]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de> | |
5 | * | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neil@brown.name> | |
23 | * | |
24 | * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2 | |
25 | * (July 28 2006). Reused by permission of SNIA. | |
26 | */ | |
27 | ||
28 | #define HAVE_STDINT_H 1 | |
29 | #include "mdadm.h" | |
30 | #include "mdmon.h" | |
31 | #include "sha1.h" | |
32 | #include <values.h> | |
33 | ||
34 | /* a non-official T10 name for creation GUIDs */ | |
35 | static char T10[] = "Linux-MD"; | |
36 | ||
37 | /* DDF timestamps are 1980 based, so we need to add | |
38 | * second-in-decade-of-seventies to convert to linux timestamps. | |
39 | * 10 years with 2 leap years. | |
40 | */ | |
41 | #define DECADE (3600*24*(365*10+2)) | |
42 | unsigned long crc32( | |
43 | unsigned long crc, | |
44 | const unsigned char *buf, | |
45 | unsigned len); | |
46 | ||
47 | #define DDF_NOTFOUND (~0U) | |
48 | #define DDF_CONTAINER (DDF_NOTFOUND-1) | |
49 | ||
50 | /* Default for safe_mode_delay. Same value as for IMSM. | |
51 | */ | |
52 | static const int DDF_SAFE_MODE_DELAY = 4000; | |
53 | ||
54 | /* The DDF metadata handling. | |
55 | * DDF metadata lives at the end of the device. | |
56 | * The last 512 byte block provides an 'anchor' which is used to locate | |
57 | * the rest of the metadata which usually lives immediately behind the anchor. | |
58 | * | |
59 | * Note: | |
60 | * - all multibyte numeric fields are bigendian. | |
61 | * - all strings are space padded. | |
62 | * | |
63 | */ | |
64 | ||
65 | typedef struct __be16 { | |
66 | __u16 _v16; | |
67 | } be16; | |
68 | #define be16_eq(x, y) ((x)._v16 == (y)._v16) | |
69 | #define be16_and(x, y) ((x)._v16 & (y)._v16) | |
70 | #define be16_or(x, y) ((x)._v16 | (y)._v16) | |
71 | #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16) | |
72 | #define be16_set(x, y) ((x)._v16 |= (y)._v16) | |
73 | ||
74 | typedef struct __be32 { | |
75 | __u32 _v32; | |
76 | } be32; | |
77 | #define be32_eq(x, y) ((x)._v32 == (y)._v32) | |
78 | ||
79 | typedef struct __be64 { | |
80 | __u64 _v64; | |
81 | } be64; | |
82 | #define be64_eq(x, y) ((x)._v64 == (y)._v64) | |
83 | ||
84 | #define be16_to_cpu(be) __be16_to_cpu((be)._v16) | |
85 | static inline be16 cpu_to_be16(__u16 x) | |
86 | { | |
87 | be16 be = { ._v16 = __cpu_to_be16(x) }; | |
88 | return be; | |
89 | } | |
90 | ||
91 | #define be32_to_cpu(be) __be32_to_cpu((be)._v32) | |
92 | static inline be32 cpu_to_be32(__u32 x) | |
93 | { | |
94 | be32 be = { ._v32 = __cpu_to_be32(x) }; | |
95 | return be; | |
96 | } | |
97 | ||
98 | #define be64_to_cpu(be) __be64_to_cpu((be)._v64) | |
99 | static inline be64 cpu_to_be64(__u64 x) | |
100 | { | |
101 | be64 be = { ._v64 = __cpu_to_be64(x) }; | |
102 | return be; | |
103 | } | |
104 | ||
105 | /* Primary Raid Level (PRL) */ | |
106 | #define DDF_RAID0 0x00 | |
107 | #define DDF_RAID1 0x01 | |
108 | #define DDF_RAID3 0x03 | |
109 | #define DDF_RAID4 0x04 | |
110 | #define DDF_RAID5 0x05 | |
111 | #define DDF_RAID1E 0x11 | |
112 | #define DDF_JBOD 0x0f | |
113 | #define DDF_CONCAT 0x1f | |
114 | #define DDF_RAID5E 0x15 | |
115 | #define DDF_RAID5EE 0x25 | |
116 | #define DDF_RAID6 0x06 | |
117 | ||
118 | /* Raid Level Qualifier (RLQ) */ | |
119 | #define DDF_RAID0_SIMPLE 0x00 | |
120 | #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */ | |
121 | #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */ | |
122 | #define DDF_RAID3_0 0x00 /* parity in first extent */ | |
123 | #define DDF_RAID3_N 0x01 /* parity in last extent */ | |
124 | #define DDF_RAID4_0 0x00 /* parity in first extent */ | |
125 | #define DDF_RAID4_N 0x01 /* parity in last extent */ | |
126 | /* these apply to raid5e and raid5ee as well */ | |
127 | #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */ | |
128 | #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */ | |
129 | #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */ | |
130 | #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */ | |
131 | ||
132 | #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */ | |
133 | #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */ | |
134 | ||
135 | /* Secondary RAID Level (SRL) */ | |
136 | #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */ | |
