state) { | |
6362 | dprintf("imsm: mark failed\n"); | |
6363 | end_migration(dev, map_state); | |
6364 | super->updates_pending++; | |
6365 | a->last_checkpoint = 0; | |
6366 | } else if (is_gen_migration(dev)) { | |
6367 | dprintf("imsm: Detected General Migration in state: "); | |
6368 | if (map_state == IMSM_T_STATE_NORMAL) { | |
6369 | end_migration(dev, map_state); | |
6370 | map = get_imsm_map(dev, 0); | |
6371 | map->failed_disk_num = ~0; | |
6372 | dprintf("normal\n"); | |
6373 | } else { | |
6374 | if (map_state == IMSM_T_STATE_DEGRADED) { | |
6375 | printf("degraded\n"); | |
6376 | end_migration(dev, map_state); | |
6377 | } else { | |
6378 | dprintf("failed\n"); | |
6379 | } | |
6380 | map->map_state = map_state; | |
6381 | } | |
6382 | super->updates_pending++; | |
6383 | } | |
6384 | } | |
6385 | ||
6386 | static int store_imsm_mpb(int fd, struct imsm_super *mpb) | |
6387 | { | |
6388 | void *buf = mpb; | |
6389 | __u32 mpb_size = __le32_to_cpu(mpb->mpb_size); | |
6390 | unsigned long long dsize; | |
6391 | unsigned long long sectors; | |
6392 | ||
6393 | get_dev_size(fd, NULL, &dsize); | |
6394 | ||
6395 | if (mpb_size > 512) { | |
6396 | /* -1 to account for anchor */ | |
6397 | sectors = mpb_sectors(mpb) - 1; | |
6398 | ||
6399 | /* write the extended mpb to the sectors preceeding the anchor */ | |
6400 | if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) | |
6401 | return 1; | |
6402 | ||
6403 | if ((unsigned long long)write(fd, buf + 512, 512 * sectors) | |
6404 | != 512 * sectors) | |
6405 | return 1; | |
6406 | } | |
6407 | ||
6408 | /* first block is stored on second to last sector of the disk */ | |
6409 | if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) | |
6410 | return 1; | |
6411 | ||
6412 | if (write(fd, buf, 512) != 512) | |
6413 | return 1; | |
6414 | ||
6415 | return 0; | |
6416 | } | |
6417 | ||
6418 | static void imsm_sync_metadata(struct supertype *container) | |
6419 | { | |
6420 | struct intel_super *super = container->sb; | |
6421 | ||
6422 | dprintf("sync metadata: %d\n", super->updates_pending); | |
6423 | if (!super->updates_pending) | |
6424 | return; | |
6425 | ||
6426 | write_super_imsm(container, 0); | |
6427 | ||
6428 | super->updates_pending = 0; | |
6429 | } | |
6430 | ||
6431 | static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a) | |
6432 | { | |
6433 | struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member); | |
6434 | int i = get_imsm_disk_idx(dev, idx, -1); | |
6435 | struct dl *dl; | |
6436 | ||
6437 | for (dl = super->disks; dl; dl = dl->next) | |
6438 | if (dl->index == i) | |
6439 | break; | |
6440 | ||
6441 | if (dl && is_failed(&dl->disk)) | |
6442 | dl = NULL; | |
6443 | ||
6444 | if (dl) | |
6445 | dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor); | |
6446 | ||
6447 | return dl; | |
6448 | } | |
6449 | ||
6450 | static struct dl *imsm_add_spare(struct intel_super *super, int slot, | |
6451 | struct active_array *a, int activate_new, | |
6452 | struct mdinfo *additional_test_list) | |
6453 | { | |
6454 | struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member); | |
6455 | int idx = get_imsm_disk_idx(dev, slot, -1); | |
6456 | struct imsm_super *mpb = super->anchor; | |
6457 | struct imsm_map *map; | |
6458 | unsigned long long pos; | |
6459 | struct mdinfo *d; | |
6460 | struct extent *ex; | |
6461 | int i, j; | |
6462 | int found; | |
6463 | __u32 array_start = 0; | |
6464 | __u32 array_end = 0; | |
6465 | struct dl *dl; | |
6466 | struct mdinfo *test_list; | |
6467 | ||
6468 | for (dl = super->disks; dl; dl = dl->next) { | |
6469 | /* If in this array, skip */ | |
6470 | for (d = a->info.devs ; d ; d = d->next) | |
6471 | if (d->state_fd >= 0 && | |
6472 | d->disk.major == dl->major && | |
6473 | d->disk.minor == dl->minor) { | |
6474 | dprintf("%x:%x already in array\n", | |
6475 | dl->major, dl->minor); | |
6476 | break; | |
6477 | } | |
6478 | if (d) | |
6479 | continue; | |
6480 | test_list = additional_test_list; | |
6481 | while (test_list) { | |
6482 | if (test_list->disk.major == dl->major && | |
6483 | test_list->disk.minor == dl->minor) { | |
6484 | dprintf("%x:%x already in additional test list\n", | |
6485 | dl->major, dl->minor); | |
6486 | break; | |
6487 | } | |
6488 | test_list = test_list->next; | |
6489 | } | |
6490 | if (test_list) | |
6491 | continue; | |
6492 | ||
6493 | /* skip in use or failed drives */ | |
6494 | if (is_failed(&dl->disk) || idx == dl->index || | |
6495 | dl->index == -2) { | |
6496 | dprintf("%x:%x status (failed: %d index: %d)\n", | |
6497 | dl->major, dl->minor, is_failed(&dl->disk), idx); | |
6498 | continue; | |
6499 | } | |
6500 | ||
6501 | /* skip pure spares when we are looking for partially | |
6502 | * assimilated drives | |
6503 | */ | |
6504 | if (dl->index == -1 && !activate_new) | |
6505 | continue; | |
6506 | ||
6507 | /* Does this unused device have the requisite free space? | |
6508 | * It needs to be able to cover all member volumes | |
6509 | */ | |
6510 | ex = get_extents(super, dl); | |
6511 | if (!ex) { | |
6512 | dprintf("cannot get extents\n"); | |
6513 | continue; | |
6514 | } | |
6515 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
6516 | dev = get_imsm_dev(super, i); | |
6517 | map = get_imsm_map(dev, 0); | |
6518 | ||
6519 | /* check if this disk is already a member of | |
6520 | * this array | |
6521 | */ | |
6522 | if (get_imsm_disk_slot(map, dl->index) >= 0) | |
6523 | continue; | |
6524 | ||
6525 | found = 0; | |
6526 | j = 0; | |
6527 | pos = 0; | |
6528 | array_start = __le32_to_cpu(map->pba_of_lba0); | |
6529 | array_end = array_start + | |
6530 | __le32_to_cpu(map->blocks_per_member) - 1; | |
6531 | ||
6532 | do { | |
6533 | /* check that we can start at pba_of_lba0 with | |
6534 | * blocks_per_member of space | |
6535 | */ | |
6536 | if (array_start >= pos && array_end < ex[j].start) { | |
6537 | found = 1; | |
6538 | break; | |
6539 | } | |
6540 | pos = ex[j].start + ex[j].size; | |
6541 | j++; | |
6542 | } while (ex[j-1].size); | |
6543 | ||
6544 | if (!found) | |
6545 | break; | |
6546 | } | |
6547 | ||
6548 | free(ex); | |
6549 | if (i < mpb->num_raid_devs) { | |
6550 | dprintf("%x:%x does not have %u to %u available\n", | |
6551 | dl->major, dl->minor, array_start, array_end); | |
6552 | /* No room */ | |
6553 | continue; | |
6554 | } | |
6555 | return dl; | |
6556 | } | |
6557 | ||
6558 | return dl; | |
6559 | } | |
6560 | ||
6561 | ||
6562 | static int imsm_rebuild_allowed(struct supertype *cont, int dev_idx, int failed) | |
6563 | { | |
6564 | struct imsm_dev *dev2; | |
6565 | struct imsm_map *map; | |
6566 | struct dl *idisk; | |
6567 | int slot; | |
6568 | int idx; | |
6569 | __u8 state; | |
6570 | ||
6571 | dev2 = get_imsm_dev(cont->sb, dev_idx); | |
6572 | if (dev2) { | |
6573 | state = imsm_check_degraded(cont->sb, dev2, failed); | |
6574 | if (state == IMSM_T_STATE_FAILED) { | |
6575 | map = get_imsm_map(dev2, 0); | |
6576 | if (!map) | |
6577 | return 1; | |
6578 | for (slot = 0; slot < map->num_members; slot++) { | |
6579 | /* | |
6580 | * Check if failed disks are deleted from intel | |
6581 | * disk list or are marked to be deleted | |
6582 | */ | |
6583 | idx = get_imsm_disk_idx(dev2, slot, -1); | |
6584 | idisk = get_imsm_dl_disk(cont->sb, idx); | |
6585 | /* | |
6586 | * Do not rebuild the array if failed disks | |
6587 | * from failed sub-array are not removed from | |
6588 | * container. | |
6589 | */ | |
6590 | if (idisk && | |
6591 | is_failed(&idisk->disk) && | |
6592 | (idisk->action != DISK_REMOVE)) | |
6593 | return 0; | |
6594 | } | |
6595 | } | |
6596 | } | |
6597 | return 1; | |
6598 | } | |
6599 | ||
6600 | static struct mdinfo *imsm_activate_spare(struct active_array *a, | |
6601 | struct metadata_update **updates) | |
6602 | { | |
6603 | /** | |
6604 | * Find a device with unused free space and use it to replace a | |
6605 | * failed/vacant region in an array. We replace failed regions one a | |
6606 | * array at a time. The result is that a new spare disk will be added | |
6607 | * to the first failed array and after the monitor has finished | |
6608 | * propagating failures the remainder will be consumed. | |
6609 | * | |
6610 | * FIXME add a capability for mdmon to request spares from another | |
6611 | * container. | |
6612 | */ | |
6613 | ||
6614 | struct intel_super *super = a->container->sb; | |
6615 | int inst = a->info.container_member; | |
6616 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
6617 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6618 | int failed = a->info.array.raid_disks; | |
6619 | struct mdinfo *rv = NULL; | |
6620 | struct mdinfo *d; | |
6621 | struct mdinfo *di; | |
6622 | struct metadata_update *mu; | |
6623 | struct dl *dl; | |
6624 | struct imsm_update_activate_spare *u; | |
6625 | int num_spares = 0; | |
6626 | int i; | |
6627 | int allowed; | |
6628 | ||
6629 | for (d = a->info.devs ; d ; d = d->next) { | |
6630 | if ((d->curr_state & DS_FAULTY) && | |
6631 | d->state_fd >= 0) | |
6632 | /* wait for Removal to happen */ | |
6633 | return NULL; | |
6634 | if (d->state_fd >= 0) | |
6635 | failed--; | |
6636 | } | |
6637 | ||
6638 | dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n", | |
6639 | inst, failed, a->info.array.raid_disks, a->info.array.level); | |
6640 | ||
6641 | if (imsm_reshape_blocks_arrays_changes(super)) | |
6642 | return NULL; | |
6643 | ||
6644 | if (a->info.array.level == 4) | |
6645 | /* No repair for takeovered array | |
6646 | * imsm doesn't support raid4 | |
6647 | */ | |
6648 | return NULL; | |
6649 | ||
6650 | if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED) | |
6651 | return NULL; | |
6652 | ||
6653 | /* | |
6654 | * If there are any failed disks check state of the other volume. | |
6655 | * Block rebuild if the another one is failed until failed disks | |
6656 | * are removed from container. | |
6657 | */ | |
6658 | if (failed) { | |
6659 | dprintf("found failed disks in %.*s, check if there another" | |
6660 | "failed sub-array.\n", | |
6661 | MAX_RAID_SERIAL_LEN, dev->volume); | |
6662 | /* check if states of the other volumes allow for rebuild */ | |
6663 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
6664 | if (i != inst) { | |
6665 | allowed = imsm_rebuild_allowed(a->container, | |
6666 | i, failed); | |
6667 | if (!allowed) | |
6668 | return NULL; | |
6669 | } | |
6670 | } | |
6671 | } | |
6672 | ||
6673 | /* For each slot, if it is not working, find a spare */ | |
6674 | for (i = 0; i < a->info.array.raid_disks; i++) { | |
6675 | for (d = a->info.devs ; d ; d = d->next) | |
6676 | if (d->disk.raid_disk == i) | |
6677 | break; | |
6678 | dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0); | |
6679 | if (d && (d->state_fd >= 0)) | |
6680 | continue; | |
6681 | ||
6682 | /* | |
6683 | * OK, this device needs recovery. Try to re-add the | |
6684 | * previous occupant of this slot, if this fails see if | |
6685 | * we can continue the assimilation of a spare that was | |
6686 | * partially assimilated, finally try to activate a new | |
6687 | * spare. | |
6688 | */ | |
6689 | dl = imsm_readd(super, i, a); | |
6690 | if (!dl) | |
6691 | dl = imsm_add_spare(super, i, a, 0, rv); | |
6692 | if (!dl) | |
6693 | dl = imsm_add_spare(super, i, a, 1, rv); | |
6694 | if (!dl) | |
6695 | continue; | |
6696 | ||
6697 | /* found a usable disk with enough space */ | |
6698 | di = malloc(sizeof(*di)); | |
6699 | if (!di) | |
6700 | continue; | |
6701 | memset(di, 0, sizeof(*di)); | |
6702 | ||
6703 | /* dl->index will be -1 in the case we are activating a | |
6704 | * pristine spare. imsm_process_update() will create a | |
6705 | * new index in this case. Once a disk is found to be | |
6706 | * failed in all member arrays it is kicked from the | |
6707 | * metadata | |
6708 | */ | |
6709 | di->disk.number = dl->index; | |
6710 | ||
6711 | /* (ab)use di->devs to store a pointer to the device | |
6712 | * we chose | |
6713 | */ | |
6714 | di->devs = (struct mdinfo *) dl; | |
6715 | ||
6716 | di->disk.raid_disk = i; | |
6717 | di->disk.major = dl->major; | |
6718 | di->disk.minor = dl->minor; | |
6719 | di->disk.state = 0; | |
6720 | di->recovery_start = 0; | |
6721 | di->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
6722 | di->component_size = a->info.component_size; | |
6723 | di->container_member = inst; | |
6724 | super->random = random32(); | |
6725 | di->next = rv; | |
6726 | rv = di; | |
6727 | num_spares++; | |
6728 | dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor, | |
6729 | i, di->data_offset); | |
6730 | } | |
6731 | ||
6732 | if (!rv) | |
6733 | /* No spares found */ | |
6734 | return rv; | |
6735 | /* Now 'rv' has a list of devices to return. | |
6736 | * Create a metadata_update record to update the | |
6737 | * disk_ord_tbl for the array | |
6738 | */ | |
6739 | mu = malloc(sizeof(*mu)); | |
6740 | if (mu) { | |
6741 | mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares); | |
6742 | if (mu->buf == NULL) { | |
6743 | free(mu); | |
6744 | mu = NULL; | |
6745 | } | |
6746 | } | |
6747 | if (!mu) { | |
6748 | while (rv) { | |
6749 | struct mdinfo *n = rv->next; | |
6750 | ||
