]>
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 | #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000) | |
45 | #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000) | |
46 | #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000) | |
47 | #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001) | |
48 | #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002) | |
49 | #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004) | |
50 | #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008) | |
51 | #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010) | |
52 | #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020) | |
53 | ||
54 | #define MPB_SECTOR_CNT 2210 | |
55 | #define IMSM_RESERVED_SECTORS 4096 | |
56 | #define SECT_PER_MB_SHIFT 11 | |
57 | ||
58 | /* Disk configuration info. */ | |
59 | #define IMSM_MAX_DEVICES 255 | |
60 | struct imsm_disk { | |
61 | __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */ | |
62 | __u32 total_blocks; /* 0xE8 - 0xEB total blocks */ | |
63 | __u32 scsi_id; /* 0xEC - 0xEF scsi ID */ | |
64 | #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */ | |
65 | #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */ | |
66 | #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */ | |
67 | __u32 status; /* 0xF0 - 0xF3 */ | |
68 | __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */ | |
69 | #define IMSM_DISK_FILLERS 4 | |
70 | __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */ | |
71 | }; | |
72 | ||
73 | /* RAID map configuration infos. */ | |
74 | struct imsm_map { | |
75 | __u32 pba_of_lba0; /* start address of partition */ | |
76 | __u32 blocks_per_member;/* blocks per member */ | |
77 | __u32 num_data_stripes; /* number of data stripes */ | |
78 | __u16 blocks_per_strip; | |
79 | __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */ | |
80 | #define IMSM_T_STATE_NORMAL 0 | |
81 | #define IMSM_T_STATE_UNINITIALIZED 1 | |
82 | #define IMSM_T_STATE_DEGRADED 2 | |
83 | #define IMSM_T_STATE_FAILED 3 | |
84 | __u8 raid_level; | |
85 | #define IMSM_T_RAID0 0 | |
86 | #define IMSM_T_RAID1 1 | |
87 | #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */ | |
88 | __u8 num_members; /* number of member disks */ | |
89 | __u8 num_domains; /* number of parity domains */ | |
90 | __u8 failed_disk_num; /* valid only when state is degraded */ | |
91 | __u8 ddf; | |
92 | __u32 filler[7]; /* expansion area */ | |
93 | #define IMSM_ORD_REBUILD (1 << 24) | |
94 | __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members], | |
95 | * top byte contains some flags | |
96 | */ | |
97 | } __attribute__ ((packed)); | |
98 | ||
99 | struct imsm_vol { | |
100 | __u32 curr_migr_unit; | |
101 | __u32 checkpoint_id; /* id to access curr_migr_unit */ | |
102 | __u8 migr_state; /* Normal or Migrating */ | |
103 | #define MIGR_INIT 0 | |
104 | #define MIGR_REBUILD 1 | |
105 | #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */ | |
106 | #define MIGR_GEN_MIGR 3 | |
107 | #define MIGR_STATE_CHANGE 4 | |
108 | #define MIGR_REPAIR 5 | |
109 | __u8 migr_type; /* Initializing, Rebuilding, ... */ | |
110 | __u8 dirty; | |
111 | __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */ | |
112 | __u16 verify_errors; /* number of mismatches */ | |
113 | __u16 bad_blocks; /* number of bad blocks during verify */ | |
114 | __u32 filler[4]; | |
115 | struct imsm_map map[1]; | |
116 | /* here comes another one if migr_state */ | |
117 | } __attribute__ ((packed)); | |
118 | ||
119 | struct imsm_dev { | |
120 | __u8 volume[MAX_RAID_SERIAL_LEN]; | |
121 | __u32 size_low; | |
122 | __u32 size_high; | |
123 | #define DEV_BOOTABLE __cpu_to_le32(0x01) | |
124 | #define DEV_BOOT_DEVICE __cpu_to_le32(0x02) | |
125 | #define DEV_READ_COALESCING __cpu_to_le32(0x04) | |
126 | #define DEV_WRITE_COALESCING __cpu_to_le32(0x08) | |
127 | #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10) | |
128 | #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20) | |
129 | #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40) | |
130 | #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80) | |
131 | #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100) | |
132 | #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200) | |
133 | #define DEV_CLONE_N_GO __cpu_to_le32(0x400) | |
134 | #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800) | |
135 | #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000) | |
136 | __u32 status; /* Persistent RaidDev status */ | |
137 | __u32 reserved_blocks; /* Reserved blocks at beginning of volume */ | |
138 | __u8 migr_priority; | |
139 | __u8 num_sub_vols; | |
140 | __u8 tid; | |
141 | __u8 cng_master_disk; | |
142 | __u16 cache_policy; | |
143 | __u8 cng_state; | |
144 | __u8 cng_sub_state; | |
145 | #define IMSM_DEV_FILLERS 10 | |
146 | __u32 filler[IMSM_DEV_FILLERS]; | |
147 | struct imsm_vol vol; | |
148 | } __attribute__ ((packed)); | |
149 | ||
150 | struct imsm_super { | |
151 | __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */ | |
152 | __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */ | |
153 | __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */ | |
154 | __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */ | |
155 | __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */ | |
156 | __u32 error_log_size; /* 0x30 - 0x33 in bytes */ | |
157 | __u32 attributes; /* 0x34 - 0x37 */ | |
158 | __u8 num_disks; /* 0x38 Number of configured disks */ | |
159 | __u8 num_raid_devs; /* 0x39 Number of configured volumes */ | |
160 | __u8 error_log_pos; /* 0x3A */ | |
161 | __u8 fill[1]; /* 0x3B */ | |
162 | __u32 cache_size; /* 0x3c - 0x40 in mb */ | |
163 | __u32 orig_family_num; /* 0x40 - 0x43 original family num */ | |
164 | __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */ | |
165 | __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */ | |
166 | #define IMSM_FILLERS 35 | |
167 | __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */ | |
168 | struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */ | |
169 | /* here comes imsm_dev[num_raid_devs] */ | |
170 | /* here comes BBM logs */ | |
171 | } __attribute__ ((packed)); | |
172 | ||
173 | #define BBM_LOG_MAX_ENTRIES 254 | |
174 | ||
175 | struct bbm_log_entry { | |
176 | __u64 defective_block_start; | |
177 | #define UNREADABLE 0xFFFFFFFF | |
178 | __u32 spare_block_offset; | |
179 | __u16 remapped_marked_count; | |
180 | __u16 disk_ordinal; | |
181 | } __attribute__ ((__packed__)); | |
182 | ||
183 | struct bbm_log { | |
184 | __u32 signature; /* 0xABADB10C */ | |
185 | __u32 entry_count; | |
186 | __u32 reserved_spare_block_count; /* 0 */ | |
187 | __u32 reserved; /* 0xFFFF */ | |
188 | __u64 first_spare_lba; | |
189 | struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES]; | |
190 | } __attribute__ ((__packed__)); | |
191 | ||
192 | ||
193 | #ifndef MDASSEMBLE | |
194 | static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" }; | |
195 | #endif | |
196 | ||
197 | #define RAID_DISK_RESERVED_BLOCKS_IMSM_HI 2209 | |
198 | ||
199 | #define GEN_MIGR_AREA_SIZE 2048 /* General Migration Copy Area size in blocks */ | |
200 | ||
201 | #define UNIT_SRC_NORMAL 0 /* Source data for curr_migr_unit must | |
202 | * be recovered using srcMap */ | |
203 | #define UNIT_SRC_IN_CP_AREA 1 /* Source data for curr_migr_unit has | |
204 | * already been migrated and must | |
205 | * be recovered from checkpoint area */ | |
206 | struct migr_record { | |
207 | __u32 rec_status; /* Status used to determine how to restart | |
208 | * migration in case it aborts | |
209 | * in some fashion */ | |
210 | __u32 curr_migr_unit; /* 0..numMigrUnits-1 */ | |
211 | __u32 family_num; /* Family number of MPB | |
212 | * containing the RaidDev | |
213 | * that is migrating */ | |
214 | __u32 ascending_migr; /* True if migrating in increasing | |
215 | * order of lbas */ | |
216 | __u32 blocks_per_unit; /* Num disk blocks per unit of operation */ | |
217 | __u32 dest_depth_per_unit; /* Num member blocks each destMap | |
218 | * member disk | |
219 | * advances per unit-of-operation */ | |
220 | __u32 ckpt_area_pba; /* Pba of first block of ckpt copy area */ | |
221 | __u32 dest_1st_member_lba; /* First member lba on first | |
222 | * stripe of destination */ | |
223 | __u32 num_migr_units; /* Total num migration units-of-op */ | |
224 | __u32 post_migr_vol_cap; /* Size of volume after | |
225 | * migration completes */ | |
226 | __u32 post_migr_vol_cap_hi; /* Expansion space for LBA64 */ | |
227 | __u32 ckpt_read_disk_num; /* Which member disk in destSubMap[0] the | |
228 | * migration ckpt record was read from | |
229 | * (for recovered migrations) */ | |
230 | } __attribute__ ((__packed__)); | |
231 | ||
232 | static __u8 migr_type(struct imsm_dev *dev) | |
233 | { | |
234 | if (dev->vol.migr_type == MIGR_VERIFY && | |
235 | dev->status & DEV_VERIFY_AND_FIX) | |
236 | return MIGR_REPAIR; | |
237 | else | |
238 | return dev->vol.migr_type; | |
239 | } | |
240 | ||
241 | static void set_migr_type(struct imsm_dev *dev, __u8 migr_type) | |
242 | { | |
243 | /* for compatibility with older oroms convert MIGR_REPAIR, into | |
244 | * MIGR_VERIFY w/ DEV_VERIFY_AND_FIX status | |
245 | */ | |
246 | if (migr_type == MIGR_REPAIR) { | |
247 | dev->vol.migr_type = MIGR_VERIFY; | |
248 | dev->status |= DEV_VERIFY_AND_FIX; | |
249 | } else { | |
250 | dev->vol.migr_type = migr_type; | |
251 | dev->status &= ~DEV_VERIFY_AND_FIX; | |
252 | } | |
253 | } | |
254 | ||
255 | static unsigned int sector_count(__u32 bytes) | |
256 | { | |
257 | return ((bytes + (512-1)) & (~(512-1))) / 512; | |
258 | } | |
259 | ||
260 | static unsigned int mpb_sectors(struct imsm_super *mpb) | |
261 | { | |
262 | return sector_count(__le32_to_cpu(mpb->mpb_size)); | |
263 | } | |
264 | ||
265 | struct intel_dev { | |
266 | struct imsm_dev *dev; | |
267 | struct intel_dev *next; | |
268 | unsigned index; | |
269 | }; | |
270 | ||
271 | struct intel_hba { | |
272 | enum sys_dev_type type; | |
273 | char *path; | |
274 | char *pci_id; | |
275 | struct intel_hba *next; | |
276 | }; | |
277 | ||
278 | enum action { | |
279 | DISK_REMOVE = 1, | |
280 | DISK_ADD | |
281 | }; | |
282 | /* internal representation of IMSM metadata */ | |
283 | struct intel_super { | |
284 | union { | |
285 | void *buf; /* O_DIRECT buffer for reading/writing metadata */ | |
286 | struct imsm_super *anchor; /* immovable parameters */ | |
287 | }; | |
288 | union { | |
289 | void *migr_rec_buf; /* buffer for I/O operations */ | |
290 | struct migr_record *migr_rec; /* migration record */ | |
291 | }; | |
292 | size_t len; /* size of the 'buf' allocation */ | |
293 | void *next_buf; /* for realloc'ing buf from the manager */ | |
294 | size_t next_len; | |
295 | int updates_pending; /* count of pending updates for mdmon */ | |
296 | int current_vol; /* index of raid device undergoing creation */ | |
297 | __u32 create_offset; /* common start for 'current_vol' */ | |
298 | __u32 random; /* random data for seeding new family numbers */ | |
299 | struct intel_dev *devlist; | |
300 | struct dl { | |
301 | struct dl *next; | |
302 | int index; | |
303 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
304 | int major, minor; | |
305 | char *devname; | |
306 | struct imsm_disk disk; | |
307 | int fd; | |
308 | int extent_cnt; | |
309 | struct extent *e; /* for determining freespace @ create */ | |
310 | int raiddisk; /* slot to fill in autolayout */ | |
311 | enum action action; | |
312 | } *disks; | |
313 | struct dl *disk_mgmt_list; /* list of disks to add/remove while mdmon | |
314 | active */ | |
315 | struct dl *missing; /* disks removed while we weren't looking */ | |
316 | struct bbm_log *bbm_log; | |
317 | struct intel_hba *hba; /* device path of the raid controller for this metadata */ | |
318 | const struct imsm_orom *orom; /* platform firmware support */ | |
319 | struct intel_super *next; /* (temp) list for disambiguating family_num */ | |
320 | }; | |
321 | ||
322 | struct intel_disk { | |
323 | struct imsm_disk disk; | |
324 | #define IMSM_UNKNOWN_OWNER (-1) | |
325 | int owner; | |
326 | struct intel_disk *next; | |
327 | }; | |
328 | ||
329 | struct extent { | |
330 | unsigned long long start, size; | |
331 | }; | |
332 | ||
333 | /* definitions of reshape process types */ | |
334 | enum imsm_reshape_type { | |
335 | CH_TAKEOVER, | |
336 | CH_MIGRATION, | |
337 | }; | |
338 | ||
339 | /* definition of messages passed to imsm_process_update */ | |
340 | enum imsm_update_type { | |
341 | update_activate_spare, | |
342 | update_create_array, | |
343 | update_kill_array, | |
344 | update_rename_array, | |
345 | update_add_remove_disk, | |
346 | update_reshape_container_disks, | |
347 | update_reshape_migration, | |
348 | update_takeover, | |
349 | update_general_migration_checkpoint, | |
350 | }; | |
351 | ||
352 | struct imsm_update_activate_spare { | |
353 | enum imsm_update_type type; | |
354 | struct dl *dl; | |
355 | int slot; | |
356 | int array; | |
357 | struct imsm_update_activate_spare *next; | |
358 | }; | |
359 | ||
360 | struct geo_params { | |
361 | int dev_id; | |
362 | char *dev_name; | |
363 | long long size; | |
364 | int level; | |
365 | int layout; | |
366 | int chunksize; | |
367 | int raid_disks; | |
368 | }; | |
369 | ||
370 | enum takeover_direction { | |
371 | R10_TO_R0, | |
372 | R0_TO_R10 | |
373 | }; | |
374 | struct imsm_update_takeover { | |
375 | enum imsm_update_type type; | |
376 | int subarray; | |
377 | enum takeover_direction direction; | |
378 | }; | |
379 | ||
380 | struct imsm_update_reshape { | |
381 | enum imsm_update_type type; | |
382 | int old_raid_disks; | |
383 | int new_raid_disks; | |
384 | ||
385 | int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */ | |
386 | }; | |
387 | ||
388 | struct imsm_update_reshape_migration { | |
389 | enum imsm_update_type type; | |
390 | int old_raid_disks; | |
391 | int new_raid_disks; | |
392 | /* fields for array migration changes | |
393 | */ | |
394 | int subdev; | |
395 | int new_level; | |
396 | int new_layout; | |
397 | int new_chunksize; | |
398 | ||
399 | int new_disks[1]; /* new_raid_disks - old_raid_disks makedev number */ | |
400 | }; | |
401 | ||
402 | struct imsm_update_general_migration_checkpoint { | |
403 | enum imsm_update_type type; | |
404 | __u32 curr_migr_unit; | |
405 | }; | |
406 | ||
407 | struct disk_info { | |
408 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
409 | }; | |
410 | ||
411 | struct imsm_update_create_array { | |
412 | enum imsm_update_type type; | |
413 | int dev_idx; | |
414 | struct imsm_dev dev; | |
415 | }; | |
416 | ||
417 | struct imsm_update_kill_array { | |
418 | enum imsm_update_type type; | |
419 | int dev_idx; | |
420 | }; | |
421 | ||
422 | struct imsm_update_rename_array { | |
423 | enum imsm_update_type type; | |
424 | __u8 name[MAX_RAID_SERIAL_LEN]; | |
425 | int dev_idx; | |
426 | }; | |
427 | ||
428 | struct imsm_update_add_remove_disk { | |
429 | enum imsm_update_type type; | |
430 | }; | |
431 | ||
432 | ||
433 | static const char *_sys_dev_type[] = { | |
434 | [SYS_DEV_UNKNOWN] = "Unknown", | |
435 | [SYS_DEV_SAS] = "SAS", | |
436 | [SYS_DEV_SATA] = "SATA" | |
437 | }; | |
438 | ||
439 | const char *get_sys_dev_type(enum sys_dev_type type) | |
440 | { | |
441 | if (type >= SYS_DEV_MAX) | |
442 | type = SYS_DEV_UNKNOWN; | |
443 | ||
444 | return _sys_dev_type[type]; | |
445 | } | |
446 | ||
447 | static struct intel_hba * alloc_intel_hba(struct sys_dev *device) | |
448 | { | |
449 | struct intel_hba *result = malloc(sizeof(*result)); | |
450 | if (result) { | |
451 | result->type = device->type; | |
452 | result->path = strdup(device->path); | |
453 | result->next = NULL; | |
454 | if (result->path && (result->pci_id = strrchr(result->path, '/')) != NULL) | |
455 | result->pci_id++; | |
456 | } | |
457 | return result; | |
458 | } | |
459 | ||
460 | static struct intel_hba * find_intel_hba(struct intel_hba *hba, struct sys_dev *device) | |
461 | { | |
462 | struct intel_hba *result=NULL; | |
463 | for (result = hba; result; result = result->next) { | |
464 | if (result->type == device->type && strcmp(result->path, device->path) == 0) | |
465 | break; | |
466 | } | |
467 | return result; | |
468 | } | |
469 | ||
470 | static int attach_hba_to_super(struct intel_super *super, struct sys_dev *device) | |
471 | { | |
472 | struct intel_hba *hba; | |
473 | ||
474 | /* check if disk attached to Intel HBA */ | |
475 | hba = find_intel_hba(super->hba, device); | |
476 | if (hba != NULL) | |
477 | return 1; | |
478 | /* Check if HBA is already attached to super */ | |
479 | if (super->hba == NULL) { | |
480 | super->hba = alloc_intel_hba(device); | |
481 | return 1; | |
482 | } | |
483 | ||
484 | hba = super->hba; | |
485 | /* Intel metadata allows for all disks attached to the same type HBA. | |
486 | * Do not sypport odf HBA types mixing | |
487 | */ | |
488 | if (device->type != hba->type) | |
489 | return 2; | |
490 | ||
491 | while (hba->next) | |
492 | hba = hba->next; | |
493 | ||
494 | hba->next = alloc_intel_hba(device); | |
495 | return 1; | |
496 | } | |
497 | ||
498 | static struct sys_dev* find_disk_attached_hba(int fd, const char *devname) | |
499 | { | |
500 | struct sys_dev *list, *elem, *prev; | |
501 | char *disk_path; | |
502 | ||
503 | if ((list = find_intel_devices()) == NULL) | |
504 | return 0; | |
505 | ||
506 | if (fd < 0) | |
507 | disk_path = (char *) devname; | |
508 | else | |
509 | disk_path = diskfd_to_devpath(fd); | |
510 | ||
511 | if (!disk_path) { | |
512 | free_sys_dev(&list); | |
513 | return 0; | |
514 | } | |
515 | ||
516 | for (prev = NULL, elem = list; elem; prev = elem, elem = elem->next) { | |
517 | if (path_attached_to_hba(disk_path, elem->path)) { | |
518 | if (prev == NULL) | |
519 | list = list->next; | |
520 | else | |
521 | prev->next = elem->next; | |
522 | elem->next = NULL; | |
523 | if (disk_path != devname) | |
524 | free(disk_path); | |
525 | free_sys_dev(&list); | |
526 | return elem; | |
527 | } | |
528 | } | |
529 | if (disk_path != devname) | |
530 | free(disk_path); | |
531 | free_sys_dev(&list); | |
532 | ||
533 | return NULL; | |
534 | } | |
535 | ||
536 | ||
537 | static int find_intel_hba_capability(int fd, struct intel_super *super, | |
538 | char *devname); | |
539 | ||
540 | static struct supertype *match_metadata_desc_imsm(char *arg) | |
541 | { | |
542 | struct supertype *st; | |
543 | ||
544 | if (strcmp(arg, "imsm") != 0 && | |
545 | strcmp(arg, "default") != 0 | |
546 | ) | |
547 | return NULL; | |
548 | ||
549 | st = malloc(sizeof(*st)); | |
550 | if (!st) | |
551 | return NULL; | |
552 | memset(st, 0, sizeof(*st)); | |
553 | st->container_dev = NoMdDev; | |
554 | st->ss = &super_imsm; | |
555 | st->max_devs = IMSM_MAX_DEVICES; | |
556 | st->minor_version = 0; | |
557 | st->sb = NULL; | |
558 | return st; | |
559 | } | |
560 | ||
561 | #ifndef MDASSEMBLE | |
562 | static __u8 *get_imsm_version(struct imsm_super *mpb) | |
563 | { | |
564 | return &mpb->sig[MPB_SIG_LEN]; | |
565 | } | |
566 | #endif | |
567 | ||
568 | /* retrieve a disk directly from the anchor when the anchor is known to be | |
569 | * up-to-date, currently only at load time | |
570 | */ | |
571 | static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index) | |
572 | { | |
573 | if (index >= mpb->num_disks) | |
574 | return NULL; | |
575 | return &mpb->disk[index]; | |
576 | } | |
577 | ||
578 | /* retrieve the disk description based on a index of the disk | |
579 | * in the sub-array | |
580 | */ | |
581 | static struct dl *get_imsm_dl_disk(struct intel_super *super, __u8 index) | |
582 | { | |
583 | struct dl *d; | |
584 | ||
585 | for (d = super->disks; d; d = d->next) | |
586 | if (d->index == index) | |
587 | return d; | |
588 | ||
589 | return NULL; | |
590 | } | |
591 | /* retrieve a disk from the parsed metadata */ | |
592 | static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index) | |
593 | { | |
594 | struct dl *dl; | |
595 | ||
596 | dl = get_imsm_dl_disk(super, index); | |
597 | if (dl) | |
598 | return &dl->disk; | |
599 | ||
600 | return NULL; | |
601 | } | |
602 | ||
603 | /* generate a checksum directly from the anchor when the anchor is known to be | |
604 | * up-to-date, currently only at load or write_super after coalescing | |
605 | */ | |
606 | static __u32 __gen_imsm_checksum(struct imsm_super *mpb) | |
607 | { | |
608 | __u32 end = mpb->mpb_size / sizeof(end); | |
609 | __u32 *p = (__u32 *) mpb; | |
610 | __u32 sum = 0; | |
611 | ||
612 | while (end--) { | |
613 | sum += __le32_to_cpu(*p); | |
614 | p++; | |
615 | } | |
616 | ||
617 | return sum - __le32_to_cpu(mpb->check_sum); | |
618 | } | |
619 | ||
620 | static size_t sizeof_imsm_map(struct imsm_map *map) | |
621 | { | |
622 | return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1); | |
623 | } | |
624 | ||
625 | struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map) | |
626 | { | |
627 | /* A device can have 2 maps if it is in the middle of a migration. | |
628 | * If second_map is: | |
629 | * 0 - we return the first map | |
630 | * 1 - we return the second map if it exists, else NULL | |
631 | * -1 - we return the second map if it exists, else the first | |
632 | */ | |
633 | struct imsm_map *map = &dev->vol.map[0]; | |
634 | ||
635 | if (second_map == 1 && !dev->vol.migr_state) | |
636 | return NULL; | |
637 | else if (second_map == 1 || | |
638 | (second_map < 0 && dev->vol.migr_state)) { | |
639 | void *ptr = map; | |
640 | ||
641 | return ptr + sizeof_imsm_map(map); | |
642 | } else | |
643 | return map; | |
644 | ||
645 | } | |
646 | ||
647 | /* return the size of the device. | |
648 | * migr_state increases the returned size if map[0] were to be duplicated | |
649 | */ | |
650 | static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state) | |
651 | { | |
652 | size_t size = sizeof(*dev) - sizeof(struct imsm_map) + | |
653 | sizeof_imsm_map(get_imsm_map(dev, 0)); | |
654 | ||
655 | /* migrating means an additional map */ | |
656 | if (dev->vol.migr_state) | |
657 | size += sizeof_imsm_map(get_imsm_map(dev, 1)); | |
658 | else if (migr_state) | |
659 | size += sizeof_imsm_map(get_imsm_map(dev, 0)); | |
660 | ||
661 | return size; | |
662 | } | |
663 | ||
664 | #ifndef MDASSEMBLE | |
665 | /* retrieve disk serial number list from a metadata update */ | |
666 | static struct disk_info *get_disk_info(struct imsm_update_create_array *update) | |
667 | { | |
668 | void *u = update; | |
669 | struct disk_info *inf; | |
670 | ||
671 | inf = u + sizeof(*update) - sizeof(struct imsm_dev) + | |
672 | sizeof_imsm_dev(&update->dev, 0); | |
673 | ||
674 | return inf; | |
675 | } | |
676 | #endif | |
677 | ||
678 | static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index) | |
679 | { | |
680 | int offset; | |
681 | int i; | |
682 | void *_mpb = mpb; | |
683 | ||
684 | if (index >= mpb->num_raid_devs) | |
685 | return NULL; | |
686 | ||
687 | /* devices start after all disks */ | |
688 | offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb; | |
689 | ||
690 | for (i = 0; i <= index; i++) | |
691 | if (i == index) | |
692 | return _mpb + offset; | |
693 | else | |
694 | offset += sizeof_imsm_dev(_mpb + offset, 0); | |
695 | ||
696 | return NULL; | |
697 | } | |
698 | ||
699 | static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index) | |
700 | { | |
701 | struct intel_dev *dv; | |
702 | ||
703 | if (index >= super->anchor->num_raid_devs) | |
704 | return NULL; | |
705 | for (dv = super->devlist; dv; dv = dv->next) | |
706 | if (dv->index == index) | |
707 | return dv->dev; | |
708 | return NULL; | |
709 | } | |
710 | ||
711 | /* | |
712 | * for second_map: | |
713 | * == 0 get first map | |
714 | * == 1 get second map | |
715 | * == -1 than get map according to the current migr_state | |
716 | */ | |
717 | static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, | |
718 | int slot, | |
719 | int second_map) | |
720 | { | |
721 | struct imsm_map *map; | |
722 | ||
723 | map = get_imsm_map(dev, second_map); | |
724 | ||
725 | /* top byte identifies disk under rebuild */ | |
726 | return __le32_to_cpu(map->disk_ord_tbl[slot]); | |
727 | } | |
728 | ||
729 | #define ord_to_idx(ord) (((ord) << 8) >> 8) | |
730 | static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot, int second_map) | |
731 | { | |
732 | __u32 ord = get_imsm_ord_tbl_ent(dev, slot, second_map); | |
733 | ||
734 | return ord_to_idx(ord); | |
735 | } | |
736 | ||
737 | static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord) | |
738 | { | |
739 | map->disk_ord_tbl[slot] = __cpu_to_le32(ord); | |
740 | } | |
741 | ||
742 | static int get_imsm_disk_slot(struct imsm_map *map, unsigned idx) | |
743 | { | |
744 | int slot; | |
745 | __u32 ord; | |
746 | ||
747 | for (slot = 0; slot < map->num_members; slot++) { | |
748 | ord = __le32_to_cpu(map->disk_ord_tbl[slot]); | |
749 | if (ord_to_idx(ord) == idx) | |
750 | return slot; | |
751 | } | |
752 | ||
753 | return -1; | |
754 | } | |
755 | ||
756 | static int get_imsm_raid_level(struct imsm_map *map) | |
757 | { | |
758 | if (map->raid_level == 1) { | |
759 | if (map->num_members == 2) | |
760 | return 1; | |
761 | else | |
762 | return 10; | |
763 | } | |
764 | ||
765 | return map->raid_level; | |
766 | } | |
767 | ||
768 | static int cmp_extent(const void *av, const void *bv) | |
769 | { | |
770 | const struct extent *a = av; | |
771 | const struct extent *b = bv; | |
772 | if (a->start < b->start) | |
773 | return -1; | |
774 | if (a->start > b->start) | |
775 | return 1; | |
776 | return 0; | |
777 | } | |
778 | ||
779 | static int count_memberships(struct dl *dl, struct intel_super *super) | |
780 | { | |
781 | int memberships = 0; | |
782 | int i; | |
783 | ||
784 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
785 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
786 | struct imsm_map *map = get_imsm_map(dev, 0); | |
787 | ||
788 | if (get_imsm_disk_slot(map, dl->index) >= 0) | |
789 | memberships++; | |
790 | } | |
791 | ||
792 | return memberships; | |
793 | } | |
794 | ||
795 | static struct extent *get_extents(struct intel_super *super, struct dl *dl) | |
796 | { | |
797 | /* find a list of used extents on the given physical device */ | |
798 | struct extent *rv, *e; | |
799 | int i; | |
800 | int memberships = count_memberships(dl, super); | |
801 | __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
802 | ||
803 | rv = malloc(sizeof(struct extent) * (memberships + 1)); | |
804 | if (!rv) | |
805 | return NULL; | |
806 | e = rv; | |
807 | ||
808 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
809 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
810 | struct imsm_map *map = get_imsm_map(dev, 0); | |
811 | ||
812 | if (get_imsm_disk_slot(map, dl->index) >= 0) { | |
813 | e->start = __le32_to_cpu(map->pba_of_lba0); | |
814 | e->size = __le32_to_cpu(map->blocks_per_member); | |
815 | e++; | |
816 | } | |
817 | } | |
818 | qsort(rv, memberships, sizeof(*rv), cmp_extent); | |
819 | ||
820 | /* determine the start of the metadata | |
821 | * when no raid devices are defined use the default | |
822 | * ...otherwise allow the metadata to truncate the value | |
823 | * as is the case with older versions of imsm | |
824 | */ | |
825 | if (memberships) { | |
826 | struct extent *last = &rv[memberships - 1]; | |
827 | __u32 remainder; | |
828 | ||
829 | remainder = __le32_to_cpu(dl->disk.total_blocks) - | |
830 | (last->start + last->size); | |
831 | /* round down to 1k block to satisfy precision of the kernel | |
832 | * 'size' interface | |
833 | */ | |
834 | remainder &= ~1UL; | |
835 | /* make sure remainder is still sane */ | |
836 | if (remainder < (unsigned)ROUND_UP(super->len, 512) >> 9) | |
837 | remainder = ROUND_UP(super->len, 512) >> 9; | |
838 | if (reservation > remainder) | |
839 | reservation = remainder; | |
840 | } | |
841 | e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation; | |
842 | e->size = 0; | |
843 | return rv; | |
844 | } | |
845 | ||
846 | /* try to determine how much space is reserved for metadata from | |
847 | * the last get_extents() entry, otherwise fallback to the | |
848 | * default | |
849 | */ | |
850 | static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl) | |
851 | { | |
852 | struct extent *e; | |
853 | int i; | |
854 | __u32 rv; | |
855 | ||
856 | /* for spares just return a minimal reservation which will grow | |
857 | * once the spare is picked up by an array | |
858 | */ | |
859 | if (dl->index == -1) | |
860 | return MPB_SECTOR_CNT; | |
861 | ||
862 | e = get_extents(super, dl); | |
863 | if (!e) | |
864 | return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
865 | ||
866 | /* scroll to last entry */ | |
867 | for (i = 0; e[i].size; i++) | |
868 | continue; | |
869 | ||
870 | rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start; | |
871 | ||
872 | free(e); | |
873 | ||
874 | return rv; | |
875 | } | |
876 | ||
877 | static int is_spare(struct imsm_disk *disk) | |
878 | { | |
879 | return (disk->status & SPARE_DISK) == SPARE_DISK; | |
880 | } | |
881 | ||
882 | static int is_configured(struct imsm_disk *disk) | |
883 | { | |
884 | return (disk->status & CONFIGURED_DISK) == CONFIGURED_DISK; | |
885 | } | |
886 | ||
887 | static int is_failed(struct imsm_disk *disk) | |
888 | { | |
889 | return (disk->status & FAILED_DISK) == FAILED_DISK; | |
890 | } | |
891 | ||
892 | /* Return minimum size of a spare that can be used in this array*/ | |
893 | static unsigned long long min_acceptable_spare_size_imsm(struct supertype *st) | |
894 | { | |
895 | struct intel_super *super = st->sb; | |
896 | struct dl *dl; | |
897 | struct extent *e; | |
898 | int i; | |
899 | unsigned long long rv = 0; | |
900 | ||
901 | if (!super) | |
902 | return rv; | |
903 | /* find first active disk in array */ | |
904 | dl = super->disks; | |
905 | while (dl && (is_failed(&dl->disk) || dl->index == -1)) | |
906 | dl = dl->next; | |
907 | if (!dl) | |
908 | return rv; | |
909 | /* find last lba used by subarrays */ | |
910 | e = get_extents(super, dl); | |
911 | if (!e) | |
912 | return rv; | |
913 | for (i = 0; e[i].size; i++) | |
914 | continue; | |
915 | if (i > 0) | |
916 | rv = e[i-1].start + e[i-1].size; | |
917 | free(e); | |
918 | /* add the amount of space needed for metadata */ | |
919 | rv = rv + MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
920 | return rv * 512; | |
921 | } | |
922 | ||
923 | #ifndef MDASSEMBLE | |
924 | static __u64 blocks_per_migr_unit(struct intel_super *super, | |
925 | struct imsm_dev *dev); | |
926 | ||
927 | static void print_imsm_dev(struct intel_super *super, | |
928 | struct imsm_dev *dev, | |
929 | char *uuid, | |
930 | int disk_idx) | |
931 | { | |
932 | __u64 sz; | |
933 | int slot, i; | |
934 | struct imsm_map *map = get_imsm_map(dev, 0); | |
935 | struct imsm_map *map2 = get_imsm_map(dev, 1); | |
936 | __u32 ord; | |
937 | ||
938 | printf("\n"); | |
939 | printf("[%.16s]:\n", dev->volume); | |
940 | printf(" UUID : %s\n", uuid); | |
941 | printf(" RAID Level : %d", get_imsm_raid_level(map)); | |
942 | if (map2) | |
943 | printf(" <-- %d", get_imsm_raid_level(map2)); | |
944 | printf("\n"); | |
945 | printf(" Members : %d", map->num_members); | |
946 | if (map2) | |
947 | printf(" <-- %d", map2->num_members); | |
948 | printf("\n"); | |
949 | printf(" Slots : ["); | |
950 | for (i = 0; i < map->num_members; i++) { | |
951 | ord = get_imsm_ord_tbl_ent(dev, i, 0); | |
952 | printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U"); | |
953 | } | |
954 | printf("]"); | |
955 | if (map2) { | |
956 | printf(" <-- ["); | |
957 | for (i = 0; i < map2->num_members; i++) { | |
958 | ord = get_imsm_ord_tbl_ent(dev, i, 1); | |
959 | printf("%s", ord & IMSM_ORD_REBUILD ? "_" : "U"); | |
960 | } | |
961 | printf("]"); | |
962 | } | |
963 | printf("\n"); | |
964 | printf(" Failed disk : "); | |
965 | if (map->failed_disk_num == 0xff) | |
966 | printf("none"); | |
967 | else | |
968 | printf("%i", map->failed_disk_num); | |
969 | printf("\n"); | |
970 | slot = get_imsm_disk_slot(map, disk_idx); | |
971 | if (slot >= 0) { | |
972 | ord = get_imsm_ord_tbl_ent(dev, slot, -1); | |
973 | printf(" This Slot : %d%s\n", slot, | |
974 | ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : ""); | |
975 | } else | |
976 | printf(" This Slot : ?\n"); | |
977 | sz = __le32_to_cpu(dev->size_high); | |
978 | sz <<= 32; | |
979 | sz += __le32_to_cpu(dev->size_low); | |
980 | printf(" Array Size : %llu%s\n", (unsigned long long)sz, | |
981 | human_size(sz * 512)); | |
982 | sz = __le32_to_cpu(map->blocks_per_member); | |
983 | printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz, | |
984 | human_size(sz * 512)); | |
985 | printf(" Sector Offset : %u\n", | |
986 | __le32_to_cpu(map->pba_of_lba0)); | |
987 | printf(" Num Stripes : %u\n", | |
988 | __le32_to_cpu(map->num_data_stripes)); | |
989 | printf(" Chunk Size : %u KiB", | |
990 | __le16_to_cpu(map->blocks_per_strip) / 2); | |
991 | if (map2) | |
992 | printf(" <-- %u KiB", | |
993 | __le16_to_cpu(map2->blocks_per_strip) / 2); | |
994 | printf("\n"); | |
995 | printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks)); | |
996 | printf(" Migrate State : "); | |
997 | if (dev->vol.migr_state) { | |
998 | if (migr_type(dev) == MIGR_INIT) | |
999 | printf("initialize\n"); | |
1000 | else if (migr_type(dev) == MIGR_REBUILD) | |
1001 | printf("rebuild\n"); | |
1002 | else if (migr_type(dev) == MIGR_VERIFY) | |
1003 | printf("check\n"); | |
1004 | else if (migr_type(dev) == MIGR_GEN_MIGR) | |
1005 | printf("general migration\n"); | |
1006 | else if (migr_type(dev) == MIGR_STATE_CHANGE) | |
1007 | printf("state change\n"); | |
1008 | else if (migr_type(dev) == MIGR_REPAIR) | |
1009 | printf("repair\n"); | |
1010 | else | |
1011 | printf("<unknown:%d>\n", migr_type(dev)); | |
1012 | } else | |
1013 | printf("idle\n"); | |
1014 | printf(" Map State : %s", map_state_str[map->map_state]); | |
1015 | if (dev->vol.migr_state) { | |
1016 | struct imsm_map *map = get_imsm_map(dev, 1); | |
1017 | ||
1018 | printf(" <-- %s", map_state_str[map->map_state]); | |
1019 | printf("\n Checkpoint : %u (%llu)", | |
1020 | __le32_to_cpu(dev->vol.curr_migr_unit), | |
1021 | (unsigned long long)blocks_per_migr_unit(super, dev)); | |
1022 | } | |
1023 | printf("\n"); | |
1024 | printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean"); | |
1025 | } | |
1026 | ||
1027 | static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved) | |
1028 | { | |
1029 | struct imsm_disk *disk = __get_imsm_disk(mpb, index); | |
1030 | char str[MAX_RAID_SERIAL_LEN + 1]; | |
1031 | __u64 sz; | |
1032 | ||
1033 | if (index < 0 || !disk) | |
1034 | return; | |
1035 | ||
1036 | printf("\n"); | |
1037 | snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial); | |
1038 | printf(" Disk%02d Serial : %s\n", index, str); | |
1039 | printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "", | |
1040 | is_configured(disk) ? " active" : "", | |
1041 | is_failed(disk) ? " failed" : ""); | |
1042 | printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id)); | |
1043 | sz = __le32_to_cpu(disk->total_blocks) - reserved; | |
1044 | printf(" Usable Size : %llu%s\n", (unsigned long long)sz, | |
1045 | human_size(sz * 512)); | |
1046 | } | |
1047 | ||
1048 | static int is_gen_migration(struct imsm_dev *dev); | |
1049 | ||
1050 | void examine_migr_rec_imsm(struct intel_super *super) | |
1051 | { | |
1052 | struct migr_record *migr_rec = super->migr_rec; | |
1053 | struct imsm_super *mpb = super->anchor; | |
1054 | int i; | |
1055 | ||
1056 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
1057 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
1058 | if (is_gen_migration(dev) == 0) | |
1059 | continue; | |
1060 | ||
1061 | printf("\nMigration Record Information:"); | |
1062 | if (super->disks->index > 1) { | |
1063 | printf(" Empty\n "); | |
1064 | printf("Examine one of first two disks in array\n"); | |
1065 | break; | |
1066 | } | |
1067 | printf("\n Status : "); | |
1068 | if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL) | |
1069 | printf("Normal\n"); | |
1070 | else | |
1071 | printf("Contains Data\n"); | |
1072 | printf(" Current Unit : %u\n", | |
1073 | __le32_to_cpu(migr_rec->curr_migr_unit)); | |
1074 | printf(" Family : %u\n", | |
1075 | __le32_to_cpu(migr_rec->family_num)); | |
1076 | printf(" Ascending : %u\n", | |
1077 | __le32_to_cpu(migr_rec->ascending_migr)); | |
1078 | printf(" Blocks Per Unit : %u\n", | |
1079 | __le32_to_cpu(migr_rec->blocks_per_unit)); | |
1080 | printf(" Dest. Depth Per Unit : %u\n", | |
1081 | __le32_to_cpu(migr_rec->dest_depth_per_unit)); | |
1082 | printf(" Checkpoint Area pba : %u\n", | |
1083 | __le32_to_cpu(migr_rec->ckpt_area_pba)); | |
1084 | printf(" First member lba : %u\n", | |
1085 | __le32_to_cpu(migr_rec->dest_1st_member_lba)); | |
1086 | printf(" Total Number of Units : %u\n", | |
1087 | __le32_to_cpu(migr_rec->num_migr_units)); | |
1088 | printf(" Size of volume : %u\n", | |
1089 | __le32_to_cpu(migr_rec->post_migr_vol_cap)); | |
1090 | printf(" Expansion space for LBA64 : %u\n", | |
1091 | __le32_to_cpu(migr_rec->post_migr_vol_cap_hi)); | |
1092 | printf(" Record was read from : %u\n", | |
1093 | __le32_to_cpu(migr_rec->ckpt_read_disk_num)); | |
1094 | ||
1095 | break; | |
1096 | } | |
1097 | } | |
1098 | ||
1099 | static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map); | |
1100 | ||
1101 | static void examine_super_imsm(struct supertype *st, char *homehost) | |
1102 | { | |
1103 | struct intel_super *super = st->sb; | |
1104 | struct imsm_super *mpb = super->anchor; | |
1105 | char str[MAX_SIGNATURE_LENGTH]; | |
1106 | int i; | |
1107 | struct mdinfo info; | |
1108 | char nbuf[64]; | |
1109 | __u32 sum; | |
1110 | __u32 reserved = imsm_reserved_sectors(super, super->disks); | |
1111 | struct dl *dl; | |
1112 | ||
1113 | snprintf(str, MPB_SIG_LEN, "%s", mpb->sig); | |
1114 | printf(" Magic : %s\n", str); | |
1115 | snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb)); | |
1116 | printf(" Version : %s\n", get_imsm_version(mpb)); | |
1117 | printf(" Orig Family : %08x\n", __le32_to_cpu(mpb->orig_family_num)); | |
1118 | printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num)); | |
1119 | printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num)); | |
1120 | getinfo_super_imsm(st, &info, NULL); | |
1121 | fname_from_uuid(st, &info, nbuf, ':'); | |
1122 | printf(" UUID : %s\n", nbuf + 5); | |
1123 | sum = __le32_to_cpu(mpb->check_sum); | |
1124 | printf(" Checksum : %08x %s\n", sum, | |
1125 | __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect"); | |
1126 | printf(" MPB Sectors : %d\n", mpb_sectors(mpb)); | |
1127 | printf(" Disks : %d\n", mpb->num_disks); | |
1128 | printf(" RAID Devices : %d\n", mpb->num_raid_devs); | |
1129 | print_imsm_disk(mpb, super->disks->index, reserved); | |