137 | #define DDF_2MIRRORED 0x01 | |
138 | #define DDF_2CONCAT 0x02 | |
139 | #define DDF_2SPANNED 0x03 /* This is also weird - be careful */ | |
140 | ||
141 | /* Magic numbers */ | |
142 | #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11) | |
143 | #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111) | |
144 | #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222) | |
145 | #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333) | |
146 | #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD) | |
147 | #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE) | |
148 | #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555) | |
149 | #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888) | |
150 | #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0) | |
151 | #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C) | |
152 | ||
153 | #define DDF_GUID_LEN 24 | |
154 | #define DDF_REVISION_0 "01.00.00" | |
155 | #define DDF_REVISION_2 "01.02.00" | |
156 | ||
157 | struct ddf_header { | |
158 | be32 magic; /* DDF_HEADER_MAGIC */ | |
159 | be32 crc; | |
160 | char guid[DDF_GUID_LEN]; | |
161 | char revision[8]; /* 01.02.00 */ | |
162 | be32 seq; /* starts at '1' */ | |
163 | be32 timestamp; | |
164 | __u8 openflag; | |
165 | __u8 foreignflag; | |
166 | __u8 enforcegroups; | |
167 | __u8 pad0; /* 0xff */ | |
168 | __u8 pad1[12]; /* 12 * 0xff */ | |
169 | /* 64 bytes so far */ | |
170 | __u8 header_ext[32]; /* reserved: fill with 0xff */ | |
171 | be64 primary_lba; | |
172 | be64 secondary_lba; | |
173 | __u8 type; | |
174 | __u8 pad2[3]; /* 0xff */ | |
175 | be32 workspace_len; /* sectors for vendor space - | |
176 | * at least 32768(sectors) */ | |
177 | be64 workspace_lba; | |
178 | be16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */ | |
179 | be16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */ | |
180 | be16 max_partitions; /* i.e. max num of configuration | |
181 | record entries per disk */ | |
182 | be16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries | |
183 | *12/512) */ | |
184 | be16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */ | |
185 | __u8 pad3[54]; /* 0xff */ | |
186 | /* 192 bytes so far */ | |
187 | be32 controller_section_offset; | |
188 | be32 controller_section_length; | |
189 | be32 phys_section_offset; | |
190 | be32 phys_section_length; | |
191 | be32 virt_section_offset; | |
192 | be32 virt_section_length; | |
193 | be32 config_section_offset; | |
194 | be32 config_section_length; | |
195 | be32 data_section_offset; | |
196 | be32 data_section_length; | |
197 | be32 bbm_section_offset; | |
198 | be32 bbm_section_length; | |
199 | be32 diag_space_offset; | |
200 | be32 diag_space_length; | |
201 | be32 vendor_offset; | |
202 | be32 vendor_length; | |
203 | /* 256 bytes so far */ | |
204 | __u8 pad4[256]; /* 0xff */ | |
205 | }; | |
206 | ||
207 | /* type field */ | |
208 | #define DDF_HEADER_ANCHOR 0x00 | |
209 | #define DDF_HEADER_PRIMARY 0x01 | |
210 | #define DDF_HEADER_SECONDARY 0x02 | |
211 | ||
212 | /* The content of the 'controller section' - global scope */ | |
213 | struct ddf_controller_data { | |
214 | be32 magic; /* DDF_CONTROLLER_MAGIC */ | |
215 | be32 crc; | |
216 | char guid[DDF_GUID_LEN]; | |
217 | struct controller_type { | |
218 | be16 vendor_id; | |
219 | be16 device_id; | |
220 | be16 sub_vendor_id; | |
221 | be16 sub_device_id; | |
222 | } type; | |
223 | char product_id[16]; | |
224 | __u8 pad[8]; /* 0xff */ | |
225 | __u8 vendor_data[448]; | |
226 | }; | |
227 | ||
228 | /* The content of phys_section - global scope */ | |
229 | struct phys_disk { | |
230 | be32 magic; /* DDF_PHYS_RECORDS_MAGIC */ | |
231 | be32 crc; | |
232 | be16 used_pdes; | |
233 | be16 max_pdes; | |
234 | __u8 pad[52]; | |
235 | struct phys_disk_entry { | |
236 | char guid[DDF_GUID_LEN]; | |
237 | be32 refnum; | |
238 | be16 type; | |
239 | be16 state; | |
240 | be64 config_size; /* DDF structures must be after here */ | |
241 | char path[18]; /* another horrible structure really */ | |
242 | __u8 pad[6]; | |
243 | } entries[0]; | |
244 | }; | |
245 | ||
246 | /* phys_disk_entry.type is a bitmap - bigendian remember */ | |
247 | #define DDF_Forced_PD_GUID 1 | |
248 | #define DDF_Active_in_VD 2 | |