6751 | free(rv); | |
6752 | rv = n; | |
6753 | } | |
6754 | return NULL; | |
6755 | } | |
6756 | ||
6757 | mu->space = NULL; | |
6758 | mu->space_list = NULL; | |
6759 | mu->len = sizeof(struct imsm_update_activate_spare) * num_spares; | |
6760 | mu->next = *updates; | |
6761 | u = (struct imsm_update_activate_spare *) mu->buf; | |
6762 | ||
6763 | for (di = rv ; di ; di = di->next) { | |
6764 | u->type = update_activate_spare; | |
6765 | u->dl = (struct dl *) di->devs; | |
6766 | di->devs = NULL; | |
6767 | u->slot = di->disk.raid_disk; | |
6768 | u->array = inst; | |
6769 | u->next = u + 1; | |
6770 | u++; | |
6771 | } | |
6772 | (u-1)->next = NULL; | |
6773 | *updates = mu; | |
6774 | ||
6775 | return rv; | |
6776 | } | |
6777 | ||
6778 | static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u) | |
6779 | { | |
6780 | struct imsm_dev *dev = get_imsm_dev(super, idx); | |
6781 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6782 | struct imsm_map *new_map = get_imsm_map(&u->dev, 0); | |
6783 | struct disk_info *inf = get_disk_info(u); | |
6784 | struct imsm_disk *disk; | |
6785 | int i; | |
6786 | int j; | |
6787 | ||
6788 | for (i = 0; i < map->num_members; i++) { | |
6789 | disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i, -1)); | |
6790 | for (j = 0; j < new_map->num_members; j++) | |
6791 | if (serialcmp(disk->serial, inf[j].serial) == 0) | |
6792 | return 1; | |
6793 | } | |
6794 | ||
6795 | return 0; | |
6796 | } | |
6797 | ||
6798 | ||
6799 | static struct dl *get_disk_super(struct intel_super *super, int major, int minor) | |
6800 | { | |
6801 | struct dl *dl = NULL; | |
6802 | for (dl = super->disks; dl; dl = dl->next) | |
6803 | if ((dl->major == major) && (dl->minor == minor)) | |
6804 | return dl; | |
6805 | return NULL; | |
6806 | } | |
6807 | ||
6808 | static int remove_disk_super(struct intel_super *super, int major, int minor) | |
6809 | { | |
6810 | struct dl *prev = NULL; | |
6811 | struct dl *dl; | |
6812 | ||
6813 | prev = NULL; | |
6814 | for (dl = super->disks; dl; dl = dl->next) { | |
6815 | if ((dl->major == major) && (dl->minor == minor)) { | |
6816 | /* remove */ | |
6817 | if (prev) | |
6818 | prev->next = dl->next; | |
6819 | else | |
6820 | super->disks = dl->next; | |
6821 | dl->next = NULL; | |
6822 | __free_imsm_disk(dl); | |
6823 | dprintf("%s: removed %x:%x\n", | |
6824 | __func__, major, minor); | |
6825 | break; | |
6826 | } | |
6827 | prev = dl; | |
6828 | } | |
6829 | return 0; | |
6830 | } | |
6831 | ||
6832 | static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index); | |
6833 | ||
6834 | static int add_remove_disk_update(struct intel_super *super) | |
6835 | { | |
6836 | int check_degraded = 0; | |
6837 | struct dl *disk = NULL; | |
6838 | /* add/remove some spares to/from the metadata/contrainer */ | |
6839 | while (super->disk_mgmt_list) { | |
6840 | struct dl *disk_cfg; | |
6841 | ||
6842 | disk_cfg = super->disk_mgmt_list; | |
6843 | super->disk_mgmt_list = disk_cfg->next; | |
6844 | disk_cfg->next = NULL; | |
6845 | ||
6846 | if (disk_cfg->action == DISK_ADD) { | |
6847 | disk_cfg->next = super->disks; | |
6848 | super->disks = disk_cfg; | |
6849 | check_degraded = 1; | |
6850 | dprintf("%s: added %x:%x\n", | |
6851 | __func__, disk_cfg->major, | |
6852 | disk_cfg->minor); | |
6853 | } else if (disk_cfg->action == DISK_REMOVE) { | |
6854 | dprintf("Disk remove action processed: %x.%x\n", | |
6855 | disk_cfg->major, disk_cfg->minor); | |
6856 | disk = get_disk_super(super, | |
6857 | disk_cfg->major, | |
6858 | disk_cfg->minor); | |
6859 | if (disk) { | |
6860 | /* store action status */ | |
6861 | disk->action = DISK_REMOVE; | |
6862 | /* remove spare disks only */ | |
6863 | if (disk->index == -1) { | |
6864 | remove_disk_super(super, | |
6865 | disk_cfg->major, | |
6866 | disk_cfg->minor); | |
6867 | } | |
6868 | } | |
6869 | /* release allocate disk structure */ | |
6870 | __free_imsm_disk(disk_cfg); | |
6871 | } | |
6872 | } | |
6873 | return check_degraded; | |
6874 | } | |
6875 | ||
6876 | ||
6877 | static int apply_reshape_migration_update(struct imsm_update_reshape_migration *u, | |
6878 | struct intel_super *super, | |
6879 | void ***space_list) | |
6880 | { | |
6881 | struct intel_dev *id; | |
6882 | void **tofree = NULL; | |
6883 | int ret_val = 0; | |
6884 | ||
6885 | dprintf("apply_reshape_migration_update()\n"); | |
6886 | if ((u->subdev < 0) || | |
6887 | (u->subdev > 1)) { | |
6888 | dprintf("imsm: Error: Wrong subdev: %i\n", u->subdev); | |
6889 | return ret_val; | |
6890 | } | |
6891 | if ((space_list == NULL) || (*space_list == NULL)) { | |
6892 | dprintf("imsm: Error: Memory is not allocated\n"); | |
6893 | return ret_val; | |
6894 | } | |
6895 | ||
6896 | for (id = super->devlist ; id; id = id->next) { | |
6897 | if (id->index == (unsigned)u->subdev) { | |
6898 | struct imsm_dev *dev = get_imsm_dev(super, u->subdev); | |
6899 | struct imsm_map *map; | |
6900 | struct imsm_dev *new_dev = | |
6901 | (struct imsm_dev *)*space_list; | |
6902 | struct imsm_map *migr_map = get_imsm_map(dev, 1); | |
6903 | int to_state; | |
6904 | struct dl *new_disk; | |
6905 | ||
6906 | if (new_dev == NULL) | |
6907 | return ret_val; | |
6908 | *space_list = **space_list; | |
6909 | memcpy(new_dev, dev, sizeof_imsm_dev(dev, 0)); | |
6910 | map = get_imsm_map(new_dev, 0); | |
6911 | if (migr_map) { | |
6912 | dprintf("imsm: Error: migration in progress"); | |
6913 | return ret_val; | |
6914 | } | |
6915 | ||
6916 | to_state = map->map_state; | |
6917 | if ((u->new_level == 5) && (map->raid_level == 0)) { | |
6918 | map->num_members++; | |
6919 | /* this should not happen */ | |
6920 | if (u->new_disks[0] < 0) { | |
6921 | map->failed_disk_num = | |
6922 | map->num_members - 1; | |
6923 | to_state = IMSM_T_STATE_DEGRADED; | |
6924 | } else | |
6925 | to_state = IMSM_T_STATE_NORMAL; | |
6926 | } | |
6927 | migrate(new_dev, super, to_state, MIGR_GEN_MIGR); | |
6928 | if (u->new_level > -1) | |
6929 | map->raid_level = u->new_level; | |
6930 | migr_map = get_imsm_map(new_dev, 1); | |
6931 | if ((u->new_level == 5) && | |
6932 | (migr_map->raid_level == 0)) { | |
6933 | int ord = map->num_members - 1; | |
6934 | migr_map->num_members--; | |
6935 | if (u->new_disks[0] < 0) | |
6936 | ord |= IMSM_ORD_REBUILD; | |
6937 | set_imsm_ord_tbl_ent(map, | |
6938 | map->num_members - 1, | |
6939 | ord); | |
6940 | } | |
6941 | id->dev = new_dev; | |
6942 | tofree = (void **)dev; | |
6943 | ||
6944 | /* update chunk size | |
6945 | */ | |
6946 | if (u->new_chunksize > 0) | |
6947 | map->blocks_per_strip = | |
6948 | __cpu_to_le16(u->new_chunksize * 2); | |
6949 | ||
6950 | /* add disk | |
6951 | */ | |
6952 | if ((u->new_level != 5) || | |
6953 | (migr_map->raid_level != 0) || | |
6954 | (migr_map->raid_level == map->raid_level)) | |
6955 | goto skip_disk_add; | |
6956 | ||
6957 | if (u->new_disks[0] >= 0) { | |
6958 | /* use passes spare | |
6959 | */ | |
6960 | new_disk = get_disk_super(super, | |
6961 | major(u->new_disks[0]), | |
6962 | minor(u->new_disks[0])); | |
6963 | dprintf("imsm: new disk for reshape is: %i:%i " | |
6964 | "(%p, index = %i)\n", | |
6965 | major(u->new_disks[0]), | |
6966 | minor(u->new_disks[0]), | |
6967 | new_disk, new_disk->index); | |
6968 | if (new_disk == NULL) | |
6969 | goto error_disk_add; | |
6970 | ||
6971 | new_disk->index = map->num_members - 1; | |
6972 | /* slot to fill in autolayout | |
6973 | */ | |
6974 | new_disk->raiddisk = new_disk->index; | |
6975 | new_disk->disk.status |= CONFIGURED_DISK; | |
6976 | new_disk->disk.status &= ~SPARE_DISK; | |
6977 | } else | |
6978 | goto error_disk_add; | |
6979 | ||
6980 | skip_disk_add: | |
6981 | *tofree = *space_list; | |
6982 | /* calculate new size | |
6983 | */ | |
6984 | imsm_set_array_size(new_dev); | |
6985 | ||
6986 | ret_val = 1; | |
6987 | } | |
6988 | } | |
6989 | ||
6990 | if (tofree) | |
6991 | *space_list = tofree; | |
6992 | return ret_val; | |
6993 | ||
6994 | error_disk_add: | |
6995 | dprintf("Error: imsm: Cannot find disk.\n"); | |
6996 | return ret_val; | |
6997 | } | |
6998 | ||
6999 | static int apply_update_activate_spare(struct imsm_update_activate_spare *u, | |
7000 | struct intel_super *super, | |
7001 | struct active_array *active_array) | |
7002 | { | |
7003 | struct imsm_super *mpb = super->anchor; | |
7004 | struct imsm_dev *dev = get_imsm_dev(super, u->array); | |
7005 | struct imsm_map *map = get_imsm_map(dev, 0); | |
7006 | struct imsm_map *migr_map; | |
7007 | struct active_array *a; | |
7008 | struct imsm_disk *disk; | |
7009 | __u8 to_state; | |
7010 | struct dl *dl; | |
7011 | unsigned int found; | |
7012 | int failed; | |
7013 | int victim; | |
7014 | int i; | |
7015 | int second_map_created = 0; | |
7016 | ||
7017 | for (; u; u = u->next) { | |
7018 | victim = get_imsm_disk_idx(dev, u->slot, -1); | |
7019 | ||
7020 | if (victim < 0) | |
7021 | return 0; | |
7022 | ||
7023 | for (dl = super->disks; dl; dl = dl->next) | |
7024 | if (dl == u->dl) | |
7025 | break; | |
7026 | ||
7027 | if (!dl) { | |
7028 | fprintf(stderr, "error: imsm_activate_spare passed " | |
7029 | "an unknown disk (index: %d)\n", | |
7030 | u->dl->index); | |
7031 | return 0; | |
7032 | } | |
7033 | ||
7034 | /* count failures (excluding rebuilds and the victim) | |
7035 | * to determine map[0] state | |
7036 | */ | |
7037 | failed = 0; | |
7038 | for (i = 0; i < map->num_members; i++) { | |
7039 | if (i == u->slot) | |
7040 | continue; | |
7041 | disk = get_imsm_disk(super, | |
7042 | get_imsm_disk_idx(dev, i, -1)); | |
7043 | if (!disk || is_failed(disk)) | |
7044 | failed++; | |
7045 | } | |
7046 | ||
7047 | /* adding a pristine spare, assign a new index */ | |
7048 | if (dl->index < 0) { | |
7049 | dl->index = super->anchor->num_disks; | |
7050 | super->anchor->num_disks++; | |
7051 | } | |
7052 | disk = &dl->disk; | |
7053 | disk->status |= CONFIGURED_DISK; | |
7054 | disk->status &= ~SPARE_DISK; | |
7055 | ||
7056 | /* mark rebuild */ | |
7057 | to_state = imsm_check_degraded(super, dev, failed); | |
7058 | if (!second_map_created) { | |
7059 | second_map_created = 1; | |
7060 | map->map_state = IMSM_T_STATE_DEGRADED; | |
7061 | migrate(dev, super, to_state, MIGR_REBUILD); | |
7062 | } else | |
7063 | map->map_state = to_state; | |
7064 | migr_map = get_imsm_map(dev, 1); | |
7065 | set_imsm_ord_tbl_ent(map, u->slot, dl->index); | |
7066 | set_imsm_ord_tbl_ent(migr_map, u->slot, | |
7067 | dl->index | IMSM_ORD_REBUILD); | |
7068 | ||
7069 | /* update the family_num to mark a new container | |
7070 | * generation, being careful to record the existing | |
7071 | * family_num in orig_family_num to clean up after | |
7072 | * earlier mdadm versions that neglected to set it. | |
7073 | */ | |
7074 | if (mpb->orig_family_num == 0) | |
7075 | mpb->orig_family_num = mpb->family_num; | |
7076 | mpb->family_num += super->random; | |
7077 | ||
7078 | /* count arrays using the victim in the metadata */ | |
7079 | found = 0; | |
7080 | for (a = active_array; a ; a = a->next) { | |
7081 | dev = get_imsm_dev(super, a->info.container_member); | |
7082 | map = get_imsm_map(dev, 0); | |
7083 | ||
7084 | if (get_imsm_disk_slot(map, victim) >= 0) | |
7085 | found++; | |
7086 | } | |
7087 | ||
7088 | /* delete the victim if it is no longer being | |
7089 | * utilized anywhere | |
7090 | */ | |
7091 | if (!found) { | |
7092 | struct dl **dlp; | |
7093 | ||
7094 | /* We know that 'manager' isn't touching anything, | |
7095 | * so it is safe to delete | |
7096 | */ | |
7097 | for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next) | |
7098 | if ((*dlp)->index == victim) | |
7099 | break; | |
7100 | ||
7101 | /* victim may be on the missing list */ | |
7102 | if (!*dlp) | |
7103 | for (dlp = &super->missing; *dlp; | |
7104 | dlp = &(*dlp)->next) | |
7105 | if ((*dlp)->index == victim) | |
7106 | break; | |
7107 | imsm_delete(super, dlp, victim); | |
7108 | } | |
7109 | } | |
7110 | ||
7111 | return 1; | |
7112 | } | |
7113 | ||
7114 | static int apply_reshape_container_disks_update(struct imsm_update_reshape *u, | |
7115 | struct intel_super *super, | |
7116 | void ***space_list) | |
7117 | { | |
7118 | struct dl *new_disk; | |
7119 | struct intel_dev *id; | |
7120 | int i; | |
7121 | int delta_disks = u->new_raid_disks - u->old_raid_disks; | |
7122 | int disk_count = u->old_raid_disks; | |
7123 | void **tofree = NULL; | |
7124 | int devices_to_reshape = 1; | |
7125 | struct imsm_super *mpb = super->anchor; | |
7126 | int ret_val = 0; | |
7127 | unsigned int dev_id; | |
7128 | ||
7129 | dprintf("imsm: apply_reshape_container_disks_update()\n"); | |
7130 | ||
7131 | /* enable spares to use in array */ | |
7132 | for (i = 0; i < delta_disks; i++) { | |
7133 | new_disk = get_disk_super(super, | |
7134 | major(u->new_disks[i]), | |
7135 | minor(u->new_disks[i])); | |
7136 | dprintf("imsm: new disk for reshape is: %i:%i " | |
7137 | "(%p, index = %i)\n", | |
7138 | major(u->new_disks[i]), minor(u->new_disks[i]), | |
7139 | new_disk, new_disk->index); | |
7140 | if ((new_disk == NULL) || | |
7141 | ((new_disk->index >= 0) && | |
7142 | (new_disk->index < u->old_raid_disks))) | |
7143 | goto update_reshape_exit; | |