1130 | if (super->bbm_log) { | |
1131 | struct bbm_log *log = super->bbm_log; | |
1132 | ||
1133 | printf("\n"); | |
1134 | printf("Bad Block Management Log:\n"); | |
1135 | printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size)); | |
1136 | printf(" Signature : %x\n", __le32_to_cpu(log->signature)); | |
1137 | printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count)); | |
1138 | printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count)); | |
1139 | printf(" First Spare : %llx\n", | |
1140 | (unsigned long long) __le64_to_cpu(log->first_spare_lba)); | |
1141 | } | |
1142 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
1143 | struct mdinfo info; | |
1144 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
1145 | ||
1146 | super->current_vol = i; | |
1147 | getinfo_super_imsm(st, &info, NULL); | |
1148 | fname_from_uuid(st, &info, nbuf, ':'); | |
1149 | print_imsm_dev(super, dev, nbuf + 5, super->disks->index); | |
1150 | } | |
1151 | for (i = 0; i < mpb->num_disks; i++) { | |
1152 | if (i == super->disks->index) | |
1153 | continue; | |
1154 | print_imsm_disk(mpb, i, reserved); | |
1155 | } | |
1156 | for (dl = super->disks ; dl; dl = dl->next) { | |
1157 | struct imsm_disk *disk; | |
1158 | char str[MAX_RAID_SERIAL_LEN + 1]; | |
1159 | __u64 sz; | |
1160 | ||
1161 | if (dl->index >= 0) | |
1162 | continue; | |
1163 | ||
1164 | disk = &dl->disk; | |
1165 | printf("\n"); | |
1166 | snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial); | |
1167 | printf(" Disk Serial : %s\n", str); | |
1168 | printf(" State :%s%s%s\n", is_spare(disk) ? " spare" : "", | |
1169 | is_configured(disk) ? " active" : "", | |
1170 | is_failed(disk) ? " failed" : ""); | |
1171 | printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id)); | |
1172 | sz = __le32_to_cpu(disk->total_blocks) - reserved; | |
1173 | printf(" Usable Size : %llu%s\n", (unsigned long long)sz, | |
1174 | human_size(sz * 512)); | |
1175 | } | |
1176 | ||
1177 | examine_migr_rec_imsm(super); | |
1178 | } | |
1179 | ||
1180 | static void brief_examine_super_imsm(struct supertype *st, int verbose) | |
1181 | { | |
1182 | /* We just write a generic IMSM ARRAY entry */ | |
1183 | struct mdinfo info; | |
1184 | char nbuf[64]; | |
1185 | struct intel_super *super = st->sb; | |
1186 | ||
1187 | if (!super->anchor->num_raid_devs) { | |
1188 | printf("ARRAY metadata=imsm\n"); | |
1189 | return; | |
1190 | } | |
1191 | ||
1192 | getinfo_super_imsm(st, &info, NULL); | |
1193 | fname_from_uuid(st, &info, nbuf, ':'); | |
1194 | printf("ARRAY metadata=imsm UUID=%s\n", nbuf + 5); | |
1195 | } | |
1196 | ||
1197 | static void brief_examine_subarrays_imsm(struct supertype *st, int verbose) | |
1198 | { | |
1199 | /* We just write a generic IMSM ARRAY entry */ | |
1200 | struct mdinfo info; | |
1201 | char nbuf[64]; | |
1202 | char nbuf1[64]; | |
1203 | struct intel_super *super = st->sb; | |
1204 | int i; | |
1205 | ||
1206 | if (!super->anchor->num_raid_devs) | |
1207 | return; | |
1208 | ||
1209 | getinfo_super_imsm(st, &info, NULL); | |
1210 | fname_from_uuid(st, &info, nbuf, ':'); | |
1211 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
1212 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
1213 | ||
1214 | super->current_vol = i; | |
1215 | getinfo_super_imsm(st, &info, NULL); | |
1216 | fname_from_uuid(st, &info, nbuf1, ':'); | |
1217 | printf("ARRAY /dev/md/%.16s container=%s member=%d UUID=%s\n", | |
1218 | dev->volume, nbuf + 5, i, nbuf1 + 5); | |
1219 | } | |
1220 | } | |
1221 | ||
1222 | static void export_examine_super_imsm(struct supertype *st) | |
1223 | { | |
1224 | struct intel_super *super = st->sb; | |
1225 | struct imsm_super *mpb = super->anchor; | |
1226 | struct mdinfo info; | |
1227 | char nbuf[64]; | |
1228 | ||
1229 | getinfo_super_imsm(st, &info, NULL); | |
1230 | fname_from_uuid(st, &info, nbuf, ':'); | |
1231 | printf("MD_METADATA=imsm\n"); | |
1232 | printf("MD_LEVEL=container\n"); | |
1233 | printf("MD_UUID=%s\n", nbuf+5); | |
1234 | printf("MD_DEVICES=%u\n", mpb->num_disks); | |
1235 | } | |
1236 | ||
1237 | static void detail_super_imsm(struct supertype *st, char *homehost) | |
1238 | { | |
1239 | struct mdinfo info; | |
1240 | char nbuf[64]; | |
1241 | ||
1242 | getinfo_super_imsm(st, &info, NULL); | |
1243 | fname_from_uuid(st, &info, nbuf, ':'); | |
1244 | printf("\n UUID : %s\n", nbuf + 5); | |
1245 | } | |
1246 | ||
1247 | static void brief_detail_super_imsm(struct supertype *st) | |
1248 | { | |
1249 | struct mdinfo info; | |
1250 | char nbuf[64]; | |
1251 | getinfo_super_imsm(st, &info, NULL); | |
1252 | fname_from_uuid(st, &info, nbuf, ':'); | |
1253 | printf(" UUID=%s", nbuf + 5); | |
1254 | } | |
1255 | ||
1256 | static int imsm_read_serial(int fd, char *devname, __u8 *serial); | |
1257 | static void fd2devname(int fd, char *name); | |
1258 | ||
1259 | static int ahci_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose) | |
1260 | { | |
1261 | /* dump an unsorted list of devices attached to AHCI Intel storage | |
1262 | * controller, as well as non-connected ports | |
1263 | */ | |
1264 | int hba_len = strlen(hba_path) + 1; | |
1265 | struct dirent *ent; | |
1266 | DIR *dir; | |
1267 | char *path = NULL; | |
1268 | int err = 0; | |
1269 | unsigned long port_mask = (1 << port_count) - 1; | |
1270 | ||
1271 | if (port_count > (int)sizeof(port_mask) * 8) { | |
1272 | if (verbose) | |
1273 | fprintf(stderr, Name ": port_count %d out of range\n", port_count); | |
1274 | return 2; | |
1275 | } | |
1276 | ||
1277 | /* scroll through /sys/dev/block looking for devices attached to | |
1278 | * this hba | |
1279 | */ | |
1280 | dir = opendir("/sys/dev/block"); | |
1281 | for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) { | |
1282 | int fd; | |
1283 | char model[64]; | |
1284 | char vendor[64]; | |
1285 | char buf[1024]; | |
1286 | int major, minor; | |
1287 | char *device; | |
1288 | char *c; | |
1289 | int port; | |
1290 | int type; | |
1291 | ||
1292 | if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2) | |
1293 | continue; | |
1294 | path = devt_to_devpath(makedev(major, minor)); | |
1295 | if (!path) | |
1296 | continue; | |
1297 | if (!path_attached_to_hba(path, hba_path)) { | |
1298 | free(path); | |
1299 | path = NULL; | |
1300 | continue; | |
1301 | } | |
1302 | ||
1303 | /* retrieve the scsi device type */ | |
1304 | if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) { | |
1305 | if (verbose) | |
1306 | fprintf(stderr, Name ": failed to allocate 'device'\n"); | |
1307 | err = 2; | |
1308 | break; | |
1309 | } | |
1310 | sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor); | |
1311 | if (load_sys(device, buf) != 0) { | |
1312 | if (verbose) | |
1313 | fprintf(stderr, Name ": failed to read device type for %s\n", | |
1314 | path); | |
1315 | err = 2; | |
1316 | free(device); | |
1317 | break; | |
1318 | } | |
1319 | type = strtoul(buf, NULL, 10); | |
1320 | ||
1321 | /* if it's not a disk print the vendor and model */ | |
1322 | if (!(type == 0 || type == 7 || type == 14)) { | |
1323 | vendor[0] = '\0'; | |
1324 | model[0] = '\0'; | |
1325 | sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor); | |
1326 | if (load_sys(device, buf) == 0) { | |
1327 | strncpy(vendor, buf, sizeof(vendor)); | |
1328 | vendor[sizeof(vendor) - 1] = '\0'; | |
1329 | c = (char *) &vendor[sizeof(vendor) - 1]; | |
1330 | while (isspace(*c) || *c == '\0') | |
1331 | *c-- = '\0'; | |
1332 | ||
1333 | } | |
1334 | sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor); | |
1335 | if (load_sys(device, buf) == 0) { | |
1336 | strncpy(model, buf, sizeof(model)); | |
1337 | model[sizeof(model) - 1] = '\0'; | |
1338 | c = (char *) &model[sizeof(model) - 1]; | |
1339 | while (isspace(*c) || *c == '\0') | |
1340 | *c-- = '\0'; | |
1341 | } | |
1342 | ||
1343 | if (vendor[0] && model[0]) | |
1344 | sprintf(buf, "%.64s %.64s", vendor, model); | |
1345 | else | |
1346 | switch (type) { /* numbers from hald/linux/device.c */ | |
1347 | case 1: sprintf(buf, "tape"); break; | |
1348 | case 2: sprintf(buf, "printer"); break; | |
1349 | case 3: sprintf(buf, "processor"); break; | |
1350 | case 4: | |
1351 | case 5: sprintf(buf, "cdrom"); break; | |
1352 | case 6: sprintf(buf, "scanner"); break; | |
1353 | case 8: sprintf(buf, "media_changer"); break; | |
1354 | case 9: sprintf(buf, "comm"); break; | |
1355 | case 12: sprintf(buf, "raid"); break; | |
1356 | default: sprintf(buf, "unknown"); | |
1357 | } | |
1358 | } else | |
1359 | buf[0] = '\0'; | |
1360 | free(device); | |
1361 | ||
1362 | /* chop device path to 'host%d' and calculate the port number */ | |
1363 | c = strchr(&path[hba_len], '/'); | |
1364 | if (!c) { | |
1365 | if (verbose) | |
1366 | fprintf(stderr, Name ": %s - invalid path name\n", path + hba_len); | |
1367 | err = 2; | |
1368 | break; | |
1369 | } | |
1370 | *c = '\0'; | |
1371 | if (sscanf(&path[hba_len], "host%d", &port) == 1) | |
1372 | port -= host_base; | |
1373 | else { | |
1374 | if (verbose) { | |
1375 | *c = '/'; /* repair the full string */ | |
1376 | fprintf(stderr, Name ": failed to determine port number for %s\n", | |
1377 | path); | |
1378 | } | |
1379 | err = 2; | |
1380 | break; | |
1381 | } | |
1382 | ||
1383 | /* mark this port as used */ | |
1384 | port_mask &= ~(1 << port); | |
1385 | ||
1386 | /* print out the device information */ | |
1387 | if (buf[0]) { | |
1388 | printf(" Port%d : - non-disk device (%s) -\n", port, buf); | |
1389 | continue; | |
1390 | } | |
1391 | ||
1392 | fd = dev_open(ent->d_name, O_RDONLY); | |
1393 | if (fd < 0) | |
1394 | printf(" Port%d : - disk info unavailable -\n", port); | |
1395 | else { | |
1396 | fd2devname(fd, buf); | |
1397 | printf(" Port%d : %s", port, buf); | |
1398 | if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0) | |
1399 | printf(" (%s)\n", buf); | |
1400 | else | |
1401 | printf("()\n"); | |
1402 | } | |
1403 | close(fd); | |
1404 | free(path); | |
1405 | path = NULL; | |
1406 | } | |
1407 | if (path) | |
1408 | free(path); | |
1409 | if (dir) | |
1410 | closedir(dir); | |
1411 | if (err == 0) { | |
1412 | int i; | |
1413 | ||
1414 | for (i = 0; i < port_count; i++) | |
1415 | if (port_mask & (1 << i)) | |
1416 | printf(" Port%d : - no device attached -\n", i); | |
1417 | } | |
1418 | ||
1419 | return err; | |
1420 | } | |
1421 | ||
1422 | ||
1423 | ||
1424 | static void print_found_intel_controllers(struct sys_dev *elem) | |
1425 | { | |
1426 | for (; elem; elem = elem->next) { | |
1427 | fprintf(stderr, Name ": found Intel(R) "); | |
1428 | if (elem->type == SYS_DEV_SATA) | |
1429 | fprintf(stderr, "SATA "); | |
1430 | else if (elem->type == SYS_DEV_SAS) | |
1431 | fprintf(stderr, "SAS "); | |
1432 | fprintf(stderr, "RAID controller"); | |
1433 | if (elem->pci_id) | |
1434 | fprintf(stderr, " at %s", elem->pci_id); | |
1435 | fprintf(stderr, ".\n"); | |
1436 | } | |
1437 | fflush(stderr); | |
1438 | } | |
1439 | ||
1440 | static int ahci_get_port_count(const char *hba_path, int *port_count) | |
1441 | { | |
1442 | struct dirent *ent; | |
1443 | DIR *dir; | |
1444 | int host_base = -1; | |
1445 | ||
1446 | *port_count = 0; | |
1447 | if ((dir = opendir(hba_path)) == NULL) | |
1448 | return -1; | |
1449 | ||
1450 | for (ent = readdir(dir); ent; ent = readdir(dir)) { | |
1451 | int host; | |
1452 | ||
1453 | if (sscanf(ent->d_name, "host%d", &host) != 1) | |
1454 | continue; | |
1455 | if (*port_count == 0) | |
1456 | host_base = host; | |
1457 | else if (host < host_base) | |
1458 | host_base = host; | |
1459 | ||
1460 | if (host + 1 > *port_count + host_base) | |
1461 | *port_count = host + 1 - host_base; | |
1462 | } | |
1463 | closedir(dir); | |
1464 | return host_base; | |
1465 | } | |
1466 | ||
1467 | static void print_imsm_capability(const struct imsm_orom *orom) | |
1468 | { | |
1469 | printf(" Platform : Intel(R) Matrix Storage Manager\n"); | |
1470 | printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver, | |
1471 | orom->hotfix_ver, orom->build); | |
1472 | printf(" RAID Levels :%s%s%s%s%s\n", | |
1473 | imsm_orom_has_raid0(orom) ? " raid0" : "", | |
1474 | imsm_orom_has_raid1(orom) ? " raid1" : "", | |
1475 | imsm_orom_has_raid1e(orom) ? " raid1e" : "", | |
1476 | imsm_orom_has_raid10(orom) ? " raid10" : "", | |
1477 | imsm_orom_has_raid5(orom) ? " raid5" : ""); | |
1478 | printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", | |
1479 | imsm_orom_has_chunk(orom, 2) ? " 2k" : "", | |
1480 | imsm_orom_has_chunk(orom, 4) ? " 4k" : "", | |
1481 | imsm_orom_has_chunk(orom, 8) ? " 8k" : "", | |
1482 | imsm_orom_has_chunk(orom, 16) ? " 16k" : "", | |
1483 | imsm_orom_has_chunk(orom, 32) ? " 32k" : "", | |
1484 | imsm_orom_has_chunk(orom, 64) ? " 64k" : "", | |
1485 | imsm_orom_has_chunk(orom, 128) ? " 128k" : "", | |
1486 | imsm_orom_has_chunk(orom, 256) ? " 256k" : "", | |
1487 | imsm_orom_has_chunk(orom, 512) ? " 512k" : "", | |
1488 | imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "", | |
1489 | imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "", | |
1490 | imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "", | |
1491 | imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "", | |
1492 | imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "", | |
1493 | imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "", | |
1494 | imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : ""); | |
1495 | printf(" Max Disks : %d\n", orom->tds); | |
1496 | printf(" Max Volumes : %d\n", orom->vpa); | |
1497 | return; | |
1498 | } | |
1499 | ||
1500 | static int detail_platform_imsm(int verbose, int enumerate_only) | |
1501 | { | |
1502 | /* There are two components to imsm platform support, the ahci SATA | |
1503 | * controller and the option-rom. To find the SATA controller we | |
1504 | * simply look in /sys/bus/pci/drivers/ahci to see if an ahci | |
1505 | * controller with the Intel vendor id is present. This approach | |
1506 | * allows mdadm to leverage the kernel's ahci detection logic, with the | |
1507 | * caveat that if ahci.ko is not loaded mdadm will not be able to | |
1508 | * detect platform raid capabilities. The option-rom resides in a | |
1509 | * platform "Adapter ROM". We scan for its signature to retrieve the | |
1510 | * platform capabilities. If raid support is disabled in the BIOS the | |
1511 | * option-rom capability structure will not be available. | |
1512 | */ | |
1513 | const struct imsm_orom *orom; | |
1514 | struct sys_dev *list, *hba; | |
1515 | int host_base = 0; | |
1516 | int port_count = 0; | |
1517 | int result=0; | |
1518 | ||
1519 | if (enumerate_only) { | |
1520 | if (check_env("IMSM_NO_PLATFORM")) | |
1521 | return 0; | |
1522 | list = find_intel_devices(); | |
1523 | if (!list) | |
1524 | return 2; | |
1525 | for (hba = list; hba; hba = hba->next) { | |
1526 | orom = find_imsm_capability(hba->type); | |
1527 | if (!orom) { | |
1528 | result = 2; | |
1529 | break; | |
1530 | } | |
1531 | } | |
1532 | free_sys_dev(&list); | |
1533 | return result; | |
1534 | } | |
1535 | ||
1536 | list = find_intel_devices(); | |
1537 | if (!list) { | |
1538 | if (verbose) | |
1539 | fprintf(stderr, Name ": no active Intel(R) RAID " | |
1540 | "controller found.\n"); | |
1541 | free_sys_dev(&list); | |
1542 | return 2; | |
1543 | } else if (verbose) | |
1544 | print_found_intel_controllers(list); | |
1545 | ||
1546 | for (hba = list; hba; hba = hba->next) { | |
1547 | orom = find_imsm_capability(hba->type); | |
1548 | if (!orom) | |
1549 | fprintf(stderr, Name ": imsm capabilities not found for controller: %s (type %s)\n", | |
1550 | hba->path, get_sys_dev_type(hba->type)); | |
1551 | else | |
1552 | print_imsm_capability(orom); | |
1553 | } | |
1554 | ||
1555 | for (hba = list; hba; hba = hba->next) { | |
1556 | printf(" I/O Controller : %s (%s)\n", | |
1557 | hba->path, get_sys_dev_type(hba->type)); | |
1558 | ||
1559 | if (hba->type == SYS_DEV_SATA) { | |
1560 | host_base = ahci_get_port_count(hba->path, &port_count); | |
1561 | if (ahci_enumerate_ports(hba->path, port_count, host_base, verbose)) { | |
1562 | if (verbose) | |
1563 | fprintf(stderr, Name ": failed to enumerate " | |
1564 | "ports on SATA controller at %s.", hba->pci_id); | |
1565 | result |= 2; | |
1566 | } | |
1567 | } | |
1568 | } | |
1569 | ||
1570 | free_sys_dev(&list); | |
1571 | return result; | |
1572 | } | |
1573 | #endif | |
1574 | ||
1575 | static int match_home_imsm(struct supertype *st, char *homehost) | |
1576 | { | |
1577 | /* the imsm metadata format does not specify any host | |
1578 | * identification information. We return -1 since we can never | |
1579 | * confirm nor deny whether a given array is "meant" for this | |
1580 | * host. We rely on compare_super and the 'family_num' fields to | |
1581 | * exclude member disks that do not belong, and we rely on | |
1582 | * mdadm.conf to specify the arrays that should be assembled. | |
1583 | * Auto-assembly may still pick up "foreign" arrays. | |
1584 | */ | |
1585 | ||
1586 | return -1; | |
1587 | } | |
1588 | ||
1589 | static void uuid_from_super_imsm(struct supertype *st, int uuid[4]) | |
1590 | { | |
1591 | /* The uuid returned here is used for: | |
1592 | * uuid to put into bitmap file (Create, Grow) | |
1593 | * uuid for backup header when saving critical section (Grow) | |
1594 | * comparing uuids when re-adding a device into an array | |
1595 | * In these cases the uuid required is that of the data-array, | |
1596 | * not the device-set. | |
1597 | * uuid to recognise same set when adding a missing device back | |
1598 | * to an array. This is a uuid for the device-set. | |
1599 | * | |
1600 | * For each of these we can make do with a truncated | |
1601 | * or hashed uuid rather than the original, as long as | |
1602 | * everyone agrees. | |
1603 | * In each case the uuid required is that of the data-array, | |
1604 | * not the device-set. | |
1605 | */ | |
1606 | /* imsm does not track uuid's so we synthesis one using sha1 on | |
1607 | * - The signature (Which is constant for all imsm array, but no matter) | |
1608 | * - the orig_family_num of the container | |
1609 | * - the index number of the volume | |
1610 | * - the 'serial' number of the volume. | |
1611 | * Hopefully these are all constant. | |
1612 | */ | |
1613 | struct intel_super *super = st->sb; | |
1614 | ||
1615 | char buf[20]; | |
1616 | struct sha1_ctx ctx; | |
1617 | struct imsm_dev *dev = NULL; | |
1618 | __u32 family_num; | |
1619 | ||
1620 | /* some mdadm versions failed to set ->orig_family_num, in which | |
1621 | * case fall back to ->family_num. orig_family_num will be | |
1622 | * fixed up with the first metadata update. | |
1623 | */ | |
1624 | family_num = super->anchor->orig_family_num; | |
1625 | if (family_num == 0) | |
1626 | family_num = super->anchor->family_num; | |
1627 | sha1_init_ctx(&ctx); | |
1628 | sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx); | |
1629 | sha1_process_bytes(&family_num, sizeof(__u32), &ctx); | |
1630 | if (super->current_vol >= 0) | |
1631 | dev = get_imsm_dev(super, super->current_vol); | |
1632 | if (dev) { | |
1633 | __u32 vol = super->current_vol; | |
1634 | sha1_process_bytes(&vol, sizeof(vol), &ctx); | |
1635 | sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx); | |
1636 | } | |
1637 | sha1_finish_ctx(&ctx, buf); | |
1638 | memcpy(uuid, buf, 4*4); | |
1639 | } | |
1640 | ||
1641 | #if 0 | |
1642 | static void | |
1643 | get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p) | |
1644 | { | |
1645 | __u8 *v = get_imsm_version(mpb); | |
1646 | __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH; | |
1647 | char major[] = { 0, 0, 0 }; | |
1648 | char minor[] = { 0 ,0, 0 }; | |
1649 | char patch[] = { 0, 0, 0 }; | |
1650 | char *ver_parse[] = { major, minor, patch }; | |
1651 | int i, j; | |
1652 | ||
1653 | i = j = 0; | |
1654 | while (*v != '\0' && v < end) { | |
1655 | if (*v != '.' && j < 2) | |
1656 | ver_parse[i][j++] = *v; | |
1657 | else { | |
1658 | i++; | |
1659 | j = 0; | |
1660 | } | |
1661 | v++; | |
1662 | } | |
1663 | ||
1664 | *m = strtol(minor, NULL, 0); | |
1665 | *p = strtol(patch, NULL, 0); | |
1666 | } | |
1667 | #endif | |
1668 | ||
1669 | static __u32 migr_strip_blocks_resync(struct imsm_dev *dev) | |
1670 | { | |
1671 | /* migr_strip_size when repairing or initializing parity */ | |
1672 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1673 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1674 | ||
1675 | switch (get_imsm_raid_level(map)) { | |
1676 | case 5: | |
1677 | case 10: | |
1678 | return chunk; | |
1679 | default: | |
1680 | return 128*1024 >> 9; | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev) | |
1685 | { | |
1686 | /* migr_strip_size when rebuilding a degraded disk, no idea why | |
1687 | * this is different than migr_strip_size_resync(), but it's good | |
1688 | * to be compatible | |
1689 | */ | |
1690 | struct imsm_map *map = get_imsm_map(dev, 1); | |
1691 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1692 | ||
1693 | switch (get_imsm_raid_level(map)) { | |
1694 | case 1: | |
1695 | case 10: | |
1696 | if (map->num_members % map->num_domains == 0) | |
1697 | return 128*1024 >> 9; | |
1698 | else | |
1699 | return chunk; | |
1700 | case 5: | |
1701 | return max((__u32) 64*1024 >> 9, chunk); | |
1702 | default: | |
1703 | return 128*1024 >> 9; | |
1704 | } | |
1705 | } | |
1706 | ||
1707 | static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev) | |
1708 | { | |
1709 | struct imsm_map *lo = get_imsm_map(dev, 0); | |
1710 | struct imsm_map *hi = get_imsm_map(dev, 1); | |
1711 | __u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip); | |
1712 | __u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip); | |
1713 | ||
1714 | return max((__u32) 1, hi_chunk / lo_chunk); | |
1715 | } | |
1716 | ||
1717 | static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev) | |
1718 | { | |
1719 | struct imsm_map *lo = get_imsm_map(dev, 0); | |
1720 | int level = get_imsm_raid_level(lo); | |
1721 | ||
1722 | if (level == 1 || level == 10) { | |
1723 | struct imsm_map *hi = get_imsm_map(dev, 1); | |
1724 | ||
1725 | return hi->num_domains; | |
1726 | } else | |
1727 | return num_stripes_per_unit_resync(dev); | |
1728 | } | |
1729 | ||
1730 | static __u8 imsm_num_data_members(struct imsm_dev *dev, int second_map) | |
1731 | { | |
1732 | /* named 'imsm_' because raid0, raid1 and raid10 | |
1733 | * counter-intuitively have the same number of data disks | |
1734 | */ | |
1735 | struct imsm_map *map = get_imsm_map(dev, second_map); | |
1736 | ||
1737 | switch (get_imsm_raid_level(map)) { | |
1738 | case 0: | |
1739 | case 1: | |
1740 | case 10: | |
1741 | return map->num_members; | |
1742 | case 5: | |
1743 | return map->num_members - 1; | |
1744 | default: | |
1745 | dprintf("%s: unsupported raid level\n", __func__); | |
1746 | return 0; | |
1747 | } | |
1748 | } | |
1749 | ||
1750 | static __u32 parity_segment_depth(struct imsm_dev *dev) | |
1751 | { | |
1752 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1753 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1754 | ||
1755 | switch(get_imsm_raid_level(map)) { | |
1756 | case 1: | |
1757 | case 10: | |
1758 | return chunk * map->num_domains; | |
1759 | case 5: | |
1760 | return chunk * map->num_members; | |
1761 | default: | |
1762 | return chunk; | |
1763 | } | |
1764 | } | |
1765 | ||
1766 | static __u32 map_migr_block(struct imsm_dev *dev, __u32 block) | |
1767 | { | |
1768 | struct imsm_map *map = get_imsm_map(dev, 1); | |
1769 | __u32 chunk = __le32_to_cpu(map->blocks_per_strip); | |
1770 | __u32 strip = block / chunk; | |
1771 | ||
1772 | switch (get_imsm_raid_level(map)) { | |
1773 | case 1: | |
1774 | case 10: { | |
1775 | __u32 vol_strip = (strip * map->num_domains) + 1; | |
1776 | __u32 vol_stripe = vol_strip / map->num_members; | |
1777 | ||
1778 | return vol_stripe * chunk + block % chunk; | |
1779 | } case 5: { | |
1780 | __u32 stripe = strip / (map->num_members - 1); | |
1781 | ||
1782 | return stripe * chunk + block % chunk; | |
1783 | } | |
1784 | default: | |
1785 | return 0; | |
1786 | } | |
1787 | } | |
1788 | ||
1789 | static __u64 blocks_per_migr_unit(struct intel_super *super, | |
1790 | struct imsm_dev *dev) | |
1791 | { | |
1792 | /* calculate the conversion factor between per member 'blocks' | |
1793 | * (md/{resync,rebuild}_start) and imsm migration units, return | |
1794 | * 0 for the 'not migrating' and 'unsupported migration' cases | |
1795 | */ | |
1796 | if (!dev->vol.migr_state) | |
1797 | return 0; | |
1798 | ||
1799 | switch (migr_type(dev)) { | |
1800 | case MIGR_GEN_MIGR: { | |
1801 | struct migr_record *migr_rec = super->migr_rec; | |
1802 | return __le32_to_cpu(migr_rec->blocks_per_unit); | |
1803 | } | |
1804 | case MIGR_VERIFY: | |
1805 | case MIGR_REPAIR: | |
1806 | case MIGR_INIT: { | |
1807 | struct imsm_map *map = get_imsm_map(dev, 0); | |
1808 | __u32 stripes_per_unit; | |
1809 | __u32 blocks_per_unit; | |
1810 | __u32 parity_depth; | |
1811 | __u32 migr_chunk; | |
1812 | __u32 block_map; | |
1813 | __u32 block_rel; | |
1814 | __u32 segment; | |
1815 | __u32 stripe; | |
1816 | __u8 disks; | |
1817 | ||
1818 | /* yes, this is really the translation of migr_units to | |
1819 | * per-member blocks in the 'resync' case | |
1820 | */ | |
1821 | stripes_per_unit = num_stripes_per_unit_resync(dev); | |
1822 | migr_chunk = migr_strip_blocks_resync(dev); | |
1823 | disks = imsm_num_data_members(dev, 0); | |
1824 | blocks_per_unit = stripes_per_unit * migr_chunk * disks; | |
1825 | stripe = __le16_to_cpu(map->blocks_per_strip) * disks; | |
1826 | segment = blocks_per_unit / stripe; | |
1827 | block_rel = blocks_per_unit - segment * stripe; | |
1828 | parity_depth = parity_segment_depth(dev); | |
1829 | block_map = map_migr_block(dev, block_rel); | |
1830 | return block_map + parity_depth * segment; | |
1831 | } | |
1832 | case MIGR_REBUILD: { | |
1833 | __u32 stripes_per_unit; | |
1834 | __u32 migr_chunk; | |
1835 | ||
1836 | stripes_per_unit = num_stripes_per_unit_rebuild(dev); | |
1837 | migr_chunk = migr_strip_blocks_rebuild(dev); | |
1838 | return migr_chunk * stripes_per_unit; | |
1839 | } | |
1840 | case MIGR_STATE_CHANGE: | |
1841 | default: | |
1842 | return 0; | |
1843 | } | |
1844 | } | |
1845 | ||
1846 | static int imsm_level_to_layout(int level) | |
1847 | { | |
1848 | switch (level) { | |
1849 | case 0: | |
1850 | case 1: | |
1851 | return 0; | |
1852 | case 5: | |
1853 | case 6: | |
1854 | return ALGORITHM_LEFT_ASYMMETRIC; | |
1855 | case 10: | |
1856 | return 0x102; | |
1857 | } | |
1858 | return UnSet; | |
1859 | } | |
1860 | ||
1861 | /******************************************************************************* | |
1862 | * Function: read_imsm_migr_rec | |
1863 | * Description: Function reads imsm migration record from last sector of disk | |
1864 | * Parameters: | |
1865 | * fd : disk descriptor | |
1866 | * super : metadata info | |
1867 | * Returns: | |
1868 | * 0 : success, | |
1869 | * -1 : fail | |
1870 | ******************************************************************************/ | |
1871 | static int read_imsm_migr_rec(int fd, struct intel_super *super) | |
1872 | { | |
1873 | int ret_val = -1; | |
1874 | unsigned long long dsize; | |
1875 | ||
1876 | get_dev_size(fd, NULL, &dsize); | |
1877 | if (lseek64(fd, dsize - 512, SEEK_SET) < 0) { | |
1878 | fprintf(stderr, | |
1879 | Name ": Cannot seek to anchor block: %s\n", | |
1880 | strerror(errno)); | |
1881 | goto out; | |
1882 | } | |
1883 | if (read(fd, super->migr_rec_buf, 512) != 512) { | |
1884 | fprintf(stderr, | |
1885 | Name ": Cannot read migr record block: %s\n", | |
1886 | strerror(errno)); | |
1887 | goto out; | |
1888 | } | |
1889 | ret_val = 0; | |
1890 | ||
1891 | out: | |
1892 | return ret_val; | |
1893 | } | |
1894 | ||
1895 | /******************************************************************************* | |
1896 | * Function: load_imsm_migr_rec | |
1897 | * Description: Function reads imsm migration record (it is stored at the last | |
1898 | * sector of disk) | |
1899 | * Parameters: | |
1900 | * super : imsm internal array info | |
1901 | * info : general array info | |
1902 | * Returns: | |
1903 | * 0 : success | |
1904 | * -1 : fail | |
1905 | ******************************************************************************/ | |
1906 | static int load_imsm_migr_rec(struct intel_super *super, struct mdinfo *info) | |
1907 | { | |
1908 | struct mdinfo *sd; | |
1909 | struct dl *dl = NULL; | |
1910 | char nm[30]; | |
1911 | int retval = -1; | |
1912 | int fd = -1; | |
1913 | ||
1914 | if (info) { | |
1915 | for (sd = info->devs ; sd ; sd = sd->next) { | |
1916 | /* read only from one of the first two slots */ | |
1917 | if ((sd->disk.raid_disk > 1) || | |
1918 | (sd->disk.raid_disk < 0)) | |
1919 | continue; | |
1920 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
1921 | fd = dev_open(nm, O_RDONLY); | |
1922 | if (fd >= 0) | |
1923 | break; | |
1924 | } | |
1925 | } | |
1926 | if (fd < 0) { | |
1927 | for (dl = super->disks; dl; dl = dl->next) { | |
1928 | /* read only from one of the first two slots */ | |
1929 | if (dl->index > 1) | |
1930 | continue; | |
1931 | sprintf(nm, "%d:%d", dl->major, dl->minor); | |
1932 | fd = dev_open(nm, O_RDONLY); | |
1933 | if (fd >= 0) | |
1934 | break; | |
1935 | } | |
1936 | } | |
1937 | if (fd < 0) | |
1938 | goto out; | |
1939 | retval = read_imsm_migr_rec(fd, super); | |
1940 | ||
1941 | out: | |
1942 | if (fd >= 0) | |
1943 | close(fd); | |
1944 | return retval; | |
1945 | } | |
1946 | ||
1947 | /******************************************************************************* | |
1948 | * function: imsm_create_metadata_checkpoint_update | |
1949 | * Description: It creates update for checkpoint change. | |
1950 | * Parameters: | |
1951 | * super : imsm internal array info | |
1952 | * u : pointer to prepared update | |
1953 | * Returns: | |
1954 | * Uptate length. | |
1955 | * If length is equal to 0, input pointer u contains no update | |
1956 | ******************************************************************************/ | |
1957 | static int imsm_create_metadata_checkpoint_update( | |
1958 | struct intel_super *super, | |
1959 | struct imsm_update_general_migration_checkpoint **u) | |
1960 | { | |
1961 | ||
1962 | int update_memory_size = 0; | |
1963 | ||
1964 | dprintf("imsm_create_metadata_checkpoint_update(enter)\n"); | |
1965 | ||
1966 | if (u == NULL) | |
1967 | return 0; | |
1968 | *u = NULL; | |
1969 | ||
1970 | /* size of all update data without anchor */ | |
1971 | update_memory_size = | |
1972 | sizeof(struct imsm_update_general_migration_checkpoint); | |
1973 | ||
1974 | *u = calloc(1, update_memory_size); | |
1975 | if (*u == NULL) { | |
1976 | dprintf("error: cannot get memory for " | |
1977 | "imsm_create_metadata_checkpoint_update update\n"); | |
1978 | return 0; | |
1979 | } | |
1980 | (*u)->type = update_general_migration_checkpoint; | |
1981 | (*u)->curr_migr_unit = __le32_to_cpu(super->migr_rec->curr_migr_unit); | |
1982 | dprintf("imsm_create_metadata_checkpoint_update: prepared for %u\n", | |
1983 | (*u)->curr_migr_unit); | |
1984 | ||
1985 | return update_memory_size; | |
1986 | } | |
1987 | ||
1988 | ||
1989 | static void imsm_update_metadata_locally(struct supertype *st, | |
1990 | void *buf, int len); | |
1991 | ||
1992 | /******************************************************************************* | |
1993 | * Function: write_imsm_migr_rec | |
1994 | * Description: Function writes imsm migration record | |
1995 | * (at the last sector of disk) | |
1996 | * Parameters: | |
1997 | * super : imsm internal array info | |
1998 | * Returns: | |
1999 | * 0 : success | |
2000 | * -1 : if fail | |
2001 | ******************************************************************************/ | |
2002 | static int write_imsm_migr_rec(struct supertype *st) | |
2003 | { | |
2004 | struct intel_super *super = st->sb; | |
2005 | unsigned long long dsize; | |
2006 | char nm[30]; | |
2007 | int fd = -1; | |
2008 | int retval = -1; | |
2009 | struct dl *sd; | |
2010 | int len; | |
2011 | struct imsm_update_general_migration_checkpoint *u; | |
2012 | ||
2013 | for (sd = super->disks ; sd ; sd = sd->next) { | |
2014 | /* write to 2 first slots only */ | |
2015 | if ((sd->index < 0) || (sd->index > 1)) | |
2016 | continue; | |
2017 | sprintf(nm, "%d:%d", sd->major, sd->minor); | |
2018 | fd = dev_open(nm, O_RDWR); | |
2019 | if (fd < 0) | |
2020 | continue; | |
2021 | get_dev_size(fd, NULL, &dsize); | |
2022 | if (lseek64(fd, dsize - 512, SEEK_SET) < 0) { | |
2023 | fprintf(stderr, | |
2024 | Name ": Cannot seek to anchor block: %s\n", | |
2025 | strerror(errno)); | |
2026 | goto out; | |
2027 | } | |
2028 | if (write(fd, super->migr_rec_buf, 512) != 512) { | |
2029 | fprintf(stderr, | |
2030 | Name ": Cannot write migr record block: %s\n", | |
2031 | strerror(errno)); | |
2032 | goto out; | |
2033 | } | |
2034 | close(fd); | |
2035 | fd = -1; | |
2036 | } | |
2037 | /* update checkpoint information in metadata */ | |
2038 | len = imsm_create_metadata_checkpoint_update(super, &u); | |
2039 | ||
2040 | if (len <= 0) { | |
2041 | dprintf("imsm: Cannot prepare update\n"); | |
2042 | goto out; | |
2043 | } | |
2044 | /* update metadata locally */ | |
2045 | imsm_update_metadata_locally(st, u, len); | |
2046 | /* and possibly remotely */ | |
2047 | if (st->update_tail) { | |
2048 | append_metadata_update(st, u, len); | |
2049 | /* during reshape we do all work inside metadata handler | |
2050 | * manage_reshape(), so metadata update has to be triggered | |
2051 | * insida it | |
2052 | */ | |
2053 | flush_metadata_updates(st); | |
2054 | st->update_tail = &st->updates; | |
2055 | } else | |
2056 | free(u); | |
2057 | ||
2058 | retval = 0; | |
2059 | out: | |
2060 | if (fd >= 0) | |
2061 | close(fd); | |
2062 | return retval; | |
2063 | } | |
2064 | ||
2065 | static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info, char *dmap) | |
2066 | { | |
2067 | struct intel_super *super = st->sb; | |
2068 | struct migr_record *migr_rec = super->migr_rec; | |
2069 | struct imsm_dev *dev = get_imsm_dev(super, super->current_vol); | |
2070 | struct imsm_map *map = get_imsm_map(dev, 0); | |
2071 | struct imsm_map *prev_map = get_imsm_map(dev, 1); | |
2072 | struct imsm_map *map_to_analyse = map; | |
2073 | struct dl *dl; | |
2074 | char *devname; | |
2075 | unsigned int component_size_alligment; | |
2076 | int map_disks = info->array.raid_disks; | |
2077 | ||
2078 | memset(info, 0, sizeof(*info)); | |
2079 | if (prev_map) | |
2080 | map_to_analyse = prev_map; | |
2081 | ||
2082 | dl = super->disks; | |
2083 | ||
2084 | info->container_member = super->current_vol; | |
2085 | info->array.raid_disks = map->num_members; | |
2086 | info->array.level = get_imsm_raid_level(map_to_analyse); | |
2087 | info->array.layout = imsm_level_to_layout(info->array.level); | |
2088 | info->array.md_minor = -1; | |
2089 | info->array.ctime = 0; | |
2090 | info->array.utime = 0; | |
2091 | info->array.chunk_size = | |
2092 | __le16_to_cpu(map_to_analyse->blocks_per_strip) << 9; | |
2093 | info->array.state = !dev->vol.dirty; | |
2094 | info->custom_array_size = __le32_to_cpu(dev->size_high); | |
2095 | info->custom_array_size <<= 32; | |
2096 | info->custom_array_size |= __le32_to_cpu(dev->size_low); | |
2097 | if (prev_map && map->map_state == prev_map->map_state) { | |
2098 | info->reshape_active = 1; | |
2099 | info->new_level = get_imsm_raid_level(map); | |
2100 | info->new_layout = imsm_level_to_layout(info->new_level); | |
2101 | info->new_chunk = __le16_to_cpu(map->blocks_per_strip) << 9; | |
2102 | info->delta_disks = map->num_members - prev_map->num_members; | |
2103 | if (info->delta_disks) { | |
2104 | /* this needs to be applied to every array | |
2105 | * in the container. | |
2106 | */ | |
2107 | info->reshape_active = 2; | |
2108 | } | |
2109 | /* We shape information that we give to md might have to be | |
2110 | * modify to cope with md's requirement for reshaping arrays. | |
2111 | * For example, when reshaping a RAID0, md requires it to be | |
2112 | * presented as a degraded RAID4. | |
2113 | * Also if a RAID0 is migrating to a RAID5 we need to specify | |
2114 | * the array as already being RAID5, but the 'before' layout | |
2115 | * is a RAID4-like layout. | |
2116 | */ | |
2117 | switch (info->array.level) { | |
2118 | case 0: | |
2119 | switch(info->new_level) { | |
2120 | case 0: | |
2121 | /* conversion is happening as RAID4 */ | |
2122 | info->array.level = 4; | |
2123 | info->array.raid_disks += 1; | |
2124 | break; | |
2125 | case 5: | |
2126 | /* conversion is happening as RAID5 */ | |
2127 | info->array.level = 5; | |
2128 | info->array.layout = ALGORITHM_PARITY_N; | |
2129 | info->array.raid_disks += 1; | |
2130 | info->delta_disks -= 1; | |
2131 | break; | |
2132 | default: | |
2133 | /* FIXME error message */ | |
2134 | info->array.level = UnSet; | |
2135 | break; | |
2136 | } | |
2137 | break; | |
2138 | } | |
2139 | } else { | |
2140 | info->new_level = UnSet; | |
2141 | info->new_layout = UnSet; | |
2142 | info->new_chunk = info->array.chunk_size; | |
2143 | info->delta_disks = 0; | |
2144 | } | |
2145 | info->disk.major = 0; | |
2146 | info->disk.minor = 0; | |
2147 | if (dl) { | |
2148 | info->disk.major = dl->major; | |
2149 | info->disk.minor = dl->minor; | |
2150 | } | |
2151 | ||
2152 | info->data_offset = __le32_to_cpu(map_to_analyse->pba_of_lba0); | |
2153 | info->component_size = | |
2154 | __le32_to_cpu(map_to_analyse->blocks_per_member); | |
2155 | ||
2156 | /* check component size aligment | |
2157 | */ | |
2158 | component_size_alligment = | |
2159 | info->component_size % (info->array.chunk_size/512); | |
2160 | ||
2161 | if (component_size_alligment && | |
2162 | (info->array.level != 1) && (info->array.level != UnSet)) { | |
2163 | dprintf("imsm: reported component size alligned from %llu ", | |
2164 | info->component_size); | |
2165 | info->component_size -= component_size_alligment; | |
2166 | dprintf("to %llu (%i).\n", | |
2167 | info->component_size, component_size_alligment); | |
2168 | } | |
2169 | ||
2170 | memset(info->uuid, 0, sizeof(info->uuid)); | |
2171 | info->recovery_start = MaxSector; | |
2172 | ||
2173 | info->reshape_progress = 0; | |
2174 | info->resync_start = MaxSector; | |
2175 | if (map_to_analyse->map_state == IMSM_T_STATE_UNINITIALIZED || | |
2176 | dev->vol.dirty) { | |
2177 | info->resync_start = 0; | |
2178 | } | |
2179 | if (dev->vol.migr_state) { | |
2180 | switch (migr_type(dev)) { | |
2181 | case MIGR_REPAIR: | |
2182 | case MIGR_INIT: { | |
2183 | __u64 blocks_per_unit = blocks_per_migr_unit(super, | |
2184 | dev); | |
2185 | __u64 units = __le32_to_cpu(dev->vol.curr_migr_unit); | |
2186 | ||
2187 | info->resync_start = blocks_per_unit * units; | |
2188 | break; | |
2189 | } | |