249 | #define DDF_Global_Spare 4 /* VD_CONF records are ignored */ | |
250 | #define DDF_Spare 8 /* overrides Global_spare */ | |
251 | #define DDF_Foreign 16 | |
252 | #define DDF_Legacy 32 /* no DDF on this device */ | |
253 | ||
254 | #define DDF_Interface_mask 0xf00 | |
255 | #define DDF_Interface_SCSI 0x100 | |
256 | #define DDF_Interface_SAS 0x200 | |
257 | #define DDF_Interface_SATA 0x300 | |
258 | #define DDF_Interface_FC 0x400 | |
259 | ||
260 | /* phys_disk_entry.state is a bigendian bitmap */ | |
261 | #define DDF_Online 1 | |
262 | #define DDF_Failed 2 /* overrides 1,4,8 */ | |
263 | #define DDF_Rebuilding 4 | |
264 | #define DDF_Transition 8 | |
265 | #define DDF_SMART 16 | |
266 | #define DDF_ReadErrors 32 | |
267 | #define DDF_Missing 64 | |
268 | ||
269 | /* The content of the virt_section global scope */ | |
270 | struct virtual_disk { | |
271 | be32 magic; /* DDF_VIRT_RECORDS_MAGIC */ | |
272 | be32 crc; | |
273 | be16 populated_vdes; | |
274 | be16 max_vdes; | |
275 | __u8 pad[52]; | |
276 | struct virtual_entry { | |
277 | char guid[DDF_GUID_LEN]; | |
278 | be16 unit; | |
279 | __u16 pad0; /* 0xffff */ | |
280 | be16 guid_crc; | |
281 | be16 type; | |
282 | __u8 state; | |
283 | __u8 init_state; | |
284 | __u8 pad1[14]; | |
285 | char name[16]; | |
286 | } entries[0]; | |
287 | }; | |
288 | ||
289 | /* virtual_entry.type is a bitmap - bigendian */ | |
290 | #define DDF_Shared 1 | |
291 | #define DDF_Enforce_Groups 2 | |
292 | #define DDF_Unicode 4 | |
293 | #define DDF_Owner_Valid 8 | |
294 | ||
295 | /* virtual_entry.state is a bigendian bitmap */ | |
296 | #define DDF_state_mask 0x7 | |
297 | #define DDF_state_optimal 0x0 | |
298 | #define DDF_state_degraded 0x1 | |
299 | #define DDF_state_deleted 0x2 | |
300 | #define DDF_state_missing 0x3 | |
301 | #define DDF_state_failed 0x4 | |
302 | #define DDF_state_part_optimal 0x5 | |
303 | ||
304 | #define DDF_state_morphing 0x8 | |
305 | #define DDF_state_inconsistent 0x10 | |
306 | ||
307 | /* virtual_entry.init_state is a bigendian bitmap */ | |
308 | #define DDF_initstate_mask 0x03 | |
309 | #define DDF_init_not 0x00 | |
310 | #define DDF_init_quick 0x01 /* initialisation is progress. | |
311 | * i.e. 'state_inconsistent' */ | |
312 | #define DDF_init_full 0x02 | |
313 | ||
314 | #define DDF_access_mask 0xc0 | |
315 | #define DDF_access_rw 0x00 | |
316 | #define DDF_access_ro 0x80 | |
317 | #define DDF_access_blocked 0xc0 | |
318 | ||
319 | /* The content of the config_section - local scope | |
320 | * It has multiple records each config_record_len sectors | |
321 | * They can be vd_config or spare_assign | |
322 | */ | |
323 | ||
324 | struct vd_config { | |
325 | be32 magic; /* DDF_VD_CONF_MAGIC */ | |
326 | be32 crc; | |
327 | char guid[DDF_GUID_LEN]; | |
328 | be32 timestamp; | |
329 | be32 seqnum; | |
330 | __u8 pad0[24]; | |
331 | be16 prim_elmnt_count; | |
332 | __u8 chunk_shift; /* 0 == 512, 1==1024 etc */ | |
333 | __u8 prl; | |
334 | __u8 rlq; | |
335 | __u8 sec_elmnt_count; | |
336 | __u8 sec_elmnt_seq; | |
337 | __u8 srl; | |
338 | be64 blocks; /* blocks per component could be different | |
339 | * on different component devices...(only | |
340 | * for concat I hope) */ | |
341 | be64 array_blocks; /* blocks in array */ | |
342 | __u8 pad1[8]; | |
343 | be32 spare_refs[8]; | |
344 | __u8 cache_pol[8]; | |
345 | __u8 bg_rate; | |
346 | __u8 pad2[3]; | |
347 | __u8 pad3[52]; | |
348 | __u8 pad4[192]; | |
349 | __u8 v0[32]; /* reserved- 0xff */ | |
350 | __u8 v1[32]; /* reserved- 0xff */ | |
351 | __u8 v2[16]; /* reserved- 0xff */ | |
352 | __u8 v3[16]; /* reserved- 0xff */ | |
353 | __u8 vendor[32]; | |
354 | be32 phys_refnum[0]; /* refnum of each disk in sequence */ | |
355 | /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a | |
356 | bvd are always the same size */ | |
357 | }; | |
358 | #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe]) | |
359 | ||
360 | /* vd_config.cache_pol[7] is a bitmap */ | |
361 | #define DDF_cache_writeback 1 /* else writethrough */ | |
362 | #define DDF_cache_wadaptive 2 /* only applies if writeback */ | |
363 | #define DDF_cache_readahead 4 | |
364 | #define DDF_cache_radaptive 8 /* only if doing read-ahead */ | |
365 | #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */ | |
366 | #define DDF_cache_wallowed 32 /* enable write caching */ | |
367 | #define DDF_cache_rallowed 64 /* enable read caching */ | |
368 | ||
369 | struct spare_assign { | |
370 | be32 magic; /* DDF_SPARE_ASSIGN_MAGIC */ | |
371 | be32 crc; | |
372 | be32 timestamp; | |
373 | __u8 reserved[7]; | |
374 | __u8 type; | |
375 | be16 populated; /* SAEs used */ | |
376 | be16 max; /* max SAEs */ | |
377 | __u8 pad[8]; | |
378 | struct spare_assign_entry { | |
379 | char guid[DDF_GUID_LEN]; | |
380 | be16 secondary_element; | |
381 | __u8 pad[6]; | |
382 | } spare_ents[0]; | |
383 | }; | |
384 | /* spare_assign.type is a bitmap */ | |
385 | #define DDF_spare_dedicated 0x1 /* else global */ | |
386 | #define DDF_spare_revertible 0x2 /* else committable */ | |
387 | #define DDF_spare_active 0x4 /* else not active */ | |
388 | #define DDF_spare_affinity 0x8 /* enclosure affinity */ | |
389 | ||
390 | /* The data_section contents - local scope */ | |
391 | struct disk_data { | |
392 | be32 magic; /* DDF_PHYS_DATA_MAGIC */ | |
393 | be32 crc; | |
394 | char guid[DDF_GUID_LEN]; | |
395 | be32 refnum; /* crc of some magic drive data ... */ | |
396 | __u8 forced_ref; /* set when above was not result of magic */ | |
397 | __u8 forced_guid; /* set if guid was forced rather than magic */ | |
398 | __u8 vendor[32]; | |
399 | __u8 pad[442]; | |
400 | }; | |
401 | ||
402 | /* bbm_section content */ | |
403 | struct bad_block_log { | |
404 | be32 magic; | |
405 | be32 crc; | |
406 | be16 entry_count; | |
407 | be32 spare_count; | |
408 | __u8 pad[10]; | |
409 | be64 first_spare; | |
410 | struct mapped_block { | |
411 | be64 defective_start; | |
412 | be32 replacement_start; | |
413 | be16 remap_count; | |
414 | __u8 pad[2]; | |
415 | } entries[0]; | |
416 | }; | |
417 | ||
418 | /* Struct for internally holding ddf structures */ | |
419 | /* The DDF structure stored on each device is potentially | |
420 | * quite different, as some data is global and some is local. | |
421 | * The global data is: | |
422 | * - ddf header | |
423 | * - controller_data | |
424 | * - Physical disk records | |
425 | * - Virtual disk records | |
426 | * The local data is: | |
427 | * - Configuration records | |
428 | * - Physical Disk data section | |
429 | * ( and Bad block and vendor which I don't care about yet). | |
430 | * | |
431 | * The local data is parsed into separate lists as it is read | |
432 | * and reconstructed for writing. This means that we only need | |
433 | * to make config changes once and they are automatically | |
434 | * propagated to all devices. | |
435 | * Note that the ddf_super has space of the conf and disk data | |
436 | * for this disk and also for a list of all such data. | |
437 | * The list is only used for the superblock that is being | |
438 | * built in Create or Assemble to describe the whole array. | |
439 | */ | |
440 | struct ddf_super { | |
441 | struct ddf_header anchor, primary, secondary; | |
442 | struct ddf_controller_data controller; | |
443 | struct ddf_header *active; | |
444 | struct phys_disk *phys; | |
445 | struct virtual_disk *virt; | |
446 | char *conf; | |
447 | int pdsize, vdsize; | |
448 | unsigned int max_part, mppe, conf_rec_len; | |
449 | int currentdev; | |
450 | int updates_pending; | |
451 | struct vcl { | |
452 | union { | |
453 | char space[512]; | |
454 | struct { | |
455 | struct vcl *next; | |
456 | unsigned int vcnum; /* index into ->virt */ | |
457 | struct vd_config **other_bvds; | |
458 | __u64 *block_sizes; /* NULL if all the same */ | |
459 | }; | |
460 | }; | |
461 | struct vd_config conf; | |
462 | } *conflist, *currentconf; | |
463 | struct dl { | |
464 | union { | |
465 | char space[512]; | |
466 | struct { | |
467 | struct dl *next; | |
468 | int major, minor; | |
469 | char *devname; | |
470 | int fd; | |
471 | unsigned long long size; /* sectors */ | |
472 | be64 primary_lba; /* sectors */ | |
473 | be64 secondary_lba; /* sectors */ | |
474 | be64 workspace_lba; /* sectors */ | |
475 | int pdnum; /* index in ->phys */ | |
476 | struct spare_assign *spare; | |
477 | void *mdupdate; /* hold metadata update */ | |
478 | ||
479 | /* These fields used by auto-layout */ | |
480 | int raiddisk; /* slot to fill in autolayout */ | |
481 | __u64 esize; | |
482 | }; | |
483 | }; | |