7144 | new_disk->index = disk_count++; | |
7145 | /* slot to fill in autolayout | |
7146 | */ | |
7147 | new_disk->raiddisk = new_disk->index; | |
7148 | new_disk->disk.status |= | |
7149 | CONFIGURED_DISK; | |
7150 | new_disk->disk.status &= ~SPARE_DISK; | |
7151 | } | |
7152 | ||
7153 | dprintf("imsm: volume set mpb->num_raid_devs = %i\n", | |
7154 | mpb->num_raid_devs); | |
7155 | /* manage changes in volume | |
7156 | */ | |
7157 | for (dev_id = 0; dev_id < mpb->num_raid_devs; dev_id++) { | |
7158 | void **sp = *space_list; | |
7159 | struct imsm_dev *newdev; | |
7160 | struct imsm_map *newmap, *oldmap; | |
7161 | ||
7162 | for (id = super->devlist ; id; id = id->next) { | |
7163 | if (id->index == dev_id) | |
7164 | break; | |
7165 | } | |
7166 | if (id == NULL) | |
7167 | break; | |
7168 | if (!sp) | |
7169 | continue; | |
7170 | *space_list = *sp; | |
7171 | newdev = (void*)sp; | |
7172 | /* Copy the dev, but not (all of) the map */ | |
7173 | memcpy(newdev, id->dev, sizeof(*newdev)); | |
7174 | oldmap = get_imsm_map(id->dev, 0); | |
7175 | newmap = get_imsm_map(newdev, 0); | |
7176 | /* Copy the current map */ | |
7177 | memcpy(newmap, oldmap, sizeof_imsm_map(oldmap)); | |
7178 | /* update one device only | |
7179 | */ | |
7180 | if (devices_to_reshape) { | |
7181 | dprintf("imsm: modifying subdev: %i\n", | |
7182 | id->index); | |
7183 | devices_to_reshape--; | |
7184 | newdev->vol.migr_state = 1; | |
7185 | newdev->vol.curr_migr_unit = 0; | |
7186 | set_migr_type(newdev, MIGR_GEN_MIGR); | |
7187 | newmap->num_members = u->new_raid_disks; | |
7188 | for (i = 0; i < delta_disks; i++) { | |
7189 | set_imsm_ord_tbl_ent(newmap, | |
7190 | u->old_raid_disks + i, | |
7191 | u->old_raid_disks + i); | |
7192 | } | |
7193 | /* New map is correct, now need to save old map | |
7194 | */ | |
7195 | newmap = get_imsm_map(newdev, 1); | |
7196 | memcpy(newmap, oldmap, sizeof_imsm_map(oldmap)); | |
7197 | ||
7198 | imsm_set_array_size(newdev); | |
7199 | } | |
7200 | ||
7201 | sp = (void **)id->dev; | |
7202 | id->dev = newdev; | |
7203 | *sp = tofree; | |
7204 | tofree = sp; | |
7205 | ||
7206 | /* Clear migration record */ | |
7207 | memset(super->migr_rec, 0, sizeof(struct migr_record)); | |
7208 | } | |
7209 | if (tofree) | |
7210 | *space_list = tofree; | |
7211 | ret_val = 1; | |
7212 | ||
7213 | update_reshape_exit: | |
7214 | ||
7215 | return ret_val; | |
7216 | } | |
7217 | ||
7218 | static int apply_takeover_update(struct imsm_update_takeover *u, | |
7219 | struct intel_super *super, | |
7220 | void ***space_list) | |
7221 | { | |
7222 | struct imsm_dev *dev = NULL; | |
7223 | struct intel_dev *dv; | |
7224 | struct imsm_dev *dev_new; | |
7225 | struct imsm_map *map; | |
7226 | struct dl *dm, *du; | |
7227 | int i; | |
7228 | ||
7229 | for (dv = super->devlist; dv; dv = dv->next) | |
7230 | if (dv->index == (unsigned int)u->subarray) { | |
7231 | dev = dv->dev; | |
7232 | break; | |
7233 | } | |
7234 | ||
7235 | if (dev == NULL) | |
7236 | return 0; | |
7237 | ||
7238 | map = get_imsm_map(dev, 0); | |
7239 | ||
7240 | if (u->direction == R10_TO_R0) { | |
7241 | /* Number of failed disks must be half of initial disk number */ | |
7242 | if (imsm_count_failed(super, dev) != (map->num_members / 2)) | |
7243 | return 0; | |
7244 | ||
7245 | /* iterate through devices to mark removed disks as spare */ | |
7246 | for (dm = super->disks; dm; dm = dm->next) { | |
7247 | if (dm->disk.status & FAILED_DISK) { | |
7248 | int idx = dm->index; | |
7249 | /* update indexes on the disk list */ | |
7250 | /* FIXME this loop-with-the-loop looks wrong, I'm not convinced | |
7251 | the index values will end up being correct.... NB */ | |
7252 | for (du = super->disks; du; du = du->next) | |
7253 | if (du->index > idx) | |
7254 | du->index--; | |
7255 | /* mark as spare disk */ | |
7256 | mark_spare(dm); | |
7257 | } | |
7258 | } | |
7259 | /* update map */ | |
7260 | map->num_members = map->num_members / 2; | |
7261 | map->map_state = IMSM_T_STATE_NORMAL; | |
7262 | map->num_domains = 1; | |
7263 | map->raid_level = 0; | |
7264 | map->failed_disk_num = -1; | |
7265 | } | |
7266 | ||
7267 | if (u->direction == R0_TO_R10) { | |
7268 | void **space; | |
7269 | /* update slots in current disk list */ | |
7270 | for (dm = super->disks; dm; dm = dm->next) { | |
7271 | if (dm->index >= 0) | |
7272 | dm->index *= 2; | |
7273 | } | |
7274 | /* create new *missing* disks */ | |
7275 | for (i = 0; i < map->num_members; i++) { | |
7276 | space = *space_list; | |
7277 | if (!space) | |
7278 | continue; | |
7279 | *space_list = *space; | |
7280 | du = (void *)space; | |
7281 | memcpy(du, super->disks, sizeof(*du)); | |
7282 | du->fd = -1; | |
7283 | du->minor = 0; | |
7284 | du->major = 0; | |
7285 | du->index = (i * 2) + 1; | |
7286 | sprintf((char *)du->disk.serial, | |
7287 | " MISSING_%d", du->index); | |
7288 | sprintf((char *)du->serial, | |
7289 | "MISSING_%d", du->index); | |
7290 | du->next = super->missing; | |
7291 | super->missing = du; | |
7292 | } | |
7293 | /* create new dev and map */ | |
7294 | space = *space_list; | |
7295 | if (!space) | |
7296 | return 0; | |
7297 | *space_list = *space; | |
7298 | dev_new = (void *)space; | |
7299 | memcpy(dev_new, dev, sizeof(*dev)); | |
7300 | /* update new map */ | |
7301 | map = get_imsm_map(dev_new, 0); | |
7302 | map->num_members = map->num_members * 2; | |
7303 | map->map_state = IMSM_T_STATE_DEGRADED; | |
7304 | map->num_domains = 2; | |
7305 | map->raid_level = 1; | |
7306 | /* replace dev<->dev_new */ | |
7307 | dv->dev = dev_new; | |
7308 | } | |
7309 | /* update disk order table */ | |
7310 | for (du = super->disks; du; du = du->next) | |
7311 | if (du->index >= 0) | |
7312 | set_imsm_ord_tbl_ent(map, du->index, du->index); | |
7313 | for (du = super->missing; du; du = du->next) | |
7314 | if (du->index >= 0) { | |
7315 | set_imsm_ord_tbl_ent(map, du->index, du->index); | |
7316 | mark_missing(dv->dev, &du->disk, du->index); | |
7317 | } | |
7318 | ||
7319 | return 1; | |
7320 | } | |
7321 | ||
7322 | static void imsm_process_update(struct supertype *st, | |
7323 | struct metadata_update *update) | |
7324 | { | |
7325 | /** | |
7326 | * crack open the metadata_update envelope to find the update record | |
7327 | * update can be one of: | |
7328 | * update_reshape_container_disks - all the arrays in the container | |
7329 | * are being reshaped to have more devices. We need to mark | |
7330 | * the arrays for general migration and convert selected spares | |
7331 | * into active devices. | |
7332 | * update_activate_spare - a spare device has replaced a failed | |
7333 | * device in an array, update the disk_ord_tbl. If this disk is | |
7334 | * present in all member arrays then also clear the SPARE_DISK | |
7335 | * flag | |
7336 | * update_create_array | |
7337 | * update_kill_array | |
7338 | * update_rename_array | |
7339 | * update_add_remove_disk | |
7340 | */ | |
7341 | struct intel_super *super = st->sb; | |
7342 | struct imsm_super *mpb; | |
7343 | enum imsm_update_type type = *(enum imsm_update_type *) update->buf; | |
7344 | ||
7345 | /* update requires a larger buf but the allocation failed */ | |
7346 | if (super->next_len && !super->next_buf) { | |
7347 | super->next_len = 0; | |
7348 | return; | |
7349 | } | |
7350 | ||
7351 | if (super->next_buf) { | |
7352 | memcpy(super->next_buf, super->buf, super->len); | |
7353 | free(super->buf); | |
7354 | super->len = super->next_len; | |
7355 | super->buf = super->next_buf; | |
7356 | ||
7357 | super->next_len = 0; | |
7358 | super->next_buf = NULL; | |
7359 | } | |
7360 | ||
7361 | mpb = super->anchor; | |
7362 | ||
7363 | switch (type) { | |
7364 | case update_general_migration_checkpoint: { | |
7365 | struct intel_dev *id; | |
7366 | struct imsm_update_general_migration_checkpoint *u = | |
7367 | (void *)update->buf; | |
7368 | ||
7369 | dprintf("imsm: process_update() " | |
7370 | "for update_general_migration_checkpoint called\n"); | |
7371 | ||
7372 | /* find device under general migration */ | |
7373 | for (id = super->devlist ; id; id = id->next) { | |
7374 | if (is_gen_migration(id->dev)) { | |
7375 | id->dev->vol.curr_migr_unit = | |
7376 | __cpu_to_le32(u->curr_migr_unit); | |
7377 | super->updates_pending++; | |
7378 | } | |
7379 | } | |
7380 | break; | |
7381 | } | |
7382 | case update_takeover: { | |
7383 | struct imsm_update_takeover *u = (void *)update->buf; | |
7384 | if (apply_takeover_update(u, super, &update->space_list)) { | |
7385 | imsm_update_version_info(super); | |
7386 | super->updates_pending++; | |
7387 | } | |
7388 | break; | |
7389 | } | |
7390 | ||
7391 | case update_reshape_container_disks: { | |
7392 | struct imsm_update_reshape *u = (void *)update->buf; | |
7393 | if (apply_reshape_container_disks_update( | |
7394 | u, super, &update->space_list)) | |
7395 | super->updates_pending++; | |
7396 | break; | |
7397 | } | |
7398 | case update_reshape_migration: { | |
7399 | struct imsm_update_reshape_migration *u = (void *)update->buf; | |
7400 | if (apply_reshape_migration_update( | |
7401 | u, super, &update->space_list)) | |
7402 | super->updates_pending++; | |
7403 | break; | |
7404 | } | |
7405 | case update_activate_spare: { | |
7406 | struct imsm_update_activate_spare *u = (void *) update->buf; | |
7407 | if (apply_update_activate_spare(u, super, st->arrays)) | |
7408 | super->updates_pending++; | |
7409 | break; | |
7410 | } | |
7411 | case update_create_array: { | |
7412 | /* someone wants to create a new array, we need to be aware of | |
7413 | * a few races/collisions: | |
7414 | * 1/ 'Create' called by two separate instances of mdadm | |
7415 | * 2/ 'Create' versus 'activate_spare': mdadm has chosen | |
7416 | * devices that have since been assimilated via | |
7417 | * activate_spare. | |
7418 | * In the event this update can not be carried out mdadm will | |
7419 | * (FIX ME) notice that its update did not take hold. | |
7420 | */ | |
7421 | struct imsm_update_create_array *u = (void *) update->buf; | |
7422 | struct intel_dev *dv; | |
7423 | struct imsm_dev *dev; | |
7424 | struct imsm_map *map, *new_map; | |
7425 | unsigned long long start, end; | |
7426 | unsigned long long new_start, new_end; | |
7427 | int i; | |
7428 | struct disk_info *inf; | |
7429 | struct dl *dl; | |
7430 | ||
7431 | /* handle racing creates: first come first serve */ | |
7432 | if (u->dev_idx < mpb->num_raid_devs) { | |
7433 | dprintf("%s: subarray %d already defined\n", | |
7434 | __func__, u->dev_idx); | |
7435 | goto create_error; | |
7436 | } | |
7437 | ||
7438 | /* check update is next in sequence */ | |
7439 | if (u->dev_idx != mpb->num_raid_devs) { | |
7440 | dprintf("%s: can not create array %d expected index %d\n", | |
7441 | __func__, u->dev_idx, mpb->num_raid_devs); | |
7442 | goto create_error; | |
7443 | } | |
7444 | ||
7445 | new_map = get_imsm_map(&u->dev, 0); | |
7446 | new_start = __le32_to_cpu(new_map->pba_of_lba0); | |
7447 | new_end = new_start + __le32_to_cpu(new_map->blocks_per_member); | |
7448 | inf = get_disk_info(u); | |
7449 | ||
7450 | /* handle activate_spare versus create race: | |
7451 | * check to make sure that overlapping arrays do not include | |
7452 | * overalpping disks | |
7453 | */ | |
7454 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
7455 | dev = get_imsm_dev(super, i); | |
7456 | map = get_imsm_map(dev, 0); | |
7457 | start = __le32_to_cpu(map->pba_of_lba0); | |
7458 | end = start + __le32_to_cpu(map->blocks_per_member); | |
7459 | if ((new_start >= start && new_start <= end) || | |
7460 | (start >= new_start && start <= new_end)) | |
7461 | /* overlap */; | |
7462 | else | |
7463 | continue; | |
7464 | ||
7465 | if (disks_overlap(super, i, u)) { | |
7466 | dprintf("%s: arrays overlap\n", __func__); | |
7467 | goto create_error; | |
7468 | } | |
7469 | } | |
7470 | ||
7471 | /* check that prepare update was successful */ | |
7472 | if (!update->space) { | |
7473 | dprintf("%s: prepare update failed\n", __func__); | |
7474 | goto create_error; | |
7475 | } | |
7476 | ||
7477 | /* check that all disks are still active before committing | |
7478 | * changes. FIXME: could we instead handle this by creating a | |
7479 | * degraded array? That's probably not what the user expects, | |
7480 | * so better to drop this update on the floor. | |
7481 | */ | |
7482 | for (i = 0; i < new_map->num_members; i++) { | |
7483 | dl = serial_to_dl(inf[i].serial, super); | |
7484 | if (!dl) { | |
7485 | dprintf("%s: disk disappeared\n", __func__); | |
7486 | goto create_error; | |
7487 | } | |
7488 | } | |
7489 | ||
7490 | super->updates_pending++; | |
7491 | ||
7492 | /* convert spares to members and fixup ord_tbl */ | |
7493 | for (i = 0; i < new_map->num_members; i++) { | |
7494 | dl = serial_to_dl(inf[i].serial, super); | |
7495 | if (dl->index == -1) { | |
7496 | dl->index = mpb->num_disks; | |
7497 | mpb->num_disks++; | |
7498 | dl->disk.status |= CONFIGURED_DISK; | |
7499 | dl->disk.status &= ~SPARE_DISK; | |
7500 | } | |
7501 | set_imsm_ord_tbl_ent(new_map, i, dl->index); | |
7502 | } | |
7503 | ||
7504 | dv = update->space; | |
7505 | dev = dv->dev; | |