2190 | case MIGR_GEN_MIGR: { | |
2191 | __u64 blocks_per_unit = blocks_per_migr_unit(super, | |
2192 | dev); | |
2193 | __u64 units = __le32_to_cpu(migr_rec->curr_migr_unit); | |
2194 | unsigned long long array_blocks; | |
2195 | int used_disks; | |
2196 | ||
2197 | info->reshape_progress = blocks_per_unit * units; | |
2198 | ||
2199 | dprintf("IMSM: General Migration checkpoint : %llu " | |
2200 | "(%llu) -> read reshape progress : %llu\n", | |
2201 | units, blocks_per_unit, info->reshape_progress); | |
2202 | ||
2203 | used_disks = imsm_num_data_members(dev, 1); | |
2204 | if (used_disks > 0) { | |
2205 | array_blocks = map->blocks_per_member * | |
2206 | used_disks; | |
2207 | /* round array size down to closest MB | |
2208 | */ | |
2209 | info->custom_array_size = (array_blocks | |
2210 | >> SECT_PER_MB_SHIFT) | |
2211 | << SECT_PER_MB_SHIFT; | |
2212 | } | |
2213 | } | |
2214 | case MIGR_VERIFY: | |
2215 | /* we could emulate the checkpointing of | |
2216 | * 'sync_action=check' migrations, but for now | |
2217 | * we just immediately complete them | |
2218 | */ | |
2219 | case MIGR_REBUILD: | |
2220 | /* this is handled by container_content_imsm() */ | |
2221 | case MIGR_STATE_CHANGE: | |
2222 | /* FIXME handle other migrations */ | |
2223 | default: | |
2224 | /* we are not dirty, so... */ | |
2225 | info->resync_start = MaxSector; | |
2226 | } | |
2227 | } | |
2228 | ||
2229 | strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN); | |
2230 | info->name[MAX_RAID_SERIAL_LEN] = 0; | |
2231 | ||
2232 | info->array.major_version = -1; | |
2233 | info->array.minor_version = -2; | |
2234 | devname = devnum2devname(st->container_dev); | |
2235 | *info->text_version = '\0'; | |
2236 | if (devname) | |
2237 | sprintf(info->text_version, "/%s/%d", devname, info->container_member); | |
2238 | free(devname); | |
2239 | info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */ | |
2240 | uuid_from_super_imsm(st, info->uuid); | |
2241 | ||
2242 | if (dmap) { | |
2243 | int i, j; | |
2244 | for (i=0; i<map_disks; i++) { | |
2245 | dmap[i] = 0; | |
2246 | if (i < info->array.raid_disks) { | |
2247 | struct imsm_disk *dsk; | |
2248 | j = get_imsm_disk_idx(dev, i, -1); | |
2249 | dsk = get_imsm_disk(super, j); | |
2250 | if (dsk && (dsk->status & CONFIGURED_DISK)) | |
2251 | dmap[i] = 1; | |
2252 | } | |
2253 | } | |
2254 | } | |
2255 | } | |
2256 | ||
2257 | static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed); | |
2258 | static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev); | |
2259 | ||
2260 | static struct imsm_disk *get_imsm_missing(struct intel_super *super, __u8 index) | |
2261 | { | |
2262 | struct dl *d; | |
2263 | ||
2264 | for (d = super->missing; d; d = d->next) | |
2265 | if (d->index == index) | |
2266 | return &d->disk; | |
2267 | return NULL; | |
2268 | } | |
2269 | ||
2270 | static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info, char *map) | |
2271 | { | |
2272 | struct intel_super *super = st->sb; | |
2273 | struct imsm_disk *disk; | |
2274 | int map_disks = info->array.raid_disks; | |
2275 | int max_enough = -1; | |
2276 | int i; | |
2277 | struct imsm_super *mpb; | |
2278 | ||
2279 | if (super->current_vol >= 0) { | |
2280 | getinfo_super_imsm_volume(st, info, map); | |
2281 | return; | |
2282 | } | |
2283 | memset(info, 0, sizeof(*info)); | |
2284 | ||
2285 | /* Set raid_disks to zero so that Assemble will always pull in valid | |
2286 | * spares | |
2287 | */ | |
2288 | info->array.raid_disks = 0; | |
2289 | info->array.level = LEVEL_CONTAINER; | |
2290 | info->array.layout = 0; | |
2291 | info->array.md_minor = -1; | |
2292 | info->array.ctime = 0; /* N/A for imsm */ | |
2293 | info->array.utime = 0; | |
2294 | info->array.chunk_size = 0; | |
2295 | ||
2296 | info->disk.major = 0; | |
2297 | info->disk.minor = 0; | |
2298 | info->disk.raid_disk = -1; | |
2299 | info->reshape_active = 0; | |
2300 | info->array.major_version = -1; | |
2301 | info->array.minor_version = -2; | |
2302 | strcpy(info->text_version, "imsm"); | |
2303 | info->safe_mode_delay = 0; | |
2304 | info->disk.number = -1; | |
2305 | info->disk.state = 0; | |
2306 | info->name[0] = 0; | |
2307 | info->recovery_start = MaxSector; | |
2308 | ||
2309 | /* do we have the all the insync disks that we expect? */ | |
2310 | mpb = super->anchor; | |
2311 | ||
2312 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
2313 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
2314 | int failed, enough, j, missing = 0; | |
2315 | struct imsm_map *map; | |
2316 | __u8 state; | |
2317 | ||
2318 | failed = imsm_count_failed(super, dev); | |
2319 | state = imsm_check_degraded(super, dev, failed); | |
2320 | map = get_imsm_map(dev, dev->vol.migr_state); | |
2321 | ||
2322 | /* any newly missing disks? | |
2323 | * (catches single-degraded vs double-degraded) | |
2324 | */ | |
2325 | for (j = 0; j < map->num_members; j++) { | |
2326 | __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1); | |
2327 | __u32 idx = ord_to_idx(ord); | |
2328 | ||
2329 | if (!(ord & IMSM_ORD_REBUILD) && | |
2330 | get_imsm_missing(super, idx)) { | |
2331 | missing = 1; | |
2332 | break; | |
2333 | } | |
2334 | } | |
2335 | ||
2336 | if (state == IMSM_T_STATE_FAILED) | |
2337 | enough = -1; | |
2338 | else if (state == IMSM_T_STATE_DEGRADED && | |
2339 | (state != map->map_state || missing)) | |
2340 | enough = 0; | |
2341 | else /* we're normal, or already degraded */ | |
2342 | enough = 1; | |
2343 | ||
2344 | /* in the missing/failed disk case check to see | |
2345 | * if at least one array is runnable | |
2346 | */ | |
2347 | max_enough = max(max_enough, enough); | |
2348 | } | |
2349 | dprintf("%s: enough: %d\n", __func__, max_enough); | |
2350 | info->container_enough = max_enough; | |
2351 | ||
2352 | if (super->disks) { | |
2353 | __u32 reserved = imsm_reserved_sectors(super, super->disks); | |
2354 | ||
2355 | disk = &super->disks->disk; | |
2356 | info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved; | |
2357 | info->component_size = reserved; | |
2358 | info->disk.state = is_configured(disk) ? (1 << MD_DISK_ACTIVE) : 0; | |
2359 | /* we don't change info->disk.raid_disk here because | |
2360 | * this state will be finalized in mdmon after we have | |
2361 | * found the 'most fresh' version of the metadata | |
2362 | */ | |
2363 | info->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0; | |
2364 | info->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC); | |
2365 | } | |
2366 | ||
2367 | /* only call uuid_from_super_imsm when this disk is part of a populated container, | |
2368 | * ->compare_super may have updated the 'num_raid_devs' field for spares | |
2369 | */ | |
2370 | if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs) | |
2371 | uuid_from_super_imsm(st, info->uuid); | |
2372 | else | |
2373 | memcpy(info->uuid, uuid_zero, sizeof(uuid_zero)); | |
2374 | ||
2375 | /* I don't know how to compute 'map' on imsm, so use safe default */ | |
2376 | if (map) { | |
2377 | int i; | |
2378 | for (i = 0; i < map_disks; i++) | |
2379 | map[i] = 1; | |
2380 | } | |
2381 | ||
2382 | } | |
2383 | ||
2384 | /* allocates memory and fills disk in mdinfo structure | |
2385 | * for each disk in array */ | |
2386 | struct mdinfo *getinfo_super_disks_imsm(struct supertype *st) | |
2387 | { | |
2388 | struct mdinfo *mddev = NULL; | |
2389 | struct intel_super *super = st->sb; | |
2390 | struct imsm_disk *disk; | |
2391 | int count = 0; | |
2392 | struct dl *dl; | |
2393 | if (!super || !super->disks) | |
2394 | return NULL; | |
2395 | dl = super->disks; | |
2396 | mddev = malloc(sizeof(*mddev)); | |
2397 | if (!mddev) { | |
2398 | fprintf(stderr, Name ": Failed to allocate memory.\n"); | |
2399 | return NULL; | |
2400 | } | |
2401 | memset(mddev, 0, sizeof(*mddev)); | |
2402 | while (dl) { | |
2403 | struct mdinfo *tmp; | |
2404 | disk = &dl->disk; | |
2405 | tmp = malloc(sizeof(*tmp)); | |
2406 | if (!tmp) { | |
2407 | fprintf(stderr, Name ": Failed to allocate memory.\n"); | |
2408 | if (mddev) | |
2409 | sysfs_free(mddev); | |
2410 | return NULL; | |
2411 | } | |
2412 | memset(tmp, 0, sizeof(*tmp)); | |
2413 | if (mddev->devs) | |
2414 | tmp->next = mddev->devs; | |
2415 | mddev->devs = tmp; | |
2416 | tmp->disk.number = count++; | |
2417 | tmp->disk.major = dl->major; | |
2418 | tmp->disk.minor = dl->minor; | |
2419 | tmp->disk.state = is_configured(disk) ? | |
2420 | (1 << MD_DISK_ACTIVE) : 0; | |
2421 | tmp->disk.state |= is_failed(disk) ? (1 << MD_DISK_FAULTY) : 0; | |
2422 | tmp->disk.state |= is_spare(disk) ? 0 : (1 << MD_DISK_SYNC); | |
2423 | tmp->disk.raid_disk = -1; | |
2424 | dl = dl->next; | |
2425 | } | |
2426 | return mddev; | |
2427 | } | |
2428 | ||
2429 | static int update_super_imsm(struct supertype *st, struct mdinfo *info, | |
2430 | char *update, char *devname, int verbose, | |
2431 | int uuid_set, char *homehost) | |
2432 | { | |
2433 | /* For 'assemble' and 'force' we need to return non-zero if any | |
2434 | * change was made. For others, the return value is ignored. | |
2435 | * Update options are: | |
2436 | * force-one : This device looks a bit old but needs to be included, | |
2437 | * update age info appropriately. | |
2438 | * assemble: clear any 'faulty' flag to allow this device to | |
2439 | * be assembled. | |
2440 | * force-array: Array is degraded but being forced, mark it clean | |
2441 | * if that will be needed to assemble it. | |
2442 | * | |
2443 | * newdev: not used ???? | |
2444 | * grow: Array has gained a new device - this is currently for | |
2445 | * linear only | |
2446 | * resync: mark as dirty so a resync will happen. | |
2447 | * name: update the name - preserving the homehost | |
2448 | * uuid: Change the uuid of the array to match watch is given | |
2449 | * | |
2450 | * Following are not relevant for this imsm: | |
2451 | * sparc2.2 : update from old dodgey metadata | |
2452 | * super-minor: change the preferred_minor number | |
2453 | * summaries: update redundant counters. | |
2454 | * homehost: update the recorded homehost | |
2455 | * _reshape_progress: record new reshape_progress position. | |
2456 | */ | |
2457 | int rv = 1; | |
2458 | struct intel_super *super = st->sb; | |
2459 | struct imsm_super *mpb; | |
2460 | ||
2461 | /* we can only update container info */ | |
2462 | if (!super || super->current_vol >= 0 || !super->anchor) | |
2463 | return 1; | |
2464 | ||
2465 | mpb = super->anchor; | |
2466 | ||
2467 | if (strcmp(update, "uuid") == 0 && uuid_set && !info->update_private) | |
2468 | rv = -1; | |
2469 | else if (strcmp(update, "uuid") == 0 && uuid_set && info->update_private) { | |
2470 | mpb->orig_family_num = *((__u32 *) info->update_private); | |
2471 | rv = 0; | |
2472 | } else if (strcmp(update, "uuid") == 0) { | |
2473 | __u32 *new_family = malloc(sizeof(*new_family)); | |
2474 | ||
2475 | /* update orig_family_number with the incoming random | |
2476 | * data, report the new effective uuid, and store the | |
2477 | * new orig_family_num for future updates. | |
2478 | */ | |
2479 | if (new_family) { | |
2480 | memcpy(&mpb->orig_family_num, info->uuid, sizeof(__u32)); | |
2481 | uuid_from_super_imsm(st, info->uuid); | |
2482 | *new_family = mpb->orig_family_num; | |
2483 | info->update_private = new_family; | |
2484 | rv = 0; | |
2485 | } | |
2486 | } else if (strcmp(update, "assemble") == 0) | |
2487 | rv = 0; | |
2488 | else | |
2489 | rv = -1; | |
2490 | ||
2491 | /* successful update? recompute checksum */ | |
2492 | if (rv == 0) | |
2493 | mpb->check_sum = __le32_to_cpu(__gen_imsm_checksum(mpb)); | |
2494 | ||
2495 | return rv; | |
2496 | } | |
2497 | ||
2498 | static size_t disks_to_mpb_size(int disks) | |
2499 | { | |
2500 | size_t size; | |
2501 | ||
2502 | size = sizeof(struct imsm_super); | |
2503 | size += (disks - 1) * sizeof(struct imsm_disk); | |
2504 | size += 2 * sizeof(struct imsm_dev); | |
2505 | /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */ | |
2506 | size += (4 - 2) * sizeof(struct imsm_map); | |
2507 | /* 4 possible disk_ord_tbl's */ | |
2508 | size += 4 * (disks - 1) * sizeof(__u32); | |
2509 | ||
2510 | return size; | |
2511 | } | |
2512 | ||
2513 | static __u64 avail_size_imsm(struct supertype *st, __u64 devsize) | |
2514 | { | |
2515 | if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS)) | |
2516 | return 0; | |
2517 | ||
2518 | return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS); | |
2519 | } | |
2520 | ||
2521 | static void free_devlist(struct intel_super *super) | |
2522 | { | |
2523 | struct intel_dev *dv; | |
2524 | ||
2525 | while (super->devlist) { | |
2526 | dv = super->devlist->next; | |
2527 | free(super->devlist->dev); | |
2528 | free(super->devlist); | |
2529 | super->devlist = dv; | |
2530 | } | |
2531 | } | |
2532 | ||
2533 | static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src) | |
2534 | { | |
2535 | memcpy(dest, src, sizeof_imsm_dev(src, 0)); | |
2536 | } | |
2537 | ||
2538 | static int compare_super_imsm(struct supertype *st, struct supertype *tst) | |
2539 | { | |
2540 | /* | |
2541 | * return: | |
2542 | * 0 same, or first was empty, and second was copied | |
2543 | * 1 second had wrong number | |
2544 | * 2 wrong uuid | |
2545 | * 3 wrong other info | |
2546 | */ | |
2547 | struct intel_super *first = st->sb; | |
2548 | struct intel_super *sec = tst->sb; | |
2549 | ||
2550 | if (!first) { | |
2551 | st->sb = tst->sb; | |
2552 | tst->sb = NULL; | |
2553 | return 0; | |
2554 | } | |
2555 | /* in platform dependent environment test if the disks | |
2556 | * use the same Intel hba | |
2557 | */ | |
2558 | if (!check_env("IMSM_NO_PLATFORM")) { | |
2559 | if (!first->hba || !sec->hba || | |
2560 | (first->hba->type != sec->hba->type)) { | |
2561 | fprintf(stderr, | |
2562 | "HBAs of devices does not match %s != %s\n", | |
2563 | first->hba ? get_sys_dev_type(first->hba->type) : NULL, | |
2564 | sec->hba ? get_sys_dev_type(sec->hba->type) : NULL); | |
2565 | return 3; | |
2566 | } | |
2567 | } | |
2568 | ||
2569 | /* if an anchor does not have num_raid_devs set then it is a free | |
2570 | * floating spare | |
2571 | */ | |
2572 | if (first->anchor->num_raid_devs > 0 && | |
2573 | sec->anchor->num_raid_devs > 0) { | |
2574 | /* Determine if these disks might ever have been | |
2575 | * related. Further disambiguation can only take place | |
2576 | * in load_super_imsm_all | |
2577 | */ | |
2578 | __u32 first_family = first->anchor->orig_family_num; | |
2579 | __u32 sec_family = sec->anchor->orig_family_num; | |
2580 | ||
2581 | if (memcmp(first->anchor->sig, sec->anchor->sig, | |
2582 | MAX_SIGNATURE_LENGTH) != 0) | |
2583 | return 3; | |
2584 | ||
2585 | if (first_family == 0) | |
2586 | first_family = first->anchor->family_num; | |
2587 | if (sec_family == 0) | |
2588 | sec_family = sec->anchor->family_num; | |
2589 | ||
2590 | if (first_family != sec_family) | |
2591 | return 3; | |
2592 | ||
2593 | } | |
2594 | ||
2595 | ||
2596 | /* if 'first' is a spare promote it to a populated mpb with sec's | |
2597 | * family number | |
2598 | */ | |
2599 | if (first->anchor->num_raid_devs == 0 && | |
2600 | sec->anchor->num_raid_devs > 0) { | |
2601 | int i; | |
2602 | struct intel_dev *dv; | |
2603 | struct imsm_dev *dev; | |
2604 | ||
2605 | /* we need to copy raid device info from sec if an allocation | |
2606 | * fails here we don't associate the spare | |
2607 | */ | |
2608 | for (i = 0; i < sec->anchor->num_raid_devs; i++) { | |
2609 | dv = malloc(sizeof(*dv)); | |
2610 | if (!dv) | |
2611 | break; | |
2612 | dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1)); | |
2613 | if (!dev) { | |
2614 | free(dv); | |
2615 | break; | |
2616 | } | |
2617 | dv->dev = dev; | |
2618 | dv->index = i; | |
2619 | dv->next = first->devlist; | |
2620 | first->devlist = dv; | |
2621 | } | |
2622 | if (i < sec->anchor->num_raid_devs) { | |
2623 | /* allocation failure */ | |
2624 | free_devlist(first); | |
2625 | fprintf(stderr, "imsm: failed to associate spare\n"); | |
2626 | return 3; | |
2627 | } | |
2628 | first->anchor->num_raid_devs = sec->anchor->num_raid_devs; | |
2629 | first->anchor->orig_family_num = sec->anchor->orig_family_num; | |
2630 | first->anchor->family_num = sec->anchor->family_num; | |
2631 | memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH); | |
2632 | for (i = 0; i < sec->anchor->num_raid_devs; i++) | |
2633 | imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i)); | |
2634 | } | |
2635 | ||
2636 | return 0; | |
2637 | } | |
2638 | ||
2639 | static void fd2devname(int fd, char *name) | |
2640 | { | |
2641 | struct stat st; | |
2642 | char path[256]; | |
2643 | char dname[PATH_MAX]; | |
2644 | char *nm; | |
2645 | int rv; | |
2646 | ||
2647 | name[0] = '\0'; | |
2648 | if (fstat(fd, &st) != 0) | |
2649 | return; | |
2650 | sprintf(path, "/sys/dev/block/%d:%d", | |
2651 | major(st.st_rdev), minor(st.st_rdev)); | |
2652 | ||
2653 | rv = readlink(path, dname, sizeof(dname)); | |
2654 | if (rv <= 0) | |
2655 | return; | |
2656 | ||
2657 | dname[rv] = '\0'; | |
2658 | nm = strrchr(dname, '/'); | |
2659 | nm++; | |
2660 | snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm); | |
2661 | } | |
2662 | ||
2663 | extern int scsi_get_serial(int fd, void *buf, size_t buf_len); | |
2664 | ||
2665 | static int imsm_read_serial(int fd, char *devname, | |
2666 | __u8 serial[MAX_RAID_SERIAL_LEN]) | |
2667 | { | |
2668 | unsigned char scsi_serial[255]; | |
2669 | int rv; | |
2670 | int rsp_len; | |
2671 | int len; | |
2672 | char *dest; | |
2673 | char *src; | |
2674 | char *rsp_buf; | |
2675 | int i; | |
2676 | ||
2677 | memset(scsi_serial, 0, sizeof(scsi_serial)); | |
2678 | ||
2679 | rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial)); | |
2680 | ||
2681 | if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) { | |
2682 | memset(serial, 0, MAX_RAID_SERIAL_LEN); | |
2683 | fd2devname(fd, (char *) serial); | |
2684 | return 0; | |
2685 | } | |
2686 | ||
2687 | if (rv != 0) { | |
2688 | if (devname) | |
2689 | fprintf(stderr, | |
2690 | Name ": Failed to retrieve serial for %s\n", | |
2691 | devname); | |
2692 | return rv; | |
2693 | } | |
2694 | ||
2695 | rsp_len = scsi_serial[3]; | |
2696 | if (!rsp_len) { | |
2697 | if (devname) | |
2698 | fprintf(stderr, | |
2699 | Name ": Failed to retrieve serial for %s\n", | |
2700 | devname); | |
2701 | return 2; | |
2702 | } | |
2703 | rsp_buf = (char *) &scsi_serial[4]; | |
2704 | ||
2705 | /* trim all whitespace and non-printable characters and convert | |
2706 | * ':' to ';' | |
2707 | */ | |
2708 | for (i = 0, dest = rsp_buf; i < rsp_len; i++) { | |
2709 | src = &rsp_buf[i]; | |
2710 | if (*src > 0x20) { | |
2711 | /* ':' is reserved for use in placeholder serial | |
2712 | * numbers for missing disks | |
2713 | */ | |
2714 | if (*src == ':') | |
2715 | *dest++ = ';'; | |
2716 | else | |
2717 | *dest++ = *src; | |
2718 | } | |
2719 | } | |
2720 | len = dest - rsp_buf; | |
2721 | dest = rsp_buf; | |
2722 | ||
2723 | /* truncate leading characters */ | |
2724 | if (len > MAX_RAID_SERIAL_LEN) { | |
2725 | dest += len - MAX_RAID_SERIAL_LEN; | |
2726 | len = MAX_RAID_SERIAL_LEN; | |
2727 | } | |
2728 | ||
2729 | memset(serial, 0, MAX_RAID_SERIAL_LEN); | |
2730 | memcpy(serial, dest, len); | |
2731 | ||
2732 | return 0; | |
2733 | } | |
2734 | ||
2735 | static int serialcmp(__u8 *s1, __u8 *s2) | |
2736 | { | |
2737 | return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN); | |
2738 | } | |
2739 | ||
2740 | static void serialcpy(__u8 *dest, __u8 *src) | |
2741 | { | |
2742 | strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN); | |
2743 | } | |
2744 | ||
2745 | #ifndef MDASSEMBLE | |
2746 | static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super) | |
2747 | { | |
2748 | struct dl *dl; | |
2749 | ||
2750 | for (dl = super->disks; dl; dl = dl->next) | |
2751 | if (serialcmp(dl->serial, serial) == 0) | |
2752 | break; | |
2753 | ||
2754 | return dl; | |
2755 | } | |
2756 | #endif | |
2757 | ||
2758 | static struct imsm_disk * | |
2759 | __serial_to_disk(__u8 *serial, struct imsm_super *mpb, int *idx) | |
2760 | { | |
2761 | int i; | |
2762 | ||
2763 | for (i = 0; i < mpb->num_disks; i++) { | |
2764 | struct imsm_disk *disk = __get_imsm_disk(mpb, i); | |
2765 | ||
2766 | if (serialcmp(disk->serial, serial) == 0) { | |
2767 | if (idx) | |
2768 | *idx = i; | |
2769 | return disk; | |
2770 | } | |
2771 | } | |
2772 | ||
2773 | return NULL; | |
2774 | } | |
2775 | ||
2776 | static int | |
2777 | load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd) | |
2778 | { | |
2779 | struct imsm_disk *disk; | |
2780 | struct dl *dl; | |
2781 | struct stat stb; | |
2782 | int rv; | |
2783 | char name[40]; | |
2784 | __u8 serial[MAX_RAID_SERIAL_LEN]; | |
2785 | ||
2786 | rv = imsm_read_serial(fd, devname, serial); | |
2787 | ||
2788 | if (rv != 0) | |
2789 | return 2; | |
2790 | ||
2791 | dl = calloc(1, sizeof(*dl)); | |
2792 | if (!dl) { | |
2793 | if (devname) | |
2794 | fprintf(stderr, | |
2795 | Name ": failed to allocate disk buffer for %s\n", | |
2796 | devname); | |
2797 | return 2; | |
2798 | } | |
2799 | ||
2800 | fstat(fd, &stb); | |
2801 | dl->major = major(stb.st_rdev); | |
2802 | dl->minor = minor(stb.st_rdev); | |
2803 | dl->next = super->disks; | |
2804 | dl->fd = keep_fd ? fd : -1; | |
2805 | assert(super->disks == NULL); | |
2806 | super->disks = dl; | |
2807 | serialcpy(dl->serial, serial); | |
2808 | dl->index = -2; | |
2809 | dl->e = NULL; | |
2810 | fd2devname(fd, name); | |
2811 | if (devname) | |
2812 | dl->devname = strdup(devname); | |
2813 | else | |
2814 | dl->devname = strdup(name); | |
2815 | ||
2816 | /* look up this disk's index in the current anchor */ | |
2817 | disk = __serial_to_disk(dl->serial, super->anchor, &dl->index); | |
2818 | if (disk) { | |
2819 | dl->disk = *disk; | |
2820 | /* only set index on disks that are a member of a | |
2821 | * populated contianer, i.e. one with raid_devs | |
2822 | */ | |
2823 | if (is_failed(&dl->disk)) | |
2824 | dl->index = -2; | |
2825 | else if (is_spare(&dl->disk)) | |
2826 | dl->index = -1; | |
2827 | } | |
2828 | ||
2829 | return 0; | |
2830 | } | |
2831 | ||
2832 | #ifndef MDASSEMBLE | |
2833 | /* When migrating map0 contains the 'destination' state while map1 | |
2834 | * contains the current state. When not migrating map0 contains the | |
2835 | * current state. This routine assumes that map[0].map_state is set to | |
2836 | * the current array state before being called. | |
2837 | * | |
2838 | * Migration is indicated by one of the following states | |
2839 | * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed) | |
2840 | * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal | |
2841 | * map1state=unitialized) | |
2842 | * 3/ Repair (Resync) (migr_state=1 migr_type=MIGR_REPAIR map0state=normal | |
2843 | * map1state=normal) | |
2844 | * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal | |
2845 | * map1state=degraded) | |
2846 | * 5/ Migration (mig_state=1 migr_type=MIGR_GEN_MIGR map0state=normal | |
2847 | * map1state=normal) | |
2848 | */ | |
2849 | static void migrate(struct imsm_dev *dev, struct intel_super *super, | |
2850 | __u8 to_state, int migr_type) | |
2851 | { | |
2852 | struct imsm_map *dest; | |
2853 | struct imsm_map *src = get_imsm_map(dev, 0); | |
2854 | ||
2855 | dev->vol.migr_state = 1; | |
2856 | set_migr_type(dev, migr_type); | |
2857 | dev->vol.curr_migr_unit = 0; | |
2858 | dest = get_imsm_map(dev, 1); | |
2859 | ||
2860 | /* duplicate and then set the target end state in map[0] */ | |
2861 | memcpy(dest, src, sizeof_imsm_map(src)); | |
2862 | if ((migr_type == MIGR_REBUILD) || | |
2863 | (migr_type == MIGR_GEN_MIGR)) { | |
2864 | __u32 ord; | |
2865 | int i; | |
2866 | ||
2867 | for (i = 0; i < src->num_members; i++) { | |
2868 | ord = __le32_to_cpu(src->disk_ord_tbl[i]); | |
2869 | set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord)); | |
2870 | } | |
2871 | } | |
2872 | ||
2873 | if (migr_type == MIGR_GEN_MIGR) | |
2874 | /* Clear migration record */ | |
2875 | memset(super->migr_rec, 0, sizeof(struct migr_record)); | |
2876 | ||
2877 | src->map_state = to_state; | |
2878 | } | |
2879 | ||
2880 | static void end_migration(struct imsm_dev *dev, __u8 map_state) | |
2881 | { | |
2882 | struct imsm_map *map = get_imsm_map(dev, 0); | |
2883 | struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state); | |
2884 | int i, j; | |
2885 | ||
2886 | /* merge any IMSM_ORD_REBUILD bits that were not successfully | |
2887 | * completed in the last migration. | |
2888 | * | |
2889 | * FIXME add support for raid-level-migration | |
2890 | */ | |
2891 | for (i = 0; i < prev->num_members; i++) | |
2892 | for (j = 0; j < map->num_members; j++) | |
2893 | /* during online capacity expansion | |
2894 | * disks position can be changed if takeover is used | |
2895 | */ | |
2896 | if (ord_to_idx(map->disk_ord_tbl[j]) == | |
2897 | ord_to_idx(prev->disk_ord_tbl[i])) { | |
2898 | map->disk_ord_tbl[j] |= prev->disk_ord_tbl[i]; | |
2899 | break; | |
2900 | } | |
2901 | ||
2902 | dev->vol.migr_state = 0; | |
2903 | dev->vol.migr_type = 0; | |
2904 | dev->vol.curr_migr_unit = 0; | |
2905 | map->map_state = map_state; | |
2906 | } | |
2907 | #endif | |
2908 | ||
2909 | static int parse_raid_devices(struct intel_super *super) | |
2910 | { | |
2911 | int i; | |
2912 | struct imsm_dev *dev_new; | |
2913 | size_t len, len_migr; | |
2914 | size_t max_len = 0; | |
2915 | size_t space_needed = 0; | |
2916 | struct imsm_super *mpb = super->anchor; | |
2917 | ||
2918 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
2919 | struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i); | |
2920 | struct intel_dev *dv; | |
2921 | ||
2922 | len = sizeof_imsm_dev(dev_iter, 0); | |
2923 | len_migr = sizeof_imsm_dev(dev_iter, 1); | |
2924 | if (len_migr > len) | |
2925 | space_needed += len_migr - len; | |
2926 | ||
2927 | dv = malloc(sizeof(*dv)); | |
2928 | if (!dv) | |
2929 | return 1; | |
2930 | if (max_len < len_migr) | |
2931 | max_len = len_migr; | |
2932 | if (max_len > len_migr) | |
2933 | space_needed += max_len - len_migr; | |
2934 | dev_new = malloc(max_len); | |
2935 | if (!dev_new) { | |
2936 | free(dv); | |
2937 | return 1; | |
2938 | } | |
2939 | imsm_copy_dev(dev_new, dev_iter); | |
2940 | dv->dev = dev_new; | |
2941 | dv->index = i; | |
2942 | dv->next = super->devlist; | |
2943 | super->devlist = dv; | |
2944 | } | |
2945 | ||
2946 | /* ensure that super->buf is large enough when all raid devices | |
2947 | * are migrating | |
2948 | */ | |
2949 | if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) { | |
2950 | void *buf; | |
2951 | ||
2952 | len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512); | |
2953 | if (posix_memalign(&buf, 512, len) != 0) | |
2954 | return 1; | |
2955 | ||
2956 | memcpy(buf, super->buf, super->len); | |
2957 | memset(buf + super->len, 0, len - super->len); | |
2958 | free(super->buf); | |
2959 | super->buf = buf; | |
2960 | super->len = len; | |
2961 | } | |
2962 | ||
2963 | return 0; | |
2964 | } | |
2965 | ||
2966 | /* retrieve a pointer to the bbm log which starts after all raid devices */ | |
2967 | struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb) | |
2968 | { | |
2969 | void *ptr = NULL; | |
2970 | ||
2971 | if (__le32_to_cpu(mpb->bbm_log_size)) { | |
2972 | ptr = mpb; | |
2973 | ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size); | |
2974 | } | |
2975 | ||
2976 | return ptr; | |
2977 | } | |
2978 | ||
2979 | /******************************************************************************* | |
2980 | * Function: check_mpb_migr_compatibility | |
2981 | * Description: Function checks for unsupported migration features: | |
2982 | * - migration optimization area (pba_of_lba0) | |
2983 | * - descending reshape (ascending_migr) | |
2984 | * Parameters: | |
2985 | * super : imsm metadata information | |
2986 | * Returns: | |
2987 | * 0 : migration is compatible | |
2988 | * -1 : migration is not compatible | |
2989 | ******************************************************************************/ | |
2990 | int check_mpb_migr_compatibility(struct intel_super *super) | |
2991 | { | |
2992 | struct imsm_map *map0, *map1; | |
2993 | struct migr_record *migr_rec = super->migr_rec; | |
2994 | int i; | |
2995 | ||
2996 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
2997 | struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i); | |
2998 | ||
2999 | if (dev_iter && | |
3000 | dev_iter->vol.migr_state == 1 && | |
3001 | dev_iter->vol.migr_type == MIGR_GEN_MIGR) { | |
3002 | /* This device is migrating */ | |
3003 | map0 = get_imsm_map(dev_iter, 0); | |
3004 | map1 = get_imsm_map(dev_iter, 1); | |
3005 | if (map0->pba_of_lba0 != map1->pba_of_lba0) | |
3006 | /* migration optimization area was used */ | |
3007 | return -1; | |
3008 | if (migr_rec->ascending_migr == 0 | |
3009 | && migr_rec->dest_depth_per_unit > 0) | |
3010 | /* descending reshape not supported yet */ | |
3011 | return -1; | |
3012 | } | |
3013 | } | |
3014 | return 0; | |
3015 | } | |
3016 | ||
3017 | static void __free_imsm(struct intel_super *super, int free_disks); | |
3018 | ||
3019 | /* load_imsm_mpb - read matrix metadata | |
3020 | * allocates super->mpb to be freed by free_imsm | |
3021 | */ | |
3022 | static int load_imsm_mpb(int fd, struct intel_super *super, char *devname) | |
3023 | { | |
3024 | unsigned long long dsize; | |
3025 | unsigned long long sectors; | |
3026 | struct stat; | |
3027 | struct imsm_super *anchor; | |
3028 | __u32 check_sum; | |
3029 | ||
3030 | get_dev_size(fd, NULL, &dsize); | |
3031 | if (dsize < 1024) { | |
3032 | if (devname) | |
3033 | fprintf(stderr, | |
3034 | Name ": %s: device to small for imsm\n", | |
3035 | devname); | |
3036 | return 1; | |
3037 | } | |
3038 | ||
3039 | if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) { | |
3040 | if (devname) | |
3041 | fprintf(stderr, Name | |
3042 | ": Cannot seek to anchor block on %s: %s\n", | |
3043 | devname, strerror(errno)); | |
3044 | return 1; | |
3045 | } | |
3046 | ||
3047 | if (posix_memalign((void**)&anchor, 512, 512) != 0) { | |
3048 | if (devname) | |
3049 | fprintf(stderr, | |
3050 | Name ": Failed to allocate imsm anchor buffer" | |
3051 | " on %s\n", devname); | |
3052 | return 1; | |
3053 | } | |
3054 | if (read(fd, anchor, 512) != 512) { | |
3055 | if (devname) | |
3056 | fprintf(stderr, | |
3057 | Name ": Cannot read anchor block on %s: %s\n", | |
3058 | devname, strerror(errno)); | |
3059 | free(anchor); | |
3060 | return 1; | |
3061 | } | |
3062 | ||
3063 | if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) { | |
3064 | if (devname) | |
3065 | fprintf(stderr, | |
3066 | Name ": no IMSM anchor on %s\n", devname); | |
3067 | free(anchor); | |
3068 | return 2; | |
3069 | } | |
3070 | ||
3071 | __free_imsm(super, 0); | |
3072 | /* reload capability and hba */ | |
3073 | ||
3074 | /* capability and hba must be updated with new super allocation */ | |
3075 | find_intel_hba_capability(fd, super, devname); | |
3076 | super->len = ROUND_UP(anchor->mpb_size, 512); | |
3077 | if (posix_memalign(&super->buf, 512, super->len) != 0) { | |
3078 | if (devname) | |
3079 | fprintf(stderr, | |
3080 | Name ": unable to allocate %zu byte mpb buffer\n", | |
3081 | super->len); | |
3082 | free(anchor); | |
3083 | return 2; | |
3084 | } | |
3085 | memcpy(super->buf, anchor, 512); | |
3086 | ||
3087 | sectors = mpb_sectors(anchor) - 1; | |
3088 | free(anchor); | |
3089 | ||
3090 | if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) { | |
3091 | fprintf(stderr, Name | |
3092 | ": %s could not allocate migr_rec buffer\n", __func__); | |
3093 | free(super->buf); | |
3094 | return 2; | |
3095 | } | |
3096 | ||
3097 | if (!sectors) { | |
3098 | check_sum = __gen_imsm_checksum(super->anchor); | |
3099 | if (check_sum != __le32_to_cpu(super->anchor->check_sum)) { | |
3100 | if (devname) | |
3101 | fprintf(stderr, | |
3102 | Name ": IMSM checksum %x != %x on %s\n", | |
3103 | check_sum, | |
3104 | __le32_to_cpu(super->anchor->check_sum), | |
3105 | devname); | |
3106 | return 2; | |
3107 | } | |
3108 | ||
3109 | return 0; | |
3110 | } | |
3111 | ||
3112 | /* read the extended mpb */ | |
3113 | if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) { | |
3114 | if (devname) | |
3115 | fprintf(stderr, | |
3116 | Name ": Cannot seek to extended mpb on %s: %s\n", | |
3117 | devname, strerror(errno)); | |
3118 | return 1; | |
3119 | } | |
3120 | ||
3121 | if ((unsigned)read(fd, super->buf + 512, super->len - 512) != super->len - 512) { | |
3122 | if (devname) | |
3123 | fprintf(stderr, | |
3124 | Name ": Cannot read extended mpb on %s: %s\n", | |
3125 | devname, strerror(errno)); | |
3126 | return 2; | |
3127 | } | |
3128 | ||
3129 | check_sum = __gen_imsm_checksum(super->anchor); | |
3130 | if (check_sum != __le32_to_cpu(super->anchor->check_sum)) { | |
3131 | if (devname) | |
3132 | fprintf(stderr, | |
3133 | Name ": IMSM checksum %x != %x on %s\n", | |
3134 | check_sum, __le32_to_cpu(super->anchor->check_sum), | |
3135 | devname); | |
3136 | return 3; | |
3137 | } | |
3138 | ||
3139 | /* FIXME the BBM log is disk specific so we cannot use this global | |
3140 | * buffer for all disks. Ok for now since we only look at the global | |
3141 | * bbm_log_size parameter to gate assembly | |
3142 | */ | |
3143 | super->bbm_log = __get_imsm_bbm_log(super->anchor); | |
3144 | ||
3145 | return 0; | |
3146 | } | |
3147 | ||
3148 | static int read_imsm_migr_rec(int fd, struct intel_super *super); | |
3149 | ||
3150 | static int | |
3151 | load_and_parse_mpb(int fd, struct intel_super *super, char *devname, int keep_fd) | |
3152 | { | |
3153 | int err; | |
3154 | ||
3155 | err = load_imsm_mpb(fd, super, devname); | |
3156 | if (err) | |
3157 | return err; | |
3158 | err = load_imsm_disk(fd, super, devname, keep_fd); | |
3159 | if (err) | |
3160 | return err; | |
3161 | err = parse_raid_devices(super); | |
3162 | ||
3163 | return err; | |
3164 | } | |
3165 | ||
3166 | static void __free_imsm_disk(struct dl *d) | |
3167 | { | |
3168 | if (d->fd >= 0) | |
3169 | close(d->fd); | |
3170 | if (d->devname) | |
3171 | free(d->devname); | |
3172 | if (d->e) | |
3173 | free(d->e); | |
3174 | free(d); | |
3175 | ||
3176 | } | |
3177 | ||
3178 | static void free_imsm_disks(struct intel_super *super) | |
3179 | { | |
3180 | struct dl *d; | |
3181 | ||
3182 | while (super->disks) { | |
3183 | d = super->disks; | |
3184 | super->disks = d->next; | |
3185 | __free_imsm_disk(d); | |
3186 | } | |
3187 | while (super->disk_mgmt_list) { | |
3188 | d = super->disk_mgmt_list; | |
3189 | super->disk_mgmt_list = d->next; | |
3190 | __free_imsm_disk(d); | |
3191 | } | |
3192 | while (super->missing) { | |
3193 | d = super->missing; | |
3194 | super->missing = d->next; | |
3195 | __free_imsm_disk(d); | |
3196 | } | |
3197 | ||
3198 | } | |
3199 | ||
3200 | /* free all the pieces hanging off of a super pointer */ | |
3201 | static void __free_imsm(struct intel_super *super, int free_disks) | |
3202 | { | |
3203 | struct intel_hba *elem, *next; | |
3204 | ||
3205 | if (super->buf) { | |
3206 | free(super->buf); | |
3207 | super->buf = NULL; | |
3208 | } | |
3209 | /* unlink capability description */ | |
3210 | super->orom = NULL; | |
3211 | if (super->migr_rec_buf) { | |
3212 | free(super->migr_rec_buf); | |
3213 | super->migr_rec_buf = NULL; | |
3214 | } | |
3215 | if (free_disks) | |
3216 | free_imsm_disks(super); | |
3217 | free_devlist(super); | |
3218 | elem = super->hba; | |
3219 | while (elem) { | |
3220 | if (elem->path) | |
3221 | free((void *)elem->path); | |
3222 | next = elem->next; | |
3223 | free(elem); | |
3224 | elem = next; | |
3225 | } | |
3226 | super->hba = NULL; | |
3227 | } | |
3228 | ||
3229 | static void free_imsm(struct intel_super *super) | |
3230 | { | |
3231 | __free_imsm(super, 1); | |
3232 | free(super); | |
3233 | } | |
3234 | ||
3235 | static void free_super_imsm(struct supertype *st) | |
3236 | { | |
3237 | struct intel_super *super = st->sb; | |
3238 | ||
3239 | if (!super) | |
3240 | return; | |
3241 | ||
3242 | free_imsm(super); | |
3243 | st->sb = NULL; | |
3244 | } | |
3245 | ||
3246 | static struct intel_super *alloc_super(void) | |
3247 | { | |
3248 | struct intel_super *super = malloc(sizeof(*super)); | |
3249 | ||
3250 | if (super) { | |
3251 | memset(super, 0, sizeof(*super)); | |
3252 | super->current_vol = -1; | |
3253 | super->create_offset = ~((__u32 ) 0); | |
3254 | } | |
3255 | return super; | |
3256 | } | |
3257 | ||
3258 | /* | |
3259 | * find and allocate hba and OROM/EFI based on valid fd of RAID component device | |
3260 | */ | |
3261 | static int find_intel_hba_capability(int fd, struct intel_super *super, char *devname) | |
3262 | { | |
3263 | struct sys_dev *hba_name; | |
3264 | int rv = 0; | |
3265 | ||
3266 | if ((fd < 0) || check_env("IMSM_NO_PLATFORM")) { | |
3267 | super->orom = NULL; | |
3268 | super->hba = NULL; | |
3269 | return 0; | |
3270 | } | |
3271 | hba_name = find_disk_attached_hba(fd, NULL); | |
3272 | if (!hba_name) { | |
3273 | if (devname) | |
3274 | fprintf(stderr, | |
3275 | Name ": %s is not attached to Intel(R) RAID controller.\n", | |
3276 | devname); | |
3277 | return 1; | |
3278 | } | |
3279 | rv = attach_hba_to_super(super, hba_name); | |
3280 | if (rv == 2) { | |
3281 | if (devname) { | |
3282 | struct intel_hba *hba = super->hba; | |
3283 | ||
3284 | fprintf(stderr, Name ": %s is attached to Intel(R) %s RAID " | |
3285 | "controller (%s),\n" | |
3286 | " but the container is assigned to Intel(R) " | |
3287 | "%s RAID controller (", | |
3288 | devname, | |
3289 | hba_name->path, | |
3290 | hba_name->pci_id ? : "Err!", | |
3291 | get_sys_dev_type(hba_name->type)); | |
3292 | ||
3293 | while (hba) { | |
3294 | fprintf(stderr, "%s", hba->pci_id ? : "Err!"); | |
3295 | if (hba->next) | |
3296 | fprintf(stderr, ", "); | |
3297 | hba = hba->next; | |
3298 | } | |
3299 | ||
3300 | fprintf(stderr, ").\n" | |
3301 | " Mixing devices attached to different controllers " | |
3302 | "is not allowed.\n"); | |
3303 | } | |
3304 | free_sys_dev(&hba_name); | |
3305 | return 2; | |
3306 | } | |
3307 | super->orom = find_imsm_capability(hba_name->type); | |
3308 | free_sys_dev(&hba_name); | |
3309 | if (!super->orom) | |
3310 | return 3; | |
3311 | return 0; | |
3312 | } | |
3313 | ||
3314 | #ifndef MDASSEMBLE | |
3315 | /* find_missing - helper routine for load_super_imsm_all that identifies | |
3316 | * disks that have disappeared from the system. This routine relies on | |
3317 | * the mpb being uptodate, which it is at load time. | |
3318 | */ | |
3319 | static int find_missing(struct intel_super *super) | |
3320 | { | |
3321 | int i; | |
3322 | struct imsm_super *mpb = super->anchor; | |
3323 | struct dl *dl; | |
3324 | struct imsm_disk *disk; | |
3325 | ||
3326 | for (i = 0; i < mpb->num_disks; i++) { | |
3327 | disk = __get_imsm_disk(mpb, i); | |