484 | struct disk_data disk; | |
485 | struct vcl *vlist[0]; /* max_part in size */ | |
486 | } *dlist, *add_list; | |
487 | }; | |
488 | ||
489 | #ifndef offsetof | |
490 | #define offsetof(t,f) ((size_t)&(((t*)0)->f)) | |
491 | #endif | |
492 | ||
493 | #if DEBUG | |
494 | static int all_ff(const char *guid); | |
495 | static void pr_state(struct ddf_super *ddf, const char *msg) | |
496 | { | |
497 | unsigned int i; | |
498 | dprintf("%s/%s: ", __func__, msg); | |
499 | for (i = 0; i < be16_to_cpu(ddf->active->max_vd_entries); i++) { | |
500 | if (all_ff(ddf->virt->entries[i].guid)) | |
501 | continue; | |
502 | dprintf("%u(s=%02x i=%02x) ", i, | |
503 | ddf->virt->entries[i].state, | |
504 | ddf->virt->entries[i].init_state); | |
505 | } | |
506 | dprintf("\n"); | |
507 | } | |
508 | #else | |
509 | static void pr_state(const struct ddf_super *ddf, const char *msg) {} | |
510 | #endif | |
511 | ||
512 | static void _ddf_set_updates_pending(struct ddf_super *ddf, const char *func) | |
513 | { | |
514 | ddf->updates_pending = 1; | |
515 | ddf->active->seq = cpu_to_be32((be32_to_cpu(ddf->active->seq)+1)); | |
516 | pr_state(ddf, func); | |
517 | } | |
518 | ||
519 | #define ddf_set_updates_pending(x) _ddf_set_updates_pending((x), __func__) | |
520 | ||
521 | static unsigned int get_pd_index_from_refnum(const struct vcl *vc, | |
522 | be32 refnum, unsigned int nmax, | |
523 | const struct vd_config **bvd, | |
524 | unsigned int *idx); | |
525 | ||
526 | static be32 calc_crc(void *buf, int len) | |
527 | { | |
528 | /* crcs are always at the same place as in the ddf_header */ | |
529 | struct ddf_header *ddf = buf; | |
530 | be32 oldcrc = ddf->crc; | |
531 | __u32 newcrc; | |
532 | ddf->crc = cpu_to_be32(0xffffffff); | |
533 | ||
534 | newcrc = crc32(0, buf, len); | |
535 | ddf->crc = oldcrc; | |
536 | /* The crc is store (like everything) bigendian, so convert | |
537 | * here for simplicity | |
538 | */ | |
539 | return cpu_to_be32(newcrc); | |
540 | } | |
541 | ||
542 | #define DDF_INVALID_LEVEL 0xff | |
543 | #define DDF_NO_SECONDARY 0xff | |
544 | static int err_bad_md_layout(const mdu_array_info_t *array) | |
545 | { | |
546 | pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n", | |
547 | array->level, array->layout, array->raid_disks); | |
548 | return -1; | |
549 | } | |
550 | ||
551 | static int layout_md2ddf(const mdu_array_info_t *array, | |
552 | struct vd_config *conf) | |
553 | { | |
554 | be16 prim_elmnt_count = cpu_to_be16(array->raid_disks); | |
555 | __u8 prl = DDF_INVALID_LEVEL, rlq = 0; | |
556 | __u8 sec_elmnt_count = 1; | |
557 | __u8 srl = DDF_NO_SECONDARY; | |
558 | ||
559 | switch (array->level) { | |
560 | case LEVEL_LINEAR: | |
561 | prl = DDF_CONCAT; | |
562 | break; | |
563 | case 0: | |
564 | rlq = DDF_RAID0_SIMPLE; | |
565 | prl = DDF_RAID0; | |
566 | break; | |
567 | case 1: | |
568 | switch (array->raid_disks) { | |
569 | case 2: | |
570 | rlq = DDF_RAID1_SIMPLE; | |
571 | break; | |
572 | case 3: | |
573 | rlq = DDF_RAID1_MULTI; | |
574 | break; | |
575 | default: | |
576 | return err_bad_md_layout(array); | |
577 | } | |
578 | prl = DDF_RAID1; | |
579 | break; | |
580 | case 4: | |
581 | if (array->layout != 0) | |
582 | return err_bad_md_layout(array); | |
583 | rlq = DDF_RAID4_N; | |
584 | prl = DDF_RAID4; | |
585 | break; | |
586 | case 5: | |
587 | switch (array->layout) { | |
588 | case ALGORITHM_LEFT_ASYMMETRIC: | |
589 | rlq = DDF_RAID5_N_RESTART; | |
590 | break; | |
591 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
592 | rlq = DDF_RAID5_0_RESTART; | |
593 | break; | |
594 | case ALGORITHM_LEFT_SYMMETRIC: | |
595 | rlq = DDF_RAID5_N_CONTINUE; | |
596 | break; | |
597 | case ALGORITHM_RIGHT_SYMMETRIC: | |
598 | /* not mentioned in standard */ | |
599 | default: | |
600 | return err_bad_md_layout(array); | |
601 | } | |
602 | prl = DDF_RAID5; | |
603 | break; | |
604 | case 6: | |
605 | switch (array->layout) { | |
606 | case ALGORITHM_ROTATING_N_RESTART: | |
607 | rlq = DDF_RAID5_N_RESTART; | |
608 | break; | |
609 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
610 | rlq = DDF_RAID6_0_RESTART; | |
611 | break; | |
612 | case ALGORITHM_ROTATING_N_CONTINUE: | |
613 | rlq = DDF_RAID5_N_CONTINUE; | |
614 | break; | |
615 | default: | |