7506 | update->space = NULL; | |
7507 | imsm_copy_dev(dev, &u->dev); | |
7508 | dv->index = u->dev_idx; | |
7509 | dv->next = super->devlist; | |
7510 | super->devlist = dv; | |
7511 | mpb->num_raid_devs++; | |
7512 | ||
7513 | imsm_update_version_info(super); | |
7514 | break; | |
7515 | create_error: | |
7516 | /* mdmon knows how to release update->space, but not | |
7517 | * ((struct intel_dev *) update->space)->dev | |
7518 | */ | |
7519 | if (update->space) { | |
7520 | dv = update->space; | |
7521 | free(dv->dev); | |
7522 | } | |
7523 | break; | |
7524 | } | |
7525 | case update_kill_array: { | |
7526 | struct imsm_update_kill_array *u = (void *) update->buf; | |
7527 | int victim = u->dev_idx; | |
7528 | struct active_array *a; | |
7529 | struct intel_dev **dp; | |
7530 | struct imsm_dev *dev; | |
7531 | ||
7532 | /* sanity check that we are not affecting the uuid of | |
7533 | * active arrays, or deleting an active array | |
7534 | * | |
7535 | * FIXME when immutable ids are available, but note that | |
7536 | * we'll also need to fixup the invalidated/active | |
7537 | * subarray indexes in mdstat | |
7538 | */ | |
7539 | for (a = st->arrays; a; a = a->next) | |
7540 | if (a->info.container_member >= victim) | |
7541 | break; | |
7542 | /* by definition if mdmon is running at least one array | |
7543 | * is active in the container, so checking | |
7544 | * mpb->num_raid_devs is just extra paranoia | |
7545 | */ | |
7546 | dev = get_imsm_dev(super, victim); | |
7547 | if (a || !dev || mpb->num_raid_devs == 1) { | |
7548 | dprintf("failed to delete subarray-%d\n", victim); | |
7549 | break; | |
7550 | } | |
7551 | ||
7552 | for (dp = &super->devlist; *dp;) | |
7553 | if ((*dp)->index == (unsigned)super->current_vol) { | |
7554 | *dp = (*dp)->next; | |
7555 | } else { | |
7556 | if ((*dp)->index > (unsigned)victim) | |
7557 | (*dp)->index--; | |
7558 | dp = &(*dp)->next; | |
7559 | } | |
7560 | mpb->num_raid_devs--; | |
7561 | super->updates_pending++; | |
7562 | break; | |
7563 | } | |
7564 | case update_rename_array: { | |
7565 | struct imsm_update_rename_array *u = (void *) update->buf; | |
7566 | char name[MAX_RAID_SERIAL_LEN+1]; | |
7567 | int target = u->dev_idx; | |
7568 | struct active_array *a; | |
7569 | struct imsm_dev *dev; | |
7570 | ||
7571 | /* sanity check that we are not affecting the uuid of | |
7572 | * an active array | |
7573 | */ | |
7574 | snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name); | |
7575 | name[MAX_RAID_SERIAL_LEN] = '\0'; | |
7576 | for (a = st->arrays; a; a = a->next) | |
7577 | if (a->info.container_member == target) | |
7578 | break; | |
7579 | dev = get_imsm_dev(super, u->dev_idx); | |
7580 | if (a || !dev || !check_name(super, name, 1)) { | |
7581 | dprintf("failed to rename subarray-%d\n", target); | |
7582 | break; | |
7583 | } | |
7584 | ||
7585 | snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name); | |
7586 | super->updates_pending++; | |
7587 | break; | |
7588 | } | |
7589 | case update_add_remove_disk: { | |
7590 | /* we may be able to repair some arrays if disks are | |
7591 | * being added, check teh status of add_remove_disk | |
7592 | * if discs has been added. | |
7593 | */ | |
7594 | if (add_remove_disk_update(super)) { | |
7595 | struct active_array *a; | |
7596 | ||
7597 | super->updates_pending++; | |
7598 | for (a = st->arrays; a; a = a->next) | |
7599 | a->check_degraded = 1; | |
7600 | } | |
7601 | break; | |
7602 | } | |
7603 | default: | |
7604 | fprintf(stderr, "error: unsuported process update type:" | |
7605 | "(type: %d)\n", type); | |
7606 | } | |
7607 | } | |
7608 | ||
7609 | static struct mdinfo *get_spares_for_grow(struct supertype *st); | |
7610 | ||
7611 | static void imsm_prepare_update(struct supertype *st, | |
7612 | struct metadata_update *update) | |
7613 | { | |
7614 | /** | |
7615 | * Allocate space to hold new disk entries, raid-device entries or a new | |
7616 | * mpb if necessary. The manager synchronously waits for updates to | |
7617 | * complete in the monitor, so new mpb buffers allocated here can be | |
7618 | * integrated by the monitor thread without worrying about live pointers | |
7619 | * in the manager thread. | |
7620 | */ | |
7621 | enum imsm_update_type type = *(enum imsm_update_type *) update->buf; | |
7622 | struct intel_super *super = st->sb; | |
7623 | struct imsm_super *mpb = super->anchor; | |
7624 | size_t buf_len; | |
7625 | size_t len = 0; | |
7626 | ||
7627 | switch (type) { | |
7628 | case update_general_migration_checkpoint: | |
7629 | dprintf("imsm: prepare_update() " | |
7630 | "for update_general_migration_checkpoint called\n"); | |
7631 | break; | |
7632 | case update_takeover: { | |
7633 | struct imsm_update_takeover *u = (void *)update->buf; | |
7634 | if (u->direction == R0_TO_R10) { | |
7635 | void **tail = (void **)&update->space_list; | |
7636 | struct imsm_dev *dev = get_imsm_dev(super, u->subarray); | |
7637 | struct imsm_map *map = get_imsm_map(dev, 0); | |
7638 | int num_members = map->num_members; | |
7639 | void *space; | |
7640 | int size, i; | |
7641 | int err = 0; | |
7642 | /* allocate memory for added disks */ | |
7643 | for (i = 0; i < num_members; i++) { | |
7644 | size = sizeof(struct dl); | |
7645 | space = malloc(size); | |
7646 | if (!space) { | |
7647 | err++; | |
7648 | break; | |
7649 | } | |
7650 | *tail = space; | |
7651 | tail = space; | |
7652 | *tail = NULL; | |
7653 | } | |
7654 | /* allocate memory for new device */ | |
7655 | size = sizeof_imsm_dev(super->devlist->dev, 0) + | |
7656 | (num_members * sizeof(__u32)); | |
7657 | space = malloc(size); | |
7658 | if (!space) | |
7659 | err++; | |
7660 | else { | |
7661 | *tail = space; | |
7662 | tail = space; | |
7663 | *tail = NULL; | |
7664 | } | |
7665 | if (!err) { | |
7666 | len = disks_to_mpb_size(num_members * 2); | |
7667 | } else { | |
7668 | /* if allocation didn't success, free buffer */ | |
7669 | while (update->space_list) { | |
7670 | void **sp = update->space_list; | |
7671 | update->space_list = *sp; | |
7672 | free(sp); | |
7673 | } | |
7674 | } | |
7675 | } | |
7676 | ||
7677 | break; | |
7678 | } | |
7679 | case update_reshape_container_disks: { | |
7680 | /* Every raid device in the container is about to | |
7681 | * gain some more devices, and we will enter a | |
7682 | * reconfiguration. | |
7683 | * So each 'imsm_map' will be bigger, and the imsm_vol | |
7684 | * will now hold 2 of them. | |
7685 | * Thus we need new 'struct imsm_dev' allocations sized | |
7686 | * as sizeof_imsm_dev but with more devices in both maps. | |
7687 | */ | |
7688 | struct imsm_update_reshape *u = (void *)update->buf; | |
7689 | struct intel_dev *dl; | |
7690 | void **space_tail = (void**)&update->space_list; | |
7691 | ||
7692 | dprintf("imsm: imsm_prepare_update() for update_reshape\n"); | |
7693 | ||
7694 | for (dl = super->devlist; dl; dl = dl->next) { | |
7695 | int size = sizeof_imsm_dev(dl->dev, 1); | |
7696 | void *s; | |
7697 | if (u->new_raid_disks > u->old_raid_disks) | |
7698 | size += sizeof(__u32)*2* | |
7699 | (u->new_raid_disks - u->old_raid_disks); | |
7700 | s = malloc(size); | |
7701 | if (!s) | |
7702 | break; | |
7703 | *space_tail = s; | |
7704 | space_tail = s; | |
7705 | *space_tail = NULL; | |
7706 | } | |
7707 | ||
7708 | len = disks_to_mpb_size(u->new_raid_disks); | |
7709 | dprintf("New anchor length is %llu\n", (unsigned long long)len); | |
7710 | break; | |
7711 | } | |
7712 | case update_reshape_migration: { | |
7713 | /* for migration level 0->5 we need to add disks | |
7714 | * so the same as for container operation we will copy | |
7715 | * device to the bigger location. | |
7716 | * in memory prepared device and new disk area are prepared | |
7717 | * for usage in process update | |
7718 | */ | |
7719 | struct imsm_update_reshape_migration *u = (void *)update->buf; | |
7720 | struct intel_dev *id; | |
7721 | void **space_tail = (void **)&update->space_list; | |
7722 | int size; | |
7723 | void *s; | |
7724 | int current_level = -1; | |
7725 | ||
7726 | dprintf("imsm: imsm_prepare_update() for update_reshape\n"); | |
7727 | ||
7728 | /* add space for bigger array in update | |
7729 | */ | |
7730 | for (id = super->devlist; id; id = id->next) { | |
7731 | if (id->index == (unsigned)u->subdev) { | |
7732 | size = sizeof_imsm_dev(id->dev, 1); | |
7733 | if (u->new_raid_disks > u->old_raid_disks) | |
7734 | size += sizeof(__u32)*2* | |
7735 | (u->new_raid_disks - u->old_raid_disks); | |
7736 | s = malloc(size); | |
7737 | if (!s) | |
7738 | break; | |
7739 | *space_tail = s; | |
7740 | space_tail = s; | |
7741 | *space_tail = NULL; | |
7742 | break; | |
7743 | } | |
7744 | } | |
7745 | if (update->space_list == NULL) | |
7746 | break; | |
7747 | ||
7748 | /* add space for disk in update | |
7749 | */ | |
7750 | size = sizeof(struct dl); | |
7751 | s = malloc(size); | |
7752 | if (!s) { | |
7753 | free(update->space_list); | |
7754 | update->space_list = NULL; | |
7755 | break; | |
7756 | } | |
7757 | *space_tail = s; | |
7758 | space_tail = s; | |
7759 | *space_tail = NULL; | |
7760 | ||
7761 | /* add spare device to update | |
7762 | */ | |
7763 | for (id = super->devlist ; id; id = id->next) | |
7764 | if (id->index == (unsigned)u->subdev) { | |
7765 | struct imsm_dev *dev; | |
7766 | struct imsm_map *map; | |
7767 | ||
7768 | dev = get_imsm_dev(super, u->subdev); | |
7769 | map = get_imsm_map(dev, 0); | |
7770 | current_level = map->raid_level; | |
7771 | break; | |
7772 | } | |
7773 | if ((u->new_level == 5) && (u->new_level != current_level)) { | |
7774 | struct mdinfo *spares; | |
7775 | ||
7776 | spares = get_spares_for_grow(st); | |
7777 | if (spares) { | |
7778 | struct dl *dl; | |
7779 | struct mdinfo *dev; | |
7780 | ||
7781 | dev = spares->devs; | |
7782 | if (dev) { | |
7783 | u->new_disks[0] = | |
7784 | makedev(dev->disk.major, | |
7785 | dev->disk.minor); | |
7786 | dl = get_disk_super(super, | |
7787 | dev->disk.major, | |
7788 | dev->disk.minor); | |
7789 | dl->index = u->old_raid_disks; | |
7790 | dev = dev->next; | |
7791 | } | |
7792 | sysfs_free(spares); | |
7793 | } | |
7794 | } | |
7795 | len = disks_to_mpb_size(u->new_raid_disks); | |
7796 | dprintf("New anchor length is %llu\n", (unsigned long long)len); | |
7797 | break; | |
7798 | } | |
7799 | case update_create_array: { | |
7800 | struct imsm_update_create_array *u = (void *) update->buf; | |
7801 | struct intel_dev *dv; | |
7802 | struct imsm_dev *dev = &u->dev; | |
7803 | struct imsm_map *map = get_imsm_map(dev, 0); | |
7804 | struct dl *dl; | |
7805 | struct disk_info *inf; | |
7806 | int i; | |
7807 | int activate = 0; | |
7808 | ||
7809 | inf = get_disk_info(u); | |
7810 | len = sizeof_imsm_dev(dev, 1); | |
7811 | /* allocate a new super->devlist entry */ | |
7812 | dv = malloc(sizeof(*dv)); | |
7813 | if (dv) { | |
7814 | dv->dev = malloc(len); | |
7815 | if (dv->dev) | |
7816 | update->space = dv; | |
7817 | else { | |
7818 | free(dv); | |
7819 | update->space = NULL; | |
7820 | } | |
7821 | } | |
7822 | ||
7823 | /* count how many spares will be converted to members */ | |
7824 | for (i = 0; i < map->num_members; i++) { | |
7825 | dl = serial_to_dl(inf[i].serial, super); | |
7826 | if (!dl) { | |
7827 | /* hmm maybe it failed?, nothing we can do about | |
7828 | * it here | |
7829 | */ | |
7830 | continue; | |
7831 | } | |
7832 | if (count_memberships(dl, super) == 0) | |
7833 | activate++; | |
7834 | } | |
7835 | len += activate * sizeof(struct imsm_disk); | |
7836 | break; | |
7837 | default: | |
7838 | break; | |
7839 | } | |
7840 | } | |
7841 | ||
7842 | /* check if we need a larger metadata buffer */ | |
7843 | if (super->next_buf) | |
7844 | buf_len = super->next_len; | |
7845 | else | |
7846 | buf_len = super->len; | |
7847 | ||
7848 | if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) { | |
7849 | /* ok we need a larger buf than what is currently allocated | |
7850 | * if this allocation fails process_update will notice that | |
7851 | * ->next_len is set and ->next_buf is NULL | |
7852 | */ | |
7853 | buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512); | |
7854 | if (super->next_buf) | |
7855 | free(super->next_buf); | |
7856 | ||
7857 | super->next_len = buf_len; | |
7858 | if (posix_memalign(&super->next_buf, 512, buf_len) == 0) | |
7859 | memset(super->next_buf, 0, buf_len); | |
7860 | else | |
7861 | super->next_buf = NULL; | |
7862 | } | |
7863 | } | |
7864 | ||
7865 | /* must be called while manager is quiesced */ | |
7866 | static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index) | |
7867 | { | |
7868 | struct imsm_super *mpb = super->anchor; | |
7869 | struct dl *iter; | |
7870 | struct imsm_dev *dev; | |
7871 | struct imsm_map *map; | |
7872 | int i, j, num_members; | |
7873 | __u32 ord; | |
7874 | ||
7875 | dprintf("%s: deleting device[%d] from imsm_super\n", | |
7876 | __func__, index); | |
7877 | ||
7878 | /* shift all indexes down one */ | |
7879 | for (iter = super->disks; iter; iter = iter->next) | |
7880 | if (iter->index > (int)index) | |
7881 | iter->index--; | |
7882 | for (iter = super->missing; iter; iter = iter->next) | |
7883 | if (iter->index > (int)index) | |