3328 | dl = serial_to_dl(disk->serial, super); | |
3329 | if (dl) | |
3330 | continue; | |
3331 | ||
3332 | dl = malloc(sizeof(*dl)); | |
3333 | if (!dl) | |
3334 | return 1; | |
3335 | dl->major = 0; | |
3336 | dl->minor = 0; | |
3337 | dl->fd = -1; | |
3338 | dl->devname = strdup("missing"); | |
3339 | dl->index = i; | |
3340 | serialcpy(dl->serial, disk->serial); | |
3341 | dl->disk = *disk; | |
3342 | dl->e = NULL; | |
3343 | dl->next = super->missing; | |
3344 | super->missing = dl; | |
3345 | } | |
3346 | ||
3347 | return 0; | |
3348 | } | |
3349 | ||
3350 | static struct intel_disk *disk_list_get(__u8 *serial, struct intel_disk *disk_list) | |
3351 | { | |
3352 | struct intel_disk *idisk = disk_list; | |
3353 | ||
3354 | while (idisk) { | |
3355 | if (serialcmp(idisk->disk.serial, serial) == 0) | |
3356 | break; | |
3357 | idisk = idisk->next; | |
3358 | } | |
3359 | ||
3360 | return idisk; | |
3361 | } | |
3362 | ||
3363 | static int __prep_thunderdome(struct intel_super **table, int tbl_size, | |
3364 | struct intel_super *super, | |
3365 | struct intel_disk **disk_list) | |
3366 | { | |
3367 | struct imsm_disk *d = &super->disks->disk; | |
3368 | struct imsm_super *mpb = super->anchor; | |
3369 | int i, j; | |
3370 | ||
3371 | for (i = 0; i < tbl_size; i++) { | |
3372 | struct imsm_super *tbl_mpb = table[i]->anchor; | |
3373 | struct imsm_disk *tbl_d = &table[i]->disks->disk; | |
3374 | ||
3375 | if (tbl_mpb->family_num == mpb->family_num) { | |
3376 | if (tbl_mpb->check_sum == mpb->check_sum) { | |
3377 | dprintf("%s: mpb from %d:%d matches %d:%d\n", | |
3378 | __func__, super->disks->major, | |
3379 | super->disks->minor, | |
3380 | table[i]->disks->major, | |
3381 | table[i]->disks->minor); | |
3382 | break; | |
3383 | } | |
3384 | ||
3385 | if (((is_configured(d) && !is_configured(tbl_d)) || | |
3386 | is_configured(d) == is_configured(tbl_d)) && | |
3387 | tbl_mpb->generation_num < mpb->generation_num) { | |
3388 | /* current version of the mpb is a | |
3389 | * better candidate than the one in | |
3390 | * super_table, but copy over "cross | |
3391 | * generational" status | |
3392 | */ | |
3393 | struct intel_disk *idisk; | |
3394 | ||
3395 | dprintf("%s: mpb from %d:%d replaces %d:%d\n", | |
3396 | __func__, super->disks->major, | |
3397 | super->disks->minor, | |
3398 | table[i]->disks->major, | |
3399 | table[i]->disks->minor); | |
3400 | ||
3401 | idisk = disk_list_get(tbl_d->serial, *disk_list); | |
3402 | if (idisk && is_failed(&idisk->disk)) | |
3403 | tbl_d->status |= FAILED_DISK; | |
3404 | break; | |
3405 | } else { | |
3406 | struct intel_disk *idisk; | |
3407 | struct imsm_disk *disk; | |
3408 | ||
3409 | /* tbl_mpb is more up to date, but copy | |
3410 | * over cross generational status before | |
3411 | * returning | |
3412 | */ | |
3413 | disk = __serial_to_disk(d->serial, mpb, NULL); | |
3414 | if (disk && is_failed(disk)) | |
3415 | d->status |= FAILED_DISK; | |
3416 | ||
3417 | idisk = disk_list_get(d->serial, *disk_list); | |
3418 | if (idisk) { | |
3419 | idisk->owner = i; | |
3420 | if (disk && is_configured(disk)) | |
3421 | idisk->disk.status |= CONFIGURED_DISK; | |
3422 | } | |
3423 | ||
3424 | dprintf("%s: mpb from %d:%d prefer %d:%d\n", | |
3425 | __func__, super->disks->major, | |
3426 | super->disks->minor, | |
3427 | table[i]->disks->major, | |
3428 | table[i]->disks->minor); | |
3429 | ||
3430 | return tbl_size; | |
3431 | } | |
3432 | } | |
3433 | } | |
3434 | ||
3435 | if (i >= tbl_size) | |
3436 | table[tbl_size++] = super; | |
3437 | else | |
3438 | table[i] = super; | |
3439 | ||
3440 | /* update/extend the merged list of imsm_disk records */ | |
3441 | for (j = 0; j < mpb->num_disks; j++) { | |
3442 | struct imsm_disk *disk = __get_imsm_disk(mpb, j); | |
3443 | struct intel_disk *idisk; | |
3444 | ||
3445 | idisk = disk_list_get(disk->serial, *disk_list); | |
3446 | if (idisk) { | |
3447 | idisk->disk.status |= disk->status; | |
3448 | if (is_configured(&idisk->disk) || | |
3449 | is_failed(&idisk->disk)) | |
3450 | idisk->disk.status &= ~(SPARE_DISK); | |
3451 | } else { | |
3452 | idisk = calloc(1, sizeof(*idisk)); | |
3453 | if (!idisk) | |
3454 | return -1; | |
3455 | idisk->owner = IMSM_UNKNOWN_OWNER; | |
3456 | idisk->disk = *disk; | |
3457 | idisk->next = *disk_list; | |
3458 | *disk_list = idisk; | |
3459 | } | |
3460 | ||
3461 | if (serialcmp(idisk->disk.serial, d->serial) == 0) | |
3462 | idisk->owner = i; | |
3463 | } | |
3464 | ||
3465 | return tbl_size; | |
3466 | } | |
3467 | ||
3468 | static struct intel_super * | |
3469 | validate_members(struct intel_super *super, struct intel_disk *disk_list, | |
3470 | const int owner) | |
3471 | { | |
3472 | struct imsm_super *mpb = super->anchor; | |
3473 | int ok_count = 0; | |
3474 | int i; | |
3475 | ||
3476 | for (i = 0; i < mpb->num_disks; i++) { | |
3477 | struct imsm_disk *disk = __get_imsm_disk(mpb, i); | |
3478 | struct intel_disk *idisk; | |
3479 | ||
3480 | idisk = disk_list_get(disk->serial, disk_list); | |
3481 | if (idisk) { | |
3482 | if (idisk->owner == owner || | |
3483 | idisk->owner == IMSM_UNKNOWN_OWNER) | |
3484 | ok_count++; | |
3485 | else | |
3486 | dprintf("%s: '%.16s' owner %d != %d\n", | |
3487 | __func__, disk->serial, idisk->owner, | |
3488 | owner); | |
3489 | } else { | |
3490 | dprintf("%s: unknown disk %x [%d]: %.16s\n", | |
3491 | __func__, __le32_to_cpu(mpb->family_num), i, | |
3492 | disk->serial); | |
3493 | break; | |
3494 | } | |
3495 | } | |
3496 | ||
3497 | if (ok_count == mpb->num_disks) | |
3498 | return super; | |
3499 | return NULL; | |
3500 | } | |
3501 | ||
3502 | static void show_conflicts(__u32 family_num, struct intel_super *super_list) | |
3503 | { | |
3504 | struct intel_super *s; | |
3505 | ||
3506 | for (s = super_list; s; s = s->next) { | |
3507 | if (family_num != s->anchor->family_num) | |
3508 | continue; | |
3509 | fprintf(stderr, "Conflict, offlining family %#x on '%s'\n", | |
3510 | __le32_to_cpu(family_num), s->disks->devname); | |
3511 | } | |
3512 | } | |
3513 | ||
3514 | static struct intel_super * | |
3515 | imsm_thunderdome(struct intel_super **super_list, int len) | |
3516 | { | |
3517 | struct intel_super *super_table[len]; | |
3518 | struct intel_disk *disk_list = NULL; | |
3519 | struct intel_super *champion, *spare; | |
3520 | struct intel_super *s, **del; | |
3521 | int tbl_size = 0; | |
3522 | int conflict; | |
3523 | int i; | |
3524 | ||
3525 | memset(super_table, 0, sizeof(super_table)); | |
3526 | for (s = *super_list; s; s = s->next) | |
3527 | tbl_size = __prep_thunderdome(super_table, tbl_size, s, &disk_list); | |
3528 | ||
3529 | for (i = 0; i < tbl_size; i++) { | |
3530 | struct imsm_disk *d; | |
3531 | struct intel_disk *idisk; | |
3532 | struct imsm_super *mpb = super_table[i]->anchor; | |
3533 | ||
3534 | s = super_table[i]; | |
3535 | d = &s->disks->disk; | |
3536 | ||
3537 | /* 'd' must appear in merged disk list for its | |
3538 | * configuration to be valid | |
3539 | */ | |
3540 | idisk = disk_list_get(d->serial, disk_list); | |
3541 | if (idisk && idisk->owner == i) | |
3542 | s = validate_members(s, disk_list, i); | |
3543 | else | |
3544 | s = NULL; | |
3545 | ||
3546 | if (!s) | |
3547 | dprintf("%s: marking family: %#x from %d:%d offline\n", | |
3548 | __func__, mpb->family_num, | |
3549 | super_table[i]->disks->major, | |
3550 | super_table[i]->disks->minor); | |
3551 | super_table[i] = s; | |
3552 | } | |
3553 | ||
3554 | /* This is where the mdadm implementation differs from the Windows | |
3555 | * driver which has no strict concept of a container. We can only | |
3556 | * assemble one family from a container, so when returning a prodigal | |
3557 | * array member to this system the code will not be able to disambiguate | |
3558 | * the container contents that should be assembled ("foreign" versus | |
3559 | * "local"). It requires user intervention to set the orig_family_num | |
3560 | * to a new value to establish a new container. The Windows driver in | |
3561 | * this situation fixes up the volume name in place and manages the | |
3562 | * foreign array as an independent entity. | |
3563 | */ | |
3564 | s = NULL; | |
3565 | spare = NULL; | |
3566 | conflict = 0; | |
3567 | for (i = 0; i < tbl_size; i++) { | |
3568 | struct intel_super *tbl_ent = super_table[i]; | |
3569 | int is_spare = 0; | |
3570 | ||
3571 | if (!tbl_ent) | |
3572 | continue; | |
3573 | ||
3574 | if (tbl_ent->anchor->num_raid_devs == 0) { | |
3575 | spare = tbl_ent; | |
3576 | is_spare = 1; | |
3577 | } | |
3578 | ||
3579 | if (s && !is_spare) { | |
3580 | show_conflicts(tbl_ent->anchor->family_num, *super_list); | |
3581 | conflict++; | |
3582 | } else if (!s && !is_spare) | |
3583 | s = tbl_ent; | |
3584 | } | |
3585 | ||
3586 | if (!s) | |
3587 | s = spare; | |
3588 | if (!s) { | |
3589 | champion = NULL; | |
3590 | goto out; | |
3591 | } | |
3592 | champion = s; | |
3593 | ||
3594 | if (conflict) | |
3595 | fprintf(stderr, "Chose family %#x on '%s', " | |
3596 | "assemble conflicts to new container with '--update=uuid'\n", | |
3597 | __le32_to_cpu(s->anchor->family_num), s->disks->devname); | |
3598 | ||
3599 | /* collect all dl's onto 'champion', and update them to | |
3600 | * champion's version of the status | |
3601 | */ | |
3602 | for (s = *super_list; s; s = s->next) { | |
3603 | struct imsm_super *mpb = champion->anchor; | |
3604 | struct dl *dl = s->disks; | |
3605 | ||
3606 | if (s == champion) | |
3607 | continue; | |
3608 | ||
3609 | for (i = 0; i < mpb->num_disks; i++) { | |
3610 | struct imsm_disk *disk; | |
3611 | ||
3612 | disk = __serial_to_disk(dl->serial, mpb, &dl->index); | |
3613 | if (disk) { | |
3614 | dl->disk = *disk; | |
3615 | /* only set index on disks that are a member of | |
3616 | * a populated contianer, i.e. one with | |
3617 | * raid_devs | |
3618 | */ | |
3619 | if (is_failed(&dl->disk)) | |
3620 | dl->index = -2; | |
3621 | else if (is_spare(&dl->disk)) | |
3622 | dl->index = -1; | |
3623 | break; | |
3624 | } | |
3625 | } | |
3626 | ||
3627 | if (i >= mpb->num_disks) { | |
3628 | struct intel_disk *idisk; | |
3629 | ||
3630 | idisk = disk_list_get(dl->serial, disk_list); | |
3631 | if (idisk && is_spare(&idisk->disk) && | |
3632 | !is_failed(&idisk->disk) && !is_configured(&idisk->disk)) | |
3633 | dl->index = -1; | |
3634 | else { | |
3635 | dl->index = -2; | |
3636 | continue; | |
3637 | } | |
3638 | } | |
3639 | ||
3640 | dl->next = champion->disks; | |
3641 | champion->disks = dl; | |
3642 | s->disks = NULL; | |
3643 | } | |
3644 | ||
3645 | /* delete 'champion' from super_list */ | |
3646 | for (del = super_list; *del; ) { | |
3647 | if (*del == champion) { | |
3648 | *del = (*del)->next; | |
3649 | break; | |
3650 | } else | |
3651 | del = &(*del)->next; | |
3652 | } | |
3653 | champion->next = NULL; | |
3654 | ||
3655 | out: | |
3656 | while (disk_list) { | |
3657 | struct intel_disk *idisk = disk_list; | |
3658 | ||
3659 | disk_list = disk_list->next; | |
3660 | free(idisk); | |
3661 | } | |
3662 | ||
3663 | return champion; | |
3664 | } | |
3665 | ||
3666 | static int load_super_imsm_all(struct supertype *st, int fd, void **sbp, | |
3667 | char *devname) | |
3668 | { | |
3669 | struct mdinfo *sra; | |
3670 | struct intel_super *super_list = NULL; | |
3671 | struct intel_super *super = NULL; | |
3672 | int devnum = fd2devnum(fd); | |
3673 | struct mdinfo *sd; | |
3674 | int retry; | |
3675 | int err = 0; | |
3676 | int i; | |
3677 | ||
3678 | /* check if 'fd' an opened container */ | |
3679 | sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE); | |
3680 | if (!sra) | |
3681 | return 1; | |
3682 | ||
3683 | if (sra->array.major_version != -1 || | |
3684 | sra->array.minor_version != -2 || | |
3685 | strcmp(sra->text_version, "imsm") != 0) { | |
3686 | err = 1; | |
3687 | goto error; | |
3688 | } | |
3689 | /* load all mpbs */ | |
3690 | for (sd = sra->devs, i = 0; sd; sd = sd->next, i++) { | |
3691 | struct intel_super *s = alloc_super(); | |
3692 | char nm[32]; | |
3693 | int dfd; | |
3694 | int rv; | |
3695 | ||
3696 | err = 1; | |
3697 | if (!s) | |
3698 | goto error; | |
3699 | s->next = super_list; | |
3700 | super_list = s; | |
3701 | ||
3702 | err = 2; | |
3703 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
3704 | dfd = dev_open(nm, O_RDWR); | |
3705 | if (dfd < 0) | |
3706 | goto error; | |
3707 | ||
3708 | rv = find_intel_hba_capability(dfd, s, devname); | |
3709 | /* no orom/efi or non-intel hba of the disk */ | |
3710 | if (rv != 0) | |
3711 | goto error; | |
3712 | ||
3713 | err = load_and_parse_mpb(dfd, s, NULL, 1); | |
3714 | ||
3715 | /* retry the load if we might have raced against mdmon */ | |
3716 | if (err == 3 && mdmon_running(devnum)) | |
3717 | for (retry = 0; retry < 3; retry++) { | |
3718 | usleep(3000); | |
3719 | err = load_and_parse_mpb(dfd, s, NULL, 1); | |
3720 | if (err != 3) | |
3721 | break; | |
3722 | } | |
3723 | if (err) | |
3724 | goto error; | |
3725 | } | |
3726 | ||
3727 | /* all mpbs enter, maybe one leaves */ | |
3728 | super = imsm_thunderdome(&super_list, i); | |
3729 | if (!super) { | |
3730 | err = 1; | |
3731 | goto error; | |
3732 | } | |
3733 | ||
3734 | if (find_missing(super) != 0) { | |
3735 | free_imsm(super); | |
3736 | err = 2; | |
3737 | goto error; | |
3738 | } | |
3739 | ||
3740 | /* load migration record */ | |
3741 | err = load_imsm_migr_rec(super, NULL); | |
3742 | if (err) { | |
3743 | err = 4; | |
3744 | goto error; | |
3745 | } | |
3746 | ||
3747 | /* Check migration compatibility */ | |
3748 | if (check_mpb_migr_compatibility(super) != 0) { | |
3749 | fprintf(stderr, Name ": Unsupported migration detected"); | |
3750 | if (devname) | |
3751 | fprintf(stderr, " on %s\n", devname); | |
3752 | else | |
3753 | fprintf(stderr, " (IMSM).\n"); | |
3754 | ||
3755 | err = 5; | |
3756 | goto error; | |
3757 | } | |
3758 | ||
3759 | err = 0; | |
3760 | ||
3761 | error: | |
3762 | while (super_list) { | |
3763 | struct intel_super *s = super_list; | |
3764 | ||
3765 | super_list = super_list->next; | |
3766 | free_imsm(s); | |
3767 | } | |
3768 | sysfs_free(sra); | |
3769 | ||
3770 | if (err) | |
3771 | return err; | |
3772 | ||
3773 | *sbp = super; | |
3774 | st->container_dev = devnum; | |
3775 | if (err == 0 && st->ss == NULL) { | |
3776 | st->ss = &super_imsm; | |
3777 | st->minor_version = 0; | |
3778 | st->max_devs = IMSM_MAX_DEVICES; | |
3779 | } | |
3780 | return 0; | |
3781 | } | |
3782 | ||
3783 | static int load_container_imsm(struct supertype *st, int fd, char *devname) | |
3784 | { | |
3785 | return load_super_imsm_all(st, fd, &st->sb, devname); | |
3786 | } | |
3787 | #endif | |
3788 | ||
3789 | static int load_super_imsm(struct supertype *st, int fd, char *devname) | |
3790 | { | |
3791 | struct intel_super *super; | |
3792 | int rv; | |
3793 | ||
3794 | if (test_partition(fd)) | |
3795 | /* IMSM not allowed on partitions */ | |
3796 | return 1; | |
3797 | ||
3798 | free_super_imsm(st); | |
3799 | ||
3800 | super = alloc_super(); | |
3801 | if (!super) { | |
3802 | fprintf(stderr, | |
3803 | Name ": malloc of %zu failed.\n", | |
3804 | sizeof(*super)); | |
3805 | return 1; | |
3806 | } | |
3807 | /* Load hba and capabilities if they exist. | |
3808 | * But do not preclude loading metadata in case capabilities or hba are | |
3809 | * non-compliant and ignore_hw_compat is set. | |
3810 | */ | |
3811 | rv = find_intel_hba_capability(fd, super, devname); | |
3812 | /* no orom/efi or non-intel hba of the disk */ | |
3813 | if ((rv != 0) && (st->ignore_hw_compat == 0)) { | |
3814 | if (devname) | |
3815 | fprintf(stderr, | |
3816 | Name ": No OROM/EFI properties for %s\n", devname); | |
3817 | free_imsm(super); | |
3818 | return 2; | |
3819 | } | |
3820 | rv = load_and_parse_mpb(fd, super, devname, 0); | |
3821 | ||
3822 | if (rv) { | |
3823 | if (devname) | |
3824 | fprintf(stderr, | |
3825 | Name ": Failed to load all information " | |
3826 | "sections on %s\n", devname); | |
3827 | free_imsm(super); | |
3828 | return rv; | |
3829 | } | |
3830 | ||
3831 | st->sb = super; | |
3832 | if (st->ss == NULL) { | |
3833 | st->ss = &super_imsm; | |
3834 | st->minor_version = 0; | |
3835 | st->max_devs = IMSM_MAX_DEVICES; | |
3836 | } | |
3837 | ||
3838 | /* load migration record */ | |
3839 | if (load_imsm_migr_rec(super, NULL) == 0) { | |
3840 | /* Check for unsupported migration features */ | |
3841 | if (check_mpb_migr_compatibility(super) != 0) { | |
3842 | fprintf(stderr, | |
3843 | Name ": Unsupported migration detected"); | |
3844 | if (devname) | |
3845 | fprintf(stderr, " on %s\n", devname); | |
3846 | else | |
3847 | fprintf(stderr, " (IMSM).\n"); | |
3848 | return 3; | |
3849 | } | |
3850 | } | |
3851 | ||
3852 | return 0; | |
3853 | } | |
3854 | ||
3855 | static __u16 info_to_blocks_per_strip(mdu_array_info_t *info) | |
3856 | { | |
3857 | if (info->level == 1) | |
3858 | return 128; | |
3859 | return info->chunk_size >> 9; | |
3860 | } | |
3861 | ||
3862 | static __u32 info_to_num_data_stripes(mdu_array_info_t *info, int num_domains) | |
3863 | { | |
3864 | __u32 num_stripes; | |
3865 | ||
3866 | num_stripes = (info->size * 2) / info_to_blocks_per_strip(info); | |
3867 | num_stripes /= num_domains; | |
3868 | ||
3869 | return num_stripes; | |
3870 | } | |
3871 | ||
3872 | static __u32 info_to_blocks_per_member(mdu_array_info_t *info) | |
3873 | { | |
3874 | if (info->level == 1) | |
3875 | return info->size * 2; | |
3876 | else | |
3877 | return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1); | |
3878 | } | |
3879 | ||
3880 | static void imsm_update_version_info(struct intel_super *super) | |
3881 | { | |
3882 | /* update the version and attributes */ | |
3883 | struct imsm_super *mpb = super->anchor; | |
3884 | char *version; | |
3885 | struct imsm_dev *dev; | |
3886 | struct imsm_map *map; | |
3887 | int i; | |
3888 | ||
3889 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
3890 | dev = get_imsm_dev(super, i); | |
3891 | map = get_imsm_map(dev, 0); | |
3892 | if (__le32_to_cpu(dev->size_high) > 0) | |
3893 | mpb->attributes |= MPB_ATTRIB_2TB; | |
3894 | ||
3895 | /* FIXME detect when an array spans a port multiplier */ | |
3896 | #if 0 | |
3897 | mpb->attributes |= MPB_ATTRIB_PM; | |
3898 | #endif | |
3899 | ||
3900 | if (mpb->num_raid_devs > 1 || | |
3901 | mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) { | |
3902 | version = MPB_VERSION_ATTRIBS; | |
3903 | switch (get_imsm_raid_level(map)) { | |
3904 | case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break; | |
3905 | case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break; | |
3906 | case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break; | |
3907 | case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break; | |
3908 | } | |
3909 | } else { | |
3910 | if (map->num_members >= 5) | |
3911 | version = MPB_VERSION_5OR6_DISK_ARRAY; | |
3912 | else if (dev->status == DEV_CLONE_N_GO) | |
3913 | version = MPB_VERSION_CNG; | |
3914 | else if (get_imsm_raid_level(map) == 5) | |
3915 | version = MPB_VERSION_RAID5; | |
3916 | else if (map->num_members >= 3) | |
3917 | version = MPB_VERSION_3OR4_DISK_ARRAY; | |
3918 | else if (get_imsm_raid_level(map) == 1) | |
3919 | version = MPB_VERSION_RAID1; | |
3920 | else | |
3921 | version = MPB_VERSION_RAID0; | |
3922 | } | |
3923 | strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version); | |
3924 | } | |
3925 | } | |
3926 | ||
3927 | static int check_name(struct intel_super *super, char *name, int quiet) | |
3928 | { | |
3929 | struct imsm_super *mpb = super->anchor; | |
3930 | char *reason = NULL; | |
3931 | int i; | |
3932 | ||
3933 | if (strlen(name) > MAX_RAID_SERIAL_LEN) | |
3934 | reason = "must be 16 characters or less"; | |
3935 | ||
3936 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
3937 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
3938 | ||
3939 | if (strncmp((char *) dev->volume, name, MAX_RAID_SERIAL_LEN) == 0) { | |
3940 | reason = "already exists"; | |
3941 | break; | |
3942 | } | |
3943 | } | |
3944 | ||
3945 | if (reason && !quiet) | |
3946 | fprintf(stderr, Name ": imsm volume name %s\n", reason); | |
3947 | ||
3948 | return !reason; | |
3949 | } | |
3950 | ||
3951 | static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info, | |
3952 | unsigned long long size, char *name, | |
3953 | char *homehost, int *uuid) | |
3954 | { | |
3955 | /* We are creating a volume inside a pre-existing container. | |
3956 | * so st->sb is already set. | |
3957 | */ | |
3958 | struct intel_super *super = st->sb; | |
3959 | struct imsm_super *mpb = super->anchor; | |
3960 | struct intel_dev *dv; | |
3961 | struct imsm_dev *dev; | |
3962 | struct imsm_vol *vol; | |
3963 | struct imsm_map *map; | |
3964 | int idx = mpb->num_raid_devs; | |
3965 | int i; | |
3966 | unsigned long long array_blocks; | |
3967 | size_t size_old, size_new; | |
3968 | __u32 num_data_stripes; | |
3969 | ||
3970 | if (super->orom && mpb->num_raid_devs >= super->orom->vpa) { | |
3971 | fprintf(stderr, Name": This imsm-container already has the " | |
3972 | "maximum of %d volumes\n", super->orom->vpa); | |
3973 | return 0; | |
3974 | } | |
3975 | ||
3976 | /* ensure the mpb is large enough for the new data */ | |
3977 | size_old = __le32_to_cpu(mpb->mpb_size); | |
3978 | size_new = disks_to_mpb_size(info->nr_disks); | |
3979 | if (size_new > size_old) { | |
3980 | void *mpb_new; | |
3981 | size_t size_round = ROUND_UP(size_new, 512); | |
3982 | ||
3983 | if (posix_memalign(&mpb_new, 512, size_round) != 0) { | |
3984 | fprintf(stderr, Name": could not allocate new mpb\n"); | |
3985 | return 0; | |
3986 | } | |
3987 | if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) { | |
3988 | fprintf(stderr, Name | |
3989 | ": %s could not allocate migr_rec buffer\n", | |
3990 | __func__); | |
3991 | free(super->buf); | |
3992 | free(super); | |
3993 | return 0; | |
3994 | } | |
3995 | memcpy(mpb_new, mpb, size_old); | |
3996 | free(mpb); | |
3997 | mpb = mpb_new; | |
3998 | super->anchor = mpb_new; | |
3999 | mpb->mpb_size = __cpu_to_le32(size_new); | |
4000 | memset(mpb_new + size_old, 0, size_round - size_old); | |
4001 | } | |
4002 | super->current_vol = idx; | |
4003 | /* when creating the first raid device in this container set num_disks | |
4004 | * to zero, i.e. delete this spare and add raid member devices in | |
4005 | * add_to_super_imsm_volume() | |
4006 | */ | |
4007 | if (super->current_vol == 0) | |
4008 | mpb->num_disks = 0; | |
4009 | ||
4010 | if (!check_name(super, name, 0)) | |
4011 | return 0; | |
4012 | dv = malloc(sizeof(*dv)); | |
4013 | if (!dv) { | |
4014 | fprintf(stderr, Name ": failed to allocate device list entry\n"); | |
4015 | return 0; | |
4016 | } | |
4017 | dev = calloc(1, sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1)); | |
4018 | if (!dev) { | |
4019 | free(dv); | |
4020 | fprintf(stderr, Name": could not allocate raid device\n"); | |
4021 | return 0; | |
4022 | } | |
4023 | ||
4024 | strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN); | |
4025 | if (info->level == 1) | |
4026 | array_blocks = info_to_blocks_per_member(info); | |
4027 | else | |
4028 | array_blocks = calc_array_size(info->level, info->raid_disks, | |
4029 | info->layout, info->chunk_size, | |
4030 | info->size*2); | |
4031 | /* round array size down to closest MB */ | |
4032 | array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT; | |
4033 | ||
4034 | dev->size_low = __cpu_to_le32((__u32) array_blocks); | |
4035 | dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32)); | |
4036 | dev->status = (DEV_READ_COALESCING | DEV_WRITE_COALESCING); | |
4037 | vol = &dev->vol; | |
4038 | vol->migr_state = 0; | |
4039 | set_migr_type(dev, MIGR_INIT); | |
4040 | vol->dirty = 0; | |
4041 | vol->curr_migr_unit = 0; | |
4042 | map = get_imsm_map(dev, 0); | |
4043 | map->pba_of_lba0 = __cpu_to_le32(super->create_offset); | |
4044 | map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info)); | |
4045 | map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info)); | |
4046 | map->failed_disk_num = ~0; | |
4047 | map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED : | |
4048 | IMSM_T_STATE_NORMAL; | |
4049 | map->ddf = 1; | |
4050 | ||
4051 | if (info->level == 1 && info->raid_disks > 2) { | |
4052 | free(dev); | |
4053 | free(dv); | |
4054 | fprintf(stderr, Name": imsm does not support more than 2 disks" | |
4055 | "in a raid1 volume\n"); | |
4056 | return 0; | |
4057 | } | |
4058 | ||
4059 | map->raid_level = info->level; | |
4060 | if (info->level == 10) { | |
4061 | map->raid_level = 1; | |
4062 | map->num_domains = info->raid_disks / 2; | |
4063 | } else if (info->level == 1) | |
4064 | map->num_domains = info->raid_disks; | |
4065 | else | |
4066 | map->num_domains = 1; | |
4067 | ||
4068 | num_data_stripes = info_to_num_data_stripes(info, map->num_domains); | |
4069 | map->num_data_stripes = __cpu_to_le32(num_data_stripes); | |
4070 | ||
4071 | map->num_members = info->raid_disks; | |
4072 | for (i = 0; i < map->num_members; i++) { | |
4073 | /* initialized in add_to_super */ | |
4074 | set_imsm_ord_tbl_ent(map, i, IMSM_ORD_REBUILD); | |
4075 | } | |
4076 | mpb->num_raid_devs++; | |
4077 | ||
4078 | dv->dev = dev; | |
4079 | dv->index = super->current_vol; | |
4080 | dv->next = super->devlist; | |
4081 | super->devlist = dv; | |
4082 | ||
4083 | imsm_update_version_info(super); | |
4084 | ||
4085 | return 1; | |
4086 | } | |
4087 | ||
4088 | static int init_super_imsm(struct supertype *st, mdu_array_info_t *info, | |
4089 | unsigned long long size, char *name, | |
4090 | char *homehost, int *uuid) | |
4091 | { | |
4092 | /* This is primarily called by Create when creating a new array. | |
4093 | * We will then get add_to_super called for each component, and then | |
4094 | * write_init_super called to write it out to each device. | |
4095 | * For IMSM, Create can create on fresh devices or on a pre-existing | |
4096 | * array. | |
4097 | * To create on a pre-existing array a different method will be called. | |
4098 | * This one is just for fresh drives. | |
4099 | */ | |
4100 | struct intel_super *super; | |
4101 | struct imsm_super *mpb; | |
4102 | size_t mpb_size; | |
4103 | char *version; | |
4104 | ||
4105 | if (st->sb) | |
4106 | return init_super_imsm_volume(st, info, size, name, homehost, uuid); | |
4107 | ||
4108 | if (info) | |
4109 | mpb_size = disks_to_mpb_size(info->nr_disks); | |
4110 | else | |
4111 | mpb_size = 512; | |
4112 | ||
4113 | super = alloc_super(); | |
4114 | if (super && posix_memalign(&super->buf, 512, mpb_size) != 0) { | |
4115 | free(super); | |
4116 | super = NULL; | |
4117 | } | |
4118 | if (!super) { | |
4119 | fprintf(stderr, Name | |
4120 | ": %s could not allocate superblock\n", __func__); | |
4121 | return 0; | |
4122 | } | |
4123 | if (posix_memalign(&super->migr_rec_buf, 512, 512) != 0) { | |
4124 | fprintf(stderr, Name | |
4125 | ": %s could not allocate migr_rec buffer\n", __func__); | |
4126 | free(super->buf); | |
4127 | free(super); | |
4128 | return 0; | |
4129 | } | |
4130 | memset(super->buf, 0, mpb_size); | |
4131 | mpb = super->buf; | |
4132 | mpb->mpb_size = __cpu_to_le32(mpb_size); | |
4133 | st->sb = super; | |
4134 | ||
4135 | if (info == NULL) { | |
4136 | /* zeroing superblock */ | |
4137 | return 0; | |
4138 | } | |
4139 | ||
4140 | mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY; | |
4141 | ||
4142 | version = (char *) mpb->sig; | |
4143 | strcpy(version, MPB_SIGNATURE); | |
4144 | version += strlen(MPB_SIGNATURE); | |
4145 | strcpy(version, MPB_VERSION_RAID0); | |
4146 | ||
4147 | return 1; | |
4148 | } | |
4149 | ||
4150 | #ifndef MDASSEMBLE | |
4151 | static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk, | |
4152 | int fd, char *devname) | |
4153 | { | |
4154 | struct intel_super *super = st->sb; | |
4155 | struct imsm_super *mpb = super->anchor; | |
4156 | struct dl *dl; | |
4157 | struct imsm_dev *dev; | |
4158 | struct imsm_map *map; | |
4159 | int slot; | |
4160 | ||
4161 | dev = get_imsm_dev(super, super->current_vol); | |
4162 | map = get_imsm_map(dev, 0); | |
4163 | ||
4164 | if (! (dk->state & (1<<MD_DISK_SYNC))) { | |
4165 | fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n", | |
4166 | devname); | |
4167 | return 1; | |
4168 | } | |
4169 | ||
4170 | if (fd == -1) { | |
4171 | /* we're doing autolayout so grab the pre-marked (in | |
4172 | * validate_geometry) raid_disk | |
4173 | */ | |
4174 | for (dl = super->disks; dl; dl = dl->next) | |
4175 | if (dl->raiddisk == dk->raid_disk) | |
4176 | break; | |
4177 | } else { | |
4178 | for (dl = super->disks; dl ; dl = dl->next) | |
4179 | if (dl->major == dk->major && | |
4180 | dl->minor == dk->minor) | |
4181 | break; | |
4182 | } | |
4183 | ||
4184 | if (!dl) { | |
4185 | fprintf(stderr, Name ": %s is not a member of the same container\n", devname); | |
4186 | return 1; | |
4187 | } | |
4188 | ||
4189 | /* add a pristine spare to the metadata */ | |
4190 | if (dl->index < 0) { | |
4191 | dl->index = super->anchor->num_disks; | |
4192 | super->anchor->num_disks++; | |
4193 | } | |
4194 | /* Check the device has not already been added */ | |
4195 | slot = get_imsm_disk_slot(map, dl->index); | |
4196 | if (slot >= 0 && | |
4197 | (get_imsm_ord_tbl_ent(dev, slot, -1) & IMSM_ORD_REBUILD) == 0) { | |
4198 | fprintf(stderr, Name ": %s has been included in this array twice\n", | |
4199 | devname); | |
4200 | return 1; | |
4201 | } | |
4202 | set_imsm_ord_tbl_ent(map, dk->number, dl->index); | |
4203 | dl->disk.status = CONFIGURED_DISK; | |
4204 | ||
4205 | /* if we are creating the first raid device update the family number */ | |
4206 | if (super->current_vol == 0) { | |
4207 | __u32 sum; | |
4208 | struct imsm_dev *_dev = __get_imsm_dev(mpb, 0); | |
4209 | struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index); | |
4210 | ||
4211 | if (!_dev || !_disk) { | |
4212 | fprintf(stderr, Name ": BUG mpb setup error\n"); | |
4213 | return 1; | |
4214 | } | |
4215 | *_dev = *dev; | |
4216 | *_disk = dl->disk; | |
4217 | sum = random32(); | |
4218 | sum += __gen_imsm_checksum(mpb); | |
4219 | mpb->family_num = __cpu_to_le32(sum); | |
4220 | mpb->orig_family_num = mpb->family_num; | |
4221 | } | |
4222 | ||
4223 | return 0; | |
4224 | } | |
4225 | ||
4226 | ||
4227 | static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk, | |
4228 | int fd, char *devname) | |
4229 | { | |
4230 | struct intel_super *super = st->sb; | |
4231 | struct dl *dd; | |
4232 | unsigned long long size; | |
4233 | __u32 id; | |
4234 | int rv; | |
4235 | struct stat stb; | |
4236 | ||
4237 | /* If we are on an RAID enabled platform check that the disk is | |
4238 | * attached to the raid controller. | |
4239 | * We do not need to test disks attachment for container based additions, | |
4240 | * they shall be already tested when container was created/assembled. | |
4241 | */ | |
4242 | rv = find_intel_hba_capability(fd, super, devname); | |
4243 | /* no orom/efi or non-intel hba of the disk */ | |
4244 | if (rv != 0) { | |
4245 | dprintf("capability: %p fd: %d ret: %d\n", | |
4246 | super->orom, fd, rv); | |
4247 | return 1; | |
4248 | } | |
4249 | ||
4250 | if (super->current_vol >= 0) | |
4251 | return add_to_super_imsm_volume(st, dk, fd, devname); | |
4252 | ||
4253 | fstat(fd, &stb); | |
4254 | dd = malloc(sizeof(*dd)); | |
4255 | if (!dd) { | |
4256 | fprintf(stderr, | |
4257 | Name ": malloc failed %s:%d.\n", __func__, __LINE__); | |
4258 | return 1; | |
4259 | } | |
4260 | memset(dd, 0, sizeof(*dd)); | |
4261 | dd->major = major(stb.st_rdev); | |
4262 | dd->minor = minor(stb.st_rdev); | |
4263 | dd->index = -1; | |
4264 | dd->devname = devname ? strdup(devname) : NULL; | |
4265 | dd->fd = fd; | |
4266 | dd->e = NULL; | |
4267 | dd->action = DISK_ADD; | |
4268 | rv = imsm_read_serial(fd, devname, dd->serial); | |
4269 | if (rv) { | |
4270 | fprintf(stderr, | |
4271 | Name ": failed to retrieve scsi serial, aborting\n"); | |
4272 | free(dd); | |
4273 | abort(); | |
4274 | } | |
4275 | ||
4276 | get_dev_size(fd, NULL, &size); | |
4277 | size /= 512; | |
4278 | serialcpy(dd->disk.serial, dd->serial); | |
4279 | dd->disk.total_blocks = __cpu_to_le32(size); | |
4280 | dd->disk.status = SPARE_DISK; | |
4281 | if (sysfs_disk_to_scsi_id(fd, &id) == 0) | |
4282 | dd->disk.scsi_id = __cpu_to_le32(id); | |
4283 | else | |
4284 | dd->disk.scsi_id = __cpu_to_le32(0); | |
4285 | ||
4286 | if (st->update_tail) { | |
4287 | dd->next = super->disk_mgmt_list; | |
4288 | super->disk_mgmt_list = dd; | |
4289 | } else { | |
4290 | dd->next = super->disks; | |
4291 | super->disks = dd; | |
4292 | super->updates_pending++; | |
4293 | } | |
4294 | ||
4295 | return 0; | |
4296 | } | |
4297 | ||
4298 | ||
4299 | static int remove_from_super_imsm(struct supertype *st, mdu_disk_info_t *dk) | |
4300 | { | |
4301 | struct intel_super *super = st->sb; | |
4302 | struct dl *dd; | |
4303 | ||
4304 | /* remove from super works only in mdmon - for communication | |
4305 | * manager - monitor. Check if communication memory buffer | |
4306 | * is prepared. | |
4307 | */ | |
4308 | if (!st->update_tail) { | |
4309 | fprintf(stderr, | |
4310 | Name ": %s shall be used in mdmon context only" | |
4311 | "(line %d).\n", __func__, __LINE__); | |
4312 | return 1; | |
4313 | } | |
4314 | dd = malloc(sizeof(*dd)); | |
4315 | if (!dd) { | |
4316 | fprintf(stderr, | |
4317 | Name ": malloc failed %s:%d.\n", __func__, __LINE__); | |
4318 | return 1; | |
4319 | } | |
4320 | memset(dd, 0, sizeof(*dd)); | |
4321 | dd->major = dk->major; | |
4322 | dd->minor = dk->minor; | |
4323 | dd->index = -1; | |
4324 | dd->fd = -1; | |
4325 | dd->disk.status = SPARE_DISK; | |
4326 | dd->action = DISK_REMOVE; | |
4327 | ||
4328 | dd->next = super->disk_mgmt_list; | |
4329 | super->disk_mgmt_list = dd; | |
4330 | ||
4331 | ||
4332 | return 0; | |
4333 | } | |
4334 | ||
4335 | static int store_imsm_mpb(int fd, struct imsm_super *mpb); | |
4336 | ||
4337 | static union { | |
4338 | char buf[512]; | |
4339 | struct imsm_super anchor; | |
4340 | } spare_record __attribute__ ((aligned(512))); | |
4341 | ||
4342 | /* spare records have their own family number and do not have any defined raid | |
4343 | * devices | |
4344 | */ | |
4345 | static int write_super_imsm_spares(struct intel_super *super, int doclose) | |
4346 | { | |
4347 | struct imsm_super *mpb = super->anchor; | |
4348 | struct imsm_super *spare = &spare_record.anchor; | |
4349 | __u32 sum; | |
4350 | struct dl *d; | |
4351 | ||
4352 | spare->mpb_size = __cpu_to_le32(sizeof(struct imsm_super)), | |
4353 | spare->generation_num = __cpu_to_le32(1UL), | |
4354 | spare->attributes = MPB_ATTRIB_CHECKSUM_VERIFY; | |
4355 | spare->num_disks = 1, | |
4356 | spare->num_raid_devs = 0, | |
4357 | spare->cache_size = mpb->cache_size, | |
4358 | spare->pwr_cycle_count = __cpu_to_le32(1), | |
4359 | ||
4360 | snprintf((char *) spare->sig, MAX_SIGNATURE_LENGTH, | |
4361 | MPB_SIGNATURE MPB_VERSION_RAID0); | |
4362 | ||
4363 | for (d = super->disks; d; d = d->next) { | |
4364 | if (d->index != -1) | |
4365 | continue; | |
4366 | ||
4367 | spare->disk[0] = d->disk; | |
4368 | sum = __gen_imsm_checksum(spare); | |
4369 | spare->family_num = __cpu_to_le32(sum); | |
4370 | spare->orig_family_num = 0; | |
4371 | sum = __gen_imsm_checksum(spare); | |
4372 | spare->check_sum = __cpu_to_le32(sum); | |
4373 | ||
4374 | if (store_imsm_mpb(d->fd, spare)) { | |
4375 | fprintf(stderr, "%s: failed for device %d:%d %s\n", | |
4376 | __func__, d->major, d->minor, strerror(errno)); | |
4377 | return 1; | |
4378 | } | |
4379 | if (doclose) { | |
4380 | close(d->fd); | |
4381 | d->fd = -1; | |
4382 | } | |
4383 | } | |
4384 | ||
4385 | return 0; | |
4386 | } | |
4387 | ||
4388 | static int is_gen_migration(struct imsm_dev *dev); | |
4389 | ||
4390 | static int write_super_imsm(struct supertype *st, int doclose) | |
4391 | { | |
4392 | struct intel_super *super = st->sb; | |
4393 | struct imsm_super *mpb = super->anchor; | |
4394 | struct dl *d; | |
4395 | __u32 generation; | |
4396 | __u32 sum; | |
4397 | int spares = 0; | |
4398 | int i; | |
4399 | __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk); | |
4400 | int num_disks = 0; | |
4401 | int clear_migration_record = 1; | |
4402 | ||
4403 | /* 'generation' is incremented everytime the metadata is written */ | |
4404 | generation = __le32_to_cpu(mpb->generation_num); | |
4405 | generation++; | |
4406 | mpb->generation_num = __cpu_to_le32(generation); | |
4407 | ||
4408 | /* fix up cases where previous mdadm releases failed to set | |
4409 | * orig_family_num | |
4410 | */ | |
4411 | if (mpb->orig_family_num == 0) | |
4412 | mpb->orig_family_num = mpb->family_num; | |
4413 | ||
4414 | for (d = super->disks; d; d = d->next) { | |
4415 | if (d->index == -1) | |
4416 | spares++; | |
4417 | else { | |
4418 | mpb->disk[d->index] = d->disk; | |
4419 | num_disks++; | |
4420 | } | |
4421 | } | |
4422 | for (d = super->missing; d; d = d->next) { | |
4423 | mpb->disk[d->index] = d->disk; | |
4424 | num_disks++; | |
4425 | } | |
4426 | mpb->num_disks = num_disks; | |
4427 | mpb_size += sizeof(struct imsm_disk) * mpb->num_disks; | |
4428 | ||
4429 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
4430 | struct imsm_dev *dev = __get_imsm_dev(mpb, i); | |
4431 | struct imsm_dev *dev2 = get_imsm_dev(super, i); | |
4432 | if (dev && dev2) { | |
4433 | imsm_copy_dev(dev, dev2); | |
4434 | mpb_size += sizeof_imsm_dev(dev, 0); | |
4435 | } | |
4436 | if (is_gen_migration(dev2)) | |
4437 | clear_migration_record = 0; | |
4438 | } | |
4439 | mpb_size += __le32_to_cpu(mpb->bbm_log_size); | |
4440 | mpb->mpb_size = __cpu_to_le32(mpb_size); | |
4441 | ||
4442 | /* recalculate checksum */ | |
4443 | sum = __gen_imsm_checksum(mpb); | |
4444 | mpb->check_sum = __cpu_to_le32(sum); | |
4445 | ||
4446 | if (clear_migration_record) | |
4447 | memset(super->migr_rec_buf, 0, 512); | |
4448 | ||
4449 | /* write the mpb for disks that compose raid devices */ | |
4450 | for (d = super->disks; d ; d = d->next) { | |
4451 | if (d->index < 0) | |
4452 | continue; | |
4453 | if (store_imsm_mpb(d->fd, mpb)) | |
4454 | fprintf(stderr, "%s: failed for device %d:%d %s\n", | |
4455 | __func__, d->major, d->minor, strerror(errno)); | |
4456 | if (clear_migration_record) { | |
4457 | unsigned long long dsize; | |
4458 | ||
4459 | get_dev_size(d->fd, NULL, &dsize); | |
4460 | if (lseek64(d->fd, dsize - 512, SEEK_SET) >= 0) { | |
4461 | write(d->fd, super->migr_rec_buf, 512); | |
4462 | } | |
4463 | } | |
4464 | if (doclose) { | |
4465 | close(d->fd); | |
4466 | d->fd = -1; | |
4467 | } | |
4468 | } | |
4469 | ||
4470 | if (spares) | |
4471 | return write_super_imsm_spares(super, doclose); | |
4472 | ||
4473 | return 0; | |
4474 | } | |
4475 | ||
4476 | ||
4477 | static int create_array(struct supertype *st, int dev_idx) | |
4478 | { | |
4479 | size_t len; | |