616 | return err_bad_md_layout(array); | |
617 | } | |
618 | prl = DDF_RAID6; | |
619 | break; | |
620 | case 10: | |
621 | if (array->raid_disks % 2 == 0 && array->layout == 0x102) { | |
622 | rlq = DDF_RAID1_SIMPLE; | |
623 | prim_elmnt_count = cpu_to_be16(2); | |
624 | sec_elmnt_count = array->raid_disks / 2; | |
625 | } else if (array->raid_disks % 3 == 0 | |
626 | && array->layout == 0x103) { | |
627 | rlq = DDF_RAID1_MULTI; | |
628 | prim_elmnt_count = cpu_to_be16(3); | |
629 | sec_elmnt_count = array->raid_disks / 3; | |
630 | } else | |
631 | return err_bad_md_layout(array); | |
632 | srl = DDF_2SPANNED; | |
633 | prl = DDF_RAID1; | |
634 | break; | |
635 | default: | |
636 | return err_bad_md_layout(array); | |
637 | } | |
638 | conf->prl = prl; | |
639 | conf->prim_elmnt_count = prim_elmnt_count; | |
640 | conf->rlq = rlq; | |
641 | conf->srl = srl; | |
642 | conf->sec_elmnt_count = sec_elmnt_count; | |
643 | return 0; | |
644 | } | |
645 | ||
646 | static int err_bad_ddf_layout(const struct vd_config *conf) | |
647 | { | |
648 | pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n", | |
649 | conf->prl, conf->rlq, be16_to_cpu(conf->prim_elmnt_count)); | |
650 | return -1; | |
651 | } | |
652 | ||
653 | static int layout_ddf2md(const struct vd_config *conf, | |
654 | mdu_array_info_t *array) | |
655 | { | |
656 | int level = LEVEL_UNSUPPORTED; | |
657 | int layout = 0; | |
658 | int raiddisks = be16_to_cpu(conf->prim_elmnt_count); | |
659 | ||
660 | if (conf->sec_elmnt_count > 1) { | |
661 | /* see also check_secondary() */ | |
662 | if (conf->prl != DDF_RAID1 || | |
663 | (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED)) { | |
664 | pr_err("Unsupported secondary RAID level %u/%u\n", | |
665 | conf->prl, conf->srl); | |
666 | return -1; | |
667 | } | |
668 | if (raiddisks == 2 && conf->rlq == DDF_RAID1_SIMPLE) | |
669 | layout = 0x102; | |
670 | else if (raiddisks == 3 && conf->rlq == DDF_RAID1_MULTI) | |
671 | layout = 0x103; | |
672 | else | |
673 | return err_bad_ddf_layout(conf); | |
674 | raiddisks *= conf->sec_elmnt_count; | |
675 | level = 10; | |
676 | goto good; | |
677 | } | |
678 | ||
679 | switch (conf->prl) { | |
680 | case DDF_CONCAT: | |
681 | level = LEVEL_LINEAR; | |
682 | break; | |
683 | case DDF_RAID0: | |
684 | if (conf->rlq != DDF_RAID0_SIMPLE) | |
685 | return err_bad_ddf_layout(conf); | |
686 | level = 0; | |
687 | break; | |
688 | case DDF_RAID1: | |
689 | if (!((conf->rlq == DDF_RAID1_SIMPLE && raiddisks == 2) || | |
690 | (conf->rlq == DDF_RAID1_MULTI && raiddisks == 3))) | |
691 | return err_bad_ddf_layout(conf); | |
692 | level = 1; | |
693 | break; | |
694 | case DDF_RAID4: | |
695 | if (conf->rlq != DDF_RAID4_N) | |
696 | return err_bad_ddf_layout(conf); | |
697 | level = 4; | |
698 | break; | |
699 | case DDF_RAID5: | |
700 | switch (conf->rlq) { | |
701 | case DDF_RAID5_N_RESTART: | |
702 | layout = ALGORITHM_LEFT_ASYMMETRIC; | |
703 | break; | |
704 | case DDF_RAID5_0_RESTART: | |
705 | layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
706 | break; | |
707 | case DDF_RAID5_N_CONTINUE: | |
708 | layout = ALGORITHM_LEFT_SYMMETRIC; | |
709 | break; | |
710 | default: | |
711 | return err_bad_ddf_layout(conf); | |
712 | } | |
713 | level = 5; | |
714 | break; | |
715 | case DDF_RAID6: | |
716 | switch (conf->rlq) { | |
717 | case DDF_RAID5_N_RESTART: | |
718 | layout = ALGORITHM_ROTATING_N_RESTART; | |
719 | break; | |
720 | case DDF_RAID6_0_RESTART: | |
721 | layout = ALGORITHM_ROTATING_ZERO_RESTART; | |
722 | break; | |
723 | case DDF_RAID5_N_CONTINUE: | |
724 | layout = ALGORITHM_ROTATING_N_CONTINUE; | |
725 | break; | |
726 | default: | |
727 | return err_bad_ddf_layout(conf); | |
728 | } | |
729 | level = 6; | |
730 | break; | |
731 | default: | |
732 | return err_bad_ddf_layout(conf); | |
733 | }; | |
734 | ||
735 | good: | |
736 | array->level = level; | |
737 | array->layout = layout; | |
738 | array->raid_disks = raiddisks; | |
739 | return 0; | |
740 | } | |
741 | ||
742 | static int load_ddf_header(int fd, unsigned long long lba, | |
743 | unsigned long long size, | |
744 | int type, | |
745 | struct ddf_header *hdr, struct ddf_header *anchor) | |
746 | { | |
747 | /* read a ddf header (primary or secondary) from fd/lba | |