7884 | iter->index--; | |
7885 | ||
7886 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
7887 | dev = get_imsm_dev(super, i); | |
7888 | map = get_imsm_map(dev, 0); | |
7889 | num_members = map->num_members; | |
7890 | for (j = 0; j < num_members; j++) { | |
7891 | /* update ord entries being careful not to propagate | |
7892 | * ord-flags to the first map | |
7893 | */ | |
7894 | ord = get_imsm_ord_tbl_ent(dev, j, -1); | |
7895 | ||
7896 | if (ord_to_idx(ord) <= index) | |
7897 | continue; | |
7898 | ||
7899 | map = get_imsm_map(dev, 0); | |
7900 | set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1)); | |
7901 | map = get_imsm_map(dev, 1); | |
7902 | if (map) | |
7903 | set_imsm_ord_tbl_ent(map, j, ord - 1); | |
7904 | } | |
7905 | } | |
7906 | ||
7907 | mpb->num_disks--; | |
7908 | super->updates_pending++; | |
7909 | if (*dlp) { | |
7910 | struct dl *dl = *dlp; | |
7911 | ||
7912 | *dlp = (*dlp)->next; | |
7913 | __free_imsm_disk(dl); | |
7914 | } | |
7915 | } | |
7916 | #endif /* MDASSEMBLE */ | |
7917 | /******************************************************************************* | |
7918 | * Function: open_backup_targets | |
7919 | * Description: Function opens file descriptors for all devices given in | |
7920 | * info->devs | |
7921 | * Parameters: | |
7922 | * info : general array info | |
7923 | * raid_disks : number of disks | |
7924 | * raid_fds : table of device's file descriptors | |
7925 | * Returns: | |
7926 | * 0 : success | |
7927 | * -1 : fail | |
7928 | ******************************************************************************/ | |
7929 | int open_backup_targets(struct mdinfo *info, int raid_disks, int *raid_fds) | |
7930 | { | |
7931 | struct mdinfo *sd; | |
7932 | ||
7933 | for (sd = info->devs ; sd ; sd = sd->next) { | |
7934 | char *dn; | |
7935 | ||
7936 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) { | |
7937 | dprintf("disk is faulty!!\n"); | |
7938 | continue; | |
7939 | } | |
7940 | ||
7941 | if ((sd->disk.raid_disk >= raid_disks) || | |
7942 | (sd->disk.raid_disk < 0)) | |
7943 | continue; | |
7944 | ||
7945 | dn = map_dev(sd->disk.major, | |
7946 | sd->disk.minor, 1); | |
7947 | raid_fds[sd->disk.raid_disk] = dev_open(dn, O_RDWR); | |
7948 | if (raid_fds[sd->disk.raid_disk] < 0) { | |
7949 | fprintf(stderr, "cannot open component\n"); | |
7950 | return -1; | |
7951 | } | |
7952 | } | |
7953 | return 0; | |
7954 | } | |
7955 | ||
7956 | #ifndef MDASSEMBLE | |
7957 | /******************************************************************************* | |
7958 | * Function: init_migr_record_imsm | |
7959 | * Description: Function inits imsm migration record | |
7960 | * Parameters: | |
7961 | * super : imsm internal array info | |
7962 | * dev : device under migration | |
7963 | * info : general array info to find the smallest device | |
7964 | * Returns: | |
7965 | * none | |
7966 | ******************************************************************************/ | |
7967 | void init_migr_record_imsm(struct supertype *st, struct imsm_dev *dev, | |
7968 | struct mdinfo *info) | |
7969 | { | |
7970 | struct intel_super *super = st->sb; | |
7971 | struct migr_record *migr_rec = super->migr_rec; | |
7972 | int new_data_disks; | |
7973 | unsigned long long dsize, dev_sectors; | |
7974 | long long unsigned min_dev_sectors = -1LLU; | |
7975 | struct mdinfo *sd; | |
7976 | char nm[30]; | |
7977 | int fd; | |
7978 | struct imsm_map *map_dest = get_imsm_map(dev, 0); | |
7979 | struct imsm_map *map_src = get_imsm_map(dev, 1); | |
7980 | unsigned long long num_migr_units; | |
7981 | unsigned long long array_blocks; | |
7982 | ||
7983 | memset(migr_rec, 0, sizeof(struct migr_record)); | |
7984 | migr_rec->family_num = __cpu_to_le32(super->anchor->family_num); | |
7985 | ||
7986 | /* only ascending reshape supported now */ | |
7987 | migr_rec->ascending_migr = __cpu_to_le32(1); | |
7988 | ||
7989 | migr_rec->dest_depth_per_unit = GEN_MIGR_AREA_SIZE / | |
7990 | max(map_dest->blocks_per_strip, map_src->blocks_per_strip); | |
7991 | migr_rec->dest_depth_per_unit *= map_dest->blocks_per_strip; | |
7992 | new_data_disks = imsm_num_data_members(dev, 0); | |
7993 | migr_rec->blocks_per_unit = | |
7994 | __cpu_to_le32(migr_rec->dest_depth_per_unit * new_data_disks); | |
7995 | migr_rec->dest_depth_per_unit = | |
7996 | __cpu_to_le32(migr_rec->dest_depth_per_unit); | |
7997 | array_blocks = info->component_size * new_data_disks; | |
7998 | num_migr_units = | |
7999 | array_blocks / __le32_to_cpu(migr_rec->blocks_per_unit); | |
8000 | ||
8001 | if (array_blocks % __le32_to_cpu(migr_rec->blocks_per_unit)) | |
8002 | num_migr_units++; | |
8003 | migr_rec->num_migr_units = __cpu_to_le32(num_migr_units); | |
8004 | ||
8005 | migr_rec->post_migr_vol_cap = dev->size_low; | |
8006 | migr_rec->post_migr_vol_cap_hi = dev->size_high; | |
8007 | ||
8008 | ||
8009 | /* Find the smallest dev */ | |
8010 | for (sd = info->devs ; sd ; sd = sd->next) { | |
8011 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
8012 | fd = dev_open(nm, O_RDONLY); | |
8013 | if (fd < 0) | |
8014 | continue; | |
8015 | get_dev_size(fd, NULL, &dsize); | |
8016 | dev_sectors = dsize / 512; | |
8017 | if (dev_sectors < min_dev_sectors) | |
8018 | min_dev_sectors = dev_sectors; | |
8019 | close(fd); | |
8020 | } | |
8021 | migr_rec->ckpt_area_pba = __cpu_to_le32(min_dev_sectors - | |
8022 | RAID_DISK_RESERVED_BLOCKS_IMSM_HI); | |
8023 | ||
8024 | write_imsm_migr_rec(st); | |
8025 | ||
8026 | return; | |
8027 | } | |
8028 | ||
8029 | /******************************************************************************* | |
8030 | * Function: save_backup_imsm | |
8031 | * Description: Function saves critical data stripes to Migration Copy Area | |
8032 | * and updates the current migration unit status. | |
8033 | * Use restore_stripes() to form a destination stripe, | |
8034 | * and to write it to the Copy Area. | |
8035 | * Parameters: | |
8036 | * st : supertype information | |
8037 | * dev : imsm device that backup is saved for | |
8038 | * info : general array info | |
8039 | * buf : input buffer | |
8040 | * length : length of data to backup (blocks_per_unit) | |
8041 | * Returns: | |
8042 | * 0 : success | |
8043 | *, -1 : fail | |
8044 | ******************************************************************************/ | |
8045 | int save_backup_imsm(struct supertype *st, | |
8046 | struct imsm_dev *dev, | |
8047 | struct mdinfo *info, | |
8048 | void *buf, | |
8049 | int length) | |
8050 | { | |
8051 | int rv = -1; | |
8052 | struct intel_super *super = st->sb; | |
8053 | unsigned long long *target_offsets = NULL; | |
8054 | int *targets = NULL; | |
8055 | int i; | |
8056 | struct imsm_map *map_dest = get_imsm_map(dev, 0); | |
8057 | int new_disks = map_dest->num_members; | |
8058 | int dest_layout = 0; | |
8059 | int dest_chunk; | |
8060 | unsigned long long start; | |
8061 | int data_disks = imsm_num_data_members(dev, 0); | |
8062 | ||
8063 | targets = malloc(new_disks * sizeof(int)); | |
8064 | if (!targets) | |
8065 | goto abort; | |
8066 | ||
8067 | for (i = 0; i < new_disks; i++) | |
8068 | targets[i] = -1; | |
8069 | ||
8070 | target_offsets = malloc(new_disks * sizeof(unsigned long long)); | |
8071 | if (!target_offsets) | |
8072 | goto abort; | |
8073 | ||
8074 | start = info->reshape_progress * 512; | |
8075 | for (i = 0; i < new_disks; i++) { | |
8076 | target_offsets[i] = (unsigned long long) | |
8077 | __le32_to_cpu(super->migr_rec->ckpt_area_pba) * 512; | |
8078 | /* move back copy area adderss, it will be moved forward | |
8079 | * in restore_stripes() using start input variable | |
8080 | */ | |
8081 | target_offsets[i] -= start/data_disks; | |
8082 | } | |
8083 | ||
8084 | if (open_backup_targets(info, new_disks, targets)) | |
8085 | goto abort; | |
8086 | ||
8087 | dest_layout = imsm_level_to_layout(map_dest->raid_level); | |
8088 | dest_chunk = __le16_to_cpu(map_dest->blocks_per_strip) * 512; | |
8089 | ||
8090 | if (restore_stripes(targets, /* list of dest devices */ | |
8091 | target_offsets, /* migration record offsets */ | |
8092 | new_disks, | |
8093 | dest_chunk, | |
8094 | map_dest->raid_level, | |
8095 | dest_layout, | |
8096 | -1, /* source backup file descriptor */ | |
8097 | 0, /* input buf offset | |
8098 | * always 0 buf is already offseted */ | |
8099 | start, | |
8100 | length, | |
8101 | buf) != 0) { | |
8102 | fprintf(stderr, Name ": Error restoring stripes\n"); | |
8103 | goto abort; | |
8104 | } | |
8105 | ||
8106 | rv = 0; | |
8107 | ||
8108 | abort: | |
8109 | if (targets) { | |
8110 | for (i = 0; i < new_disks; i++) | |
8111 | if (targets[i] >= 0) | |
8112 | close(targets[i]); | |
8113 | free(targets); | |
8114 | } | |
8115 | free(target_offsets); | |
8116 | ||
8117 | return rv; | |
8118 | } | |
8119 | ||
8120 | /******************************************************************************* | |
8121 | * Function: save_checkpoint_imsm | |
8122 | * Description: Function called for current unit status update | |
8123 | * in the migration record. It writes it to disk. | |
8124 | * Parameters: | |
8125 | * super : imsm internal array info | |
8126 | * info : general array info | |
8127 | * Returns: | |
8128 | * 0: success | |
8129 | * 1: failure | |
8130 | * 2: failure, means no valid migration record | |
8131 | * / no general migration in progress / | |
8132 | ******************************************************************************/ | |
8133 | int save_checkpoint_imsm(struct supertype *st, struct mdinfo *info, int state) | |
8134 | { | |
8135 | struct intel_super *super = st->sb; | |
8136 | unsigned long long blocks_per_unit; | |
8137 | unsigned long long curr_migr_unit; | |
8138 | ||
8139 | if (load_imsm_migr_rec(super, info) != 0) { | |
8140 | dprintf("imsm: ERROR: Cannot read migration record " | |
8141 | "for checkpoint save.\n"); | |
8142 | return 1; | |
8143 | } | |
8144 | ||
8145 | blocks_per_unit = __le32_to_cpu(super->migr_rec->blocks_per_unit); | |
8146 | if (blocks_per_unit == 0) { | |
8147 | dprintf("imsm: no migration in progress.\n"); | |
8148 | return 2; | |
8149 | } | |
8150 | curr_migr_unit = info->reshape_progress / blocks_per_unit; | |
8151 | /* check if array is alligned to copy area | |
8152 | * if it is not alligned, add one to current migration unit value | |
8153 | * this can happend on array reshape finish only | |
8154 | */ | |
8155 | if (info->reshape_progress % blocks_per_unit) | |
8156 | curr_migr_unit++; | |
8157 | ||
8158 | super->migr_rec->curr_migr_unit = | |
8159 | __cpu_to_le32(curr_migr_unit); | |
8160 | super->migr_rec->rec_status = __cpu_to_le32(state); | |
8161 | super->migr_rec->dest_1st_member_lba = | |
8162 | __cpu_to_le32(curr_migr_unit * | |
8163 | __le32_to_cpu(super->migr_rec->dest_depth_per_unit)); | |
8164 | if (write_imsm_migr_rec(st) < 0) { | |
8165 | dprintf("imsm: Cannot write migration record " | |
8166 | "outside backup area\n"); | |
8167 | return 1; | |
8168 | } | |
8169 | ||
8170 | return 0; | |
8171 | } | |
8172 | ||
8173 | /******************************************************************************* | |
8174 | * Function: recover_backup_imsm | |
8175 | * Description: Function recovers critical data from the Migration Copy Area | |
8176 | * while assembling an array. | |
8177 | * Parameters: | |
8178 | * super : imsm internal array info | |
8179 | * info : general array info | |
8180 | * Returns: | |
8181 | * 0 : success (or there is no data to recover) | |
8182 | * 1 : fail | |
8183 | ******************************************************************************/ | |
8184 | int recover_backup_imsm(struct supertype *st, struct mdinfo *info) | |
8185 | { | |
8186 | struct intel_super *super = st->sb; | |
8187 | struct migr_record *migr_rec = super->migr_rec; | |
8188 | struct imsm_map *map_dest = NULL; | |
8189 | struct intel_dev *id = NULL; | |
8190 | unsigned long long read_offset; | |
8191 | unsigned long long write_offset; | |
8192 | unsigned unit_len; | |
8193 | int *targets = NULL; | |
8194 | int new_disks, i, err; | |
8195 | char *buf = NULL; | |
8196 | int retval = 1; | |
8197 | unsigned long curr_migr_unit = __le32_to_cpu(migr_rec->curr_migr_unit); | |
8198 | unsigned long num_migr_units = __le32_to_cpu(migr_rec->num_migr_units); | |
8199 | char buffer[20]; | |
8200 | int skipped_disks = 0; | |
8201 | int max_degradation; | |
8202 | ||
8203 | err = sysfs_get_str(info, NULL, "array_state", (char *)buffer, 20); | |
8204 | if (err < 1) | |
8205 | return 1; | |
8206 | ||
8207 | /* recover data only during assemblation */ | |
8208 | if (strncmp(buffer, "inactive", 8) != 0) | |
8209 | return 0; | |
8210 | /* no data to recover */ | |
8211 | if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL) | |
8212 | return 0; | |
8213 | if (curr_migr_unit >= num_migr_units) | |
8214 | return 1; | |
8215 | ||
8216 | /* find device during reshape */ | |
8217 | for (id = super->devlist; id; id = id->next) | |
8218 | if (is_gen_migration(id->dev)) | |
8219 | break; | |
8220 | if (id == NULL) | |
8221 | return 1; | |
8222 | ||
8223 | map_dest = get_imsm_map(id->dev, 0); | |
8224 | new_disks = map_dest->num_members; | |