4480 | struct imsm_update_create_array *u; | |
4481 | struct intel_super *super = st->sb; | |
4482 | struct imsm_dev *dev = get_imsm_dev(super, dev_idx); | |
4483 | struct imsm_map *map = get_imsm_map(dev, 0); | |
4484 | struct disk_info *inf; | |
4485 | struct imsm_disk *disk; | |
4486 | int i; | |
4487 | ||
4488 | len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) + | |
4489 | sizeof(*inf) * map->num_members; | |
4490 | u = malloc(len); | |
4491 | if (!u) { | |
4492 | fprintf(stderr, "%s: failed to allocate update buffer\n", | |
4493 | __func__); | |
4494 | return 1; | |
4495 | } | |
4496 | ||
4497 | u->type = update_create_array; | |
4498 | u->dev_idx = dev_idx; | |
4499 | imsm_copy_dev(&u->dev, dev); | |
4500 | inf = get_disk_info(u); | |
4501 | for (i = 0; i < map->num_members; i++) { | |
4502 | int idx = get_imsm_disk_idx(dev, i, -1); | |
4503 | ||
4504 | disk = get_imsm_disk(super, idx); | |
4505 | serialcpy(inf[i].serial, disk->serial); | |
4506 | } | |
4507 | append_metadata_update(st, u, len); | |
4508 | ||
4509 | return 0; | |
4510 | } | |
4511 | ||
4512 | static int mgmt_disk(struct supertype *st) | |
4513 | { | |
4514 | struct intel_super *super = st->sb; | |
4515 | size_t len; | |
4516 | struct imsm_update_add_remove_disk *u; | |
4517 | ||
4518 | if (!super->disk_mgmt_list) | |
4519 | return 0; | |
4520 | ||
4521 | len = sizeof(*u); | |
4522 | u = malloc(len); | |
4523 | if (!u) { | |
4524 | fprintf(stderr, "%s: failed to allocate update buffer\n", | |
4525 | __func__); | |
4526 | return 1; | |
4527 | } | |
4528 | ||
4529 | u->type = update_add_remove_disk; | |
4530 | append_metadata_update(st, u, len); | |
4531 | ||
4532 | return 0; | |
4533 | } | |
4534 | ||
4535 | static int write_init_super_imsm(struct supertype *st) | |
4536 | { | |
4537 | struct intel_super *super = st->sb; | |
4538 | int current_vol = super->current_vol; | |
4539 | ||
4540 | /* we are done with current_vol reset it to point st at the container */ | |
4541 | super->current_vol = -1; | |
4542 | ||
4543 | if (st->update_tail) { | |
4544 | /* queue the recently created array / added disk | |
4545 | * as a metadata update */ | |
4546 | int rv; | |
4547 | ||
4548 | /* determine if we are creating a volume or adding a disk */ | |
4549 | if (current_vol < 0) { | |
4550 | /* in the mgmt (add/remove) disk case we are running | |
4551 | * in mdmon context, so don't close fd's | |
4552 | */ | |
4553 | return mgmt_disk(st); | |
4554 | } else | |
4555 | rv = create_array(st, current_vol); | |
4556 | ||
4557 | return rv; | |
4558 | } else { | |
4559 | struct dl *d; | |
4560 | for (d = super->disks; d; d = d->next) | |
4561 | Kill(d->devname, NULL, 0, 1, 1); | |
4562 | return write_super_imsm(st, 1); | |
4563 | } | |
4564 | } | |
4565 | #endif | |
4566 | ||
4567 | static int store_super_imsm(struct supertype *st, int fd) | |
4568 | { | |
4569 | struct intel_super *super = st->sb; | |
4570 | struct imsm_super *mpb = super ? super->anchor : NULL; | |
4571 | ||
4572 | if (!mpb) | |
4573 | return 1; | |
4574 | ||
4575 | #ifndef MDASSEMBLE | |
4576 | return store_imsm_mpb(fd, mpb); | |
4577 | #else | |
4578 | return 1; | |
4579 | #endif | |
4580 | } | |
4581 | ||
4582 | static int imsm_bbm_log_size(struct imsm_super *mpb) | |
4583 | { | |
4584 | return __le32_to_cpu(mpb->bbm_log_size); | |
4585 | } | |
4586 | ||
4587 | #ifndef MDASSEMBLE | |
4588 | static int validate_geometry_imsm_container(struct supertype *st, int level, | |
4589 | int layout, int raiddisks, int chunk, | |
4590 | unsigned long long size, char *dev, | |
4591 | unsigned long long *freesize, | |
4592 | int verbose) | |
4593 | { | |
4594 | int fd; | |
4595 | unsigned long long ldsize; | |
4596 | struct intel_super *super=NULL; | |
4597 | int rv = 0; | |
4598 | ||
4599 | if (level != LEVEL_CONTAINER) | |
4600 | return 0; | |
4601 | if (!dev) | |
4602 | return 1; | |
4603 | ||
4604 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
4605 | if (fd < 0) { | |
4606 | if (verbose) | |
4607 | fprintf(stderr, Name ": imsm: Cannot open %s: %s\n", | |
4608 | dev, strerror(errno)); | |
4609 | return 0; | |
4610 | } | |
4611 | if (!get_dev_size(fd, dev, &ldsize)) { | |
4612 | close(fd); | |
4613 | return 0; | |
4614 | } | |
4615 | ||
4616 | /* capabilities retrieve could be possible | |
4617 | * note that there is no fd for the disks in array. | |
4618 | */ | |
4619 | super = alloc_super(); | |
4620 | if (!super) { | |
4621 | fprintf(stderr, | |
4622 | Name ": malloc of %zu failed.\n", | |
4623 | sizeof(*super)); | |
4624 | close(fd); | |
4625 | return 0; | |
4626 | } | |
4627 | ||
4628 | rv = find_intel_hba_capability(fd, super, verbose ? dev : NULL); | |
4629 | if (rv != 0) { | |
4630 | #if DEBUG | |
4631 | char str[256]; | |
4632 | fd2devname(fd, str); | |
4633 | dprintf("validate_geometry_imsm_container: fd: %d %s orom: %p rv: %d raiddisk: %d\n", | |
4634 | fd, str, super->orom, rv, raiddisks); | |
4635 | #endif | |
4636 | /* no orom/efi or non-intel hba of the disk */ | |
4637 | close(fd); | |
4638 | free_imsm(super); | |
4639 | return 0; | |
4640 | } | |
4641 | close(fd); | |
4642 | if (super->orom && raiddisks > super->orom->tds) { | |
4643 | if (verbose) | |
4644 | fprintf(stderr, Name ": %d exceeds maximum number of" | |
4645 | " platform supported disks: %d\n", | |
4646 | raiddisks, super->orom->tds); | |
4647 | ||
4648 | free_imsm(super); | |
4649 | return 0; | |
4650 | } | |
4651 | ||
4652 | *freesize = avail_size_imsm(st, ldsize >> 9); | |
4653 | free_imsm(super); | |
4654 | ||
4655 | return 1; | |
4656 | } | |
4657 | ||
4658 | static unsigned long long find_size(struct extent *e, int *idx, int num_extents) | |
4659 | { | |
4660 | const unsigned long long base_start = e[*idx].start; | |
4661 | unsigned long long end = base_start + e[*idx].size; | |
4662 | int i; | |
4663 | ||
4664 | if (base_start == end) | |
4665 | return 0; | |
4666 | ||
4667 | *idx = *idx + 1; | |
4668 | for (i = *idx; i < num_extents; i++) { | |
4669 | /* extend overlapping extents */ | |
4670 | if (e[i].start >= base_start && | |
4671 | e[i].start <= end) { | |
4672 | if (e[i].size == 0) | |
4673 | return 0; | |
4674 | if (e[i].start + e[i].size > end) | |
4675 | end = e[i].start + e[i].size; | |
4676 | } else if (e[i].start > end) { | |
4677 | *idx = i; | |
4678 | break; | |
4679 | } | |
4680 | } | |
4681 | ||
4682 | return end - base_start; | |
4683 | } | |
4684 | ||
4685 | static unsigned long long merge_extents(struct intel_super *super, int sum_extents) | |
4686 | { | |
4687 | /* build a composite disk with all known extents and generate a new | |
4688 | * 'maxsize' given the "all disks in an array must share a common start | |
4689 | * offset" constraint | |
4690 | */ | |
4691 | struct extent *e = calloc(sum_extents, sizeof(*e)); | |
4692 | struct dl *dl; | |
4693 | int i, j; | |
4694 | int start_extent; | |
4695 | unsigned long long pos; | |
4696 | unsigned long long start = 0; | |
4697 | unsigned long long maxsize; | |
4698 | unsigned long reserve; | |
4699 | ||
4700 | if (!e) | |
4701 | return 0; | |
4702 | ||
4703 | /* coalesce and sort all extents. also, check to see if we need to | |
4704 | * reserve space between member arrays | |
4705 | */ | |
4706 | j = 0; | |
4707 | for (dl = super->disks; dl; dl = dl->next) { | |
4708 | if (!dl->e) | |
4709 | continue; | |
4710 | for (i = 0; i < dl->extent_cnt; i++) | |
4711 | e[j++] = dl->e[i]; | |
4712 | } | |
4713 | qsort(e, sum_extents, sizeof(*e), cmp_extent); | |
4714 | ||
4715 | /* merge extents */ | |
4716 | i = 0; | |
4717 | j = 0; | |
4718 | while (i < sum_extents) { | |
4719 | e[j].start = e[i].start; | |
4720 | e[j].size = find_size(e, &i, sum_extents); | |
4721 | j++; | |
4722 | if (e[j-1].size == 0) | |
4723 | break; | |
4724 | } | |
4725 | ||
4726 | pos = 0; | |
4727 | maxsize = 0; | |
4728 | start_extent = 0; | |
4729 | i = 0; | |
4730 | do { | |
4731 | unsigned long long esize; | |
4732 | ||
4733 | esize = e[i].start - pos; | |
4734 | if (esize >= maxsize) { | |
4735 | maxsize = esize; | |
4736 | start = pos; | |
4737 | start_extent = i; | |
4738 | } | |
4739 | pos = e[i].start + e[i].size; | |
4740 | i++; | |
4741 | } while (e[i-1].size); | |
4742 | free(e); | |
4743 | ||
4744 | if (maxsize == 0) | |
4745 | return 0; | |
4746 | ||
4747 | /* FIXME assumes volume at offset 0 is the first volume in a | |
4748 | * container | |
4749 | */ | |
4750 | if (start_extent > 0) | |
4751 | reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */ | |
4752 | else | |
4753 | reserve = 0; | |
4754 | ||
4755 | if (maxsize < reserve) | |
4756 | return 0; | |
4757 | ||
4758 | super->create_offset = ~((__u32) 0); | |
4759 | if (start + reserve > super->create_offset) | |
4760 | return 0; /* start overflows create_offset */ | |
4761 | super->create_offset = start + reserve; | |
4762 | ||
4763 | return maxsize - reserve; | |
4764 | } | |
4765 | ||
4766 | static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks) | |
4767 | { | |
4768 | if (level < 0 || level == 6 || level == 4) | |
4769 | return 0; | |
4770 | ||
4771 | /* if we have an orom prevent invalid raid levels */ | |
4772 | if (orom) | |
4773 | switch (level) { | |
4774 | case 0: return imsm_orom_has_raid0(orom); | |
4775 | case 1: | |
4776 | if (raiddisks > 2) | |
4777 | return imsm_orom_has_raid1e(orom); | |
4778 | return imsm_orom_has_raid1(orom) && raiddisks == 2; | |
4779 | case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4; | |
4780 | case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2; | |
4781 | } | |
4782 | else | |
4783 | return 1; /* not on an Intel RAID platform so anything goes */ | |
4784 | ||
4785 | return 0; | |
4786 | } | |
4787 | ||
4788 | ||
4789 | #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg)) | |
4790 | /* | |
4791 | * validate volume parameters with OROM/EFI capabilities | |
4792 | */ | |
4793 | static int | |
4794 | validate_geometry_imsm_orom(struct intel_super *super, int level, int layout, | |
4795 | int raiddisks, int *chunk, int verbose) | |
4796 | { | |
4797 | #if DEBUG | |
4798 | verbose = 1; | |
4799 | #endif | |
4800 | /* validate container capabilities */ | |
4801 | if (super->orom && raiddisks > super->orom->tds) { | |
4802 | if (verbose) | |
4803 | fprintf(stderr, Name ": %d exceeds maximum number of" | |
4804 | " platform supported disks: %d\n", | |
4805 | raiddisks, super->orom->tds); | |
4806 | return 0; | |
4807 | } | |
4808 | ||
4809 | /* capabilities of OROM tested - copied from validate_geometry_imsm_volume */ | |
4810 | if (super->orom && (!is_raid_level_supported(super->orom, level, | |
4811 | raiddisks))) { | |
4812 | pr_vrb(": platform does not support raid%d with %d disk%s\n", | |
4813 | level, raiddisks, raiddisks > 1 ? "s" : ""); | |
4814 | return 0; | |
4815 | } | |
4816 | if (super->orom && level != 1) { | |
4817 | if (chunk && (*chunk == 0 || *chunk == UnSet)) | |
4818 | *chunk = imsm_orom_default_chunk(super->orom); | |
4819 | else if (chunk && !imsm_orom_has_chunk(super->orom, *chunk)) { | |
4820 | pr_vrb(": platform does not support a chunk size of: " | |
4821 | "%d\n", *chunk); | |
4822 | return 0; | |
4823 | } | |
4824 | } | |
4825 | if (layout != imsm_level_to_layout(level)) { | |
4826 | if (level == 5) | |
4827 | pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n"); | |
4828 | else if (level == 10) | |
4829 | pr_vrb(": imsm raid 10 only supports the n2 layout\n"); | |
4830 | else | |
4831 | pr_vrb(": imsm unknown layout %#x for this raid level %d\n", | |
4832 | layout, level); | |
4833 | return 0; | |
4834 | } | |
4835 | return 1; | |
4836 | } | |
4837 | ||
4838 | /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd | |
4839 | * FIX ME add ahci details | |
4840 | */ | |
4841 | static int validate_geometry_imsm_volume(struct supertype *st, int level, | |
4842 | int layout, int raiddisks, int *chunk, | |
4843 | unsigned long long size, char *dev, | |
4844 | unsigned long long *freesize, | |
4845 | int verbose) | |
4846 | { | |
4847 | struct stat stb; | |
4848 | struct intel_super *super = st->sb; | |
4849 | struct imsm_super *mpb = super->anchor; | |
4850 | struct dl *dl; | |
4851 | unsigned long long pos = 0; | |
4852 | unsigned long long maxsize; | |
4853 | struct extent *e; | |
4854 | int i; | |
4855 | ||
4856 | /* We must have the container info already read in. */ | |
4857 | if (!super) | |
4858 | return 0; | |
4859 | ||
4860 | if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose)) { | |
4861 | fprintf(stderr, Name ": RAID gemetry validation failed. " | |
4862 | "Cannot proceed with the action(s).\n"); | |
4863 | return 0; | |
4864 | } | |
4865 | if (!dev) { | |
4866 | /* General test: make sure there is space for | |
4867 | * 'raiddisks' device extents of size 'size' at a given | |
4868 | * offset | |
4869 | */ | |
4870 | unsigned long long minsize = size; | |
4871 | unsigned long long start_offset = MaxSector; | |
4872 | int dcnt = 0; | |
4873 | if (minsize == 0) | |
4874 | minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS; | |
4875 | for (dl = super->disks; dl ; dl = dl->next) { | |
4876 | int found = 0; | |
4877 | ||
4878 | pos = 0; | |
4879 | i = 0; | |
4880 | e = get_extents(super, dl); | |
4881 | if (!e) continue; | |
4882 | do { | |
4883 | unsigned long long esize; | |
4884 | esize = e[i].start - pos; | |
4885 | if (esize >= minsize) | |
4886 | found = 1; | |
4887 | if (found && start_offset == MaxSector) { | |
4888 | start_offset = pos; | |
4889 | break; | |
4890 | } else if (found && pos != start_offset) { | |
4891 | found = 0; | |
4892 | break; | |
4893 | } | |
4894 | pos = e[i].start + e[i].size; | |
4895 | i++; | |
4896 | } while (e[i-1].size); | |
4897 | if (found) | |
4898 | dcnt++; | |
4899 | free(e); | |
4900 | } | |
4901 | if (dcnt < raiddisks) { | |
4902 | if (verbose) | |
4903 | fprintf(stderr, Name ": imsm: Not enough " | |
4904 | "devices with space for this array " | |
4905 | "(%d < %d)\n", | |
4906 | dcnt, raiddisks); | |
4907 | return 0; | |
4908 | } | |
4909 | return 1; | |
4910 | } | |
4911 | ||
4912 | /* This device must be a member of the set */ | |
4913 | if (stat(dev, &stb) < 0) | |
4914 | return 0; | |
4915 | if ((S_IFMT & stb.st_mode) != S_IFBLK) | |
4916 | return 0; | |
4917 | for (dl = super->disks ; dl ; dl = dl->next) { | |
4918 | if (dl->major == (int)major(stb.st_rdev) && | |
4919 | dl->minor == (int)minor(stb.st_rdev)) | |
4920 | break; | |
4921 | } | |
4922 | if (!dl) { | |
4923 | if (verbose) | |
4924 | fprintf(stderr, Name ": %s is not in the " | |
4925 | "same imsm set\n", dev); | |
4926 | return 0; | |
4927 | } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) { | |
4928 | /* If a volume is present then the current creation attempt | |
4929 | * cannot incorporate new spares because the orom may not | |
4930 | * understand this configuration (all member disks must be | |
4931 | * members of each array in the container). | |
4932 | */ | |
4933 | fprintf(stderr, Name ": %s is a spare and a volume" | |
4934 | " is already defined for this container\n", dev); | |
4935 | fprintf(stderr, Name ": The option-rom requires all member" | |
4936 | " disks to be a member of all volumes\n"); | |
4937 | return 0; | |
4938 | } | |
4939 | ||
4940 | /* retrieve the largest free space block */ | |
4941 | e = get_extents(super, dl); | |
4942 | maxsize = 0; | |
4943 | i = 0; | |
4944 | if (e) { | |
4945 | do { | |
4946 | unsigned long long esize; | |
4947 | ||
4948 | esize = e[i].start - pos; | |
4949 | if (esize >= maxsize) | |
4950 | maxsize = esize; | |
4951 | pos = e[i].start + e[i].size; | |
4952 | i++; | |
4953 | } while (e[i-1].size); | |
4954 | dl->e = e; | |
4955 | dl->extent_cnt = i; | |
4956 | } else { | |
4957 | if (verbose) | |
4958 | fprintf(stderr, Name ": unable to determine free space for: %s\n", | |
4959 | dev); | |
4960 | return 0; | |
4961 | } | |
4962 | if (maxsize < size) { | |
4963 | if (verbose) | |
4964 | fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n", | |
4965 | dev, maxsize, size); | |
4966 | return 0; | |
4967 | } | |
4968 | ||
4969 | /* count total number of extents for merge */ | |
4970 | i = 0; | |
4971 | for (dl = super->disks; dl; dl = dl->next) | |
4972 | if (dl->e) | |
4973 | i += dl->extent_cnt; | |
4974 | ||
4975 | maxsize = merge_extents(super, i); | |
4976 | if (maxsize < size || maxsize == 0) { | |
4977 | if (verbose) | |
4978 | fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n", | |
4979 | maxsize, size); | |
4980 | return 0; | |
4981 | } | |
4982 | ||
4983 | *freesize = maxsize; | |
4984 | ||
4985 | return 1; | |
4986 | } | |
4987 | ||
4988 | static int reserve_space(struct supertype *st, int raiddisks, | |
4989 | unsigned long long size, int chunk, | |
4990 | unsigned long long *freesize) | |
4991 | { | |
4992 | struct intel_super *super = st->sb; | |
4993 | struct imsm_super *mpb = super->anchor; | |
4994 | struct dl *dl; | |
4995 | int i; | |
4996 | int extent_cnt; | |
4997 | struct extent *e; | |
4998 | unsigned long long maxsize; | |
4999 | unsigned long long minsize; | |
5000 | int cnt; | |
5001 | int used; | |
5002 | ||
5003 | /* find the largest common start free region of the possible disks */ | |
5004 | used = 0; | |
5005 | extent_cnt = 0; | |
5006 | cnt = 0; | |
5007 | for (dl = super->disks; dl; dl = dl->next) { | |
5008 | dl->raiddisk = -1; | |
5009 | ||
5010 | if (dl->index >= 0) | |
5011 | used++; | |
5012 | ||
5013 | /* don't activate new spares if we are orom constrained | |
5014 | * and there is already a volume active in the container | |
5015 | */ | |
5016 | if (super->orom && dl->index < 0 && mpb->num_raid_devs) | |
5017 | continue; | |
5018 | ||
5019 | e = get_extents(super, dl); | |
5020 | if (!e) | |
5021 | continue; | |
5022 | for (i = 1; e[i-1].size; i++) | |
5023 | ; | |
5024 | dl->e = e; | |
5025 | dl->extent_cnt = i; | |
5026 | extent_cnt += i; | |
5027 | cnt++; | |
5028 | } | |
5029 | ||
5030 | maxsize = merge_extents(super, extent_cnt); | |
5031 | minsize = size; | |
5032 | if (size == 0) | |
5033 | /* chunk is in K */ | |
5034 | minsize = chunk * 2; | |
5035 | ||
5036 | if (cnt < raiddisks || | |
5037 | (super->orom && used && used != raiddisks) || | |
5038 | maxsize < minsize || | |
5039 | maxsize == 0) { | |
5040 | fprintf(stderr, Name ": not enough devices with space to create array.\n"); | |
5041 | return 0; /* No enough free spaces large enough */ | |
5042 | } | |
5043 | ||
5044 | if (size == 0) { | |
5045 | size = maxsize; | |
5046 | if (chunk) { | |
5047 | size /= 2 * chunk; | |
5048 | size *= 2 * chunk; | |
5049 | } | |
5050 | } | |
5051 | ||
5052 | cnt = 0; | |
5053 | for (dl = super->disks; dl; dl = dl->next) | |
5054 | if (dl->e) | |
5055 | dl->raiddisk = cnt++; | |
5056 | ||
5057 | *freesize = size; | |
5058 | ||
5059 | return 1; | |
5060 | } | |
5061 | ||
5062 | static int validate_geometry_imsm(struct supertype *st, int level, int layout, | |
5063 | int raiddisks, int *chunk, unsigned long long size, | |
5064 | char *dev, unsigned long long *freesize, | |
5065 | int verbose) | |
5066 | { | |
5067 | int fd, cfd; | |
5068 | struct mdinfo *sra; | |
5069 | int is_member = 0; | |
5070 | ||
5071 | /* load capability | |
5072 | * if given unused devices create a container | |
5073 | * if given given devices in a container create a member volume | |
5074 | */ | |
5075 | if (level == LEVEL_CONTAINER) { | |
5076 | /* Must be a fresh device to add to a container */ | |
5077 | return validate_geometry_imsm_container(st, level, layout, | |
5078 | raiddisks, | |
5079 | chunk?*chunk:0, size, | |
5080 | dev, freesize, | |
5081 | verbose); | |
5082 | } | |
5083 | ||
5084 | if (!dev) { | |
5085 | if (st->sb && freesize) { | |
5086 | /* we are being asked to automatically layout a | |
5087 | * new volume based on the current contents of | |
5088 | * the container. If the the parameters can be | |
5089 | * satisfied reserve_space will record the disks, | |
5090 | * start offset, and size of the volume to be | |
5091 | * created. add_to_super and getinfo_super | |
5092 | * detect when autolayout is in progress. | |
5093 | */ | |
5094 | if (!validate_geometry_imsm_orom(st->sb, level, layout, | |
5095 | raiddisks, chunk, | |
5096 | verbose)) | |
5097 | return 0; | |
5098 | return reserve_space(st, raiddisks, size, | |
5099 | chunk?*chunk:0, freesize); | |
5100 | } | |
5101 | return 1; | |
5102 | } | |
5103 | if (st->sb) { | |
5104 | /* creating in a given container */ | |
5105 | return validate_geometry_imsm_volume(st, level, layout, | |
5106 | raiddisks, chunk, size, | |
5107 | dev, freesize, verbose); | |
5108 | } | |
5109 | ||
5110 | /* This device needs to be a device in an 'imsm' container */ | |
5111 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
5112 | if (fd >= 0) { | |
5113 | if (verbose) | |
5114 | fprintf(stderr, | |
5115 | Name ": Cannot create this array on device %s\n", | |
5116 | dev); | |
5117 | close(fd); | |
5118 | return 0; | |
5119 | } | |
5120 | if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) { | |
5121 | if (verbose) | |
5122 | fprintf(stderr, Name ": Cannot open %s: %s\n", | |
5123 | dev, strerror(errno)); | |
5124 | return 0; | |
5125 | } | |
5126 | /* Well, it is in use by someone, maybe an 'imsm' container. */ | |
5127 | cfd = open_container(fd); | |
5128 | close(fd); | |
5129 | if (cfd < 0) { | |
5130 | if (verbose) | |
5131 | fprintf(stderr, Name ": Cannot use %s: It is busy\n", | |
5132 | dev); | |
5133 | return 0; | |
5134 | } | |
5135 | sra = sysfs_read(cfd, 0, GET_VERSION); | |
5136 | if (sra && sra->array.major_version == -1 && | |
5137 | strcmp(sra->text_version, "imsm") == 0) | |
5138 | is_member = 1; | |
5139 | sysfs_free(sra); | |
5140 | if (is_member) { | |
5141 | /* This is a member of a imsm container. Load the container | |
5142 | * and try to create a volume | |
5143 | */ | |
5144 | struct intel_super *super; | |
5145 | ||
5146 | if (load_super_imsm_all(st, cfd, (void **) &super, NULL) == 0) { | |
5147 | st->sb = super; | |
5148 | st->container_dev = fd2devnum(cfd); | |
5149 | close(cfd); | |
5150 | return validate_geometry_imsm_volume(st, level, layout, | |
5151 | raiddisks, chunk, | |
5152 | size, dev, | |
5153 | freesize, verbose); | |
5154 | } | |
5155 | } | |
5156 | ||
5157 | if (verbose) | |
5158 | fprintf(stderr, Name ": failed container membership check\n"); | |
5159 | ||
5160 | close(cfd); | |
5161 | return 0; | |
5162 | } | |
5163 | ||
5164 | static void default_geometry_imsm(struct supertype *st, int *level, int *layout, int *chunk) | |
5165 | { | |
5166 | struct intel_super *super = st->sb; | |
5167 | ||
5168 | if (level && *level == UnSet) | |
5169 | *level = LEVEL_CONTAINER; | |
5170 | ||
5171 | if (level && layout && *layout == UnSet) | |
5172 | *layout = imsm_level_to_layout(*level); | |
5173 | ||
5174 | if (chunk && (*chunk == UnSet || *chunk == 0) && | |
5175 | super && super->orom) | |
5176 | *chunk = imsm_orom_default_chunk(super->orom); | |
5177 | } | |
5178 | ||
5179 | static void handle_missing(struct intel_super *super, struct imsm_dev *dev); | |
5180 | ||
5181 | static int kill_subarray_imsm(struct supertype *st) | |
5182 | { | |
5183 | /* remove the subarray currently referenced by ->current_vol */ | |
5184 | __u8 i; | |
5185 | struct intel_dev **dp; | |
5186 | struct intel_super *super = st->sb; | |
5187 | __u8 current_vol = super->current_vol; | |
5188 | struct imsm_super *mpb = super->anchor; | |
5189 | ||
5190 | if (super->current_vol < 0) | |
5191 | return 2; | |
5192 | super->current_vol = -1; /* invalidate subarray cursor */ | |
5193 | ||
5194 | /* block deletions that would change the uuid of active subarrays | |
5195 | * | |
5196 | * FIXME when immutable ids are available, but note that we'll | |
5197 | * also need to fixup the invalidated/active subarray indexes in | |
5198 | * mdstat | |
5199 | */ | |
5200 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5201 | char subarray[4]; | |
5202 | ||
5203 | if (i < current_vol) | |
5204 | continue; | |
5205 | sprintf(subarray, "%u", i); | |
5206 | if (is_subarray_active(subarray, st->devname)) { | |
5207 | fprintf(stderr, | |
5208 | Name ": deleting subarray-%d would change the UUID of active subarray-%d, aborting\n", | |
5209 | current_vol, i); | |
5210 | ||
5211 | return 2; | |
5212 | } | |
5213 | } | |
5214 | ||
5215 | if (st->update_tail) { | |
5216 | struct imsm_update_kill_array *u = malloc(sizeof(*u)); | |
5217 | ||
5218 | if (!u) | |
5219 | return 2; | |
5220 | u->type = update_kill_array; | |
5221 | u->dev_idx = current_vol; | |
5222 | append_metadata_update(st, u, sizeof(*u)); | |
5223 | ||
5224 | return 0; | |
5225 | } | |
5226 | ||
5227 | for (dp = &super->devlist; *dp;) | |
5228 | if ((*dp)->index == current_vol) { | |
5229 | *dp = (*dp)->next; | |
5230 | } else { | |
5231 | handle_missing(super, (*dp)->dev); | |
5232 | if ((*dp)->index > current_vol) | |
5233 | (*dp)->index--; | |
5234 | dp = &(*dp)->next; | |
5235 | } | |
5236 | ||
5237 | /* no more raid devices, all active components are now spares, | |
5238 | * but of course failed are still failed | |
5239 | */ | |
5240 | if (--mpb->num_raid_devs == 0) { | |
5241 | struct dl *d; | |
5242 | ||
5243 | for (d = super->disks; d; d = d->next) | |
5244 | if (d->index > -2) { | |
5245 | d->index = -1; | |
5246 | d->disk.status = SPARE_DISK; | |
5247 | } | |
5248 | } | |
5249 | ||
5250 | super->updates_pending++; | |
5251 | ||
5252 | return 0; | |
5253 | } | |
5254 | ||
5255 | static int update_subarray_imsm(struct supertype *st, char *subarray, | |
5256 | char *update, struct mddev_ident *ident) | |
5257 | { | |
5258 | /* update the subarray currently referenced by ->current_vol */ | |
5259 | struct intel_super *super = st->sb; | |
5260 | struct imsm_super *mpb = super->anchor; | |
5261 | ||
5262 | if (strcmp(update, "name") == 0) { | |
5263 | char *name = ident->name; | |
5264 | char *ep; | |
5265 | int vol; | |
5266 | ||
5267 | if (is_subarray_active(subarray, st->devname)) { | |
5268 | fprintf(stderr, | |
5269 | Name ": Unable to update name of active subarray\n"); | |
5270 | return 2; | |
5271 | } | |
5272 | ||
5273 | if (!check_name(super, name, 0)) | |
5274 | return 2; | |
5275 | ||
5276 | vol = strtoul(subarray, &ep, 10); | |
5277 | if (*ep != '\0' || vol >= super->anchor->num_raid_devs) | |
5278 | return 2; | |
5279 | ||
5280 | if (st->update_tail) { | |
5281 | struct imsm_update_rename_array *u = malloc(sizeof(*u)); | |
5282 | ||
5283 | if (!u) | |
5284 | return 2; | |
5285 | u->type = update_rename_array; | |
5286 | u->dev_idx = vol; | |
5287 | snprintf((char *) u->name, MAX_RAID_SERIAL_LEN, "%s", name); | |
5288 | append_metadata_update(st, u, sizeof(*u)); | |
5289 | } else { | |
5290 | struct imsm_dev *dev; | |
5291 | int i; | |
5292 | ||
5293 | dev = get_imsm_dev(super, vol); | |
5294 | snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name); | |
5295 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5296 | dev = get_imsm_dev(super, i); | |
5297 | handle_missing(super, dev); | |
5298 | } | |
5299 | super->updates_pending++; | |
5300 | } | |
5301 | } else | |
5302 | return 2; | |
5303 | ||
5304 | return 0; | |
5305 | } | |
5306 | ||
5307 | static int is_gen_migration(struct imsm_dev *dev) | |
5308 | { | |
5309 | if (!dev->vol.migr_state) | |
5310 | return 0; | |
5311 | ||
5312 | if (migr_type(dev) == MIGR_GEN_MIGR) | |
5313 | return 1; | |
5314 | ||
5315 | return 0; | |
5316 | } | |
5317 | #endif /* MDASSEMBLE */ | |
5318 | ||
5319 | static int is_rebuilding(struct imsm_dev *dev) | |
5320 | { | |
5321 | struct imsm_map *migr_map; | |
5322 | ||
5323 | if (!dev->vol.migr_state) | |
5324 | return 0; | |
5325 | ||
5326 | if (migr_type(dev) != MIGR_REBUILD) | |
5327 | return 0; | |
5328 | ||
5329 | migr_map = get_imsm_map(dev, 1); | |
5330 | ||
5331 | if (migr_map->map_state == IMSM_T_STATE_DEGRADED) | |
5332 | return 1; | |
5333 | else | |
5334 | return 0; | |
5335 | } | |
5336 | ||
5337 | static void update_recovery_start(struct intel_super *super, | |
5338 | struct imsm_dev *dev, | |
5339 | struct mdinfo *array) | |
5340 | { | |
5341 | struct mdinfo *rebuild = NULL; | |
5342 | struct mdinfo *d; | |
5343 | __u32 units; | |
5344 | ||
5345 | if (!is_rebuilding(dev)) | |
5346 | return; | |
5347 | ||
5348 | /* Find the rebuild target, but punt on the dual rebuild case */ | |
5349 | for (d = array->devs; d; d = d->next) | |
5350 | if (d->recovery_start == 0) { | |
5351 | if (rebuild) | |
5352 | return; | |
5353 | rebuild = d; | |
5354 | } | |
5355 | ||
5356 | if (!rebuild) { | |
5357 | /* (?) none of the disks are marked with | |
5358 | * IMSM_ORD_REBUILD, so assume they are missing and the | |
5359 | * disk_ord_tbl was not correctly updated | |
5360 | */ | |
5361 | dprintf("%s: failed to locate out-of-sync disk\n", __func__); | |
5362 | return; | |
5363 | } | |
5364 | ||
5365 | units = __le32_to_cpu(dev->vol.curr_migr_unit); | |
5366 | rebuild->recovery_start = units * blocks_per_migr_unit(super, dev); | |
5367 | } | |
5368 | ||
5369 | static int recover_backup_imsm(struct supertype *st, struct mdinfo *info); | |
5370 | ||
5371 | static struct mdinfo *container_content_imsm(struct supertype *st, char *subarray) | |
5372 | { | |
5373 | /* Given a container loaded by load_super_imsm_all, | |
5374 | * extract information about all the arrays into | |
5375 | * an mdinfo tree. | |
5376 | * If 'subarray' is given, just extract info about that array. | |
5377 | * | |
5378 | * For each imsm_dev create an mdinfo, fill it in, | |
5379 | * then look for matching devices in super->disks | |
5380 | * and create appropriate device mdinfo. | |
5381 | */ | |
5382 | struct intel_super *super = st->sb; | |
5383 | struct imsm_super *mpb = super->anchor; | |
5384 | struct mdinfo *rest = NULL; | |
5385 | unsigned int i; | |
5386 | int bbm_errors = 0; | |
5387 | struct dl *d; | |
5388 | int spare_disks = 0; | |
5389 | ||
5390 | /* check for bad blocks */ | |
5391 | if (imsm_bbm_log_size(super->anchor)) | |
5392 | bbm_errors = 1; | |
5393 | ||
5394 | /* count spare devices, not used in maps | |
5395 | */ | |
5396 | for (d = super->disks; d; d = d->next) | |
5397 | if (d->index == -1) | |
5398 | spare_disks++; | |
5399 | ||
5400 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5401 | struct imsm_dev *dev; | |
5402 | struct imsm_map *map; | |
5403 | struct imsm_map *map2; | |
5404 | struct mdinfo *this; | |
5405 | int slot, chunk; | |
5406 | char *ep; | |
5407 | ||
5408 | if (subarray && | |
5409 | (i != strtoul(subarray, &ep, 10) || *ep != '\0')) | |
5410 | continue; | |
5411 | ||
5412 | dev = get_imsm_dev(super, i); | |
5413 | map = get_imsm_map(dev, 0); | |
5414 | map2 = get_imsm_map(dev, 1); | |
5415 | ||
5416 | /* do not publish arrays that are in the middle of an | |
5417 | * unsupported migration | |
5418 | */ | |
5419 | if (dev->vol.migr_state && | |
5420 | (migr_type(dev) == MIGR_STATE_CHANGE)) { | |
5421 | fprintf(stderr, Name ": cannot assemble volume '%.16s':" | |
5422 | " unsupported migration in progress\n", | |
5423 | dev->volume); | |
5424 | continue; | |
5425 | } | |
5426 | /* do not publish arrays that are not support by controller's | |
5427 | * OROM/EFI | |
5428 | */ | |
5429 | ||
5430 | chunk = __le16_to_cpu(map->blocks_per_strip) >> 1; | |
5431 | #ifndef MDASSEMBLE | |
5432 | if (!validate_geometry_imsm_orom(super, | |
5433 | get_imsm_raid_level(map), /* RAID level */ | |
5434 | imsm_level_to_layout(get_imsm_raid_level(map)), | |
5435 | map->num_members, /* raid disks */ | |
5436 | &chunk, | |
5437 | 1 /* verbose */)) { | |
5438 | fprintf(stderr, Name ": RAID gemetry validation failed. " | |
5439 | "Cannot proceed with the action(s).\n"); | |
5440 | continue; | |
5441 | } | |
5442 | #endif /* MDASSEMBLE */ | |
5443 | this = malloc(sizeof(*this)); | |
5444 | if (!this) { | |
5445 | fprintf(stderr, Name ": failed to allocate %zu bytes\n", | |
5446 | sizeof(*this)); | |
5447 | break; | |
5448 | } | |
5449 | ||
5450 | super->current_vol = i; | |
5451 | getinfo_super_imsm_volume(st, this, NULL); | |
5452 | this->next = rest; | |
5453 | for (slot = 0 ; slot < map->num_members; slot++) { | |
5454 | unsigned long long recovery_start; | |
5455 | struct mdinfo *info_d; | |
5456 | struct dl *d; | |
5457 | int idx; | |
5458 | int skip; | |
5459 | __u32 ord; | |
5460 | ||
5461 | skip = 0; | |
5462 | idx = get_imsm_disk_idx(dev, slot, 0); | |
5463 | ord = get_imsm_ord_tbl_ent(dev, slot, -1); | |
5464 | for (d = super->disks; d ; d = d->next) | |
5465 | if (d->index == idx) | |
5466 | break; | |
5467 | ||
5468 | recovery_start = MaxSector; | |
5469 | if (d == NULL) | |
5470 | skip = 1; | |
5471 | if (d && is_failed(&d->disk)) | |
5472 | skip = 1; | |
5473 | if (ord & IMSM_ORD_REBUILD) | |
5474 | recovery_start = 0; | |
5475 | ||
5476 | /* | |
5477 | * if we skip some disks the array will be assmebled degraded; | |
5478 | * reset resync start to avoid a dirty-degraded | |
5479 | * situation when performing the intial sync | |
5480 | * | |
5481 | * FIXME handle dirty degraded | |
5482 | */ | |
5483 | if ((skip || recovery_start == 0) && !dev->vol.dirty) | |
5484 | this->resync_start = MaxSector; | |
5485 | if (skip) | |
5486 | continue; | |
5487 | ||
5488 | info_d = calloc(1, sizeof(*info_d)); | |
5489 | if (!info_d) { | |
5490 | fprintf(stderr, Name ": failed to allocate disk" | |
5491 | " for volume %.16s\n", dev->volume); | |
5492 | info_d = this->devs; | |
5493 | while (info_d) { | |
5494 | struct mdinfo *d = info_d->next; | |
5495 | ||
5496 | free(info_d); | |
5497 | info_d = d; | |
5498 | } | |
5499 | free(this); | |
5500 | this = rest; | |
5501 | break; | |
5502 | } | |
5503 | info_d->next = this->devs; | |
5504 | this->devs = info_d; | |
5505 | ||
5506 | info_d->disk.number = d->index; | |
5507 | info_d->disk.major = d->major; | |
5508 | info_d->disk.minor = d->minor; | |
5509 | info_d->disk.raid_disk = slot; | |
5510 | info_d->recovery_start = recovery_start; | |
5511 | if (map2) { | |
5512 | if (slot < map2->num_members) | |
5513 | info_d->disk.state = (1 << MD_DISK_ACTIVE); | |
5514 | else | |
5515 | this->array.spare_disks++; | |
5516 | } else { | |
5517 | if (slot < map->num_members) | |
5518 | info_d->disk.state = (1 << MD_DISK_ACTIVE); | |
5519 | else | |
5520 | this->array.spare_disks++; | |
5521 | } | |
5522 | if (info_d->recovery_start == MaxSector) | |
5523 | this->array.working_disks++; | |
5524 | ||
5525 | info_d->events = __le32_to_cpu(mpb->generation_num); | |
5526 | info_d->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
5527 | info_d->component_size = __le32_to_cpu(map->blocks_per_member); | |
5528 | } | |
5529 | /* now that the disk list is up-to-date fixup recovery_start */ | |
5530 | update_recovery_start(super, dev, this); | |
5531 | this->array.spare_disks += spare_disks; | |
5532 | ||
5533 | /* check for reshape */ | |
5534 | if (this->reshape_active == 1) | |
5535 | recover_backup_imsm(st, this); | |
5536 | ||
5537 | rest = this; | |
5538 | } | |
5539 | ||
5540 | /* if array has bad blocks, set suitable bit in array status */ | |
5541 | if (bbm_errors) | |
5542 | rest->array.state |= (1<<MD_SB_BBM_ERRORS); | |
5543 | ||
5544 | return rest; | |
5545 | } | |
5546 | ||
5547 | ||
5548 | static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed) | |
5549 | { | |
5550 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5551 | ||
5552 | if (!failed) | |
5553 | return map->map_state == IMSM_T_STATE_UNINITIALIZED ? | |
5554 | IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL; | |
5555 | ||
5556 | switch (get_imsm_raid_level(map)) { | |
5557 | case 0: | |
5558 | return IMSM_T_STATE_FAILED; | |
5559 | break; | |
5560 | case 1: | |
5561 | if (failed < map->num_members) | |
5562 | return IMSM_T_STATE_DEGRADED; | |
5563 | else | |
5564 | return IMSM_T_STATE_FAILED; | |
5565 | break; | |
5566 | case 10: | |
5567 | { | |
5568 | /** | |
5569 | * check to see if any mirrors have failed, otherwise we | |
5570 | * are degraded. Even numbered slots are mirrored on | |
5571 | * slot+1 | |
5572 | */ | |
5573 | int i; | |
5574 | /* gcc -Os complains that this is unused */ | |
5575 | int insync = insync; | |
5576 | ||
5577 | for (i = 0; i < map->num_members; i++) { | |
5578 | __u32 ord = get_imsm_ord_tbl_ent(dev, i, -1); | |
5579 | int idx = ord_to_idx(ord); | |
5580 | struct imsm_disk *disk; | |
5581 | ||
5582 | /* reset the potential in-sync count on even-numbered | |
5583 | * slots. num_copies is always 2 for imsm raid10 | |
5584 | */ | |
5585 | if ((i & 1) == 0) | |
5586 | insync = 2; | |
5587 | ||
5588 | disk = get_imsm_disk(super, idx); | |
5589 | if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD) | |
5590 | insync--; | |
5591 | ||
5592 | /* no in-sync disks left in this mirror the | |
5593 | * array has failed | |
5594 | */ | |
5595 | if (insync == 0) | |
5596 | return IMSM_T_STATE_FAILED; | |
5597 | } | |
5598 | ||
5599 | return IMSM_T_STATE_DEGRADED; | |
5600 | } | |
5601 | case 5: | |
5602 | if (failed < 2) | |
5603 | return IMSM_T_STATE_DEGRADED; | |
5604 | else | |
5605 | return IMSM_T_STATE_FAILED; | |
5606 | break; | |
5607 | default: | |
5608 | break; | |
5609 | } | |
5610 | ||
5611 | return map->map_state; | |
5612 | } | |
5613 | ||
5614 | static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev) | |
5615 | { | |
5616 | int i; | |
5617 | int failed = 0; | |
5618 | struct imsm_disk *disk; | |
5619 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5620 | struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state); | |
5621 | __u32 ord; | |
5622 | int idx; | |
5623 | ||
5624 | /* at the beginning of migration we set IMSM_ORD_REBUILD on | |
5625 | * disks that are being rebuilt. New failures are recorded to | |
5626 | * map[0]. So we look through all the disks we started with and | |
5627 | * see if any failures are still present, or if any new ones | |
5628 | * have arrived | |
5629 | * | |
5630 | * FIXME add support for online capacity expansion and | |
5631 | * raid-level-migration | |
5632 | */ | |
5633 | for (i = 0; i < prev->num_members; i++) { | |
5634 | ord = __le32_to_cpu(prev->disk_ord_tbl[i]); | |
5635 | ord |= __le32_to_cpu(map->disk_ord_tbl[i]); | |
5636 | idx = ord_to_idx(ord); | |
5637 | ||
5638 | disk = get_imsm_disk(super, idx); | |
5639 | if (!disk || is_failed(disk) || ord & IMSM_ORD_REBUILD) | |
5640 | failed++; | |
5641 | } | |
5642 | ||
5643 | return failed; | |
5644 | } | |
5645 | ||
5646 | #ifndef MDASSEMBLE | |
5647 | static int imsm_open_new(struct supertype *c, struct active_array *a, | |
5648 | char *inst) | |
5649 | { | |
5650 | struct intel_super *super = c->sb; | |
5651 | struct imsm_super *mpb = super->anchor; | |
5652 | ||
5653 | if (atoi(inst) >= mpb->num_raid_devs) { | |
5654 | fprintf(stderr, "%s: subarry index %d, out of range\n", | |
5655 | __func__, atoi(inst)); | |
5656 | return -ENODEV; | |
5657 | } | |
5658 | ||
5659 | dprintf("imsm: open_new %s\n", inst); | |
5660 | a->info.container_member = atoi(inst); | |
5661 | return 0; | |
5662 | } | |
5663 | ||