748 | * and check that it is consistent with anchor | |
749 | * Need to check: | |
750 | * magic, crc, guid, rev, and LBA's header_type, and | |
751 | * everything after header_type must be the same | |
752 | */ | |
753 | if (lba >= size-1) | |
754 | return 0; | |
755 | ||
756 | if (lseek64(fd, lba<<9, 0) < 0) | |
757 | return 0; | |
758 | ||
759 | if (read(fd, hdr, 512) != 512) | |
760 | return 0; | |
761 | ||
762 | if (!be32_eq(hdr->magic, DDF_HEADER_MAGIC)) { | |
763 | pr_err("%s: bad header magic\n", __func__); | |
764 | return 0; | |
765 | } | |
766 | if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) { | |
767 | pr_err("%s: bad CRC\n", __func__); | |
768 | return 0; | |
769 | } | |
770 | if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 || | |
771 | memcmp(anchor->revision, hdr->revision, 8) != 0 || | |
772 | !be64_eq(anchor->primary_lba, hdr->primary_lba) || | |
773 | !be64_eq(anchor->secondary_lba, hdr->secondary_lba) || | |
774 | hdr->type != type || | |
775 | memcmp(anchor->pad2, hdr->pad2, 512 - | |
776 | offsetof(struct ddf_header, pad2)) != 0) { | |
777 | pr_err("%s: header mismatch\n", __func__); | |
778 | return 0; | |
779 | } | |
780 | ||
781 | /* Looks good enough to me... */ | |
782 | return 1; | |
783 | } | |
784 | ||
785 | static void *load_section(int fd, struct ddf_super *super, void *buf, | |
786 | be32 offset_be, be32 len_be, int check) | |
787 | { | |
788 | unsigned long long offset = be32_to_cpu(offset_be); | |
789 | unsigned long long len = be32_to_cpu(len_be); | |
790 | int dofree = (buf == NULL); | |
791 | ||
792 | if (check) | |
793 | if (len != 2 && len != 8 && len != 32 | |
794 | && len != 128 && len != 512) | |
795 | return NULL; | |
796 | ||
797 | if (len > 1024) | |
798 | return NULL; | |
799 | if (!buf && posix_memalign(&buf, 512, len<<9) != 0) | |
800 | buf = NULL; | |
801 | ||
802 | if (!buf) | |
803 | return NULL; | |
804 | ||
805 | if (super->active->type == 1) | |
806 | offset += be64_to_cpu(super->active->primary_lba); | |
807 | else | |
808 | offset += be64_to_cpu(super->active->secondary_lba); | |
809 | ||
810 | if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) { | |
811 | if (dofree) | |
812 | free(buf); | |
813 | return NULL; | |
814 | } | |
815 | if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) { | |
816 | if (dofree) | |
817 | free(buf); | |
818 | return NULL; | |
819 | } | |
820 | return buf; | |
821 | } | |
822 | ||
823 | static int load_ddf_headers(int fd, struct ddf_super *super, char *devname) | |
824 | { | |
825 | unsigned long long dsize; | |
826 | ||
827 | get_dev_size(fd, NULL, &dsize); | |
828 | ||
829 | if (lseek64(fd, dsize-512, 0) < 0) { | |
830 | if (devname) | |
831 | pr_err("Cannot seek to anchor block on %s: %s\n", | |
832 | devname, strerror(errno)); | |
833 | return 1; | |
834 | } | |
835 | if (read(fd, &super->anchor, 512) != 512) { | |
836 | if (devname) | |
837 | pr_err("Cannot read anchor block on %s: %s\n", | |
838 | devname, strerror(errno)); | |
839 | return 1; | |
840 | } | |
841 | if (!be32_eq(super->anchor.magic, DDF_HEADER_MAGIC)) { | |
842 | if (devname) | |
843 | pr_err("no DDF anchor found on %s\n", | |
844 | devname); | |
845 | return 2; | |
846 | } | |
847 | if (!be32_eq(calc_crc(&super->anchor, 512), super->anchor.crc)) { | |
848 | if (devname) | |
849 | pr_err("bad CRC on anchor on %s\n", | |
850 | devname); | |
851 | return 2; | |
852 | } | |
853 | if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 && | |
854 | memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) { | |
855 | if (devname) | |
856 | pr_err("can only support super revision" | |
857 | " %.8s and earlier, not %.8s on %s\n", | |
858 | DDF_REVISION_2, super->anchor.revision,devname); | |
859 | return 2; | |
860 | } | |
861 | super->active = NULL; | |
862 | if (load_ddf_header(fd, be64_to_cpu(super->anchor.primary_lba), | |
863 | dsize >> 9, 1, | |
864 | &super->primary, &super->anchor) == 0) { | |
865 | if (devname) | |
866 | pr_err("Failed to load primary DDF header " | |
867 | "on %s\n", devname); | |
868 | } else | |
869 | super->active = &super->primary; | |
870 | ||
871 | if (load_ddf_header(fd, be64_to_cpu(super->anchor.secondary_lba), | |
872 | dsize >> 9, 2, | |
873 | &super->secondary, &super->anchor)) { | |