8225 | max_degradation = new_disks - imsm_num_data_members(id->dev, 0); | |
8226 | ||
8227 | read_offset = (unsigned long long) | |
8228 | __le32_to_cpu(migr_rec->ckpt_area_pba) * 512; | |
8229 | ||
8230 | write_offset = ((unsigned long long) | |
8231 | __le32_to_cpu(migr_rec->dest_1st_member_lba) + | |
8232 | __le32_to_cpu(map_dest->pba_of_lba0)) * 512; | |
8233 | ||
8234 | unit_len = __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512; | |
8235 | if (posix_memalign((void **)&buf, 512, unit_len) != 0) | |
8236 | goto abort; | |
8237 | targets = malloc(new_disks * sizeof(int)); | |
8238 | if (!targets) | |
8239 | goto abort; | |
8240 | ||
8241 | open_backup_targets(info, new_disks, targets); | |
8242 | ||
8243 | for (i = 0; i < new_disks; i++) { | |
8244 | if (targets[i] < 0) { | |
8245 | skipped_disks++; | |
8246 | continue; | |
8247 | } | |
8248 | if (lseek64(targets[i], read_offset, SEEK_SET) < 0) { | |
8249 | fprintf(stderr, | |
8250 | Name ": Cannot seek to block: %s\n", | |
8251 | strerror(errno)); | |
8252 | goto abort; | |
8253 | } | |
8254 | if ((unsigned)read(targets[i], buf, unit_len) != unit_len) { | |
8255 | fprintf(stderr, | |
8256 | Name ": Cannot read copy area block: %s\n", | |
8257 | strerror(errno)); | |
8258 | goto abort; | |
8259 | } | |
8260 | if (lseek64(targets[i], write_offset, SEEK_SET) < 0) { | |
8261 | fprintf(stderr, | |
8262 | Name ": Cannot seek to block: %s\n", | |
8263 | strerror(errno)); | |
8264 | goto abort; | |
8265 | } | |
8266 | if ((unsigned)write(targets[i], buf, unit_len) != unit_len) { | |
8267 | fprintf(stderr, | |
8268 | Name ": Cannot restore block: %s\n", | |
8269 | strerror(errno)); | |
8270 | goto abort; | |
8271 | } | |
8272 | } | |
8273 | ||
8274 | if (skipped_disks > max_degradation) { | |
8275 | fprintf(stderr, | |
8276 | Name ": Cannot restore data from backup." | |
8277 | " Too many failed disks\n"); | |
8278 | goto abort; | |
8279 | } | |
8280 | ||
8281 | if (save_checkpoint_imsm(st, info, UNIT_SRC_NORMAL)) { | |
8282 | /* ignore error == 2, this can mean end of reshape here | |
8283 | */ | |
8284 | dprintf("imsm: Cannot write checkpoint to " | |
8285 | "migration record (UNIT_SRC_NORMAL) during restart\n"); | |
8286 | } else | |
8287 | retval = 0; | |
8288 | ||
8289 | abort: | |
8290 | if (targets) { | |
8291 | for (i = 0; i < new_disks; i++) | |
8292 | if (targets[i]) | |
8293 | close(targets[i]); | |
8294 | free(targets); | |
8295 | } | |
8296 | free(buf); | |
8297 | return retval; | |
8298 | } | |
8299 | ||
8300 | static char disk_by_path[] = "/dev/disk/by-path/"; | |
8301 | ||
8302 | static const char *imsm_get_disk_controller_domain(const char *path) | |
8303 | { | |
8304 | char disk_path[PATH_MAX]; | |
8305 | char *drv=NULL; | |
8306 | struct stat st; | |
8307 | ||
8308 | strncpy(disk_path, disk_by_path, PATH_MAX - 1); | |
8309 | strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1); | |
8310 | if (stat(disk_path, &st) == 0) { | |
8311 | struct sys_dev* hba; | |
8312 | char *path=NULL; | |
8313 | ||
8314 | path = devt_to_devpath(st.st_rdev); | |
8315 | if (path == NULL) | |
8316 | return "unknown"; | |
8317 | hba = find_disk_attached_hba(-1, path); | |
8318 | if (hba && hba->type == SYS_DEV_SAS) | |
8319 | drv = "isci"; | |
8320 | else if (hba && hba->type == SYS_DEV_SATA) | |
8321 | drv = "ahci"; | |
8322 | else | |
8323 | drv = "unknown"; | |
8324 | dprintf("path: %s hba: %s attached: %s\n", | |
8325 | path, (hba) ? hba->path : "NULL", drv); | |
8326 | free(path); | |
8327 | if (hba) | |
8328 | free_sys_dev(&hba); | |
8329 | } | |
8330 | return drv; | |
8331 | } | |
8332 | ||
8333 | static int imsm_find_array_minor_by_subdev(int subdev, int container, int *minor) | |
8334 | { | |
8335 | char subdev_name[20]; | |
8336 | struct mdstat_ent *mdstat; | |
8337 | ||
8338 | sprintf(subdev_name, "%d", subdev); | |
8339 | mdstat = mdstat_by_subdev(subdev_name, container); | |
8340 | if (!mdstat) | |
8341 | return -1; | |
8342 | ||
8343 | *minor = mdstat->devnum; | |
8344 | free_mdstat(mdstat); | |
8345 | return 0; | |
8346 | } | |
8347 | ||
8348 | static int imsm_reshape_is_allowed_on_container(struct supertype *st, | |
8349 | struct geo_params *geo, | |
8350 | int *old_raid_disks) | |
8351 | { | |
8352 | /* currently we only support increasing the number of devices | |
8353 | * for a container. This increases the number of device for each | |
8354 | * member array. They must all be RAID0 or RAID5. | |
8355 | */ | |
8356 | int ret_val = 0; | |
8357 | struct mdinfo *info, *member; | |
8358 | int devices_that_can_grow = 0; | |
8359 | ||
8360 | dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): " | |
8361 | "st->devnum = (%i)\n", | |
8362 | st->devnum); | |
8363 | ||
8364 | if (geo->size != -1 || | |
8365 | geo->level != UnSet || | |
8366 | geo->layout != UnSet || | |
8367 | geo->chunksize != 0 || | |
8368 | geo->raid_disks == UnSet) { | |
8369 | dprintf("imsm: Container operation is allowed for " | |
8370 | "raid disks number change only.\n"); | |
8371 | return ret_val; | |
8372 | } | |
8373 | ||
8374 | info = container_content_imsm(st, NULL); | |
8375 | for (member = info; member; member = member->next) { | |
8376 | int result; | |
8377 | int minor; | |
8378 | ||
8379 | dprintf("imsm: checking device_num: %i\n", | |
8380 | member->container_member); | |
8381 | ||
8382 | if (geo->raid_disks <= member->array.raid_disks) { | |
8383 | /* we work on container for Online Capacity Expansion | |
8384 | * only so raid_disks has to grow | |
8385 | */ | |
8386 | dprintf("imsm: for container operation raid disks " | |
8387 | "increase is required\n"); | |
8388 | break; | |
8389 | } | |
8390 | ||
8391 | if ((info->array.level != 0) && | |
8392 | (info->array.level != 5)) { | |
8393 | /* we cannot use this container with other raid level | |
8394 | */ | |
8395 | dprintf("imsm: for container operation wrong" | |
8396 | " raid level (%i) detected\n", | |
8397 | info->array.level); | |
8398 | break; | |
8399 | } else { | |
8400 | /* check for platform support | |
8401 | * for this raid level configuration | |
8402 | */ | |
8403 | struct intel_super *super = st->sb; | |
8404 | if (!is_raid_level_supported(super->orom, | |
8405 | member->array.level, | |
8406 | geo->raid_disks)) { | |
8407 | dprintf("platform does not support raid%d with" | |
8408 | " %d disk%s\n", | |
8409 | info->array.level, | |
8410 | geo->raid_disks, | |
8411 | geo->raid_disks > 1 ? "s" : ""); | |
8412 | break; | |
8413 | } | |
8414 | /* check if component size is aligned to chunk size | |
8415 | */ | |
8416 | if (info->component_size % | |
8417 | (info->array.chunk_size/512)) { | |
8418 | dprintf("Component size is not aligned to " | |
8419 | "chunk size\n"); | |
8420 | break; | |
8421 | } | |
8422 | } | |
8423 | ||
8424 | if (*old_raid_disks && | |
8425 | info->array.raid_disks != *old_raid_disks) | |
8426 | break; | |
8427 | *old_raid_disks = info->array.raid_disks; | |
8428 | ||
8429 | /* All raid5 and raid0 volumes in container | |
8430 | * have to be ready for Online Capacity Expansion | |
8431 | * so they need to be assembled. We have already | |
8432 | * checked that no recovery etc is happening. | |
8433 | */ | |
8434 | result = imsm_find_array_minor_by_subdev(member->container_member, | |
8435 | st->container_dev, | |
8436 | &minor); | |
8437 | if (result < 0) { | |
8438 | dprintf("imsm: cannot find array\n"); | |
8439 | break; | |
8440 | } | |
8441 | devices_that_can_grow++; | |
8442 | } | |
8443 | sysfs_free(info); | |
8444 | if (!member && devices_that_can_grow) | |
8445 | ret_val = 1; | |
8446 | ||
8447 | if (ret_val) | |
8448 | dprintf("\tContainer operation allowed\n"); | |
8449 | else | |
8450 | dprintf("\tError: %i\n", ret_val); | |
8451 | ||
8452 | return ret_val; | |
8453 | } | |
8454 | ||
8455 | /* Function: get_spares_for_grow | |
8456 | * Description: Allocates memory and creates list of spare devices | |
8457 | * avaliable in container. Checks if spare drive size is acceptable. | |
8458 | * Parameters: Pointer to the supertype structure | |
8459 | * Returns: Pointer to the list of spare devices (mdinfo structure) on success, | |
8460 | * NULL if fail | |
8461 | */ | |
8462 | static struct mdinfo *get_spares_for_grow(struct supertype *st) | |
8463 | { | |
8464 | unsigned long long min_size = min_acceptable_spare_size_imsm(st); | |
8465 | return container_choose_spares(st, min_size, NULL, NULL, NULL, 0); | |
8466 | } | |
8467 | ||
8468 | /****************************************************************************** | |
8469 | * function: imsm_create_metadata_update_for_reshape | |
8470 | * Function creates update for whole IMSM container. | |
8471 | * | |
8472 | ******************************************************************************/ | |
8473 | static int imsm_create_metadata_update_for_reshape( | |
8474 | struct supertype *st, | |
8475 | struct geo_params *geo, | |
8476 | int old_raid_disks, | |
8477 | struct imsm_update_reshape **updatep) | |
8478 | { | |
8479 | struct intel_super *super = st->sb; | |
8480 | struct imsm_super *mpb = super->anchor; | |
8481 | int update_memory_size = 0; | |
8482 | struct imsm_update_reshape *u = NULL; | |
8483 | struct mdinfo *spares = NULL; | |
8484 | int i; | |
8485 | int delta_disks = 0; | |
8486 | struct mdinfo *dev; | |
8487 | ||
8488 | dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n", | |
8489 | geo->raid_disks); | |
8490 | ||
8491 | delta_disks = geo->raid_disks - old_raid_disks; | |
8492 | ||
8493 | /* size of all update data without anchor */ | |
8494 | update_memory_size = sizeof(struct imsm_update_reshape); | |
8495 | ||
8496 | /* now add space for spare disks that we need to add. */ | |
8497 | update_memory_size += sizeof(u->new_disks[0]) * (delta_disks - 1); | |
8498 | ||
8499 | u = calloc(1, update_memory_size); | |
8500 | if (u == NULL) { | |
8501 | dprintf("error: " | |
8502 | "cannot get memory for imsm_update_reshape update\n"); | |
8503 | return 0; | |
8504 | } | |
8505 | u->type = update_reshape_container_disks; | |
8506 | u->old_raid_disks = old_raid_disks; | |
8507 | u->new_raid_disks = geo->raid_disks; | |
8508 | ||
8509 | /* now get spare disks list | |
8510 | */ | |
8511 | spares = get_spares_for_grow(st); | |
8512 | ||
8513 | if (spares == NULL | |
8514 | || delta_disks > spares->array.spare_disks) { | |
8515 | fprintf(stderr, Name ": imsm: ERROR: Cannot get spare devices " | |
8516 | "for %s.\n", geo->dev_name); | |
8517 | i = -1; | |
8518 | goto abort; | |
8519 | } | |
8520 | ||
8521 | /* we have got spares | |
8522 | * update disk list in imsm_disk list table in anchor | |
8523 | */ | |
8524 | dprintf("imsm: %i spares are available.\n\n", | |
8525 | spares->array.spare_disks); | |
8526 | ||
8527 | dev = spares->devs; | |
8528 | for (i = 0; i < delta_disks; i++) { | |
8529 | struct dl *dl; | |
8530 | ||
8531 | if (dev == NULL) | |
8532 | break; | |
8533 | u->new_disks[i] = makedev(dev->disk.major, | |
8534 | dev->disk.minor); | |
8535 | dl = get_disk_super(super, dev->disk.major, dev->disk.minor); | |
8536 | dl->index = mpb->num_disks; | |
8537 | mpb->num_disks++; | |
8538 | dev = dev->next; | |
8539 | } | |
8540 | ||
8541 | abort: | |
8542 | /* free spares | |
8543 | */ | |
8544 | sysfs_free(spares); | |
8545 | ||
8546 | dprintf("imsm: reshape update preparation :"); | |
8547 | if (i == delta_disks) { | |
8548 | dprintf(" OK\n"); | |
8549 | *updatep = u; | |
8550 | return update_memory_size; | |
8551 | } | |
8552 | free(u); | |
8553 | dprintf(" Error\n"); | |
8554 | ||
8555 | return 0; | |
8556 | } | |
8557 | ||
8558 | /****************************************************************************** | |
8559 | * function: imsm_create_metadata_update_for_migration() | |
8560 | * Creates update for IMSM array. | |
8561 | * | |
8562 | ******************************************************************************/ | |
8563 | static int imsm_create_metadata_update_for_migration( | |
8564 | struct supertype *st, | |
8565 | struct geo_params *geo, | |
8566 | struct imsm_update_reshape_migration **updatep) | |
8567 | { | |
8568 | struct intel_super *super = st->sb; | |
8569 | int update_memory_size = 0; | |
8570 | struct imsm_update_reshape_migration *u = NULL; | |
8571 | struct imsm_dev *dev; | |
8572 | int previous_level = -1; | |
8573 | ||
8574 | dprintf("imsm_create_metadata_update_for_migration(enter)" | |
8575 | " New Level = %i\n", geo->level); | |
8576 | ||
8577 | /* size of all update data without anchor */ | |
8578 | update_memory_size = sizeof(struct imsm_update_reshape_migration); | |
8579 | ||
8580 | u = calloc(1, update_memory_size); | |
8581 | if (u == NULL) { | |
8582 | dprintf("error: cannot get memory for " | |
8583 | "imsm_create_metadata_update_for_migration\n"); | |
8584 | return 0; | |
8585 | } | |
8586 | u->type = update_reshape_migration; | |
8587 | u->subdev = super->current_vol; | |
8588 | u->new_level = geo->level; | |
8589 | u->new_layout = geo->layout; | |
8590 | u->new_raid_disks = u->old_raid_disks = geo->raid_disks; | |
8591 | u->new_disks[0] = -1; | |
8592 | u->new_chunksize = -1; | |
8593 | ||
8594 | dev = get_imsm_dev(super, u->subdev); | |
8595 | if (dev) { | |
8596 | struct imsm_map *map; | |