5664 | static int is_resyncing(struct imsm_dev *dev) | |
5665 | { | |
5666 | struct imsm_map *migr_map; | |
5667 | ||
5668 | if (!dev->vol.migr_state) | |
5669 | return 0; | |
5670 | ||
5671 | if (migr_type(dev) == MIGR_INIT || | |
5672 | migr_type(dev) == MIGR_REPAIR) | |
5673 | return 1; | |
5674 | ||
5675 | if (migr_type(dev) == MIGR_GEN_MIGR) | |
5676 | return 0; | |
5677 | ||
5678 | migr_map = get_imsm_map(dev, 1); | |
5679 | ||
5680 | if ((migr_map->map_state == IMSM_T_STATE_NORMAL) && | |
5681 | (dev->vol.migr_type != MIGR_GEN_MIGR)) | |
5682 | return 1; | |
5683 | else | |
5684 | return 0; | |
5685 | } | |
5686 | ||
5687 | /* return true if we recorded new information */ | |
5688 | static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx) | |
5689 | { | |
5690 | __u32 ord; | |
5691 | int slot; | |
5692 | struct imsm_map *map; | |
5693 | ||
5694 | /* new failures are always set in map[0] */ | |
5695 | map = get_imsm_map(dev, 0); | |
5696 | ||
5697 | slot = get_imsm_disk_slot(map, idx); | |
5698 | if (slot < 0) | |
5699 | return 0; | |
5700 | ||
5701 | ord = __le32_to_cpu(map->disk_ord_tbl[slot]); | |
5702 | if (is_failed(disk) && (ord & IMSM_ORD_REBUILD)) | |
5703 | return 0; | |
5704 | ||
5705 | disk->status |= FAILED_DISK; | |
5706 | set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD); | |
5707 | if (map->failed_disk_num == 0xff) | |
5708 | map->failed_disk_num = slot; | |
5709 | return 1; | |
5710 | } | |
5711 | ||
5712 | static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx) | |
5713 | { | |
5714 | mark_failure(dev, disk, idx); | |
5715 | ||
5716 | if (disk->scsi_id == __cpu_to_le32(~(__u32)0)) | |
5717 | return; | |
5718 | ||
5719 | disk->scsi_id = __cpu_to_le32(~(__u32)0); | |
5720 | memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1); | |
5721 | } | |
5722 | ||
5723 | static void handle_missing(struct intel_super *super, struct imsm_dev *dev) | |
5724 | { | |
5725 | __u8 map_state; | |
5726 | struct dl *dl; | |
5727 | int failed; | |
5728 | ||
5729 | if (!super->missing) | |
5730 | return; | |
5731 | failed = imsm_count_failed(super, dev); | |
5732 | map_state = imsm_check_degraded(super, dev, failed); | |
5733 | ||
5734 | dprintf("imsm: mark missing\n"); | |
5735 | end_migration(dev, map_state); | |
5736 | for (dl = super->missing; dl; dl = dl->next) | |
5737 | mark_missing(dev, &dl->disk, dl->index); | |
5738 | super->updates_pending++; | |
5739 | } | |
5740 | ||
5741 | static unsigned long long imsm_set_array_size(struct imsm_dev *dev) | |
5742 | { | |
5743 | int used_disks = imsm_num_data_members(dev, 0); | |
5744 | unsigned long long array_blocks; | |
5745 | struct imsm_map *map; | |
5746 | ||
5747 | if (used_disks == 0) { | |
5748 | /* when problems occures | |
5749 | * return current array_blocks value | |
5750 | */ | |
5751 | array_blocks = __le32_to_cpu(dev->size_high); | |
5752 | array_blocks = array_blocks << 32; | |
5753 | array_blocks += __le32_to_cpu(dev->size_low); | |
5754 | ||
5755 | return array_blocks; | |
5756 | } | |
5757 | ||
5758 | /* set array size in metadata | |
5759 | */ | |
5760 | map = get_imsm_map(dev, 0); | |
5761 | array_blocks = map->blocks_per_member * used_disks; | |
5762 | ||
5763 | /* round array size down to closest MB | |
5764 | */ | |
5765 | array_blocks = (array_blocks >> SECT_PER_MB_SHIFT) << SECT_PER_MB_SHIFT; | |
5766 | dev->size_low = __cpu_to_le32((__u32)array_blocks); | |
5767 | dev->size_high = __cpu_to_le32((__u32)(array_blocks >> 32)); | |
5768 | ||
5769 | return array_blocks; | |
5770 | } | |
5771 | ||
5772 | static void imsm_set_disk(struct active_array *a, int n, int state); | |
5773 | ||
5774 | static void imsm_progress_container_reshape(struct intel_super *super) | |
5775 | { | |
5776 | /* if no device has a migr_state, but some device has a | |
5777 | * different number of members than the previous device, start | |
5778 | * changing the number of devices in this device to match | |
5779 | * previous. | |
5780 | */ | |
5781 | struct imsm_super *mpb = super->anchor; | |
5782 | int prev_disks = -1; | |
5783 | int i; | |
5784 | int copy_map_size; | |
5785 | ||
5786 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
5787 | struct imsm_dev *dev = get_imsm_dev(super, i); | |
5788 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5789 | struct imsm_map *map2; | |
5790 | int prev_num_members; | |
5791 | ||
5792 | if (dev->vol.migr_state) | |
5793 | return; | |
5794 | ||
5795 | if (prev_disks == -1) | |
5796 | prev_disks = map->num_members; | |
5797 | if (prev_disks == map->num_members) | |
5798 | continue; | |
5799 | ||
5800 | /* OK, this array needs to enter reshape mode. | |
5801 | * i.e it needs a migr_state | |
5802 | */ | |
5803 | ||
5804 | copy_map_size = sizeof_imsm_map(map); | |
5805 | prev_num_members = map->num_members; | |
5806 | map->num_members = prev_disks; | |
5807 | dev->vol.migr_state = 1; | |
5808 | dev->vol.curr_migr_unit = 0; | |
5809 | dev->vol.migr_type = MIGR_GEN_MIGR; | |
5810 | for (i = prev_num_members; | |
5811 | i < map->num_members; i++) | |
5812 | set_imsm_ord_tbl_ent(map, i, i); | |
5813 | map2 = get_imsm_map(dev, 1); | |
5814 | /* Copy the current map */ | |
5815 | memcpy(map2, map, copy_map_size); | |
5816 | map2->num_members = prev_num_members; | |
5817 | ||
5818 | imsm_set_array_size(dev); | |
5819 | super->updates_pending++; | |
5820 | } | |
5821 | } | |
5822 | ||
5823 | /* Handle dirty -> clean transititions, resync and reshape. Degraded and rebuild | |
5824 | * states are handled in imsm_set_disk() with one exception, when a | |
5825 | * resync is stopped due to a new failure this routine will set the | |
5826 | * 'degraded' state for the array. | |
5827 | */ | |
5828 | static int imsm_set_array_state(struct active_array *a, int consistent) | |
5829 | { | |
5830 | int inst = a->info.container_member; | |
5831 | struct intel_super *super = a->container->sb; | |
5832 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
5833 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5834 | int failed = imsm_count_failed(super, dev); | |
5835 | __u8 map_state = imsm_check_degraded(super, dev, failed); | |
5836 | __u32 blocks_per_unit; | |
5837 | ||
5838 | if (dev->vol.migr_state && | |
5839 | dev->vol.migr_type == MIGR_GEN_MIGR) { | |
5840 | /* array state change is blocked due to reshape action | |
5841 | * We might need to | |
5842 | * - abort the reshape (if last_checkpoint is 0 and action!= reshape) | |
5843 | * - finish the reshape (if last_checkpoint is big and action != reshape) | |
5844 | * - update curr_migr_unit | |
5845 | */ | |
5846 | if (a->curr_action == reshape) { | |
5847 | /* still reshaping, maybe update curr_migr_unit */ | |
5848 | goto mark_checkpoint; | |
5849 | } else { | |
5850 | if (a->last_checkpoint == 0 && a->prev_action == reshape) { | |
5851 | /* for some reason we aborted the reshape. | |
5852 | * Better clean up | |
5853 | */ | |
5854 | struct imsm_map *map2 = get_imsm_map(dev, 1); | |
5855 | dev->vol.migr_state = 0; | |
5856 | dev->vol.migr_type = 0; | |
5857 | dev->vol.curr_migr_unit = 0; | |
5858 | memcpy(map, map2, sizeof_imsm_map(map2)); | |
5859 | super->updates_pending++; | |
5860 | } | |
5861 | if (a->last_checkpoint >= a->info.component_size) { | |
5862 | unsigned long long array_blocks; | |
5863 | int used_disks; | |
5864 | struct mdinfo *mdi; | |
5865 | ||
5866 | used_disks = imsm_num_data_members(dev, 0); | |
5867 | if (used_disks > 0) { | |
5868 | array_blocks = | |
5869 | map->blocks_per_member * | |
5870 | used_disks; | |
5871 | /* round array size down to closest MB | |
5872 | */ | |
5873 | array_blocks = (array_blocks | |
5874 | >> SECT_PER_MB_SHIFT) | |
5875 | << SECT_PER_MB_SHIFT; | |
5876 | a->info.custom_array_size = array_blocks; | |
5877 | /* encourage manager to update array | |
5878 | * size | |
5879 | */ | |
5880 | ||
5881 | a->check_reshape = 1; | |
5882 | } | |
5883 | /* finalize online capacity expansion/reshape */ | |
5884 | for (mdi = a->info.devs; mdi; mdi = mdi->next) | |
5885 | imsm_set_disk(a, | |
5886 | mdi->disk.raid_disk, | |
5887 | mdi->curr_state); | |
5888 | ||
5889 | imsm_progress_container_reshape(super); | |
5890 | } | |
5891 | } | |
5892 | } | |
5893 | ||
5894 | /* before we activate this array handle any missing disks */ | |
5895 | if (consistent == 2) | |
5896 | handle_missing(super, dev); | |
5897 | ||
5898 | if (consistent == 2 && | |
5899 | (!is_resync_complete(&a->info) || | |
5900 | map_state != IMSM_T_STATE_NORMAL || | |
5901 | dev->vol.migr_state)) | |
5902 | consistent = 0; | |
5903 | ||
5904 | if (is_resync_complete(&a->info)) { | |
5905 | /* complete intialization / resync, | |
5906 | * recovery and interrupted recovery is completed in | |
5907 | * ->set_disk | |
5908 | */ | |
5909 | if (is_resyncing(dev)) { | |
5910 | dprintf("imsm: mark resync done\n"); | |
5911 | end_migration(dev, map_state); | |
5912 | super->updates_pending++; | |
5913 | a->last_checkpoint = 0; | |
5914 | } | |
5915 | } else if (!is_resyncing(dev) && !failed) { | |
5916 | /* mark the start of the init process if nothing is failed */ | |
5917 | dprintf("imsm: mark resync start\n"); | |
5918 | if (map->map_state == IMSM_T_STATE_UNINITIALIZED) | |
5919 | migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_INIT); | |
5920 | else | |
5921 | migrate(dev, super, IMSM_T_STATE_NORMAL, MIGR_REPAIR); | |
5922 | super->updates_pending++; | |
5923 | } | |
5924 | ||
5925 | mark_checkpoint: | |
5926 | /* skip checkpointing for general migration, | |
5927 | * it is controlled in mdadm | |
5928 | */ | |
5929 | if (is_gen_migration(dev)) | |
5930 | goto skip_mark_checkpoint; | |
5931 | ||
5932 | /* check if we can update curr_migr_unit from resync_start, recovery_start */ | |
5933 | blocks_per_unit = blocks_per_migr_unit(super, dev); | |
5934 | if (blocks_per_unit) { | |
5935 | __u32 units32; | |
5936 | __u64 units; | |
5937 | ||
5938 | units = a->last_checkpoint / blocks_per_unit; | |
5939 | units32 = units; | |
5940 | ||
5941 | /* check that we did not overflow 32-bits, and that | |
5942 | * curr_migr_unit needs updating | |
5943 | */ | |
5944 | if (units32 == units && | |
5945 | units32 != 0 && | |
5946 | __le32_to_cpu(dev->vol.curr_migr_unit) != units32) { | |
5947 | dprintf("imsm: mark checkpoint (%u)\n", units32); | |
5948 | dev->vol.curr_migr_unit = __cpu_to_le32(units32); | |
5949 | super->updates_pending++; | |
5950 | } | |
5951 | } | |
5952 | ||
5953 | skip_mark_checkpoint: | |
5954 | /* mark dirty / clean */ | |
5955 | if (dev->vol.dirty != !consistent) { | |
5956 | dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty"); | |
5957 | if (consistent) | |
5958 | dev->vol.dirty = 0; | |
5959 | else | |
5960 | dev->vol.dirty = 1; | |
5961 | super->updates_pending++; | |
5962 | } | |
5963 | ||
5964 | return consistent; | |
5965 | } | |
5966 | ||
5967 | static void imsm_set_disk(struct active_array *a, int n, int state) | |
5968 | { | |
5969 | int inst = a->info.container_member; | |
5970 | struct intel_super *super = a->container->sb; | |
5971 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
5972 | struct imsm_map *map = get_imsm_map(dev, 0); | |
5973 | struct imsm_disk *disk; | |
5974 | int failed; | |
5975 | __u32 ord; | |
5976 | __u8 map_state; | |
5977 | ||
5978 | if (n > map->num_members) | |
5979 | fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n", | |
5980 | n, map->num_members - 1); | |
5981 | ||
5982 | if (n < 0) | |
5983 | return; | |
5984 | ||
5985 | dprintf("imsm: set_disk %d:%x\n", n, state); | |
5986 | ||
5987 | ord = get_imsm_ord_tbl_ent(dev, n, -1); | |
5988 | disk = get_imsm_disk(super, ord_to_idx(ord)); | |
5989 | ||
5990 | /* check for new failures */ | |
5991 | if (state & DS_FAULTY) { | |
5992 | if (mark_failure(dev, disk, ord_to_idx(ord))) | |
5993 | super->updates_pending++; | |
5994 | } | |
5995 | ||
5996 | /* check if in_sync */ | |
5997 | if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) { | |
5998 | struct imsm_map *migr_map = get_imsm_map(dev, 1); | |
5999 | ||
6000 | set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord)); | |
6001 | super->updates_pending++; | |
6002 | } | |
6003 | ||
6004 | failed = imsm_count_failed(super, dev); | |
6005 | map_state = imsm_check_degraded(super, dev, failed); | |
6006 | ||
6007 | /* check if recovery complete, newly degraded, or failed */ | |
6008 | if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) { | |
6009 | end_migration(dev, map_state); | |
6010 | map = get_imsm_map(dev, 0); | |
6011 | map->failed_disk_num = ~0; | |
6012 | super->updates_pending++; | |
6013 | a->last_checkpoint = 0; | |
6014 | } else if (map_state == IMSM_T_STATE_DEGRADED && | |
6015 | map->map_state != map_state && | |
6016 | !dev->vol.migr_state) { | |
6017 | dprintf("imsm: mark degraded\n"); | |
6018 | map->map_state = map_state; | |
6019 | super->updates_pending++; | |
6020 | a->last_checkpoint = 0; | |
6021 | } else if (map_state == IMSM_T_STATE_FAILED && | |
6022 | map->map_state != map_state) { | |
6023 | dprintf("imsm: mark failed\n"); | |
6024 | end_migration(dev, map_state); | |
6025 | super->updates_pending++; | |
6026 | a->last_checkpoint = 0; | |
6027 | } else if (is_gen_migration(dev)) { | |
6028 | dprintf("imsm: Detected General Migration in state: "); | |
6029 | if (map_state == IMSM_T_STATE_NORMAL) { | |
6030 | end_migration(dev, map_state); | |
6031 | map = get_imsm_map(dev, 0); | |
6032 | map->failed_disk_num = ~0; | |
6033 | dprintf("normal\n"); | |
6034 | } else { | |
6035 | if (map_state == IMSM_T_STATE_DEGRADED) { | |
6036 | printf("degraded\n"); | |
6037 | end_migration(dev, map_state); | |
6038 | } else { | |
6039 | dprintf("failed\n"); | |
6040 | } | |
6041 | map->map_state = map_state; | |
6042 | } | |
6043 | super->updates_pending++; | |
6044 | } | |
6045 | } | |
6046 | ||
6047 | static int store_imsm_mpb(int fd, struct imsm_super *mpb) | |
6048 | { | |
6049 | void *buf = mpb; | |
6050 | __u32 mpb_size = __le32_to_cpu(mpb->mpb_size); | |
6051 | unsigned long long dsize; | |
6052 | unsigned long long sectors; | |
6053 | ||
6054 | get_dev_size(fd, NULL, &dsize); | |
6055 | ||
6056 | if (mpb_size > 512) { | |
6057 | /* -1 to account for anchor */ | |
6058 | sectors = mpb_sectors(mpb) - 1; | |
6059 | ||
6060 | /* write the extended mpb to the sectors preceeding the anchor */ | |
6061 | if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) | |
6062 | return 1; | |
6063 | ||
6064 | if ((unsigned long long)write(fd, buf + 512, 512 * sectors) | |
6065 | != 512 * sectors) | |
6066 | return 1; | |
6067 | } | |
6068 | ||
6069 | /* first block is stored on second to last sector of the disk */ | |
6070 | if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) | |
6071 | return 1; | |
6072 | ||
6073 | if (write(fd, buf, 512) != 512) | |
6074 | return 1; | |
6075 | ||
6076 | return 0; | |
6077 | } | |
6078 | ||
6079 | static void imsm_sync_metadata(struct supertype *container) | |
6080 | { | |
6081 | struct intel_super *super = container->sb; | |
6082 | ||
6083 | dprintf("sync metadata: %d\n", super->updates_pending); | |
6084 | if (!super->updates_pending) | |
6085 | return; | |
6086 | ||
6087 | write_super_imsm(container, 0); | |
6088 | ||
6089 | super->updates_pending = 0; | |
6090 | } | |
6091 | ||
6092 | static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a) | |
6093 | { | |
6094 | struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member); | |
6095 | int i = get_imsm_disk_idx(dev, idx, -1); | |
6096 | struct dl *dl; | |
6097 | ||
6098 | for (dl = super->disks; dl; dl = dl->next) | |
6099 | if (dl->index == i) | |
6100 | break; | |
6101 | ||
6102 | if (dl && is_failed(&dl->disk)) | |
6103 | dl = NULL; | |
6104 | ||
6105 | if (dl) | |
6106 | dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor); | |
6107 | ||
6108 | return dl; | |
6109 | } | |
6110 | ||
6111 | static struct dl *imsm_add_spare(struct intel_super *super, int slot, | |
6112 | struct active_array *a, int activate_new, | |
6113 | struct mdinfo *additional_test_list) | |
6114 | { | |
6115 | struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member); | |
6116 | int idx = get_imsm_disk_idx(dev, slot, -1); | |
6117 | struct imsm_super *mpb = super->anchor; | |
6118 | struct imsm_map *map; | |
6119 | unsigned long long pos; | |
6120 | struct mdinfo *d; | |
6121 | struct extent *ex; | |
6122 | int i, j; | |
6123 | int found; | |
6124 | __u32 array_start = 0; | |
6125 | __u32 array_end = 0; | |
6126 | struct dl *dl; | |
6127 | struct mdinfo *test_list; | |
6128 | ||
6129 | for (dl = super->disks; dl; dl = dl->next) { | |
6130 | /* If in this array, skip */ | |
6131 | for (d = a->info.devs ; d ; d = d->next) | |
6132 | if (d->state_fd >= 0 && | |
6133 | d->disk.major == dl->major && | |
6134 | d->disk.minor == dl->minor) { | |
6135 | dprintf("%x:%x already in array\n", | |
6136 | dl->major, dl->minor); | |
6137 | break; | |
6138 | } | |
6139 | if (d) | |
6140 | continue; | |
6141 | test_list = additional_test_list; | |
6142 | while (test_list) { | |
6143 | if (test_list->disk.major == dl->major && | |
6144 | test_list->disk.minor == dl->minor) { | |
6145 | dprintf("%x:%x already in additional test list\n", | |
6146 | dl->major, dl->minor); | |
6147 | break; | |
6148 | } | |
6149 | test_list = test_list->next; | |
6150 | } | |
6151 | if (test_list) | |
6152 | continue; | |
6153 | ||
6154 | /* skip in use or failed drives */ | |
6155 | if (is_failed(&dl->disk) || idx == dl->index || | |
6156 | dl->index == -2) { | |
6157 | dprintf("%x:%x status (failed: %d index: %d)\n", | |
6158 | dl->major, dl->minor, is_failed(&dl->disk), idx); | |
6159 | continue; | |
6160 | } | |
6161 | ||
6162 | /* skip pure spares when we are looking for partially | |
6163 | * assimilated drives | |
6164 | */ | |
6165 | if (dl->index == -1 && !activate_new) | |
6166 | continue; | |
6167 | ||
6168 | /* Does this unused device have the requisite free space? | |
6169 | * It needs to be able to cover all member volumes | |
6170 | */ | |
6171 | ex = get_extents(super, dl); | |
6172 | if (!ex) { | |
6173 | dprintf("cannot get extents\n"); | |
6174 | continue; | |
6175 | } | |
6176 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
6177 | dev = get_imsm_dev(super, i); | |
6178 | map = get_imsm_map(dev, 0); | |
6179 | ||
6180 | /* check if this disk is already a member of | |
6181 | * this array | |
6182 | */ | |
6183 | if (get_imsm_disk_slot(map, dl->index) >= 0) | |
6184 | continue; | |
6185 | ||
6186 | found = 0; | |
6187 | j = 0; | |
6188 | pos = 0; | |
6189 | array_start = __le32_to_cpu(map->pba_of_lba0); | |
6190 | array_end = array_start + | |
6191 | __le32_to_cpu(map->blocks_per_member) - 1; | |
6192 | ||
6193 | do { | |
6194 | /* check that we can start at pba_of_lba0 with | |
6195 | * blocks_per_member of space | |
6196 | */ | |
6197 | if (array_start >= pos && array_end < ex[j].start) { | |
6198 | found = 1; | |
6199 | break; | |
6200 | } | |
6201 | pos = ex[j].start + ex[j].size; | |
6202 | j++; | |
6203 | } while (ex[j-1].size); | |
6204 | ||
6205 | if (!found) | |
6206 | break; | |
6207 | } | |
6208 | ||
6209 | free(ex); | |
6210 | if (i < mpb->num_raid_devs) { | |
6211 | dprintf("%x:%x does not have %u to %u available\n", | |
6212 | dl->major, dl->minor, array_start, array_end); | |
6213 | /* No room */ | |
6214 | continue; | |
6215 | } | |
6216 | return dl; | |
6217 | } | |
6218 | ||
6219 | return dl; | |
6220 | } | |
6221 | ||
6222 | ||
6223 | static int imsm_rebuild_allowed(struct supertype *cont, int dev_idx, int failed) | |
6224 | { | |
6225 | struct imsm_dev *dev2; | |
6226 | struct imsm_map *map; | |
6227 | struct dl *idisk; | |
6228 | int slot; | |
6229 | int idx; | |
6230 | __u8 state; | |
6231 | ||
6232 | dev2 = get_imsm_dev(cont->sb, dev_idx); | |
6233 | if (dev2) { | |
6234 | state = imsm_check_degraded(cont->sb, dev2, failed); | |
6235 | if (state == IMSM_T_STATE_FAILED) { | |
6236 | map = get_imsm_map(dev2, 0); | |
6237 | if (!map) | |
6238 | return 1; | |
6239 | for (slot = 0; slot < map->num_members; slot++) { | |
6240 | /* | |
6241 | * Check if failed disks are deleted from intel | |
6242 | * disk list or are marked to be deleted | |
6243 | */ | |
6244 | idx = get_imsm_disk_idx(dev2, slot, -1); | |
6245 | idisk = get_imsm_dl_disk(cont->sb, idx); | |
6246 | /* | |
6247 | * Do not rebuild the array if failed disks | |
6248 | * from failed sub-array are not removed from | |
6249 | * container. | |
6250 | */ | |
6251 | if (idisk && | |
6252 | is_failed(&idisk->disk) && | |
6253 | (idisk->action != DISK_REMOVE)) | |
6254 | return 0; | |
6255 | } | |
6256 | } | |
6257 | } | |
6258 | return 1; | |
6259 | } | |
6260 | ||
6261 | static struct mdinfo *imsm_activate_spare(struct active_array *a, | |
6262 | struct metadata_update **updates) | |
6263 | { | |
6264 | /** | |
6265 | * Find a device with unused free space and use it to replace a | |
6266 | * failed/vacant region in an array. We replace failed regions one a | |
6267 | * array at a time. The result is that a new spare disk will be added | |
6268 | * to the first failed array and after the monitor has finished | |
6269 | * propagating failures the remainder will be consumed. | |
6270 | * | |
6271 | * FIXME add a capability for mdmon to request spares from another | |
6272 | * container. | |
6273 | */ | |
6274 | ||
6275 | struct intel_super *super = a->container->sb; | |
6276 | int inst = a->info.container_member; | |
6277 | struct imsm_dev *dev = get_imsm_dev(super, inst); | |
6278 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6279 | int failed = a->info.array.raid_disks; | |
6280 | struct mdinfo *rv = NULL; | |
6281 | struct mdinfo *d; | |
6282 | struct mdinfo *di; | |
6283 | struct metadata_update *mu; | |
6284 | struct dl *dl; | |
6285 | struct imsm_update_activate_spare *u; | |
6286 | int num_spares = 0; | |
6287 | int i; | |
6288 | int allowed; | |
6289 | ||
6290 | for (d = a->info.devs ; d ; d = d->next) { | |
6291 | if ((d->curr_state & DS_FAULTY) && | |
6292 | d->state_fd >= 0) | |
6293 | /* wait for Removal to happen */ | |
6294 | return NULL; | |
6295 | if (d->state_fd >= 0) | |
6296 | failed--; | |
6297 | } | |
6298 | ||
6299 | dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n", | |
6300 | inst, failed, a->info.array.raid_disks, a->info.array.level); | |
6301 | ||
6302 | if (dev->vol.migr_state && | |
6303 | dev->vol.migr_type == MIGR_GEN_MIGR) | |
6304 | /* No repair during migration */ | |
6305 | return NULL; | |
6306 | ||
6307 | if (a->info.array.level == 4) | |
6308 | /* No repair for takeovered array | |
6309 | * imsm doesn't support raid4 | |
6310 | */ | |
6311 | return NULL; | |
6312 | ||
6313 | if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED) | |
6314 | return NULL; | |
6315 | ||
6316 | /* | |
6317 | * If there are any failed disks check state of the other volume. | |
6318 | * Block rebuild if the another one is failed until failed disks | |
6319 | * are removed from container. | |
6320 | */ | |
6321 | if (failed) { | |
6322 | dprintf("found failed disks in %s, check if there another" | |
6323 | "failed sub-array.\n", | |
6324 | dev->volume); | |
6325 | /* check if states of the other volumes allow for rebuild */ | |
6326 | for (i = 0; i < super->anchor->num_raid_devs; i++) { | |
6327 | if (i != inst) { | |
6328 | allowed = imsm_rebuild_allowed(a->container, | |
6329 | i, failed); | |
6330 | if (!allowed) | |
6331 | return NULL; | |
6332 | } | |
6333 | } | |
6334 | } | |
6335 | ||
6336 | /* For each slot, if it is not working, find a spare */ | |
6337 | for (i = 0; i < a->info.array.raid_disks; i++) { | |
6338 | for (d = a->info.devs ; d ; d = d->next) | |
6339 | if (d->disk.raid_disk == i) | |
6340 | break; | |
6341 | dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0); | |
6342 | if (d && (d->state_fd >= 0)) | |
6343 | continue; | |
6344 | ||
6345 | /* | |
6346 | * OK, this device needs recovery. Try to re-add the | |
6347 | * previous occupant of this slot, if this fails see if | |
6348 | * we can continue the assimilation of a spare that was | |
6349 | * partially assimilated, finally try to activate a new | |
6350 | * spare. | |
6351 | */ | |
6352 | dl = imsm_readd(super, i, a); | |
6353 | if (!dl) | |
6354 | dl = imsm_add_spare(super, i, a, 0, NULL); | |
6355 | if (!dl) | |
6356 | dl = imsm_add_spare(super, i, a, 1, NULL); | |
6357 | if (!dl) | |
6358 | continue; | |
6359 | ||
6360 | /* found a usable disk with enough space */ | |
6361 | di = malloc(sizeof(*di)); | |
6362 | if (!di) | |
6363 | continue; | |
6364 | memset(di, 0, sizeof(*di)); | |
6365 | ||
6366 | /* dl->index will be -1 in the case we are activating a | |
6367 | * pristine spare. imsm_process_update() will create a | |
6368 | * new index in this case. Once a disk is found to be | |
6369 | * failed in all member arrays it is kicked from the | |
6370 | * metadata | |
6371 | */ | |
6372 | di->disk.number = dl->index; | |
6373 | ||
6374 | /* (ab)use di->devs to store a pointer to the device | |
6375 | * we chose | |
6376 | */ | |
6377 | di->devs = (struct mdinfo *) dl; | |
6378 | ||
6379 | di->disk.raid_disk = i; | |
6380 | di->disk.major = dl->major; | |
6381 | di->disk.minor = dl->minor; | |
6382 | di->disk.state = 0; | |
6383 | di->recovery_start = 0; | |
6384 | di->data_offset = __le32_to_cpu(map->pba_of_lba0); | |
6385 | di->component_size = a->info.component_size; | |
6386 | di->container_member = inst; | |
6387 | super->random = random32(); | |
6388 | di->next = rv; | |
6389 | rv = di; | |
6390 | num_spares++; | |
6391 | dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor, | |
6392 | i, di->data_offset); | |
6393 | ||
6394 | break; | |
6395 | } | |
6396 | ||
6397 | if (!rv) | |
6398 | /* No spares found */ | |
6399 | return rv; | |
6400 | /* Now 'rv' has a list of devices to return. | |
6401 | * Create a metadata_update record to update the | |
6402 | * disk_ord_tbl for the array | |
6403 | */ | |
6404 | mu = malloc(sizeof(*mu)); | |
6405 | if (mu) { | |
6406 | mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares); | |
6407 | if (mu->buf == NULL) { | |
6408 | free(mu); | |
6409 | mu = NULL; | |
6410 | } | |
6411 | } | |
6412 | if (!mu) { | |
6413 | while (rv) { | |
6414 | struct mdinfo *n = rv->next; | |
6415 | ||
6416 | free(rv); | |
6417 | rv = n; | |
6418 | } | |
6419 | return NULL; | |
6420 | } | |
6421 | ||
6422 | mu->space = NULL; | |
6423 | mu->space_list = NULL; | |
6424 | mu->len = sizeof(struct imsm_update_activate_spare) * num_spares; | |
6425 | mu->next = *updates; | |
6426 | u = (struct imsm_update_activate_spare *) mu->buf; | |
6427 | ||
6428 | for (di = rv ; di ; di = di->next) { | |
6429 | u->type = update_activate_spare; | |
6430 | u->dl = (struct dl *) di->devs; | |
6431 | di->devs = NULL; | |
6432 | u->slot = di->disk.raid_disk; | |
6433 | u->array = inst; | |
6434 | u->next = u + 1; | |
6435 | u++; | |
6436 | } | |
6437 | (u-1)->next = NULL; | |
6438 | *updates = mu; | |
6439 | ||
6440 | return rv; | |
6441 | } | |
6442 | ||
6443 | static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u) | |
6444 | { | |
6445 | struct imsm_dev *dev = get_imsm_dev(super, idx); | |
6446 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6447 | struct imsm_map *new_map = get_imsm_map(&u->dev, 0); | |
6448 | struct disk_info *inf = get_disk_info(u); | |
6449 | struct imsm_disk *disk; | |
6450 | int i; | |
6451 | int j; | |
6452 | ||
6453 | for (i = 0; i < map->num_members; i++) { | |
6454 | disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i, -1)); | |
6455 | for (j = 0; j < new_map->num_members; j++) | |
6456 | if (serialcmp(disk->serial, inf[j].serial) == 0) | |
6457 | return 1; | |
6458 | } | |
6459 | ||
6460 | return 0; | |
6461 | } | |
6462 | ||
6463 | ||
6464 | static struct dl *get_disk_super(struct intel_super *super, int major, int minor) | |
6465 | { | |
6466 | struct dl *dl = NULL; | |
6467 | for (dl = super->disks; dl; dl = dl->next) | |
6468 | if ((dl->major == major) && (dl->minor == minor)) | |
6469 | return dl; | |
6470 | return NULL; | |
6471 | } | |
6472 | ||
6473 | static int remove_disk_super(struct intel_super *super, int major, int minor) | |
6474 | { | |
6475 | struct dl *prev = NULL; | |
6476 | struct dl *dl; | |
6477 | ||
6478 | prev = NULL; | |
6479 | for (dl = super->disks; dl; dl = dl->next) { | |
6480 | if ((dl->major == major) && (dl->minor == minor)) { | |
6481 | /* remove */ | |
6482 | if (prev) | |
6483 | prev->next = dl->next; | |
6484 | else | |
6485 | super->disks = dl->next; | |
6486 | dl->next = NULL; | |
6487 | __free_imsm_disk(dl); | |
6488 | dprintf("%s: removed %x:%x\n", | |
6489 | __func__, major, minor); | |
6490 | break; | |
6491 | } | |
6492 | prev = dl; | |
6493 | } | |
6494 | return 0; | |
6495 | } | |
6496 | ||
6497 | static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index); | |
6498 | ||
6499 | static int add_remove_disk_update(struct intel_super *super) | |
6500 | { | |
6501 | int check_degraded = 0; | |
6502 | struct dl *disk = NULL; | |
6503 | /* add/remove some spares to/from the metadata/contrainer */ | |
6504 | while (super->disk_mgmt_list) { | |
6505 | struct dl *disk_cfg; | |
6506 | ||
6507 | disk_cfg = super->disk_mgmt_list; | |
6508 | super->disk_mgmt_list = disk_cfg->next; | |
6509 | disk_cfg->next = NULL; | |
6510 | ||
6511 | if (disk_cfg->action == DISK_ADD) { | |
6512 | disk_cfg->next = super->disks; | |
6513 | super->disks = disk_cfg; | |
6514 | check_degraded = 1; | |
6515 | dprintf("%s: added %x:%x\n", | |
6516 | __func__, disk_cfg->major, | |
6517 | disk_cfg->minor); | |
6518 | } else if (disk_cfg->action == DISK_REMOVE) { | |
6519 | dprintf("Disk remove action processed: %x.%x\n", | |
6520 | disk_cfg->major, disk_cfg->minor); | |
6521 | disk = get_disk_super(super, | |
6522 | disk_cfg->major, | |
6523 | disk_cfg->minor); | |
6524 | if (disk) { | |
6525 | /* store action status */ | |
6526 | disk->action = DISK_REMOVE; | |
6527 | /* remove spare disks only */ | |
6528 | if (disk->index == -1) { | |
6529 | remove_disk_super(super, | |
6530 | disk_cfg->major, | |
6531 | disk_cfg->minor); | |
6532 | } | |
6533 | } | |
6534 | /* release allocate disk structure */ | |
6535 | __free_imsm_disk(disk_cfg); | |
6536 | } | |
6537 | } | |
6538 | return check_degraded; | |
6539 | } | |
6540 | ||
6541 | ||
6542 | static int apply_reshape_migration_update(struct imsm_update_reshape_migration *u, | |
6543 | struct intel_super *super, | |
6544 | void ***space_list) | |
6545 | { | |
6546 | struct intel_dev *id; | |
6547 | void **tofree = NULL; | |
6548 | int ret_val = 0; | |
6549 | ||
6550 | dprintf("apply_reshape_migration_update()\n"); | |
6551 | if ((u->subdev < 0) || | |
6552 | (u->subdev > 1)) { | |
6553 | dprintf("imsm: Error: Wrong subdev: %i\n", u->subdev); | |
6554 | return ret_val; | |
6555 | } | |
6556 | if ((space_list == NULL) || (*space_list == NULL)) { | |
6557 | dprintf("imsm: Error: Memory is not allocated\n"); | |
6558 | return ret_val; | |
6559 | } | |
6560 | ||
6561 | for (id = super->devlist ; id; id = id->next) { | |
6562 | if (id->index == (unsigned)u->subdev) { | |
6563 | struct imsm_dev *dev = get_imsm_dev(super, u->subdev); | |
6564 | struct imsm_map *map; | |
6565 | struct imsm_dev *new_dev = | |
6566 | (struct imsm_dev *)*space_list; | |
6567 | struct imsm_map *migr_map = get_imsm_map(dev, 1); | |
6568 | int to_state; | |
6569 | struct dl *new_disk; | |
6570 | ||
6571 | if (new_dev == NULL) | |
6572 | return ret_val; | |
6573 | *space_list = **space_list; | |
6574 | memcpy(new_dev, dev, sizeof_imsm_dev(dev, 0)); | |
6575 | map = get_imsm_map(new_dev, 0); | |
6576 | if (migr_map) { | |
6577 | dprintf("imsm: Error: migration in progress"); | |
6578 | return ret_val; | |
6579 | } | |
6580 | ||
6581 | to_state = map->map_state; | |
6582 | if ((u->new_level == 5) && (map->raid_level == 0)) { | |
6583 | map->num_members++; | |
6584 | /* this should not happen */ | |
6585 | if (u->new_disks[0] < 0) { | |
6586 | map->failed_disk_num = | |
6587 | map->num_members - 1; | |
6588 | to_state = IMSM_T_STATE_DEGRADED; | |
6589 | } else | |
6590 | to_state = IMSM_T_STATE_NORMAL; | |
6591 | } | |
6592 | migrate(new_dev, super, to_state, MIGR_GEN_MIGR); | |
6593 | if (u->new_level > -1) | |
6594 | map->raid_level = u->new_level; | |
6595 | migr_map = get_imsm_map(new_dev, 1); | |
6596 | if ((u->new_level == 5) && | |
6597 | (migr_map->raid_level == 0)) { | |
6598 | int ord = map->num_members - 1; | |
6599 | migr_map->num_members--; | |
6600 | if (u->new_disks[0] < 0) | |
6601 | ord |= IMSM_ORD_REBUILD; | |
6602 | set_imsm_ord_tbl_ent(map, | |
6603 | map->num_members - 1, | |
6604 | ord); | |
6605 | } | |
6606 | id->dev = new_dev; | |
6607 | tofree = (void **)dev; | |
6608 | ||
6609 | /* update chunk size | |
6610 | */ | |
6611 | if (u->new_chunksize > 0) | |
6612 | map->blocks_per_strip = | |
6613 | __cpu_to_le16(u->new_chunksize * 2); | |
6614 | ||
6615 | /* add disk | |
6616 | */ | |
6617 | if ((u->new_level != 5) || | |
6618 | (migr_map->raid_level != 0) || | |
6619 | (migr_map->raid_level == map->raid_level)) | |
6620 | goto skip_disk_add; | |
6621 | ||
6622 | if (u->new_disks[0] >= 0) { | |
6623 | /* use passes spare | |
6624 | */ | |
6625 | new_disk = get_disk_super(super, | |
6626 | major(u->new_disks[0]), | |
6627 | minor(u->new_disks[0])); | |
6628 | dprintf("imsm: new disk for reshape is: %i:%i " | |
6629 | "(%p, index = %i)\n", | |
6630 | major(u->new_disks[0]), | |
6631 | minor(u->new_disks[0]), | |
6632 | new_disk, new_disk->index); | |
6633 | if (new_disk == NULL) | |
6634 | goto error_disk_add; | |
6635 | ||
6636 | new_disk->index = map->num_members - 1; | |
6637 | /* slot to fill in autolayout | |
6638 | */ | |
6639 | new_disk->raiddisk = new_disk->index; | |
6640 | new_disk->disk.status |= CONFIGURED_DISK; | |
6641 | new_disk->disk.status &= ~SPARE_DISK; | |
6642 | } else | |
6643 | goto error_disk_add; | |
6644 | ||
6645 | skip_disk_add: | |
6646 | *tofree = *space_list; | |
6647 | /* calculate new size | |
6648 | */ | |
6649 | imsm_set_array_size(new_dev); | |
6650 | ||
6651 | ret_val = 1; | |
6652 | } | |
6653 | } | |
6654 | ||
6655 | if (tofree) | |
6656 | *space_list = tofree; | |
6657 | return ret_val; | |
6658 | ||
6659 | error_disk_add: | |
6660 | dprintf("Error: imsm: Cannot find disk.\n"); | |
6661 | return ret_val; | |
6662 | } | |
6663 | ||
6664 | ||
6665 | static int apply_reshape_container_disks_update(struct imsm_update_reshape *u, | |
6666 | struct intel_super *super, | |
6667 | void ***space_list) | |
6668 | { | |
6669 | struct dl *new_disk; | |
6670 | struct intel_dev *id; | |
6671 | int i; | |
6672 | int delta_disks = u->new_raid_disks - u->old_raid_disks; | |
6673 | int disk_count = u->old_raid_disks; | |
6674 | void **tofree = NULL; | |
6675 | int devices_to_reshape = 1; | |
6676 | struct imsm_super *mpb = super->anchor; | |
6677 | int ret_val = 0; | |
6678 | unsigned int dev_id; | |
6679 | ||
6680 | dprintf("imsm: apply_reshape_container_disks_update()\n"); | |
6681 | ||
6682 | /* enable spares to use in array */ | |
6683 | for (i = 0; i < delta_disks; i++) { | |
6684 | new_disk = get_disk_super(super, | |
6685 | major(u->new_disks[i]), | |
6686 | minor(u->new_disks[i])); | |
6687 | dprintf("imsm: new disk for reshape is: %i:%i " | |
6688 | "(%p, index = %i)\n", | |
6689 | major(u->new_disks[i]), minor(u->new_disks[i]), | |
6690 | new_disk, new_disk->index); | |
6691 | if ((new_disk == NULL) || | |
6692 | ((new_disk->index >= 0) && | |
6693 | (new_disk->index < u->old_raid_disks))) | |
6694 | goto update_reshape_exit; | |
6695 | new_disk->index = disk_count++; | |
6696 | /* slot to fill in autolayout | |
6697 | */ | |
6698 | new_disk->raiddisk = new_disk->index; | |
6699 | new_disk->disk.status |= | |
6700 | CONFIGURED_DISK; | |
6701 | new_disk->disk.status &= ~SPARE_DISK; | |
6702 | } | |
6703 | ||
6704 | dprintf("imsm: volume set mpb->num_raid_devs = %i\n", | |
6705 | mpb->num_raid_devs); | |
6706 | /* manage changes in volume | |
6707 | */ | |
6708 | for (dev_id = 0; dev_id < mpb->num_raid_devs; dev_id++) { | |
6709 | void **sp = *space_list; | |
6710 | struct imsm_dev *newdev; | |
6711 | struct imsm_map *newmap, *oldmap; | |
6712 | ||
6713 | for (id = super->devlist ; id; id = id->next) { | |
6714 | if (id->index == dev_id) | |
6715 | break; | |
6716 | } | |
6717 | if (id == NULL) | |
6718 | break; | |
6719 | if (!sp) | |
6720 | continue; | |
6721 | *space_list = *sp; | |
6722 | newdev = (void*)sp; | |
6723 | /* Copy the dev, but not (all of) the map */ | |
6724 | memcpy(newdev, id->dev, sizeof(*newdev)); | |
6725 | oldmap = get_imsm_map(id->dev, 0); | |
6726 | newmap = get_imsm_map(newdev, 0); | |
6727 | /* Copy the current map */ | |
6728 | memcpy(newmap, oldmap, sizeof_imsm_map(oldmap)); | |
6729 | /* update one device only | |
6730 | */ | |
6731 | if (devices_to_reshape) { | |
6732 | dprintf("imsm: modifying subdev: %i\n", | |
6733 | id->index); | |
6734 | devices_to_reshape--; | |
6735 | newdev->vol.migr_state = 1; | |
6736 | newdev->vol.curr_migr_unit = 0; | |
6737 | newdev->vol.migr_type = MIGR_GEN_MIGR; | |
6738 | newmap->num_members = u->new_raid_disks; | |
6739 | for (i = 0; i < delta_disks; i++) { | |
6740 | set_imsm_ord_tbl_ent(newmap, | |
6741 | u->old_raid_disks + i, | |
6742 | u->old_raid_disks + i); | |
6743 | } | |
6744 | /* New map is correct, now need to save old map | |
6745 | */ | |
6746 | newmap = get_imsm_map(newdev, 1); | |
6747 | memcpy(newmap, oldmap, sizeof_imsm_map(oldmap)); | |
6748 | ||
6749 | imsm_set_array_size(newdev); | |
6750 | } | |
6751 | ||
6752 | sp = (void **)id->dev; | |
6753 | id->dev = newdev; | |
6754 | *sp = tofree; | |
6755 | tofree = sp; | |
6756 | ||
6757 | /* Clear migration record */ | |
6758 | memset(super->migr_rec, 0, sizeof(struct migr_record)); | |
6759 | } | |
6760 | if (tofree) | |
6761 | *space_list = tofree; | |
6762 | ret_val = 1; | |
6763 | ||
6764 | update_reshape_exit: | |
6765 | ||
6766 | return ret_val; | |
6767 | } | |
6768 | ||
6769 | static int apply_takeover_update(struct imsm_update_takeover *u, | |
6770 | struct intel_super *super, | |
6771 | void ***space_list) | |
6772 | { | |
6773 | struct imsm_dev *dev = NULL; | |
6774 | struct intel_dev *dv; | |
6775 | struct imsm_dev *dev_new; | |
6776 | struct imsm_map *map; | |
6777 | struct dl *dm, *du; | |
6778 | int i; | |
6779 | ||
6780 | for (dv = super->devlist; dv; dv = dv->next) | |
6781 | if (dv->index == (unsigned int)u->subarray) { | |
6782 | dev = dv->dev; | |
6783 | break; | |
6784 | } | |
6785 | ||
6786 | if (dev == NULL) | |
6787 | return 0; | |
6788 | ||
6789 | map = get_imsm_map(dev, 0); | |
6790 | ||
6791 | if (u->direction == R10_TO_R0) { | |
6792 | /* Number of failed disks must be half of initial disk number */ | |
6793 | if (imsm_count_failed(super, dev) != (map->num_members / 2)) | |
6794 | return 0; | |
6795 | ||
6796 | /* iterate through devices to mark removed disks as spare */ | |
6797 | for (dm = super->disks; dm; dm = dm->next) { | |
6798 | if (dm->disk.status & FAILED_DISK) { | |
6799 | int idx = dm->index; | |
6800 | /* update indexes on the disk list */ | |
6801 | /* FIXME this loop-with-the-loop looks wrong, I'm not convinced | |
6802 | the index values will end up being correct.... NB */ | |
6803 | for (du = super->disks; du; du = du->next) | |
6804 | if (du->index > idx) | |
6805 | du->index--; | |
6806 | /* mark as spare disk */ | |
6807 | dm->disk.status = SPARE_DISK; | |
6808 | dm->index = -1; | |
6809 | } | |
6810 | } | |
6811 | /* update map */ | |
6812 | map->num_members = map->num_members / 2; | |
6813 | map->map_state = IMSM_T_STATE_NORMAL; | |
6814 | map->num_domains = 1; | |
6815 | map->raid_level = 0; | |
6816 | map->failed_disk_num = -1; | |
6817 | } | |
6818 | ||
6819 | if (u->direction == R0_TO_R10) { | |
6820 | void **space; | |
6821 | /* update slots in current disk list */ | |
6822 | for (dm = super->disks; dm; dm = dm->next) { | |
6823 | if (dm->index >= 0) | |
6824 | dm->index *= 2; | |
6825 | } | |
6826 | /* create new *missing* disks */ | |
6827 | for (i = 0; i < map->num_members; i++) { | |
6828 | space = *space_list; | |
6829 | if (!space) | |
6830 | continue; | |
6831 | *space_list = *space; | |
6832 | du = (void *)space; | |
6833 | memcpy(du, super->disks, sizeof(*du)); | |
6834 | du->fd = -1; | |
6835 | du->minor = 0; | |
6836 | du->major = 0; | |
6837 | du->index = (i * 2) + 1; | |
6838 | sprintf((char *)du->disk.serial, | |
6839 | " MISSING_%d", du->index); | |
6840 | sprintf((char *)du->serial, | |
6841 | "MISSING_%d", du->index); | |
6842 | du->next = super->missing; | |
6843 | super->missing = du; | |
6844 | } | |
6845 | /* create new dev and map */ | |
6846 | space = *space_list; | |
6847 | if (!space) | |
6848 | return 0; | |
6849 | *space_list = *space; | |
6850 | dev_new = (void *)space; | |
6851 | memcpy(dev_new, dev, sizeof(*dev)); | |
6852 | /* update new map */ | |
6853 | map = get_imsm_map(dev_new, 0); | |
6854 | map->num_members = map->num_members * 2; | |
6855 | map->map_state = IMSM_T_STATE_DEGRADED; | |
6856 | map->num_domains = 2; | |
6857 | map->raid_level = 1; | |
6858 | /* replace dev<->dev_new */ | |
6859 | dv->dev = dev_new; | |
6860 | } | |
6861 | /* update disk order table */ | |
6862 | for (du = super->disks; du; du = du->next) | |
6863 | if (du->index >= 0) | |
6864 | set_imsm_ord_tbl_ent(map, du->index, du->index); | |
6865 | for (du = super->missing; du; du = du->next) | |
6866 | if (du->index >= 0) { | |
6867 | set_imsm_ord_tbl_ent(map, du->index, du->index); | |
6868 | mark_missing(dev_new, &du->disk, du->index); | |
6869 | } | |
6870 | ||
6871 | return 1; | |
6872 | } | |
6873 | ||
6874 | static void imsm_process_update(struct supertype *st, | |
6875 | struct metadata_update *update) | |
6876 | { | |
6877 | /** | |
6878 | * crack open the metadata_update envelope to find the update record | |
6879 | * update can be one of: | |
6880 | * update_reshape_container_disks - all the arrays in the container | |
6881 | * are being reshaped to have more devices. We need to mark | |
6882 | * the arrays for general migration and convert selected spares | |
6883 | * into active devices. | |
6884 | * update_activate_spare - a spare device has replaced a failed | |
6885 | * device in an array, update the disk_ord_tbl. If this disk is | |
6886 | * present in all member arrays then also clear the SPARE_DISK | |
6887 | * flag | |
6888 | * update_create_array | |
6889 | * update_kill_array | |
6890 | * update_rename_array | |
6891 | * update_add_remove_disk | |
6892 | */ | |
6893 | struct intel_super *super = st->sb; | |
6894 | struct imsm_super *mpb; | |
6895 | enum imsm_update_type type = *(enum imsm_update_type *) update->buf; | |
6896 | ||
6897 | /* update requires a larger buf but the allocation failed */ | |
6898 | if (super->next_len && !super->next_buf) { | |
6899 | super->next_len = 0; | |
6900 | return; | |
6901 | } | |
6902 | ||
6903 | if (super->next_buf) { | |
6904 | memcpy(super->next_buf, super->buf, super->len); | |
6905 | free(super->buf); | |
6906 | super->len = super->next_len; | |
6907 | super->buf = super->next_buf; | |
6908 | ||
6909 | super->next_len = 0; | |
6910 | super->next_buf = NULL; | |
6911 | } | |
6912 | ||
6913 | mpb = super->anchor; | |
6914 | ||
6915 | switch (type) { | |
6916 | case update_general_migration_checkpoint: { | |
6917 | struct intel_dev *id; | |
6918 | struct imsm_update_general_migration_checkpoint *u = | |
6919 | (void *)update->buf; | |
6920 | ||
6921 | dprintf("imsm: process_update() " | |
6922 | "for update_general_migration_checkpoint called\n"); | |
6923 | ||
6924 | /* find device under general migration */ | |
6925 | for (id = super->devlist ; id; id = id->next) { | |
6926 | if (is_gen_migration(id->dev)) { | |
6927 | id->dev->vol.curr_migr_unit = | |
6928 | __cpu_to_le32(u->curr_migr_unit); | |
6929 | super->updates_pending++; | |
6930 | } | |
6931 | } | |
6932 | break; | |
6933 | } | |
6934 | case update_takeover: { | |
6935 | struct imsm_update_takeover *u = (void *)update->buf; | |
6936 | if (apply_takeover_update(u, super, &update->space_list)) { | |
6937 | imsm_update_version_info(super); | |
6938 | super->updates_pending++; | |
6939 | } | |
6940 | break; | |
6941 | } | |
6942 | ||
6943 | case update_reshape_container_disks: { | |
6944 | struct imsm_update_reshape *u = (void *)update->buf; | |
6945 | if (apply_reshape_container_disks_update( | |
6946 | u, super, &update->space_list)) | |
6947 | super->updates_pending++; | |
6948 | break; | |
6949 | } | |
6950 | case update_reshape_migration: { | |
6951 | struct imsm_update_reshape_migration *u = (void *)update->buf; | |
6952 | if (apply_reshape_migration_update( | |
6953 | u, super, &update->space_list)) | |
6954 | super->updates_pending++; | |
6955 | break; | |
6956 | } | |
6957 | case update_activate_spare: { | |
6958 | struct imsm_update_activate_spare *u = (void *) update->buf; | |
6959 | struct imsm_dev *dev = get_imsm_dev(super, u->array); | |
6960 | struct imsm_map *map = get_imsm_map(dev, 0); | |
6961 | struct imsm_map *migr_map; | |
6962 | struct active_array *a; | |
6963 | struct imsm_disk *disk; | |
6964 | __u8 to_state; | |
6965 | struct dl *dl; | |
6966 | unsigned int found; | |
6967 | int failed; | |
6968 | int victim = get_imsm_disk_idx(dev, u->slot, -1); | |
6969 | int i; | |
6970 | ||
6971 | for (dl = super->disks; dl; dl = dl->next) | |
6972 | if (dl == u->dl) | |
6973 | break; | |
6974 | ||
6975 | if (!dl) { | |
6976 | fprintf(stderr, "error: imsm_activate_spare passed " | |
6977 | "an unknown disk (index: %d)\n", | |
6978 | u->dl->index); | |
6979 | return; | |
6980 | } | |
6981 | ||
6982 | super->updates_pending++; | |
6983 | /* count failures (excluding rebuilds and the victim) | |
6984 | * to determine map[0] state | |
6985 | */ | |
6986 | failed = 0; | |
6987 | for (i = 0; i < map->num_members; i++) { | |
6988 | if (i == u->slot) | |
6989 | continue; | |
6990 | disk = get_imsm_disk(super, | |
6991 | get_imsm_disk_idx(dev, i, -1)); | |
6992 | if (!disk || is_failed(disk)) | |
6993 | failed++; | |
6994 | } | |
6995 | ||
6996 | /* adding a pristine spare, assign a new index */ | |
6997 | if (dl->index < 0) { | |
6998 | dl->index = super->anchor->num_disks; | |
6999 | super->anchor->num_disks++; | |
7000 | } | |
7001 | disk = &dl->disk; | |
7002 | disk->status |= CONFIGURED_DISK; | |
7003 | disk->status &= ~SPARE_DISK; | |
7004 | ||
7005 | /* mark rebuild */ | |
7006 | to_state = imsm_check_degraded(super, dev, failed); | |
7007 | map->map_state = IMSM_T_STATE_DEGRADED; | |
7008 | migrate(dev, super, to_state, MIGR_REBUILD); | |
7009 | migr_map = get_imsm_map(dev, 1); | |
7010 | set_imsm_ord_tbl_ent(map, u->slot, dl->index); | |
7011 | set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD); | |
7012 | ||
7013 | /* update the family_num to mark a new container | |
7014 | * generation, being careful to record the existing | |
7015 | * family_num in orig_family_num to clean up after | |
7016 | * earlier mdadm versions that neglected to set it. | |
7017 | */ | |
7018 | if (mpb->orig_family_num == 0) | |
7019 | mpb->orig_family_num = mpb->family_num; | |
7020 | mpb->family_num += super->random; | |
7021 | ||
7022 | /* count arrays using the victim in the metadata */ | |
7023 | found = 0; | |
7024 | for (a = st->arrays; a ; a = a->next) { | |
7025 | dev = get_imsm_dev(super, a->info.container_member); | |
7026 | map = get_imsm_map(dev, 0); | |
7027 | ||
7028 | if (get_imsm_disk_slot(map, victim) >= 0) | |
7029 | found++; | |
7030 | } | |
7031 | ||
7032 | /* delete the victim if it is no longer being | |
7033 | * utilized anywhere | |
7034 | */ | |
7035 | if (!found) { | |
7036 | struct dl **dlp; | |
7037 | ||
7038 | /* We know that 'manager' isn't touching anything, | |
7039 | * so it is safe to delete | |
7040 | */ | |
7041 | for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next) | |
7042 | if ((*dlp)->index == victim) | |
7043 | break; | |
7044 | ||
7045 | /* victim may be on the missing list */ | |
7046 | if (!*dlp) | |
7047 | for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next) | |
7048 | if ((*dlp)->index == victim) | |
7049 | break; | |
7050 | imsm_delete(super, dlp, victim); | |
7051 | } | |
7052 | break; | |
7053 | } | |
7054 | case update_create_array: { | |
7055 | /* someone wants to create a new array, we need to be aware of | |
7056 | * a few races/collisions: | |
7057 | * 1/ 'Create' called by two separate instances of mdadm | |
7058 | * 2/ 'Create' versus 'activate_spare': mdadm has chosen | |
7059 | * devices that have since been assimilated via | |
7060 | * activate_spare. | |
7061 | * In the event this update can not be carried out mdadm will | |
7062 | * (FIX ME) notice that its update did not take hold. | |
7063 | */ | |
7064 | struct imsm_update_create_array *u = (void *) update->buf; | |
7065 | struct intel_dev *dv; | |
7066 | struct imsm_dev *dev; | |
7067 | struct imsm_map *map, *new_map; | |
7068 | unsigned long long start, end; | |
7069 | unsigned long long new_start, new_end; | |
7070 | int i; | |
7071 | struct disk_info *inf; | |
7072 | struct dl *dl; | |
7073 | ||
7074 | /* handle racing creates: first come first serve */ | |
7075 | if (u->dev_idx < mpb->num_raid_devs) { | |
7076 | dprintf("%s: subarray %d already defined\n", | |
7077 | __func__, u->dev_idx); | |
7078 | goto create_error; | |
7079 | } | |
7080 | ||
7081 | /* check update is next in sequence */ | |
7082 | if (u->dev_idx != mpb->num_raid_devs) { | |
7083 | dprintf("%s: can not create array %d expected index %d\n", | |
7084 | __func__, u->dev_idx, mpb->num_raid_devs); | |
7085 | goto create_error; | |
7086 | } | |
7087 | ||
7088 | new_map = get_imsm_map(&u->dev, 0); | |
7089 | new_start = __le32_to_cpu(new_map->pba_of_lba0); | |
7090 | new_end = new_start + __le32_to_cpu(new_map->blocks_per_member); | |
7091 | inf = get_disk_info(u); | |
7092 | ||
7093 | /* handle activate_spare versus create race: | |
7094 | * check to make sure that overlapping arrays do not include | |
7095 | * overalpping disks | |
7096 | */ | |
7097 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
7098 | dev = get_imsm_dev(super, i); | |
7099 | map = get_imsm_map(dev, 0); | |
7100 | start = __le32_to_cpu(map->pba_of_lba0); | |
7101 | end = start + __le32_to_cpu(map->blocks_per_member); | |
7102 | if ((new_start >= start && new_start <= end) || | |
7103 | (start >= new_start && start <= new_end)) | |
7104 | /* overlap */; | |
7105 | else | |
7106 | continue; | |
7107 | ||
7108 | if (disks_overlap(super, i, u)) { | |
7109 | dprintf("%s: arrays overlap\n", __func__); | |
7110 | goto create_error; | |
7111 | } | |
7112 | } | |
7113 | ||
7114 | /* check that prepare update was successful */ | |
7115 | if (!update->space) { | |
7116 | dprintf("%s: prepare update failed\n", __func__); | |
7117 | goto create_error; | |
7118 | } | |
7119 | ||
7120 | /* check that all disks are still active before committing | |
7121 | * changes. FIXME: could we instead handle this by creating a | |
7122 | * degraded array? That's probably not what the user expects, | |
7123 | * so better to drop this update on the floor. | |
7124 | */ | |
7125 | for (i = 0; i < new_map->num_members; i++) { | |
7126 | dl = serial_to_dl(inf[i].serial, super); | |
7127 | if (!dl) { | |
7128 | dprintf("%s: disk disappeared\n", __func__); | |
7129 | goto create_error; | |
7130 | } | |
7131 | } | |
7132 | ||
7133 | super->updates_pending++; | |
7134 | ||
7135 | /* convert spares to members and fixup ord_tbl */ | |
7136 | for (i = 0; i < new_map->num_members; i++) { | |
7137 | dl = serial_to_dl(inf[i].serial, super); | |
7138 | if (dl->index == -1) { | |
7139 | dl->index = mpb->num_disks; | |
7140 | mpb->num_disks++; | |
7141 | dl->disk.status |= CONFIGURED_DISK; | |
7142 | dl->disk.status &= ~SPARE_DISK; | |
7143 | } | |
7144 | set_imsm_ord_tbl_ent(new_map, i, dl->index); | |
7145 | } | |
7146 | ||
7147 | dv = update->space; | |
7148 | dev = dv->dev; | |
7149 | update->space = NULL; | |
7150 | imsm_copy_dev(dev, &u->dev); | |
7151 | dv->index = u->dev_idx; | |
7152 | dv->next = super->devlist; | |
7153 | super->devlist = dv; | |
7154 | mpb->num_raid_devs++; | |
7155 | ||
7156 | imsm_update_version_info(super); | |
7157 | break; | |
7158 | create_error: | |
7159 | /* mdmon knows how to release update->space, but not | |
7160 | * ((struct intel_dev *) update->space)->dev | |
7161 | */ | |
7162 | if (update->space) { | |
7163 | dv = update->space; | |
7164 | free(dv->dev); | |
7165 | } | |
7166 | break; | |
7167 | } | |
7168 | case update_kill_array: { | |
7169 | struct imsm_update_kill_array *u = (void *) update->buf; | |
7170 | int victim = u->dev_idx; | |
7171 | struct active_array *a; | |
7172 | struct intel_dev **dp; | |
7173 | struct imsm_dev *dev; | |
7174 | ||
7175 | /* sanity check that we are not affecting the uuid of | |
7176 | * active arrays, or deleting an active array | |
7177 | * | |
7178 | * FIXME when immutable ids are available, but note that | |
7179 | * we'll also need to fixup the invalidated/active | |
7180 | * subarray indexes in mdstat | |
7181 | */ | |
7182 | for (a = st->arrays; a; a = a->next) | |
7183 | if (a->info.container_member >= victim) | |
7184 | break; | |
7185 | /* by definition if mdmon is running at least one array | |
7186 | * is active in the container, so checking | |
7187 | * mpb->num_raid_devs is just extra paranoia | |
7188 | */ | |
7189 | dev = get_imsm_dev(super, victim); | |
7190 | if (a || !dev || mpb->num_raid_devs == 1) { | |
7191 | dprintf("failed to delete subarray-%d\n", victim); | |
7192 | break; | |
7193 | } | |
7194 | ||
7195 | for (dp = &super->devlist; *dp;) | |
7196 | if ((*dp)->index == (unsigned)super->current_vol) { | |
7197 | *dp = (*dp)->next; | |
7198 | } else { | |
7199 | if ((*dp)->index > (unsigned)victim) | |
7200 | (*dp)->index--; | |
7201 | dp = &(*dp)->next; | |
7202 | } | |
7203 | mpb->num_raid_devs--; | |
7204 | super->updates_pending++; | |
7205 | break; | |
7206 | } | |
7207 | case update_rename_array: { | |
7208 | struct imsm_update_rename_array *u = (void *) update->buf; | |
7209 | char name[MAX_RAID_SERIAL_LEN+1]; | |
7210 | int target = u->dev_idx; | |
7211 | struct active_array *a; | |
7212 | struct imsm_dev *dev; | |
7213 | ||
7214 | /* sanity check that we are not affecting the uuid of | |
7215 | * an active array | |
7216 | */ | |
7217 | snprintf(name, MAX_RAID_SERIAL_LEN, "%s", (char *) u->name); | |
7218 | name[MAX_RAID_SERIAL_LEN] = '\0'; | |
7219 | for (a = st->arrays; a; a = a->next) | |
7220 | if (a->info.container_member == target) | |
7221 | break; | |
7222 | dev = get_imsm_dev(super, u->dev_idx); | |
7223 | if (a || !dev || !check_name(super, name, 1)) { | |
7224 | dprintf("failed to rename subarray-%d\n", target); | |
7225 | break; | |
7226 | } | |
7227 | ||
7228 | snprintf((char *) dev->volume, MAX_RAID_SERIAL_LEN, "%s", name); | |
7229 | super->updates_pending++; | |
7230 | break; | |
7231 | } | |
7232 | case update_add_remove_disk: { | |
7233 | /* we may be able to repair some arrays if disks are | |
7234 | * being added, check teh status of add_remove_disk | |
7235 | * if discs has been added. | |
7236 | */ | |
7237 | if (add_remove_disk_update(super)) { | |
7238 | struct active_array *a; | |
7239 | ||
7240 | super->updates_pending++; | |
7241 | for (a = st->arrays; a; a = a->next) | |
7242 | a->check_degraded = 1; | |
7243 | } | |
7244 | break; | |
7245 | } | |
7246 | default: | |
7247 | fprintf(stderr, "error: unsuported process update type:" | |
7248 | "(type: %d)\n", type); | |
7249 | } | |
7250 | } | |
7251 | ||
7252 | static struct mdinfo *get_spares_for_grow(struct supertype *st); | |
7253 | ||
7254 | static void imsm_prepare_update(struct supertype *st, | |
7255 | struct metadata_update *update) | |
7256 | { | |
7257 | /** | |
7258 | * Allocate space to hold new disk entries, raid-device entries or a new | |
7259 | * mpb if necessary. The manager synchronously waits for updates to | |
7260 | * complete in the monitor, so new mpb buffers allocated here can be | |
7261 | * integrated by the monitor thread without worrying about live pointers | |
7262 | * in the manager thread. | |
7263 | */ | |
7264 | enum imsm_update_type type = *(enum imsm_update_type *) update->buf; | |
7265 | struct intel_super *super = st->sb; | |
7266 | struct imsm_super *mpb = super->anchor; | |
7267 | size_t buf_len; | |
7268 | size_t len = 0; | |
7269 | ||
7270 | switch (type) { | |
7271 | case update_general_migration_checkpoint: | |
7272 | dprintf("imsm: prepare_update() " | |
7273 | "for update_general_migration_checkpoint called\n"); | |
7274 | break; | |
7275 | case update_takeover: { | |
7276 | struct imsm_update_takeover *u = (void *)update->buf; | |
7277 | if (u->direction == R0_TO_R10) { | |
7278 | void **tail = (void **)&update->space_list; | |
7279 | struct imsm_dev *dev = get_imsm_dev(super, u->subarray); | |
7280 | struct imsm_map *map = get_imsm_map(dev, 0); | |
7281 | int num_members = map->num_members; | |
7282 | void *space; | |
7283 | int size, i; | |
7284 | int err = 0; | |
7285 | /* allocate memory for added disks */ | |
7286 | for (i = 0; i < num_members; i++) { | |
7287 | size = sizeof(struct dl); | |
7288 | space = malloc(size); | |
7289 | if (!space) { | |
7290 | err++; | |
7291 | break; | |
7292 | } | |
7293 | *tail = space; | |
7294 | tail = space; | |
7295 | *tail = NULL; | |
7296 | } | |
7297 | /* allocate memory for new device */ | |
7298 | size = sizeof_imsm_dev(super->devlist->dev, 0) + | |
7299 | (num_members * sizeof(__u32)); | |
7300 | space = malloc(size); | |
7301 | if (!space) | |
7302 | err++; | |
7303 | else { | |
7304 | *tail = space; | |
7305 | tail = space; | |
7306 | *tail = NULL; | |
7307 | } | |
7308 | if (!err) { | |
7309 | len = disks_to_mpb_size(num_members * 2); | |
7310 | } else { | |
7311 | /* if allocation didn't success, free buffer */ | |
7312 | while (update->space_list) { | |
7313 | void **sp = update->space_list; | |
7314 | update->space_list = *sp; | |
7315 | free(sp); | |
7316 | } | |
7317 | } | |
7318 | } | |
7319 | ||
7320 | break; | |
7321 | } | |
7322 | case update_reshape_container_disks: { | |
7323 | /* Every raid device in the container is about to | |
7324 | * gain some more devices, and we will enter a | |
7325 | * reconfiguration. | |
7326 | * So each 'imsm_map' will be bigger, and the imsm_vol | |
7327 | * will now hold 2 of them. | |
7328 | * Thus we need new 'struct imsm_dev' allocations sized | |
7329 | * as sizeof_imsm_dev but with more devices in both maps. | |
7330 | */ | |
7331 | struct imsm_update_reshape *u = (void *)update->buf; | |
7332 | struct intel_dev *dl; | |
7333 | void **space_tail = (void**)&update->space_list; | |
7334 | ||
7335 | dprintf("imsm: imsm_prepare_update() for update_reshape\n"); | |
7336 | ||
7337 | for (dl = super->devlist; dl; dl = dl->next) { | |
7338 | int size = sizeof_imsm_dev(dl->dev, 1); | |
7339 | void *s; | |
7340 | if (u->new_raid_disks > u->old_raid_disks) | |
7341 | size += sizeof(__u32)*2* | |
7342 | (u->new_raid_disks - u->old_raid_disks); | |
7343 | s = malloc(size); | |
7344 | if (!s) | |
7345 | break; | |
7346 | *space_tail = s; | |
7347 | space_tail = s; | |
7348 | *space_tail = NULL; | |
7349 | } | |
7350 | ||
7351 | len = disks_to_mpb_size(u->new_raid_disks); | |
7352 | dprintf("New anchor length is %llu\n", (unsigned long long)len); | |
7353 | break; | |
7354 | } | |
7355 | case update_reshape_migration: { | |
7356 | /* for migration level 0->5 we need to add disks | |
7357 | * so the same as for container operation we will copy | |
7358 | * device to the bigger location. | |
7359 | * in memory prepared device and new disk area are prepared | |
7360 | * for usage in process update | |
7361 | */ | |
7362 | struct imsm_update_reshape_migration *u = (void *)update->buf; | |
7363 | struct intel_dev *id; | |
7364 | void **space_tail = (void **)&update->space_list; | |
7365 | int size; | |
7366 | void *s; | |
7367 | int current_level = -1; | |
7368 | ||
7369 | dprintf("imsm: imsm_prepare_update() for update_reshape\n"); | |
7370 | ||
7371 | /* add space for bigger array in update | |
7372 | */ | |
7373 | for (id = super->devlist; id; id = id->next) { | |
7374 | if (id->index == (unsigned)u->subdev) { | |
7375 | size = sizeof_imsm_dev(id->dev, 1); | |
7376 | if (u->new_raid_disks > u->old_raid_disks) | |
7377 | size += sizeof(__u32)*2* | |
7378 | (u->new_raid_disks - u->old_raid_disks); | |
7379 | s = malloc(size); | |
7380 | if (!s) | |
7381 | break; | |
7382 | *space_tail = s; | |
7383 | space_tail = s; | |
7384 | *space_tail = NULL; | |
7385 | break; | |
7386 | } | |
7387 | } | |
7388 | if (update->space_list == NULL) | |
7389 | break; | |
7390 | ||
7391 | /* add space for disk in update | |
7392 | */ | |
7393 | size = sizeof(struct dl); | |
7394 | s = malloc(size); | |
7395 | if (!s) { | |
7396 | free(update->space_list); | |
7397 | update->space_list = NULL; | |
7398 | break; | |
7399 | } | |
7400 | *space_tail = s; | |
7401 | space_tail = s; | |
7402 | *space_tail = NULL; | |
7403 | ||
7404 | /* add spare device to update | |
7405 | */ | |
7406 | for (id = super->devlist ; id; id = id->next) | |
7407 | if (id->index == (unsigned)u->subdev) { | |
7408 | struct imsm_dev *dev; | |
7409 | struct imsm_map *map; | |
7410 | ||
7411 | dev = get_imsm_dev(super, u->subdev); | |
7412 | map = get_imsm_map(dev, 0); | |
7413 | current_level = map->raid_level; | |
7414 | break; | |
7415 | } | |
7416 | if ((u->new_level == 5) && (u->new_level != current_level)) { | |
7417 | struct mdinfo *spares; | |
7418 | ||
7419 | spares = get_spares_for_grow(st); | |
7420 | if (spares) { | |
7421 | struct dl *dl; | |
7422 | struct mdinfo *dev; | |
7423 | ||
7424 | dev = spares->devs; | |
7425 | if (dev) { | |
7426 | u->new_disks[0] = | |
7427 | makedev(dev->disk.major, | |
7428 | dev->disk.minor); | |
7429 | dl = get_disk_super(super, | |
7430 | dev->disk.major, | |
7431 | dev->disk.minor); | |
7432 | dl->index = u->old_raid_disks; | |
7433 | dev = dev->next; | |
7434 | } | |
7435 | sysfs_free(spares); | |
7436 | } | |
7437 | } | |
7438 | len = disks_to_mpb_size(u->new_raid_disks); | |
7439 | dprintf("New anchor length is %llu\n", (unsigned long long)len); | |
7440 | break; | |
7441 | } | |
7442 | case update_create_array: { | |
7443 | struct imsm_update_create_array *u = (void *) update->buf; | |
7444 | struct intel_dev *dv; | |
7445 | struct imsm_dev *dev = &u->dev; | |
7446 | struct imsm_map *map = get_imsm_map(dev, 0); | |
7447 | struct dl *dl; | |
7448 | struct disk_info *inf; | |
7449 | int i; | |
7450 | int activate = 0; | |
7451 | ||
7452 | inf = get_disk_info(u); | |
7453 | len = sizeof_imsm_dev(dev, 1); | |
7454 | /* allocate a new super->devlist entry */ | |
7455 | dv = malloc(sizeof(*dv)); | |
7456 | if (dv) { | |
7457 | dv->dev = malloc(len); | |
7458 | if (dv->dev) | |
7459 | update->space = dv; | |
7460 | else { | |
7461 | free(dv); | |
7462 | update->space = NULL; | |
7463 | } | |
7464 | } | |
7465 | ||
7466 | /* count how many spares will be converted to members */ | |
7467 | for (i = 0; i < map->num_members; i++) { | |
7468 | dl = serial_to_dl(inf[i].serial, super); | |
7469 | if (!dl) { | |
7470 | /* hmm maybe it failed?, nothing we can do about | |
7471 | * it here | |
7472 | */ | |
7473 | continue; | |
7474 | } | |
7475 | if (count_memberships(dl, super) == 0) | |
7476 | activate++; | |
7477 | } | |
7478 | len += activate * sizeof(struct imsm_disk); | |
7479 | break; | |
7480 | default: | |
7481 | break; | |
7482 | } | |
7483 | } | |
7484 | ||
7485 | /* check if we need a larger metadata buffer */ | |
7486 | if (super->next_buf) | |
7487 | buf_len = super->next_len; | |
7488 | else | |
7489 | buf_len = super->len; | |
7490 | ||
7491 | if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) { | |
7492 | /* ok we need a larger buf than what is currently allocated | |
7493 | * if this allocation fails process_update will notice that | |
7494 | * ->next_len is set and ->next_buf is NULL | |
7495 | */ | |
7496 | buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512); | |
7497 | if (super->next_buf) | |
7498 | free(super->next_buf); | |
7499 | ||
7500 | super->next_len = buf_len; | |
7501 | if (posix_memalign(&super->next_buf, 512, buf_len) == 0) | |
7502 | memset(super->next_buf, 0, buf_len); | |
7503 | else | |
7504 | super->next_buf = NULL; | |
7505 | } | |
7506 | } | |
7507 | ||
7508 | /* must be called while manager is quiesced */ | |
7509 | static void imsm_delete(struct intel_super *super, struct dl **dlp, unsigned index) | |
7510 | { | |
7511 | struct imsm_super *mpb = super->anchor; | |
7512 | struct dl *iter; | |
7513 | struct imsm_dev *dev; | |
7514 | struct imsm_map *map; | |
7515 | int i, j, num_members; | |
7516 | __u32 ord; | |
7517 | ||
7518 | dprintf("%s: deleting device[%d] from imsm_super\n", | |
7519 | __func__, index); | |
7520 | ||
7521 | /* shift all indexes down one */ | |
7522 | for (iter = super->disks; iter; iter = iter->next) | |
7523 | if (iter->index > (int)index) | |
7524 | iter->index--; | |
7525 | for (iter = super->missing; iter; iter = iter->next) | |
7526 | if (iter->index > (int)index) | |
7527 | iter->index--; | |
7528 | ||
7529 | for (i = 0; i < mpb->num_raid_devs; i++) { | |
7530 | dev = get_imsm_dev(super, i); | |
7531 | map = get_imsm_map(dev, 0); | |
7532 | num_members = map->num_members; | |
7533 | for (j = 0; j < num_members; j++) { | |
7534 | /* update ord entries being careful not to propagate | |
7535 | * ord-flags to the first map | |
7536 | */ | |
7537 | ord = get_imsm_ord_tbl_ent(dev, j, -1); | |
7538 | ||
7539 | if (ord_to_idx(ord) <= index) | |
7540 | continue; | |
7541 | ||
7542 | map = get_imsm_map(dev, 0); | |
7543 | set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1)); | |
7544 | map = get_imsm_map(dev, 1); | |
7545 | if (map) | |
7546 | set_imsm_ord_tbl_ent(map, j, ord - 1); | |
7547 | } | |
7548 | } | |
7549 | ||
7550 | mpb->num_disks--; | |
7551 | super->updates_pending++; | |
7552 | if (*dlp) { | |
7553 | struct dl *dl = *dlp; | |
7554 | ||
7555 | *dlp = (*dlp)->next; | |
7556 | __free_imsm_disk(dl); | |
7557 | } | |
7558 | } | |
7559 | ||
7560 | /******************************************************************************* | |
7561 | * Function: open_backup_targets | |
7562 | * Description: Function opens file descriptors for all devices given in | |
7563 | * info->devs | |
7564 | * Parameters: | |
7565 | * info : general array info | |
7566 | * raid_disks : number of disks | |
7567 | * raid_fds : table of device's file descriptors | |
7568 | * Returns: | |
7569 | * 0 : success | |
7570 | * -1 : fail | |
7571 | ******************************************************************************/ | |
7572 | int open_backup_targets(struct mdinfo *info, int raid_disks, int *raid_fds) | |
7573 | { | |
7574 | struct mdinfo *sd; | |
7575 | ||
7576 | for (sd = info->devs ; sd ; sd = sd->next) { | |
7577 | char *dn; | |
7578 | ||
7579 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) { | |
7580 | dprintf("disk is faulty!!\n"); | |
7581 | continue; | |
7582 | } | |
7583 | ||
7584 | if ((sd->disk.raid_disk >= raid_disks) || | |
7585 | (sd->disk.raid_disk < 0)) | |
7586 | continue; | |
7587 | ||
7588 | dn = map_dev(sd->disk.major, | |
7589 | sd->disk.minor, 1); | |
7590 | raid_fds[sd->disk.raid_disk] = dev_open(dn, O_RDWR); | |
7591 | if (raid_fds[sd->disk.raid_disk] < 0) { | |
7592 | fprintf(stderr, "cannot open component\n"); | |
7593 | return -1; | |
7594 | } | |
7595 | } | |
7596 | return 0; | |
7597 | } | |
7598 | ||
7599 | /******************************************************************************* | |
7600 | * Function: init_migr_record_imsm | |
7601 | * Description: Function inits imsm migration record | |
7602 | * Parameters: | |
7603 | * super : imsm internal array info | |
7604 | * dev : device under migration | |
7605 | * info : general array info to find the smallest device | |
7606 | * Returns: | |
7607 | * none | |
7608 | ******************************************************************************/ | |
7609 | void init_migr_record_imsm(struct supertype *st, struct imsm_dev *dev, | |
7610 | struct mdinfo *info) | |
7611 | { | |
7612 | struct intel_super *super = st->sb; | |
7613 | struct migr_record *migr_rec = super->migr_rec; | |
7614 | int new_data_disks; | |
7615 | unsigned long long dsize, dev_sectors; | |
7616 | long long unsigned min_dev_sectors = -1LLU; | |
7617 | struct mdinfo *sd; | |
7618 | char nm[30]; | |
7619 | int fd; | |
7620 | struct imsm_map *map_dest = get_imsm_map(dev, 0); | |
7621 | struct imsm_map *map_src = get_imsm_map(dev, 1); | |
7622 | unsigned long long num_migr_units; | |
7623 | unsigned long long array_blocks; | |
7624 | ||
7625 | memset(migr_rec, 0, sizeof(struct migr_record)); | |
7626 | migr_rec->family_num = __cpu_to_le32(super->anchor->family_num); | |
7627 | ||
7628 | /* only ascending reshape supported now */ | |
7629 | migr_rec->ascending_migr = __cpu_to_le32(1); | |
7630 | ||
7631 | migr_rec->dest_depth_per_unit = GEN_MIGR_AREA_SIZE / | |
7632 | max(map_dest->blocks_per_strip, map_src->blocks_per_strip); | |
7633 | migr_rec->dest_depth_per_unit *= map_dest->blocks_per_strip; | |
7634 | new_data_disks = imsm_num_data_members(dev, 0); | |
7635 | migr_rec->blocks_per_unit = | |
7636 | __cpu_to_le32(migr_rec->dest_depth_per_unit * new_data_disks); | |
7637 | migr_rec->dest_depth_per_unit = | |
7638 | __cpu_to_le32(migr_rec->dest_depth_per_unit); | |
7639 | array_blocks = info->component_size * new_data_disks; | |
7640 | num_migr_units = | |
7641 | array_blocks / __le32_to_cpu(migr_rec->blocks_per_unit); | |
7642 | ||
7643 | if (array_blocks % __le32_to_cpu(migr_rec->blocks_per_unit)) | |
7644 | num_migr_units++; | |
7645 | migr_rec->num_migr_units = __cpu_to_le32(num_migr_units); | |
7646 | ||
7647 | migr_rec->post_migr_vol_cap = dev->size_low; | |
7648 | migr_rec->post_migr_vol_cap_hi = dev->size_high; | |
7649 | ||
7650 | ||
7651 | /* Find the smallest dev */ | |
7652 | for (sd = info->devs ; sd ; sd = sd->next) { | |
7653 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
7654 | fd = dev_open(nm, O_RDONLY); | |
7655 | if (fd < 0) | |
7656 | continue; | |
7657 | get_dev_size(fd, NULL, &dsize); | |
7658 | dev_sectors = dsize / 512; | |
7659 | if (dev_sectors < min_dev_sectors) | |
7660 | min_dev_sectors = dev_sectors; | |
7661 | close(fd); | |
7662 | } | |
7663 | migr_rec->ckpt_area_pba = __cpu_to_le32(min_dev_sectors - | |
7664 | RAID_DISK_RESERVED_BLOCKS_IMSM_HI); | |
7665 | ||
7666 | write_imsm_migr_rec(st); | |
7667 | ||
7668 | return; | |
7669 | } | |
7670 | ||
7671 | /******************************************************************************* | |
7672 | * Function: save_backup_imsm | |
7673 | * Description: Function saves critical data stripes to Migration Copy Area | |
7674 | * and updates the current migration unit status. | |
7675 | * Use restore_stripes() to form a destination stripe, | |
7676 | * and to write it to the Copy Area. | |
7677 | * Parameters: | |
7678 | * st : supertype information | |
7679 | * dev : imsm device that backup is saved for | |
7680 | * info : general array info | |
7681 | * buf : input buffer | |
7682 | * length : length of data to backup (blocks_per_unit) | |
7683 | * Returns: | |
7684 | * 0 : success | |
7685 | *, -1 : fail | |
7686 | ******************************************************************************/ | |
7687 | int save_backup_imsm(struct supertype *st, | |
7688 | struct imsm_dev *dev, | |
7689 | struct mdinfo *info, | |
7690 | void *buf, | |
7691 | int length) | |
7692 | { | |
7693 | int rv = -1; | |
7694 | struct intel_super *super = st->sb; | |
7695 | unsigned long long *target_offsets = NULL; | |
7696 | int *targets = NULL; | |
7697 | int i; | |
7698 | struct imsm_map *map_dest = get_imsm_map(dev, 0); | |
7699 | int new_disks = map_dest->num_members; | |
7700 | int dest_layout = 0; | |
7701 | int dest_chunk; | |
7702 | ||
7703 | targets = malloc(new_disks * sizeof(int)); | |
7704 | if (!targets) | |
7705 | goto abort; | |
7706 | ||
7707 | target_offsets = malloc(new_disks * sizeof(unsigned long long)); | |
7708 | if (!target_offsets) | |
7709 | goto abort; | |
7710 | ||
7711 | for (i = 0; i < new_disks; i++) { | |
7712 | targets[i] = -1; | |
7713 | target_offsets[i] = (unsigned long long) | |
7714 | __le32_to_cpu(super->migr_rec->ckpt_area_pba) * 512; | |
7715 | } | |
7716 | ||
7717 | if (open_backup_targets(info, new_disks, targets)) | |
7718 | goto abort; | |
7719 | ||
7720 | if (map_dest->raid_level != 0) | |
7721 | dest_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
7722 | dest_chunk = __le16_to_cpu(map_dest->blocks_per_strip) * 512; | |
7723 | ||
7724 | if (restore_stripes(targets, /* list of dest devices */ | |
7725 | target_offsets, /* migration record offsets */ | |
7726 | new_disks, | |
7727 | dest_chunk, | |
7728 | map_dest->raid_level, | |
7729 | dest_layout, | |
7730 | -1, /* source backup file descriptor */ | |
7731 | 0, /* input buf offset | |
7732 | * always 0 buf is already offseted */ | |
7733 | 0, | |
7734 | length, | |
7735 | buf) != 0) { | |
7736 | fprintf(stderr, Name ": Error restoring stripes\n"); | |
7737 | goto abort; | |
7738 | } | |
7739 | ||
7740 | rv = 0; | |
7741 | ||
7742 | abort: | |
7743 | if (targets) { | |
7744 | for (i = 0; i < new_disks; i++) | |
7745 | if (targets[i] >= 0) | |
7746 | close(targets[i]); | |
7747 | free(targets); | |
7748 | } | |
7749 | free(target_offsets); | |
7750 | ||
7751 | return rv; | |
7752 | } | |
7753 | ||
7754 | /******************************************************************************* | |
7755 | * Function: save_checkpoint_imsm | |
7756 | * Description: Function called for current unit status update | |
7757 | * in the migration record. It writes it to disk. | |
7758 | * Parameters: | |
7759 | * super : imsm internal array info | |
7760 | * info : general array info | |
7761 | * Returns: | |
7762 | * 0: success | |
7763 | * 1: failure | |
7764 | * 2: failure, means no valid migration record | |
7765 | * / no general migration in progress / | |
7766 | ******************************************************************************/ | |
7767 | int save_checkpoint_imsm(struct supertype *st, struct mdinfo *info, int state) | |
7768 | { | |
7769 | struct intel_super *super = st->sb; | |
7770 | if (load_imsm_migr_rec(super, info) != 0) { | |
7771 | dprintf("imsm: ERROR: Cannot read migration record " | |
7772 | "for checkpoint save.\n"); | |
7773 | return 1; | |
7774 | } | |
7775 | ||
7776 | if (__le32_to_cpu(super->migr_rec->blocks_per_unit) == 0) { | |
7777 | dprintf("imsm: no migration in progress.\n"); | |
7778 | return 2; | |
7779 | } | |
7780 | ||
7781 | super->migr_rec->curr_migr_unit = | |
7782 | __cpu_to_le32(info->reshape_progress / | |
7783 | __le32_to_cpu(super->migr_rec->blocks_per_unit)); | |
7784 | super->migr_rec->rec_status = __cpu_to_le32(state); | |
7785 | super->migr_rec->dest_1st_member_lba = | |
7786 | __cpu_to_le32((__le32_to_cpu(super->migr_rec->curr_migr_unit)) | |
7787 | * __le32_to_cpu(super->migr_rec->dest_depth_per_unit)); | |
7788 | if (write_imsm_migr_rec(st) < 0) { | |
7789 | dprintf("imsm: Cannot write migration record " | |
7790 | "outside backup area\n"); | |
7791 | return 1; | |
7792 | } | |
7793 | ||
7794 | return 0; | |
7795 | } | |
7796 | ||
7797 | static __u64 blocks_per_migr_unit(struct intel_super *super, | |
7798 | struct imsm_dev *dev); | |
7799 | ||
7800 | /******************************************************************************* | |
7801 | * Function: recover_backup_imsm | |
7802 | * Description: Function recovers critical data from the Migration Copy Area | |
7803 | * while assembling an array. | |
7804 | * Parameters: | |
7805 | * super : imsm internal array info | |
7806 | * info : general array info | |
7807 | * Returns: | |
7808 | * 0 : success (or there is no data to recover) | |
7809 | * 1 : fail | |
7810 | ******************************************************************************/ | |
7811 | int recover_backup_imsm(struct supertype *st, struct mdinfo *info) | |
7812 | { | |
7813 | struct intel_super *super = st->sb; | |
7814 | struct migr_record *migr_rec = super->migr_rec; | |
7815 | struct imsm_map *map_dest = NULL; | |
7816 | struct intel_dev *id = NULL; | |
7817 | unsigned long long read_offset; | |
7818 | unsigned long long write_offset; | |
7819 | unsigned unit_len; | |
7820 | int *targets = NULL; | |
7821 | int new_disks, i, err; | |
7822 | char *buf = NULL; | |
7823 | int retval = 1; | |
7824 | unsigned long curr_migr_unit = __le32_to_cpu(migr_rec->curr_migr_unit); | |
7825 | unsigned long num_migr_units = __le32_to_cpu(migr_rec->num_migr_units); | |
7826 | int ascending = __le32_to_cpu(migr_rec->ascending_migr); | |
7827 | char buffer[20]; | |
7828 | int skipped_disks = 0; | |
7829 | int max_degradation; | |
7830 | ||
7831 | err = sysfs_get_str(info, NULL, "array_state", (char *)buffer, 20); | |
7832 | if (err < 1) | |
7833 | return 1; | |
7834 | ||
7835 | /* recover data only during assemblation */ | |
7836 | if (strncmp(buffer, "inactive", 8) != 0) | |
7837 | return 0; | |
7838 | /* no data to recover */ | |
7839 | if (__le32_to_cpu(migr_rec->rec_status) == UNIT_SRC_NORMAL) | |
7840 | return 0; | |
7841 | if (curr_migr_unit >= num_migr_units) | |
7842 | return 1; | |
7843 | ||
7844 | /* find device during reshape */ | |
7845 | for (id = super->devlist; id; id = id->next) | |
7846 | if (is_gen_migration(id->dev)) | |
7847 | break; | |
7848 | if (id == NULL) | |
7849 | return 1; | |
7850 | ||
7851 | map_dest = get_imsm_map(id->dev, 0); | |
7852 | new_disks = map_dest->num_members; | |
7853 | max_degradation = new_disks - imsm_num_data_members(id->dev, 0); | |
7854 | ||
7855 | read_offset = (unsigned long long) | |
7856 | __le32_to_cpu(migr_rec->ckpt_area_pba) * 512; | |
7857 | ||