8597 | ||
8598 | map = get_imsm_map(dev, 0); | |
8599 | if (map) { | |
8600 | int current_chunk_size = | |
8601 | __le16_to_cpu(map->blocks_per_strip) / 2; | |
8602 | ||
8603 | if (geo->chunksize != current_chunk_size) { | |
8604 | u->new_chunksize = geo->chunksize / 1024; | |
8605 | dprintf("imsm: " | |
8606 | "chunk size change from %i to %i\n", | |
8607 | current_chunk_size, u->new_chunksize); | |
8608 | } | |
8609 | previous_level = map->raid_level; | |
8610 | } | |
8611 | } | |
8612 | if ((geo->level == 5) && (previous_level == 0)) { | |
8613 | struct mdinfo *spares = NULL; | |
8614 | ||
8615 | u->new_raid_disks++; | |
8616 | spares = get_spares_for_grow(st); | |
8617 | if ((spares == NULL) || (spares->array.spare_disks < 1)) { | |
8618 | free(u); | |
8619 | sysfs_free(spares); | |
8620 | update_memory_size = 0; | |
8621 | dprintf("error: cannot get spare device " | |
8622 | "for requested migration"); | |
8623 | return 0; | |
8624 | } | |
8625 | sysfs_free(spares); | |
8626 | } | |
8627 | dprintf("imsm: reshape update preparation : OK\n"); | |
8628 | *updatep = u; | |
8629 | ||
8630 | return update_memory_size; | |
8631 | } | |
8632 | ||
8633 | static void imsm_update_metadata_locally(struct supertype *st, | |
8634 | void *buf, int len) | |
8635 | { | |
8636 | struct metadata_update mu; | |
8637 | ||
8638 | mu.buf = buf; | |
8639 | mu.len = len; | |
8640 | mu.space = NULL; | |
8641 | mu.space_list = NULL; | |
8642 | mu.next = NULL; | |
8643 | imsm_prepare_update(st, &mu); | |
8644 | imsm_process_update(st, &mu); | |
8645 | ||
8646 | while (mu.space_list) { | |
8647 | void **space = mu.space_list; | |
8648 | mu.space_list = *space; | |
8649 | free(space); | |
8650 | } | |
8651 | } | |
8652 | ||
8653 | /*************************************************************************** | |
8654 | * Function: imsm_analyze_change | |
8655 | * Description: Function analyze change for single volume | |
8656 | * and validate if transition is supported | |
8657 | * Parameters: Geometry parameters, supertype structure | |
8658 | * Returns: Operation type code on success, -1 if fail | |
8659 | ****************************************************************************/ | |
8660 | enum imsm_reshape_type imsm_analyze_change(struct supertype *st, | |
8661 | struct geo_params *geo) | |
8662 | { | |
8663 | struct mdinfo info; | |
8664 | int change = -1; | |
8665 | int check_devs = 0; | |
8666 | int chunk; | |
8667 | ||
8668 | getinfo_super_imsm_volume(st, &info, NULL); | |
8669 | if ((geo->level != info.array.level) && | |
8670 | (geo->level >= 0) && | |
8671 | (geo->level != UnSet)) { | |
8672 | switch (info.array.level) { | |
8673 | case 0: | |
8674 | if (geo->level == 5) { | |
8675 | change = CH_MIGRATION; | |
8676 | if (geo->layout != ALGORITHM_LEFT_ASYMMETRIC) { | |
8677 | fprintf(stderr, | |
8678 | Name " Error. Requested Layout " | |
8679 | "not supported (left-asymmetric layout " | |
8680 | "is supported only)!\n"); | |
8681 | change = -1; | |
8682 | goto analyse_change_exit; | |
8683 | } | |
8684 | check_devs = 1; | |
8685 | } | |
8686 | if (geo->level == 10) { | |
8687 | change = CH_TAKEOVER; | |
8688 | check_devs = 1; | |
8689 | } | |
8690 | break; | |
8691 | case 1: | |
8692 | if (geo->level == 0) { | |
8693 | change = CH_TAKEOVER; | |
8694 | check_devs = 1; | |
8695 | } | |
8696 | break; | |
8697 | case 10: | |
8698 | if (geo->level == 0) { | |
8699 | change = CH_TAKEOVER; | |
8700 | check_devs = 1; | |
8701 | } | |
8702 | break; | |
8703 | } | |
8704 | if (change == -1) { | |
8705 | fprintf(stderr, | |
8706 | Name " Error. Level Migration from %d to %d " | |
8707 | "not supported!\n", | |
8708 | info.array.level, geo->level); | |
8709 | goto analyse_change_exit; | |
8710 | } | |
8711 | } else | |
8712 | geo->level = info.array.level; | |
8713 | ||
8714 | if ((geo->layout != info.array.layout) | |
8715 | && ((geo->layout != UnSet) && (geo->layout != -1))) { | |
8716 | change = CH_MIGRATION; | |
8717 | if ((info.array.layout == 0) | |
8718 | && (info.array.level == 5) | |
8719 | && (geo->layout == 5)) { | |
8720 | /* reshape 5 -> 4 */ | |
8721 | } else if ((info.array.layout == 5) | |
8722 | && (info.array.level == 5) | |
8723 | && (geo->layout == 0)) { | |
8724 | /* reshape 4 -> 5 */ | |
8725 | geo->layout = 0; | |
8726 | geo->level = 5; | |
8727 | } else { | |
8728 | fprintf(stderr, | |
8729 | Name " Error. Layout Migration from %d to %d " | |
8730 | "not supported!\n", | |
8731 | info.array.layout, geo->layout); | |
8732 | change = -1; | |
8733 | goto analyse_change_exit; | |
8734 | } | |
8735 | } else | |
8736 | geo->layout = info.array.layout; | |
8737 | ||
8738 | if ((geo->chunksize > 0) && (geo->chunksize != UnSet) | |
8739 | && (geo->chunksize != info.array.chunk_size)) | |
8740 | change = CH_MIGRATION; | |
8741 | else | |
8742 | geo->chunksize = info.array.chunk_size; | |
8743 | ||
8744 | chunk = geo->chunksize / 1024; | |
8745 | if (!validate_geometry_imsm(st, | |
8746 | geo->level, | |
8747 | geo->layout, | |
8748 | geo->raid_disks, | |
8749 | &chunk, | |
8750 | geo->size, | |
8751 | 0, 0, 1)) | |
8752 | change = -1; | |
8753 | ||
8754 | if (check_devs) { | |
8755 | struct intel_super *super = st->sb; | |
8756 | struct imsm_super *mpb = super->anchor; | |
8757 | ||
8758 | if (mpb->num_raid_devs > 1) { | |
8759 | fprintf(stderr, | |
8760 | Name " Error. Cannot perform operation on %s" | |
8761 | "- for this operation it MUST be single " | |
8762 | "array in container\n", | |
8763 | geo->dev_name); | |
8764 | change = -1; | |
8765 | } | |
8766 | } | |
8767 | ||
8768 | analyse_change_exit: | |
8769 | ||
8770 | return change; | |
8771 | } | |
8772 | ||
8773 | int imsm_takeover(struct supertype *st, struct geo_params *geo) | |
8774 | { | |
8775 | struct intel_super *super = st->sb; | |
8776 | struct imsm_update_takeover *u; | |
8777 | ||
8778 | u = malloc(sizeof(struct imsm_update_takeover)); | |
8779 | if (u == NULL) | |
8780 | return 1; | |
8781 | ||
8782 | u->type = update_takeover; | |
8783 | u->subarray = super->current_vol; | |
8784 | ||
8785 | /* 10->0 transition */ | |
8786 | if (geo->level == 0) | |
8787 | u->direction = R10_TO_R0; | |
8788 | ||
8789 | /* 0->10 transition */ | |
8790 | if (geo->level == 10) | |
8791 | u->direction = R0_TO_R10; | |
8792 | ||
8793 | /* update metadata locally */ | |
8794 | imsm_update_metadata_locally(st, u, | |
8795 | sizeof(struct imsm_update_takeover)); | |
8796 | /* and possibly remotely */ | |
8797 | if (st->update_tail) | |
8798 | append_metadata_update(st, u, | |
8799 | sizeof(struct imsm_update_takeover)); | |
8800 | else | |
8801 | free(u); | |
8802 | ||
8803 | return 0; | |
8804 | } | |
8805 | ||
8806 | static int imsm_reshape_super(struct supertype *st, long long size, int level, | |
8807 | int layout, int chunksize, int raid_disks, | |
8808 | int delta_disks, char *backup, char *dev, | |
8809 | int verbose) | |
8810 | { | |
8811 | int ret_val = 1; | |
8812 | struct geo_params geo; | |
8813 | ||
8814 | dprintf("imsm: reshape_super called.\n"); | |
8815 | ||
8816 | memset(&geo, 0, sizeof(struct geo_params)); | |
8817 | ||
8818 | geo.dev_name = dev; | |
8819 | geo.dev_id = st->devnum; | |
8820 | geo.size = size; | |
8821 | geo.level = level; | |
8822 | geo.layout = layout; | |
8823 | geo.chunksize = chunksize; | |
8824 | geo.raid_disks = raid_disks; | |
8825 | if (delta_disks != UnSet) | |
8826 | geo.raid_disks += delta_disks; | |
8827 | ||
8828 | dprintf("\tfor level : %i\n", geo.level); | |
8829 | dprintf("\tfor raid_disks : %i\n", geo.raid_disks); | |
8830 | ||
8831 | if (experimental() == 0) | |
8832 | return ret_val; | |
8833 | ||
8834 | if (st->container_dev == st->devnum) { | |
8835 | /* On container level we can only increase number of devices. */ | |
8836 | dprintf("imsm: info: Container operation\n"); | |
8837 | int old_raid_disks = 0; | |
8838 | ||
8839 | if (imsm_reshape_is_allowed_on_container( | |
8840 | st, &geo, &old_raid_disks)) { | |
8841 | struct imsm_update_reshape *u = NULL; | |
8842 | int len; | |
8843 | ||
8844 | len = imsm_create_metadata_update_for_reshape( | |
8845 | st, &geo, old_raid_disks, &u); | |
8846 | ||
8847 | if (len <= 0) { | |
8848 | dprintf("imsm: Cannot prepare update\n"); | |
8849 | goto exit_imsm_reshape_super; | |
8850 | } | |
8851 | ||
8852 | ret_val = 0; | |
8853 | /* update metadata locally */ | |
8854 | imsm_update_metadata_locally(st, u, len); | |
8855 | /* and possibly remotely */ | |
8856 | if (st->update_tail) | |
8857 | append_metadata_update(st, u, len); | |
8858 | else | |
8859 | free(u); | |
8860 | ||
8861 | } else { | |
8862 | fprintf(stderr, Name ": (imsm) Operation " | |
8863 | "is not allowed on this container\n"); | |
8864 | } | |
8865 | } else { | |
8866 | /* On volume level we support following operations | |
8867 | * - takeover: raid10 -> raid0; raid0 -> raid10 | |
8868 | * - chunk size migration | |
8869 | * - migration: raid5 -> raid0; raid0 -> raid5 | |
8870 | */ | |
8871 | struct intel_super *super = st->sb; | |
8872 | struct intel_dev *dev = super->devlist; | |
8873 | int change, devnum; | |
8874 | dprintf("imsm: info: Volume operation\n"); | |
8875 | /* find requested device */ | |
8876 | while (dev) { | |
8877 | if (imsm_find_array_minor_by_subdev( | |
8878 | dev->index, st->container_dev, &devnum) == 0 | |
8879 | && devnum == geo.dev_id) | |
8880 | break; | |
8881 | dev = dev->next; | |
8882 | } | |
8883 | if (dev == NULL) { | |
8884 | fprintf(stderr, Name " Cannot find %s (%i) subarray\n", | |
8885 | geo.dev_name, geo.dev_id); | |
8886 | goto exit_imsm_reshape_super; | |
8887 | } | |
8888 | super->current_vol = dev->index; | |
8889 | change = imsm_analyze_change(st, &geo); | |
8890 | switch (change) { | |
8891 | case CH_TAKEOVER: | |
8892 | ret_val = imsm_takeover(st, &geo); | |
8893 | break; | |
8894 | case CH_MIGRATION: { | |
8895 | struct imsm_update_reshape_migration *u = NULL; | |
8896 | int len = | |
8897 | imsm_create_metadata_update_for_migration( | |
8898 | st, &geo, &u); | |
8899 | if (len < 1) { | |
8900 | dprintf("imsm: " | |
8901 | "Cannot prepare update\n"); | |
8902 | break; | |
8903 | } | |
8904 | ret_val = 0; | |
8905 | /* update metadata locally */ | |
8906 | imsm_update_metadata_locally(st, u, len); | |
8907 | /* and possibly remotely */ | |
8908 | if (st->update_tail) | |
8909 | append_metadata_update(st, u, len); | |
8910 | else | |
8911 | free(u); | |
8912 | } | |
8913 | break; | |
8914 | default: | |
8915 | ret_val = 1; | |
8916 | } | |
8917 | } | |
8918 | ||
8919 | exit_imsm_reshape_super: | |
8920 | dprintf("imsm: reshape_super Exit code = %i\n", ret_val); | |
8921 | return ret_val; | |
8922 | } | |
8923 | ||
8924 | /******************************************************************************* | |
8925 | * Function: wait_for_reshape_imsm | |
8926 | * Description: Function writes new sync_max value and waits until | |
8927 | * reshape process reach new position | |
8928 | * Parameters: | |
8929 | * sra : general array info | |
8930 | * ndata : number of disks in new array's layout | |
8931 | * Returns: | |
8932 | * 0 : success, | |
8933 | * 1 : there is no reshape in progress, | |
8934 | * -1 : fail | |
8935 | ******************************************************************************/ | |
8936 | int wait_for_reshape_imsm(struct mdinfo *sra, int ndata) | |
8937 | { | |
8938 | int fd = sysfs_get_fd(sra, NULL, "reshape_position"); | |
8939 | unsigned long long completed; | |
8940 | /* to_complete : new sync_max position */ | |
8941 | unsigned long long to_complete = sra->reshape_progress; | |
8942 | unsigned long long position_to_set = to_complete / ndata; | |
8943 | ||
8944 | if (fd < 0) { | |
8945 | dprintf("imsm: wait_for_reshape_imsm() " | |
8946 | "cannot open reshape_position\n"); | |
8947 | return 1; | |
8948 | } | |
8949 | ||
8950 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
8951 | dprintf("imsm: wait_for_reshape_imsm() " | |
8952 | "cannot read reshape_position (no reshape in progres)\n"); | |
8953 | close(fd); | |
8954 | return 0; | |
8955 | } | |
8956 | ||
8957 | if (completed > to_complete) { | |
8958 | dprintf("imsm: wait_for_reshape_imsm() " | |
8959 | "wrong next position to set %llu (%llu)\n", | |
8960 | to_complete, completed); | |
8961 | close(fd); | |
8962 | return -1; | |
8963 | } | |
8964 | dprintf("Position set: %llu\n", position_to_set); | |
8965 | if (sysfs_set_num(sra, NULL, "sync_max", | |
8966 | position_to_set) != 0) { | |
8967 | dprintf("imsm: wait_for_reshape_imsm() " | |
8968 | "cannot set reshape position to %llu\n", | |
8969 | position_to_set); | |
8970 | close(fd); | |
8971 | return -1; | |
8972 | } | |
8973 | ||
8974 | do { | |
8975 | char action[20]; | |
8976 | fd_set rfds; | |
8977 | FD_ZERO(&rfds); | |
8978 | FD_SET(fd, &rfds); | |
8979 | select(fd+1, &rfds, NULL, NULL, NULL); | |
8980 | if (sysfs_get_str(sra, NULL, "sync_action", | |
8981 | action, 20) > 0 && | |
8982 | strncmp(action, "reshape", 7) != 0) | |
8983 | break; | |
8984 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
8985 | dprintf("imsm: wait_for_reshape_imsm() " | |
8986 | "cannot read reshape_position (in loop)\n"); | |
8987 | close(fd); | |