7858 | write_offset = ((unsigned long long) | |
7859 | __le32_to_cpu(migr_rec->dest_1st_member_lba) + | |
7860 | __le32_to_cpu(map_dest->pba_of_lba0)) * 512; | |
7861 | ||
7862 | unit_len = __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512; | |
7863 | if (posix_memalign((void **)&buf, 512, unit_len) != 0) | |
7864 | goto abort; | |
7865 | targets = malloc(new_disks * sizeof(int)); | |
7866 | if (!targets) | |
7867 | goto abort; | |
7868 | ||
7869 | open_backup_targets(info, new_disks, targets); | |
7870 | ||
7871 | for (i = 0; i < new_disks; i++) { | |
7872 | if (targets[i] < 0) { | |
7873 | skipped_disks++; | |
7874 | continue; | |
7875 | } | |
7876 | if (lseek64(targets[i], read_offset, SEEK_SET) < 0) { | |
7877 | fprintf(stderr, | |
7878 | Name ": Cannot seek to block: %s\n", | |
7879 | strerror(errno)); | |
7880 | goto abort; | |
7881 | } | |
7882 | if (read(targets[i], buf, unit_len) != unit_len) { | |
7883 | fprintf(stderr, | |
7884 | Name ": Cannot read copy area block: %s\n", | |
7885 | strerror(errno)); | |
7886 | goto abort; | |
7887 | } | |
7888 | if (lseek64(targets[i], write_offset, SEEK_SET) < 0) { | |
7889 | fprintf(stderr, | |
7890 | Name ": Cannot seek to block: %s\n", | |
7891 | strerror(errno)); | |
7892 | goto abort; | |
7893 | } | |
7894 | if (write(targets[i], buf, unit_len) != unit_len) { | |
7895 | fprintf(stderr, | |
7896 | Name ": Cannot restore block: %s\n", | |
7897 | strerror(errno)); | |
7898 | goto abort; | |
7899 | } | |
7900 | } | |
7901 | ||
7902 | if (skipped_disks > max_degradation) { | |
7903 | fprintf(stderr, | |
7904 | Name ": Cannot restore data from backup." | |
7905 | " Too many failed disks\n"); | |
7906 | goto abort; | |
7907 | } | |
7908 | ||
7909 | if (ascending && curr_migr_unit < (num_migr_units-1)) | |
7910 | curr_migr_unit++; | |
7911 | ||
7912 | migr_rec->curr_migr_unit = __le32_to_cpu(curr_migr_unit); | |
7913 | super->migr_rec->rec_status = __cpu_to_le32(UNIT_SRC_NORMAL); | |
7914 | if (write_imsm_migr_rec(st) == 0) { | |
7915 | __u64 blocks_per_unit = blocks_per_migr_unit(super, id->dev); | |
7916 | info->reshape_progress = curr_migr_unit * blocks_per_unit; | |
7917 | retval = 0; | |
7918 | } | |
7919 | ||
7920 | abort: | |
7921 | if (targets) { | |
7922 | for (i = 0; i < new_disks; i++) | |
7923 | if (targets[i]) | |
7924 | close(targets[i]); | |
7925 | free(targets); | |
7926 | } | |
7927 | free(buf); | |
7928 | return retval; | |
7929 | } | |
7930 | ||
7931 | static char disk_by_path[] = "/dev/disk/by-path/"; | |
7932 | ||
7933 | static const char *imsm_get_disk_controller_domain(const char *path) | |
7934 | { | |
7935 | char disk_path[PATH_MAX]; | |
7936 | char *drv=NULL; | |
7937 | struct stat st; | |
7938 | ||
7939 | strncpy(disk_path, disk_by_path, PATH_MAX - 1); | |
7940 | strncat(disk_path, path, PATH_MAX - strlen(disk_path) - 1); | |
7941 | if (stat(disk_path, &st) == 0) { | |
7942 | struct sys_dev* hba; | |
7943 | char *path=NULL; | |
7944 | ||
7945 | path = devt_to_devpath(st.st_rdev); | |
7946 | if (path == NULL) | |
7947 | return "unknown"; | |
7948 | hba = find_disk_attached_hba(-1, path); | |
7949 | if (hba && hba->type == SYS_DEV_SAS) | |
7950 | drv = "isci"; | |
7951 | else if (hba && hba->type == SYS_DEV_SATA) | |
7952 | drv = "ahci"; | |
7953 | else | |
7954 | drv = "unknown"; | |
7955 | dprintf("path: %s hba: %s attached: %s\n", | |
7956 | path, (hba) ? hba->path : "NULL", drv); | |
7957 | free(path); | |
7958 | if (hba) | |
7959 | free_sys_dev(&hba); | |
7960 | } | |
7961 | return drv; | |
7962 | } | |
7963 | ||
7964 | static int imsm_find_array_minor_by_subdev(int subdev, int container, int *minor) | |
7965 | { | |
7966 | char subdev_name[20]; | |
7967 | struct mdstat_ent *mdstat; | |
7968 | ||
7969 | sprintf(subdev_name, "%d", subdev); | |
7970 | mdstat = mdstat_by_subdev(subdev_name, container); | |
7971 | if (!mdstat) | |
7972 | return -1; | |
7973 | ||
7974 | *minor = mdstat->devnum; | |
7975 | free_mdstat(mdstat); | |
7976 | return 0; | |
7977 | } | |
7978 | ||
7979 | static int imsm_reshape_is_allowed_on_container(struct supertype *st, | |
7980 | struct geo_params *geo, | |
7981 | int *old_raid_disks) | |
7982 | { | |
7983 | /* currently we only support increasing the number of devices | |
7984 | * for a container. This increases the number of device for each | |
7985 | * member array. They must all be RAID0 or RAID5. | |
7986 | */ | |
7987 | int ret_val = 0; | |
7988 | struct mdinfo *info, *member; | |
7989 | int devices_that_can_grow = 0; | |
7990 | ||
7991 | dprintf("imsm: imsm_reshape_is_allowed_on_container(ENTER): " | |
7992 | "st->devnum = (%i)\n", | |
7993 | st->devnum); | |
7994 | ||
7995 | if (geo->size != -1 || | |
7996 | geo->level != UnSet || | |
7997 | geo->layout != UnSet || | |
7998 | geo->chunksize != 0 || | |
7999 | geo->raid_disks == UnSet) { | |
8000 | dprintf("imsm: Container operation is allowed for " | |
8001 | "raid disks number change only.\n"); | |
8002 | return ret_val; | |
8003 | } | |
8004 | ||
8005 | info = container_content_imsm(st, NULL); | |
8006 | for (member = info; member; member = member->next) { | |
8007 | int result; | |
8008 | int minor; | |
8009 | ||
8010 | dprintf("imsm: checking device_num: %i\n", | |
8011 | member->container_member); | |
8012 | ||
8013 | if (geo->raid_disks <= member->array.raid_disks) { | |
8014 | /* we work on container for Online Capacity Expansion | |
8015 | * only so raid_disks has to grow | |
8016 | */ | |
8017 | dprintf("imsm: for container operation raid disks " | |
8018 | "increase is required\n"); | |
8019 | break; | |
8020 | } | |
8021 | ||
8022 | if ((info->array.level != 0) && | |
8023 | (info->array.level != 5)) { | |
8024 | /* we cannot use this container with other raid level | |
8025 | */ | |
8026 | dprintf("imsm: for container operation wrong" | |
8027 | " raid level (%i) detected\n", | |
8028 | info->array.level); | |
8029 | break; | |
8030 | } else { | |
8031 | /* check for platform support | |
8032 | * for this raid level configuration | |
8033 | */ | |
8034 | struct intel_super *super = st->sb; | |
8035 | if (!is_raid_level_supported(super->orom, | |
8036 | member->array.level, | |
8037 | geo->raid_disks)) { | |
8038 | dprintf("platform does not support raid%d with" | |
8039 | " %d disk%s\n", | |
8040 | info->array.level, | |
8041 | geo->raid_disks, | |
8042 | geo->raid_disks > 1 ? "s" : ""); | |
8043 | break; | |
8044 | } | |
8045 | /* check if component size is aligned to chunk size | |
8046 | */ | |
8047 | if (info->component_size % | |
8048 | (info->array.chunk_size/512)) { | |
8049 | dprintf("Component size is not aligned to " | |
8050 | "chunk size\n"); | |
8051 | break; | |
8052 | } | |
8053 | } | |
8054 | ||
8055 | if (*old_raid_disks && | |
8056 | info->array.raid_disks != *old_raid_disks) | |
8057 | break; | |
8058 | *old_raid_disks = info->array.raid_disks; | |
8059 | ||
8060 | /* All raid5 and raid0 volumes in container | |
8061 | * have to be ready for Online Capacity Expansion | |
8062 | * so they need to be assembled. We have already | |
8063 | * checked that no recovery etc is happening. | |
8064 | */ | |
8065 | result = imsm_find_array_minor_by_subdev(member->container_member, | |
8066 | st->container_dev, | |
8067 | &minor); | |
8068 | if (result < 0) { | |
8069 | dprintf("imsm: cannot find array\n"); | |
8070 | break; | |
8071 | } | |
8072 | devices_that_can_grow++; | |
8073 | } | |
8074 | sysfs_free(info); | |
8075 | if (!member && devices_that_can_grow) | |
8076 | ret_val = 1; | |
8077 | ||
8078 | if (ret_val) | |
8079 | dprintf("\tContainer operation allowed\n"); | |
8080 | else | |
8081 | dprintf("\tError: %i\n", ret_val); | |
8082 | ||
8083 | return ret_val; | |
8084 | } | |
8085 | ||
8086 | /* Function: get_spares_for_grow | |
8087 | * Description: Allocates memory and creates list of spare devices | |
8088 | * avaliable in container. Checks if spare drive size is acceptable. | |
8089 | * Parameters: Pointer to the supertype structure | |
8090 | * Returns: Pointer to the list of spare devices (mdinfo structure) on success, | |
8091 | * NULL if fail | |
8092 | */ | |
8093 | static struct mdinfo *get_spares_for_grow(struct supertype *st) | |
8094 | { | |
8095 | unsigned long long min_size = min_acceptable_spare_size_imsm(st); | |
8096 | return container_choose_spares(st, min_size, NULL, NULL, NULL, 0); | |
8097 | } | |
8098 | ||
8099 | /****************************************************************************** | |
8100 | * function: imsm_create_metadata_update_for_reshape | |
8101 | * Function creates update for whole IMSM container. | |
8102 | * | |
8103 | ******************************************************************************/ | |
8104 | static int imsm_create_metadata_update_for_reshape( | |
8105 | struct supertype *st, | |
8106 | struct geo_params *geo, | |
8107 | int old_raid_disks, | |
8108 | struct imsm_update_reshape **updatep) | |
8109 | { | |
8110 | struct intel_super *super = st->sb; | |
8111 | struct imsm_super *mpb = super->anchor; | |
8112 | int update_memory_size = 0; | |
8113 | struct imsm_update_reshape *u = NULL; | |
8114 | struct mdinfo *spares = NULL; | |
8115 | int i; | |
8116 | int delta_disks = 0; | |
8117 | struct mdinfo *dev; | |
8118 | ||
8119 | dprintf("imsm_update_metadata_for_reshape(enter) raid_disks = %i\n", | |
8120 | geo->raid_disks); | |
8121 | ||
8122 | delta_disks = geo->raid_disks - old_raid_disks; | |
8123 | ||
8124 | /* size of all update data without anchor */ | |
8125 | update_memory_size = sizeof(struct imsm_update_reshape); | |
8126 | ||
8127 | /* now add space for spare disks that we need to add. */ | |
8128 | update_memory_size += sizeof(u->new_disks[0]) * (delta_disks - 1); | |
8129 | ||
8130 | u = calloc(1, update_memory_size); | |
8131 | if (u == NULL) { | |
8132 | dprintf("error: " | |
8133 | "cannot get memory for imsm_update_reshape update\n"); | |
8134 | return 0; | |
8135 | } | |
8136 | u->type = update_reshape_container_disks; | |
8137 | u->old_raid_disks = old_raid_disks; | |
8138 | u->new_raid_disks = geo->raid_disks; | |
8139 | ||
8140 | /* now get spare disks list | |
8141 | */ | |
8142 | spares = get_spares_for_grow(st); | |
8143 | ||
8144 | if (spares == NULL | |
8145 | || delta_disks > spares->array.spare_disks) { | |
8146 | fprintf(stderr, Name ": imsm: ERROR: Cannot get spare devices " | |
8147 | "for %s.\n", geo->dev_name); | |
8148 | goto abort; | |
8149 | } | |
8150 | ||
8151 | /* we have got spares | |
8152 | * update disk list in imsm_disk list table in anchor | |
8153 | */ | |
8154 | dprintf("imsm: %i spares are available.\n\n", | |
8155 | spares->array.spare_disks); | |
8156 | ||
8157 | dev = spares->devs; | |
8158 | for (i = 0; i < delta_disks; i++) { | |
8159 | struct dl *dl; | |
8160 | ||
8161 | if (dev == NULL) | |
8162 | break; | |
8163 | u->new_disks[i] = makedev(dev->disk.major, | |
8164 | dev->disk.minor); | |
8165 | dl = get_disk_super(super, dev->disk.major, dev->disk.minor); | |
8166 | dl->index = mpb->num_disks; | |
8167 | mpb->num_disks++; | |
8168 | dev = dev->next; | |
8169 | } | |
8170 | ||
8171 | abort: | |
8172 | /* free spares | |
8173 | */ | |
8174 | sysfs_free(spares); | |
8175 | ||
8176 | dprintf("imsm: reshape update preparation :"); | |
8177 | if (i == delta_disks) { | |
8178 | dprintf(" OK\n"); | |
8179 | *updatep = u; | |
8180 | return update_memory_size; | |
8181 | } | |
8182 | free(u); | |
8183 | dprintf(" Error\n"); | |
8184 | ||
8185 | return 0; | |
8186 | } | |
8187 | ||
8188 | /****************************************************************************** | |
8189 | * function: imsm_create_metadata_update_for_migration() | |
8190 | * Creates update for IMSM array. | |
8191 | * | |
8192 | ******************************************************************************/ | |
8193 | static int imsm_create_metadata_update_for_migration( | |
8194 | struct supertype *st, | |
8195 | struct geo_params *geo, | |
8196 | struct imsm_update_reshape_migration **updatep) | |
8197 | { | |
8198 | struct intel_super *super = st->sb; | |
8199 | int update_memory_size = 0; | |
8200 | struct imsm_update_reshape_migration *u = NULL; | |
8201 | struct imsm_dev *dev; | |
8202 | int previous_level = -1; | |
8203 | ||
8204 | dprintf("imsm_create_metadata_update_for_migration(enter)" | |
8205 | " New Level = %i\n", geo->level); | |
8206 | ||
8207 | /* size of all update data without anchor */ | |
8208 | update_memory_size = sizeof(struct imsm_update_reshape_migration); | |
8209 | ||
8210 | u = calloc(1, update_memory_size); | |
8211 | if (u == NULL) { | |
8212 | dprintf("error: cannot get memory for " | |
8213 | "imsm_create_metadata_update_for_migration\n"); | |
8214 | return 0; | |
8215 | } | |
8216 | u->type = update_reshape_migration; | |
8217 | u->subdev = super->current_vol; | |
8218 | u->new_level = geo->level; | |
8219 | u->new_layout = geo->layout; | |
8220 | u->new_raid_disks = u->old_raid_disks = geo->raid_disks; | |
8221 | u->new_disks[0] = -1; | |
8222 | u->new_chunksize = -1; | |
8223 | ||
8224 | dev = get_imsm_dev(super, u->subdev); | |
8225 | if (dev) { | |
8226 | struct imsm_map *map; | |
8227 | ||
8228 | map = get_imsm_map(dev, 0); | |
8229 | if (map) { | |
8230 | int current_chunk_size = | |
8231 | __le16_to_cpu(map->blocks_per_strip) / 2; | |
8232 | ||
8233 | if (geo->chunksize != current_chunk_size) { | |
8234 | u->new_chunksize = geo->chunksize / 1024; | |
8235 | dprintf("imsm: " | |
8236 | "chunk size change from %i to %i\n", | |
8237 | current_chunk_size, u->new_chunksize); | |
8238 | } | |
8239 | previous_level = map->raid_level; | |
8240 | } | |
8241 | } | |
8242 | if ((geo->level == 5) && (previous_level == 0)) { | |
8243 | struct mdinfo *spares = NULL; | |
8244 | ||
8245 | u->new_raid_disks++; | |
8246 | spares = get_spares_for_grow(st); | |
8247 | if ((spares == NULL) || (spares->array.spare_disks < 1)) { | |
8248 | free(u); | |
8249 | sysfs_free(spares); | |
8250 | update_memory_size = 0; | |
8251 | dprintf("error: cannot get spare device " | |
8252 | "for requested migration"); | |
8253 | return 0; | |
8254 | } | |
8255 | sysfs_free(spares); | |
8256 | } | |
8257 | dprintf("imsm: reshape update preparation : OK\n"); | |
8258 | *updatep = u; | |
8259 | ||
8260 | return update_memory_size; | |
8261 | } | |
8262 | ||
8263 | static void imsm_update_metadata_locally(struct supertype *st, | |
8264 | void *buf, int len) | |
8265 | { | |
8266 | struct metadata_update mu; | |
8267 | ||
8268 | mu.buf = buf; | |
8269 | mu.len = len; | |
8270 | mu.space = NULL; | |
8271 | mu.space_list = NULL; | |
8272 | mu.next = NULL; | |
8273 | imsm_prepare_update(st, &mu); | |
8274 | imsm_process_update(st, &mu); | |
8275 | ||
8276 | while (mu.space_list) { | |
8277 | void **space = mu.space_list; | |
8278 | mu.space_list = *space; | |
8279 | free(space); | |
8280 | } | |
8281 | } | |
8282 | ||
8283 | /*************************************************************************** | |
8284 | * Function: imsm_analyze_change | |
8285 | * Description: Function analyze change for single volume | |
8286 | * and validate if transition is supported | |
8287 | * Parameters: Geometry parameters, supertype structure | |
8288 | * Returns: Operation type code on success, -1 if fail | |
8289 | ****************************************************************************/ | |
8290 | enum imsm_reshape_type imsm_analyze_change(struct supertype *st, | |
8291 | struct geo_params *geo) | |
8292 | { | |
8293 | struct mdinfo info; | |
8294 | int change = -1; | |
8295 | int check_devs = 0; | |
8296 | int chunk; | |
8297 | ||
8298 | getinfo_super_imsm_volume(st, &info, NULL); | |
8299 | if ((geo->level != info.array.level) && | |
8300 | (geo->level >= 0) && | |
8301 | (geo->level != UnSet)) { | |
8302 | switch (info.array.level) { | |
8303 | case 0: | |
8304 | if (geo->level == 5) { | |
8305 | change = CH_MIGRATION; | |
8306 | if (geo->layout != ALGORITHM_LEFT_ASYMMETRIC) { | |
8307 | fprintf(stderr, | |
8308 | Name " Error. Requested Layout " | |
8309 | "not supported (left-asymmetric layout " | |
8310 | "is supported only)!\n"); | |
8311 | change = -1; | |
8312 | goto analyse_change_exit; | |
8313 | } | |
8314 | check_devs = 1; | |
8315 | } | |
8316 | if (geo->level == 10) { | |
8317 | change = CH_TAKEOVER; | |
8318 | check_devs = 1; | |
8319 | } | |
8320 | break; | |
8321 | case 1: | |
8322 | if (geo->level == 0) { | |
8323 | change = CH_TAKEOVER; | |
8324 | check_devs = 1; | |
8325 | } | |
8326 | break; | |
8327 | case 10: | |
8328 | if (geo->level == 0) { | |
8329 | change = CH_TAKEOVER; | |
8330 | check_devs = 1; | |
8331 | } | |
8332 | break; | |
8333 | } | |
8334 | if (change == -1) { | |
8335 | fprintf(stderr, | |
8336 | Name " Error. Level Migration from %d to %d " | |
8337 | "not supported!\n", | |
8338 | info.array.level, geo->level); | |
8339 | goto analyse_change_exit; | |
8340 | } | |
8341 | } else | |
8342 | geo->level = info.array.level; | |
8343 | ||
8344 | if ((geo->layout != info.array.layout) | |
8345 | && ((geo->layout != UnSet) && (geo->layout != -1))) { | |
8346 | change = CH_MIGRATION; | |
8347 | if ((info.array.layout == 0) | |
8348 | && (info.array.level == 5) | |
8349 | && (geo->layout == 5)) { | |
8350 | /* reshape 5 -> 4 */ | |
8351 | } else if ((info.array.layout == 5) | |
8352 | && (info.array.level == 5) | |
8353 | && (geo->layout == 0)) { | |
8354 | /* reshape 4 -> 5 */ | |
8355 | geo->layout = 0; | |
8356 | geo->level = 5; | |
8357 | } else { | |
8358 | fprintf(stderr, | |
8359 | Name " Error. Layout Migration from %d to %d " | |
8360 | "not supported!\n", | |
8361 | info.array.layout, geo->layout); | |
8362 | change = -1; | |
8363 | goto analyse_change_exit; | |
8364 | } | |
8365 | } else | |
8366 | geo->layout = info.array.layout; | |
8367 | ||
8368 | if ((geo->chunksize > 0) && (geo->chunksize != UnSet) | |
8369 | && (geo->chunksize != info.array.chunk_size)) | |
8370 | change = CH_MIGRATION; | |
8371 | else | |
8372 | geo->chunksize = info.array.chunk_size; | |
8373 | ||
8374 | chunk = geo->chunksize / 1024; | |
8375 | if (!validate_geometry_imsm(st, | |
8376 | geo->level, | |
8377 | geo->layout, | |
8378 | geo->raid_disks, | |
8379 | &chunk, | |
8380 | geo->size, | |
8381 | 0, 0, 1)) | |
8382 | change = -1; | |
8383 | ||
8384 | if (check_devs) { | |
8385 | struct intel_super *super = st->sb; | |
8386 | struct imsm_super *mpb = super->anchor; | |
8387 | ||
8388 | if (mpb->num_raid_devs > 1) { | |
8389 | fprintf(stderr, | |
8390 | Name " Error. Cannot perform operation on %s" | |
8391 | "- for this operation it MUST be single " | |
8392 | "array in container\n", | |
8393 | geo->dev_name); | |
8394 | change = -1; | |
8395 | } | |
8396 | } | |
8397 | ||
8398 | analyse_change_exit: | |
8399 | ||
8400 | return change; | |
8401 | } | |
8402 | ||
8403 | int imsm_takeover(struct supertype *st, struct geo_params *geo) | |
8404 | { | |
8405 | struct intel_super *super = st->sb; | |
8406 | struct imsm_update_takeover *u; | |
8407 | ||
8408 | u = malloc(sizeof(struct imsm_update_takeover)); | |
8409 | if (u == NULL) | |
8410 | return 1; | |
8411 | ||
8412 | u->type = update_takeover; | |
8413 | u->subarray = super->current_vol; | |
8414 | ||
8415 | /* 10->0 transition */ | |
8416 | if (geo->level == 0) | |
8417 | u->direction = R10_TO_R0; | |
8418 | ||
8419 | /* 0->10 transition */ | |
8420 | if (geo->level == 10) | |
8421 | u->direction = R0_TO_R10; | |
8422 | ||
8423 | /* update metadata locally */ | |
8424 | imsm_update_metadata_locally(st, u, | |
8425 | sizeof(struct imsm_update_takeover)); | |
8426 | /* and possibly remotely */ | |
8427 | if (st->update_tail) | |
8428 | append_metadata_update(st, u, | |
8429 | sizeof(struct imsm_update_takeover)); | |
8430 | else | |
8431 | free(u); | |
8432 | ||
8433 | return 0; | |
8434 | } | |
8435 | ||
8436 | static int imsm_reshape_super(struct supertype *st, long long size, int level, | |
8437 | int layout, int chunksize, int raid_disks, | |
8438 | int delta_disks, char *backup, char *dev, | |
8439 | int verbose) | |
8440 | { | |
8441 | int ret_val = 1; | |
8442 | struct geo_params geo; | |
8443 | ||
8444 | dprintf("imsm: reshape_super called.\n"); | |
8445 | ||
8446 | memset(&geo, 0, sizeof(struct geo_params)); | |
8447 | ||
8448 | geo.dev_name = dev; | |
8449 | geo.dev_id = st->devnum; | |
8450 | geo.size = size; | |
8451 | geo.level = level; | |
8452 | geo.layout = layout; | |
8453 | geo.chunksize = chunksize; | |
8454 | geo.raid_disks = raid_disks; | |
8455 | if (delta_disks != UnSet) | |
8456 | geo.raid_disks += delta_disks; | |
8457 | ||
8458 | dprintf("\tfor level : %i\n", geo.level); | |
8459 | dprintf("\tfor raid_disks : %i\n", geo.raid_disks); | |
8460 | ||
8461 | if (experimental() == 0) | |
8462 | return ret_val; | |
8463 | ||
8464 | if (st->container_dev == st->devnum) { | |
8465 | /* On container level we can only increase number of devices. */ | |
8466 | dprintf("imsm: info: Container operation\n"); | |
8467 | int old_raid_disks = 0; | |
8468 | ||
8469 | if (imsm_reshape_is_allowed_on_container( | |
8470 | st, &geo, &old_raid_disks)) { | |
8471 | struct imsm_update_reshape *u = NULL; | |
8472 | int len; | |
8473 | ||
8474 | len = imsm_create_metadata_update_for_reshape( | |
8475 | st, &geo, old_raid_disks, &u); | |
8476 | ||
8477 | if (len <= 0) { | |
8478 | dprintf("imsm: Cannot prepare update\n"); | |
8479 | goto exit_imsm_reshape_super; | |
8480 | } | |
8481 | ||
8482 | ret_val = 0; | |
8483 | /* update metadata locally */ | |
8484 | imsm_update_metadata_locally(st, u, len); | |
8485 | /* and possibly remotely */ | |
8486 | if (st->update_tail) | |
8487 | append_metadata_update(st, u, len); | |
8488 | else | |
8489 | free(u); | |
8490 | ||
8491 | } else { | |
8492 | fprintf(stderr, Name ": (imsm) Operation " | |
8493 | "is not allowed on this container\n"); | |
8494 | } | |
8495 | } else { | |
8496 | /* On volume level we support following operations | |
8497 | * - takeover: raid10 -> raid0; raid0 -> raid10 | |
8498 | * - chunk size migration | |
8499 | * - migration: raid5 -> raid0; raid0 -> raid5 | |
8500 | */ | |
8501 | struct intel_super *super = st->sb; | |
8502 | struct intel_dev *dev = super->devlist; | |
8503 | int change, devnum; | |
8504 | dprintf("imsm: info: Volume operation\n"); | |
8505 | /* find requested device */ | |
8506 | while (dev) { | |
8507 | imsm_find_array_minor_by_subdev(dev->index, st->container_dev, &devnum); | |
8508 | if (devnum == geo.dev_id) | |
8509 | break; | |
8510 | dev = dev->next; | |
8511 | } | |
8512 | if (dev == NULL) { | |
8513 | fprintf(stderr, Name " Cannot find %s (%i) subarray\n", | |
8514 | geo.dev_name, geo.dev_id); | |
8515 | goto exit_imsm_reshape_super; | |
8516 | } | |
8517 | super->current_vol = dev->index; | |
8518 | change = imsm_analyze_change(st, &geo); | |
8519 | switch (change) { | |
8520 | case CH_TAKEOVER: | |
8521 | ret_val = imsm_takeover(st, &geo); | |
8522 | break; | |
8523 | case CH_MIGRATION: { | |
8524 | struct imsm_update_reshape_migration *u = NULL; | |
8525 | int len = | |
8526 | imsm_create_metadata_update_for_migration( | |
8527 | st, &geo, &u); | |
8528 | if (len < 1) { | |
8529 | dprintf("imsm: " | |
8530 | "Cannot prepare update\n"); | |
8531 | break; | |
8532 | } | |
8533 | ret_val = 0; | |
8534 | /* update metadata locally */ | |
8535 | imsm_update_metadata_locally(st, u, len); | |
8536 | /* and possibly remotely */ | |
8537 | if (st->update_tail) | |
8538 | append_metadata_update(st, u, len); | |
8539 | else | |
8540 | free(u); | |
8541 | } | |
8542 | break; | |
8543 | default: | |
8544 | ret_val = 1; | |
8545 | } | |
8546 | } | |
8547 | ||
8548 | exit_imsm_reshape_super: | |
8549 | dprintf("imsm: reshape_super Exit code = %i\n", ret_val); | |
8550 | return ret_val; | |
8551 | } | |
8552 | ||
8553 | /******************************************************************************* | |
8554 | * Function: wait_for_reshape_imsm | |
8555 | * Description: Function writes new sync_max value and waits until | |
8556 | * reshape process reach new position | |
8557 | * Parameters: | |
8558 | * sra : general array info | |
8559 | * to_complete : new sync_max position | |
8560 | * ndata : number of disks in new array's layout | |
8561 | * Returns: | |
8562 | * 0 : success, | |
8563 | * 1 : there is no reshape in progress, | |
8564 | * -1 : fail | |
8565 | ******************************************************************************/ | |
8566 | int wait_for_reshape_imsm(struct mdinfo *sra, unsigned long long to_complete, | |
8567 | int ndata) | |
8568 | { | |
8569 | int fd = sysfs_get_fd(sra, NULL, "reshape_position"); | |
8570 | unsigned long long completed; | |
8571 | ||
8572 | struct timeval timeout; | |
8573 | ||
8574 | if (fd < 0) | |
8575 | return 1; | |
8576 | ||
8577 | sysfs_fd_get_ll(fd, &completed); | |
8578 | ||
8579 | if (to_complete == 0) {/* reshape till the end of array */ | |
8580 | sysfs_set_str(sra, NULL, "sync_max", "max"); | |
8581 | to_complete = MaxSector; | |
8582 | } else { | |
8583 | if (completed > to_complete) { | |
8584 | close(fd); | |
8585 | return -1; | |
8586 | } | |
8587 | if (sysfs_set_num(sra, NULL, "sync_max", | |
8588 | to_complete / ndata) != 0) { | |
8589 | close(fd); | |
8590 | return -1; | |
8591 | } | |
8592 | } | |
8593 | ||
8594 | /* FIXME should not need a timeout at all */ | |
8595 | timeout.tv_sec = 30; | |
8596 | timeout.tv_usec = 0; | |
8597 | do { | |
8598 | char action[20]; | |
8599 | fd_set rfds; | |
8600 | FD_ZERO(&rfds); | |
8601 | FD_SET(fd, &rfds); | |
8602 | select(fd+1, NULL, NULL, &rfds, &timeout); | |
8603 | if (sysfs_fd_get_ll(fd, &completed) < 0) { | |
8604 | close(fd); | |
8605 | return 1; | |
8606 | } | |
8607 | if (sysfs_get_str(sra, NULL, "sync_action", | |
8608 | action, 20) > 0 && | |
8609 | strncmp(action, "reshape", 7) != 0) | |
8610 | break; | |
8611 | } while (completed < to_complete); | |
8612 | close(fd); | |
8613 | return 0; | |
8614 | ||
8615 | } | |
8616 | ||
8617 | /******************************************************************************* | |
8618 | * Function: check_degradation_change | |
8619 | * Description: Check that array hasn't become failed. | |
8620 | * Parameters: | |
8621 | * info : for sysfs access | |
8622 | * sources : source disks descriptors | |
8623 | * degraded: previous degradation level | |
8624 | * Returns: | |
8625 | * degradation level | |
8626 | ******************************************************************************/ | |
8627 | int check_degradation_change(struct mdinfo *info, | |
8628 | int *sources, | |
8629 | int degraded) | |
8630 | { | |
8631 | unsigned long long new_degraded; | |
8632 | sysfs_get_ll(info, NULL, "degraded", &new_degraded); | |
8633 | if (new_degraded != (unsigned long long)degraded) { | |
8634 | /* check each device to ensure it is still working */ | |
8635 | struct mdinfo *sd; | |
8636 | new_degraded = 0; | |
8637 | for (sd = info->devs ; sd ; sd = sd->next) { | |
8638 | if (sd->disk.state & (1<<MD_DISK_FAULTY)) | |
8639 | continue; | |
8640 | if (sd->disk.state & (1<<MD_DISK_SYNC)) { | |
8641 | char sbuf[20]; | |
8642 | if (sysfs_get_str(info, | |
8643 | sd, "state", sbuf, 20) < 0 || | |
8644 | strstr(sbuf, "faulty") || | |
8645 | strstr(sbuf, "in_sync") == NULL) { | |
8646 | /* this device is dead */ | |
8647 | sd->disk.state = (1<<MD_DISK_FAULTY); | |
8648 | if (sd->disk.raid_disk >= 0 && | |
8649 | sources[sd->disk.raid_disk] >= 0) { | |
8650 | close(sources[ | |
8651 | sd->disk.raid_disk]); | |
8652 | sources[sd->disk.raid_disk] = | |
8653 | -1; | |
8654 | } | |
8655 | new_degraded++; | |
8656 | } | |
8657 | } | |
8658 | } | |
8659 | } | |
8660 | ||
8661 | return new_degraded; | |
8662 | } | |
8663 | ||
8664 | /******************************************************************************* | |
8665 | * Function: imsm_manage_reshape | |
8666 | * Description: Function finds array under reshape and it manages reshape | |
8667 | * process. It creates stripes backups (if required) and sets | |
8668 | * checheckpoits. | |
8669 | * Parameters: | |
8670 | * afd : Backup handle (nattive) - not used | |
8671 | * sra : general array info | |
8672 | * reshape : reshape parameters - not used | |
8673 | * st : supertype structure | |
8674 | * blocks : size of critical section [blocks] | |
8675 | * fds : table of source device descriptor | |
8676 | * offsets : start of array (offest per devices) | |
8677 | * dests : not used | |
8678 | * destfd : table of destination device descriptor | |
8679 | * destoffsets : table of destination offsets (per device) | |
8680 | * Returns: | |
8681 | * 1 : success, reshape is done | |
8682 | * 0 : fail | |
8683 | ******************************************************************************/ | |
8684 | static int imsm_manage_reshape( | |
8685 | int afd, struct mdinfo *sra, struct reshape *reshape, | |
8686 | struct supertype *st, unsigned long backup_blocks, | |
8687 | int *fds, unsigned long long *offsets, | |
8688 | int dests, int *destfd, unsigned long long *destoffsets) | |
8689 | { | |
8690 | int ret_val = 0; | |
8691 | struct intel_super *super = st->sb; | |
8692 | struct intel_dev *dv = NULL; | |
8693 | struct imsm_dev *dev = NULL; | |
8694 | struct imsm_map *map_src, *map_dest; | |
8695 | int migr_vol_qan = 0; | |
8696 | int ndata, odata; /* [bytes] */ | |
8697 | int chunk; /* [bytes] */ | |
8698 | struct migr_record *migr_rec; | |
8699 | char *buf = NULL; | |
8700 | unsigned int buf_size; /* [bytes] */ | |
8701 | unsigned long long max_position; /* array size [bytes] */ | |
8702 | unsigned long long next_step; /* [blocks]/[bytes] */ | |
8703 | unsigned long long old_data_stripe_length; | |
8704 | unsigned long long new_data_stripe_length; | |
8705 | unsigned long long start_src; /* [bytes] */ | |
8706 | unsigned long long start; /* [bytes] */ | |
8707 | unsigned long long start_buf_shift; /* [bytes] */ | |
8708 | int degraded = 0; | |
8709 | int source_layout = 0; | |
8710 | ||
8711 | if (!fds || !offsets || !sra) | |
8712 | goto abort; | |
8713 | ||
8714 | /* Find volume during the reshape */ | |
8715 | for (dv = super->devlist; dv; dv = dv->next) { | |
8716 | if (dv->dev->vol.migr_type == MIGR_GEN_MIGR | |
8717 | && dv->dev->vol.migr_state == 1) { | |
8718 | dev = dv->dev; | |
8719 | migr_vol_qan++; | |
8720 | } | |
8721 | } | |
8722 | /* Only one volume can migrate at the same time */ | |
8723 | if (migr_vol_qan != 1) { | |
8724 | fprintf(stderr, Name " : %s", migr_vol_qan ? | |
8725 | "Number of migrating volumes greater than 1\n" : | |
8726 | "There is no volume during migrationg\n"); | |
8727 | goto abort; | |
8728 | } | |
8729 | ||
8730 | map_src = get_imsm_map(dev, 1); | |
8731 | if (map_src == NULL) | |
8732 | goto abort; | |
8733 | map_dest = get_imsm_map(dev, 0); | |
8734 | ||
8735 | ndata = imsm_num_data_members(dev, 0); | |
8736 | odata = imsm_num_data_members(dev, 1); | |
8737 | ||
8738 | chunk = __le16_to_cpu(map_src->blocks_per_strip) * 512; | |
8739 | old_data_stripe_length = odata * chunk; | |
8740 | ||
8741 | migr_rec = super->migr_rec; | |
8742 | ||
8743 | /* [bytes] */ | |
8744 | sra->new_chunk = __le16_to_cpu(map_dest->blocks_per_strip) * 512; | |
8745 | sra->new_level = map_dest->raid_level; | |
8746 | new_data_stripe_length = sra->new_chunk * ndata; | |
8747 | ||
8748 | /* initialize migration record for start condition */ | |
8749 | if (sra->reshape_progress == 0) | |
8750 | init_migr_record_imsm(st, dev, sra); | |
8751 | ||
8752 | /* size for data */ | |
8753 | buf_size = __le32_to_cpu(migr_rec->blocks_per_unit) * 512; | |
8754 | /* extend buffer size for parity disk */ | |
8755 | buf_size += __le32_to_cpu(migr_rec->dest_depth_per_unit) * 512; | |
8756 | /* add space for stripe aligment */ | |
8757 | buf_size += old_data_stripe_length; | |
8758 | if (posix_memalign((void **)&buf, 4096, buf_size)) { | |
8759 | dprintf("imsm: Cannot allocate checpoint buffer\n"); | |
8760 | goto abort; | |
8761 | } | |
8762 | ||
8763 | max_position = sra->component_size * ndata; | |
8764 | if (map_src->raid_level != 0) | |
8765 | source_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8766 | ||
8767 | while (__le32_to_cpu(migr_rec->curr_migr_unit) < | |
8768 | __le32_to_cpu(migr_rec->num_migr_units)) { | |
8769 | /* current reshape position [blocks] */ | |
8770 | unsigned long long current_position = | |
8771 | __le32_to_cpu(migr_rec->blocks_per_unit) | |
8772 | * __le32_to_cpu(migr_rec->curr_migr_unit); | |
8773 | unsigned long long border; | |
8774 | ||
8775 | /* Check that array hasn't become failed. | |
8776 | */ | |
8777 | degraded = check_degradation_change(sra, fds, degraded); | |
8778 | if (degraded > 1) { | |
8779 | dprintf("imsm: Abort reshape due to degradation" | |
8780 | " level (%i)\n", degraded); | |
8781 | goto abort; | |
8782 | } | |
8783 | ||
8784 | next_step = __le32_to_cpu(migr_rec->blocks_per_unit); | |
8785 | ||
8786 | if ((current_position + next_step) > max_position) | |
8787 | next_step = max_position - current_position; | |
8788 | ||
8789 | start = current_position * 512; | |
8790 | ||
8791 | /* allign reading start to old geometry */ | |
8792 | start_buf_shift = start % old_data_stripe_length; | |
8793 | start_src = start - start_buf_shift; | |
8794 | ||
8795 | border = (start_src / odata) - (start / ndata); | |
8796 | border /= 512; | |
8797 | if (border <= __le32_to_cpu(migr_rec->dest_depth_per_unit)) { | |
8798 | /* save critical stripes to buf | |
8799 | * start - start address of current unit | |
8800 | * to backup [bytes] | |
8801 | * start_src - start address of current unit | |
8802 | * to backup alligned to source array | |
8803 | * [bytes] | |
8804 | */ | |
8805 | unsigned long long next_step_filler = 0; | |
8806 | unsigned long long copy_length = next_step * 512; | |
8807 | ||
8808 | /* allign copy area length to stripe in old geometry */ | |
8809 | next_step_filler = ((copy_length + start_buf_shift) | |
8810 | % old_data_stripe_length); | |
8811 | if (next_step_filler) | |
8812 | next_step_filler = (old_data_stripe_length | |
8813 | - next_step_filler); | |
8814 | dprintf("save_stripes() parameters: start = %llu," | |
8815 | "\tstart_src = %llu,\tnext_step*512 = %llu," | |
8816 | "\tstart_in_buf_shift = %llu," | |
8817 | "\tnext_step_filler = %llu\n", | |
8818 | start, start_src, copy_length, | |
8819 | start_buf_shift, next_step_filler); | |
8820 | ||
8821 | if (save_stripes(fds, offsets, map_src->num_members, | |
8822 | chunk, map_src->raid_level, | |
8823 | source_layout, 0, NULL, start_src, | |
8824 | copy_length + | |
8825 | next_step_filler + start_buf_shift, | |
8826 | buf)) { | |
8827 | dprintf("imsm: Cannot save stripes" | |
8828 | " to buffer\n"); | |
8829 | goto abort; | |
8830 | } | |
8831 | /* Convert data to destination format and store it | |
8832 | * in backup general migration area | |
8833 | */ | |
8834 | if (save_backup_imsm(st, dev, sra, | |
8835 | buf + start_buf_shift, copy_length)) { | |
8836 | dprintf("imsm: Cannot save stripes to " | |
8837 | "target devices\n"); | |
8838 | goto abort; | |
8839 | } | |
8840 | if (save_checkpoint_imsm(st, sra, | |
8841 | UNIT_SRC_IN_CP_AREA)) { | |
8842 | dprintf("imsm: Cannot write checkpoint to " | |
8843 | "migration record (UNIT_SRC_IN_CP_AREA)\n"); | |
8844 | goto abort; | |
8845 | } | |
8846 | /* decrease backup_blocks */ | |
8847 | if (backup_blocks > (unsigned long)next_step) | |
8848 | backup_blocks -= next_step; | |
8849 | else | |
8850 | backup_blocks = 0; | |
8851 | } | |
8852 | /* When data backed up, checkpoint stored, | |
8853 | * kick the kernel to reshape unit of data | |
8854 | */ | |
8855 | next_step = next_step + sra->reshape_progress; | |
8856 | sysfs_set_num(sra, NULL, "suspend_lo", sra->reshape_progress); | |
8857 | sysfs_set_num(sra, NULL, "suspend_hi", next_step); | |
8858 | ||
8859 | /* wait until reshape finish */ | |
8860 | if (wait_for_reshape_imsm(sra, next_step, ndata) < 0) { | |
8861 | dprintf("wait_for_reshape_imsm returned error!\n"); | |
8862 | goto abort; | |
8863 | } | |
8864 | ||
8865 | sra->reshape_progress = next_step; | |
8866 | ||
8867 | if (save_checkpoint_imsm(st, sra, UNIT_SRC_NORMAL) == 1) { | |
8868 | /* ignore error == 2, this can mean end of reshape here | |
8869 | */ | |
8870 | dprintf("imsm: Cannot write checkpoint to " | |
8871 | "migration record (UNIT_SRC_NORMAL)\n"); | |
8872 | goto abort; | |
8873 | } | |
8874 | ||
8875 | } | |
8876 | ||
8877 | /* return '1' if done */ | |
8878 | ret_val = 1; | |
8879 | abort: | |
8880 | free(buf); | |
8881 | abort_reshape(sra); | |
8882 | ||
8883 | return ret_val; | |
8884 | } | |
8885 | #endif /* MDASSEMBLE */ | |
8886 | ||
8887 | struct superswitch super_imsm = { | |
8888 | #ifndef MDASSEMBLE | |
8889 | .examine_super = examine_super_imsm, | |
8890 | .brief_examine_super = brief_examine_super_imsm, | |
8891 | .brief_examine_subarrays = brief_examine_subarrays_imsm, | |
8892 | .export_examine_super = export_examine_super_imsm, | |
8893 | .detail_super = detail_super_imsm, | |
8894 | .brief_detail_super = brief_detail_super_imsm, | |
8895 | .write_init_super = write_init_super_imsm, | |
8896 | .validate_geometry = validate_geometry_imsm, | |
8897 | .add_to_super = add_to_super_imsm, | |
8898 | .remove_from_super = remove_from_super_imsm, | |
8899 | .detail_platform = detail_platform_imsm, | |
8900 | .kill_subarray = kill_subarray_imsm, | |
8901 | .update_subarray = update_subarray_imsm, | |
8902 | .load_container = load_container_imsm, | |
8903 | .default_geometry = default_geometry_imsm, | |
8904 | .get_disk_controller_domain = imsm_get_disk_controller_domain, | |
8905 | .reshape_super = imsm_reshape_super, | |
8906 | .manage_reshape = imsm_manage_reshape, | |
8907 | #endif | |
8908 | .match_home = match_home_imsm, | |
8909 | .uuid_from_super= uuid_from_super_imsm, | |
8910 | .getinfo_super = getinfo_super_imsm, | |
8911 | .getinfo_super_disks = getinfo_super_disks_imsm, | |
8912 | .update_super = update_super_imsm, | |
8913 | ||
8914 | .avail_size = avail_size_imsm, | |
8915 | .min_acceptable_spare_size = min_acceptable_spare_size_imsm, | |
8916 | ||
8917 | .compare_super = compare_super_imsm, | |
8918 | ||
8919 | .load_super = load_super_imsm, | |
8920 | .init_super = init_super_imsm, | |
8921 | .store_super = store_super_imsm, | |
8922 | .free_super = free_super_imsm, | |
8923 | .match_metadata_desc = match_metadata_desc_imsm, | |
8924 | .container_content = container_content_imsm, | |
8925 | ||
8926 | .recover_backup = recover_backup_imsm, | |
8927 | ||
8928 | .external = 1, | |
8929 | .name = "imsm", | |
8930 | ||
8931 | #ifndef MDASSEMBLE | |
8932 | /* for mdmon */ | |
8933 | .open_new = imsm_open_new, | |
8934 | .set_array_state= imsm_set_array_state, | |
8935 | .set_disk = imsm_set_disk, | |
8936 | .sync_metadata = imsm_sync_metadata, | |
8937 | .activate_spare = imsm_activate_spare, | |
8938 | .process_update = imsm_process_update, | |
8939 | .prepare_update = imsm_prepare_update, | |
8940 | #endif /* MDASSEMBLE */ | |
8941 | }; |