8988 | return 1; | |
8989 | } | |
8990 | } while (completed < to_complete); | |
8991 | close(fd); | |
8992 | return 0; | |
8993 | ||
8994 | } | |
8995 | ||
8996 | /******************************************************************************* | |
8997 | * Function: check_degradation_change | |
8998 | * Description: Check that array hasn't become failed. | |
8999 | * Parameters: | |
9000 | * info : for sysfs access | |
9001 | * sources : source disks descriptors | |
9002 | * degraded: previous degradation level | |
9003 | * Returns: | |
9004 | * degradation level | |
9005 | ******************************************************************************/ | |
9006 | int check_degradation_change(struct mdinfo *info, | |
9007 | int *sources, | |
9008 | int degraded) | |
9009 | { | |
9010 | unsigned long long new_degraded; | |
9011 | sysfs_get_ll(info, NULL, "degraded", &new_degraded); | |
9012 | if (new_degraded != (unsigned long long)degraded) { | |
9013 | /* check each device to ensure it is still working */ | |
9014 | struct mdinfo *sd; | |
9015 | new_degraded = 0; | |
9016 | for (sd = info->devs ; sd ; sd = sd->next) { | |
9017 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
9018 | continue; | |
9019 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
9020 | char sbuf[20]; | |
9021 | if (sysfs_get_str(info, | |
9022 | sd, "state", sbuf, 20) < 0 || | |
9023 | strstr(sbuf, "faulty") || | |
9024 | strstr(sbuf, "in_sync") == NULL) { | |
9025 | /* this device is dead */ | |
9026 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
9027 | if (sd->disk.raid_disk >= 0 && | |
9028 | sources[sd->disk.raid_disk] >= 0) { | |
9029 | close(sources[ | |
9030 | sd->disk.raid_disk]); | |
9031 | sources[sd->disk.raid_disk] = | |
9032 | -1; | |
9033 | } | |
9034 | new_degraded++; | |
9035 | } | |
9036 | } | |
9037 | } | |
9038 | } | |
9039 | ||
9040 | return new_degraded; | |
9041 | } | |
9042 | ||
9043 | /******************************************************************************* | |
9044 | * Function: imsm_manage_reshape | |
9045 | * Description: Function finds array under reshape and it manages reshape | |
9046 | * process. It creates stripes backups (if required) and sets | |
9047 | * checheckpoits. | |
9048 | * Parameters: | |
9049 | * afd : Backup handle (nattive) - not used | |
9050 | * sra : general array info | |
9051 | * reshape : reshape parameters - not used | |
9052 | * st : supertype structure | |
9053 | * blocks : size of critical section [blocks] | |
9054 | * fds : table of source device descriptor | |
9055 | * offsets : start of array (offest per devices) | |
9056 | * dests : not used | |
9057 | * destfd : table of destination device descriptor | |
9058 | * destoffsets : table of destination offsets (per device) | |
9059 | * Returns: | |
9060 | * 1 : success, reshape is done | |
9061 | * 0 : fail | |
9062 | ******************************************************************************/ | |
9063 | static int imsm_manage_reshape( | |
9064 | int afd, struct mdinfo *sra, struct reshape *reshape, | |
9065 | struct supertype *st, unsigned long backup_blocks, | |
9066 | int *fds, unsigned long long *offsets, | |
9067 | int dests, int *destfd, unsigned long long *destoffsets) | |
9068 | { | |
9069 | int ret_val = 0; | |
9070 | struct intel_super *super = st->sb; | |
9071 | struct intel_dev *dv = NULL; | |
9072 | struct imsm_dev *dev = NULL; | |
9073 | struct imsm_map *map_src; | |
9074 | int migr_vol_qan = 0; | |
9075 | int ndata, odata; /* [bytes] */ | |
9076 | int chunk; /* [bytes] */ | |
9077 | struct migr_record *migr_rec; | |
9078 | char *buf = NULL; | |
9079 | unsigned int buf_size; /* [bytes] */ | |
9080 | unsigned long long max_position; /* array size [bytes] */ | |
9081 | unsigned long long next_step; /* [blocks]/[bytes] */ | |
9082 | unsigned long long old_data_stripe_length; | |
9083 | unsigned long long start_src; /* [bytes] */ | |
9084 | unsigned long long start; /* [bytes] */ | |
9085 | unsigned long long start_buf_shift; /* [bytes] */ | |
9086 | int degraded = 0; | |
9087 | int source_layout = 0; | |
9088 | ||
9089 | if (!fds || !offsets || !sra) | |
9090 | goto abort; | |
9091 | ||
9092 | /* Find volume during the reshape */ | |
9093 | for (dv = super->devlist; dv; dv = dv->next) { | |
9094 | if (dv->dev->vol.migr_type == MIGR_GEN_MIGR | |
9095 | && dv->dev->vol.migr_state == 1) { | |
9096 | dev = dv->dev; | |
9097 | migr_vol_qan++; | |
9098 | } | |
9099 | } | |
9100 | /* Only one volume can migrate at the same time */ | |
9101 | if (migr_vol_qan != 1) { | |
9102 | fprintf(stderr, Name " : %s", migr_vol_qan ? | |
9103 | "Number of migrating volumes greater than 1\n" : | |
9104 | "There is no volume during migrationg\n"); | |
9105 | goto abort; | |
9106 | } | |
9107 | ||
9108 | map_src = get_imsm_map(dev, 1); | |
9109 | if (map_src == NULL) | |
9110 | goto abort; | |
9111 | ||
9112 | ndata = imsm_num_data_members(dev, 0); | |
9113 | odata = imsm_num_data_members(dev, 1); | |
9114 | ||
9115 | chunk = __le16_to_cpu(map_src->blocks_per_strip) * 512; | |
9116 | old_data_stripe_length = odata * chunk; | |
9117 | ||
9118 | migr_rec = super->migr_rec; | |
9119 | ||
9120 | /* initialize migration record for start condition */ | |
9121 | if (sra->reshape_progress == 0) | |
9122 | init_migr_record_imsm(st, dev, sra); | |
9123 | else { | |
9124 | if (__le32_to_cpu(migr_rec->rec_status) != UNIT_SRC_NORMAL) { | |
9125 | dprintf("imsm: cannot restart migration when data " | |
9126 | "are present in copy area.\n"); | |
9127 | goto abort; | |
9128 | } | |
9129 | } | |
9130 | ||
9131 | /* size for data */ | |
9132 | buf_size = __le32_to_cpu(migr_rec->blocks_per_unit) * 512; | |
9133 | /* extend buffer size for parity disk */ | |
9134 | buf_size += __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512; | |
9135 | /* add space for stripe aligment */ | |
9136 | buf_size += old_data_stripe_length; | |
9137 | if (posix_memalign((void **)&buf, 4096, buf_size)) { | |
9138 | dprintf("imsm: Cannot allocate checpoint buffer\n"); | |
9139 | goto abort; | |
9140 | } | |
9141 | ||
9142 | max_position = sra->component_size * ndata; | |
9143 | source_layout = imsm_level_to_layout(map_src->raid_level); | |
9144 | ||
9145 | while (__le32_to_cpu(migr_rec->curr_migr_unit) < | |
9146 | __le32_to_cpu(migr_rec->num_migr_units)) { | |
9147 | /* current reshape position [blocks] */ | |
9148 | unsigned long long current_position = | |
9149 | __le32_to_cpu(migr_rec->blocks_per_unit) | |
9150 | * __le32_to_cpu(migr_rec->curr_migr_unit); | |
9151 | unsigned long long border; | |
9152 | ||
9153 | /* Check that array hasn't become failed. | |
9154 | */ | |
9155 | degraded = check_degradation_change(sra, fds, degraded); | |
9156 | if (degraded > 1) { | |
9157 | dprintf("imsm: Abort reshape due to degradation" | |
9158 | " level (%i)\n", degraded); | |
9159 | goto abort; | |
9160 | } | |
9161 | ||
9162 | next_step = __le32_to_cpu(migr_rec->blocks_per_unit); | |
9163 | ||
9164 | if ((current_position + next_step) > max_position) | |
9165 | next_step = max_position - current_position; | |
9166 | ||
9167 | start = current_position * 512; | |
9168 | ||
9169 | /* allign reading start to old geometry */ | |
9170 | start_buf_shift = start % old_data_stripe_length; | |
9171 | start_src = start - start_buf_shift; | |
9172 | ||
9173 | border = (start_src / odata) - (start / ndata); | |
9174 | border /= 512; | |
9175 | if (border <= __le32_to_cpu(migr_rec->dest_depth_per_unit)) { | |
9176 | /* save critical stripes to buf | |
9177 | * start - start address of current unit | |
9178 | * to backup [bytes] | |
9179 | * start_src - start address of current unit | |
9180 | * to backup alligned to source array | |
9181 | * [bytes] | |
9182 | */ | |
9183 | unsigned long long next_step_filler = 0; | |
9184 | unsigned long long copy_length = next_step * 512; | |
9185 | ||
9186 | /* allign copy area length to stripe in old geometry */ | |
9187 | next_step_filler = ((copy_length + start_buf_shift) | |
9188 | % old_data_stripe_length); | |
9189 | if (next_step_filler) | |
9190 | next_step_filler = (old_data_stripe_length | |
9191 | - next_step_filler); | |
9192 | dprintf("save_stripes() parameters: start = %llu," | |
9193 | "\tstart_src = %llu,\tnext_step*512 = %llu," | |
9194 | "\tstart_in_buf_shift = %llu," | |
9195 | "\tnext_step_filler = %llu\n", | |
9196 | start, start_src, copy_length, | |
9197 | start_buf_shift, next_step_filler); | |
9198 | ||
9199 | if (save_stripes(fds, offsets, map_src->num_members, | |
9200 | chunk, map_src->raid_level, | |
9201 | source_layout, 0, NULL, start_src, | |
9202 | copy_length + | |
9203 | next_step_filler + start_buf_shift, | |
9204 | buf)) { | |
9205 | dprintf("imsm: Cannot save stripes" | |
9206 | " to buffer\n"); | |
9207 | goto abort; | |
9208 | } | |
9209 | /* Convert data to destination format and store it | |
9210 | * in backup general migration area | |
9211 | */ | |
9212 | if (save_backup_imsm(st, dev, sra, | |
9213 | buf + start_buf_shift, copy_length)) { | |
9214 | dprintf("imsm: Cannot save stripes to " | |
9215 | "target devices\n"); | |
9216 | goto abort; | |
9217 | } | |
9218 | if (save_checkpoint_imsm(st, sra, | |
9219 | UNIT_SRC_IN_CP_AREA)) { | |
9220 | dprintf("imsm: Cannot write checkpoint to " | |
9221 | "migration record (UNIT_SRC_IN_CP_AREA)\n"); | |
9222 | goto abort; | |
9223 | } | |
9224 | } else { | |
9225 | /* set next step to use whole border area */ | |
9226 | border /= next_step; | |
9227 | if (border > 1) | |
9228 | next_step *= border; | |
9229 | } | |
9230 | /* When data backed up, checkpoint stored, | |
9231 | * kick the kernel to reshape unit of data | |
9232 | */ | |
9233 | next_step = next_step + sra->reshape_progress; | |
9234 | /* limit next step to array max position */ | |
9235 | if (next_step > max_position) | |
9236 | next_step = max_position; | |
9237 | sysfs_set_num(sra, NULL, "suspend_lo", sra->reshape_progress); | |
9238 | sysfs_set_num(sra, NULL, "suspend_hi", next_step); | |
9239 | sra->reshape_progress = next_step; | |
9240 | ||
9241 | /* wait until reshape finish */ | |
9242 | if (wait_for_reshape_imsm(sra, ndata) < 0) { | |
9243 | dprintf("wait_for_reshape_imsm returned error!\n"); | |
9244 | goto abort; | |
9245 | } | |
9246 | ||
9247 | if (save_checkpoint_imsm(st, sra, UNIT_SRC_NORMAL) == 1) { | |
9248 | /* ignore error == 2, this can mean end of reshape here | |
9249 | */ | |
9250 | dprintf("imsm: Cannot write checkpoint to " | |
9251 | "migration record (UNIT_SRC_NORMAL)\n"); | |
9252 | goto abort; | |
9253 | } | |
9254 | ||
9255 | } | |
9256 | ||
9257 | /* return '1' if done */ | |
9258 | ret_val = 1; | |
9259 | abort: | |
9260 | free(buf); | |
9261 | abort_reshape(sra); | |
9262 | ||
9263 | return ret_val; | |
9264 | } | |
9265 | #endif /* MDASSEMBLE */ | |
9266 | ||
9267 | struct superswitch super_imsm = { | |
9268 | #ifndef MDASSEMBLE | |
9269 | .examine_super = examine_super_imsm, | |
9270 | .brief_examine_super = brief_examine_super_imsm, | |
9271 | .brief_examine_subarrays = brief_examine_subarrays_imsm, | |
9272 | .export_examine_super = export_examine_super_imsm, | |
9273 | .detail_super = detail_super_imsm, | |
9274 | .brief_detail_super = brief_detail_super_imsm, | |
9275 | .write_init_super = write_init_super_imsm, | |
9276 | .validate_geometry = validate_geometry_imsm, | |
9277 | .add_to_super = add_to_super_imsm, | |
9278 | .remove_from_super = remove_from_super_imsm, | |
9279 | .detail_platform = detail_platform_imsm, | |
9280 | .kill_subarray = kill_subarray_imsm, | |
9281 | .update_subarray = update_subarray_imsm, | |
9282 | .load_container = load_container_imsm, | |
9283 | .default_geometry = default_geometry_imsm, | |
9284 | .get_disk_controller_domain = imsm_get_disk_controller_domain, | |
9285 | .reshape_super = imsm_reshape_super, | |
9286 | .manage_reshape = imsm_manage_reshape, | |
9287 | .recover_backup = recover_backup_imsm, | |
9288 | #endif | |
9289 | .match_home = match_home_imsm, | |
9290 | .uuid_from_super= uuid_from_super_imsm, | |
9291 | .getinfo_super = getinfo_super_imsm, | |
9292 | .getinfo_super_disks = getinfo_super_disks_imsm, | |
9293 | .update_super = update_super_imsm, | |
9294 | ||
9295 | .avail_size = avail_size_imsm, | |
9296 | .min_acceptable_spare_size = min_acceptable_spare_size_imsm, | |
9297 | ||
9298 | .compare_super = compare_super_imsm, | |
9299 | ||
9300 | .load_super = load_super_imsm, | |
9301 | .init_super = init_super_imsm, | |
9302 | .store_super = store_super_imsm, | |
9303 | .free_super = free_super_imsm, | |
9304 | .match_metadata_desc = match_metadata_desc_imsm, | |
9305 | .container_content = container_content_imsm, | |
9306 | ||
9307 | ||
9308 | .external = 1, | |
9309 | .name = "imsm", | |
9310 | ||
9311 | #ifndef MDASSEMBLE | |
9312 | /* for mdmon */ | |
9313 | .open_new = imsm_open_new, | |
9314 | .set_array_state= imsm_set_array_state, | |
9315 | .set_disk = imsm_set_disk, | |
9316 | .sync_metadata = imsm_sync_metadata, | |
9317 | .activate_spare = imsm_activate_spare, | |
9318 | .process_update = imsm_process_update, | |
9319 | .prepare_update = imsm_prepare_update, | |
9320 | #endif /* MDASSEMBLE */ | |
9321 | }; |