874 | if (super->active == NULL | |
875 | || (be32_to_cpu(super->primary.seq) | |
876 | < be32_to_cpu(super->secondary.seq) && | |
877 | !super->secondary.openflag) | |
878 | || (be32_to_cpu(super->primary.seq) | |
879 | == be32_to_cpu(super->secondary.seq) && | |
880 | super->primary.openflag && !super->secondary.openflag) | |
881 | ) | |
882 | super->active = &super->secondary; | |
883 | } else if (devname) | |
884 | pr_err("Failed to load secondary DDF header on %s\n", | |
885 | devname); | |
886 | if (super->active == NULL) | |
887 | return 2; | |
888 | return 0; | |
889 | } | |
890 | ||
891 | static int load_ddf_global(int fd, struct ddf_super *super, char *devname) | |
892 | { | |
893 | void *ok; | |
894 | ok = load_section(fd, super, &super->controller, | |
895 | super->active->controller_section_offset, | |
896 | super->active->controller_section_length, | |
897 | 0); | |
898 | super->phys = load_section(fd, super, NULL, | |
899 | super->active->phys_section_offset, | |
900 | super->active->phys_section_length, | |
901 | 1); | |
902 | super->pdsize = be32_to_cpu(super->active->phys_section_length) * 512; | |
903 | ||
904 | super->virt = load_section(fd, super, NULL, | |
905 | super->active->virt_section_offset, | |
906 | super->active->virt_section_length, | |
907 | 1); | |
908 | super->vdsize = be32_to_cpu(super->active->virt_section_length) * 512; | |
909 | if (!ok || | |
910 | !super->phys || | |
911 | !super->virt) { | |
912 | free(super->phys); | |
913 | free(super->virt); | |
914 | super->phys = NULL; | |
915 | super->virt = NULL; | |
916 | return 2; | |
917 | } | |
918 | super->conflist = NULL; | |
919 | super->dlist = NULL; | |
920 | ||
921 | super->max_part = be16_to_cpu(super->active->max_partitions); | |
922 | super->mppe = be16_to_cpu(super->active->max_primary_element_entries); | |
923 | super->conf_rec_len = be16_to_cpu(super->active->config_record_len); | |
924 | return 0; | |
925 | } | |
926 | ||
927 | #define DDF_UNUSED_BVD 0xff | |
928 | static int alloc_other_bvds(const struct ddf_super *ddf, struct vcl *vcl) | |
929 | { | |
930 | unsigned int n_vds = vcl->conf.sec_elmnt_count - 1; | |
931 | unsigned int i, vdsize; | |
932 | void *p; | |
933 | if (n_vds == 0) { | |
934 | vcl->other_bvds = NULL; | |
935 | return 0; | |
936 | } | |
937 | vdsize = ddf->conf_rec_len * 512; | |
938 | if (posix_memalign(&p, 512, n_vds * | |
939 | (vdsize + sizeof(struct vd_config *))) != 0) | |
940 | return -1; | |
941 | vcl->other_bvds = (struct vd_config **) (p + n_vds * vdsize); | |
942 | for (i = 0; i < n_vds; i++) { | |
943 | vcl->other_bvds[i] = p + i * vdsize; | |
944 | memset(vcl->other_bvds[i], 0, vdsize); | |
945 | vcl->other_bvds[i]->sec_elmnt_seq = DDF_UNUSED_BVD; | |
946 | } | |
947 | return 0; | |
948 | } | |
949 | ||
950 | static void add_other_bvd(struct vcl *vcl, struct vd_config *vd, | |
951 | unsigned int len) | |
952 | { | |
953 | int i; | |
954 | for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++) | |
955 | if (vcl->other_bvds[i]->sec_elmnt_seq == vd->sec_elmnt_seq) | |
956 | break; | |
957 | ||
958 | if (i < vcl->conf.sec_elmnt_count-1) { | |
959 | if (be32_to_cpu(vd->seqnum) <= | |
960 | be32_to_cpu(vcl->other_bvds[i]->seqnum)) | |
961 | return; | |
962 | } else { | |
963 | for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++) | |
964 | if (vcl->other_bvds[i]->sec_elmnt_seq == DDF_UNUSED_BVD) | |
965 | break; | |
966 | if (i == vcl->conf.sec_elmnt_count-1) { | |
967 | pr_err("no space for sec level config %u, count is %u\n", | |
968 | vd->sec_elmnt_seq, vcl->conf.sec_elmnt_count); | |
969 | return; | |
970 | } | |
971 | } | |
972 | memcpy(vcl->other_bvds[i], vd, len); | |
973 | } | |
974 | ||
975 | static int load_ddf_local(int fd, struct ddf_super *super, | |
976 | char *devname, int keep) | |
977 | { | |
978 | struct dl *dl; | |
979 | struct stat stb; | |
980 | char *conf; | |
981 | unsigned int i; | |
982 | unsigned int confsec; | |
983 | int vnum; | |
984 | unsigned int max_virt_disks = be16_to_cpu | |
985 | (super->active->max_vd_entries); | |
986 | unsigned long long dsize; | |
987 | ||
988 | /* First the local disk info */ | |
989 | if (posix_memalign((void**)&dl, 512, | |
990 | sizeof(*dl) + | |
991 | (super->max_part) * sizeof(dl->vlist[0])) != 0) { | |
992 | pr_err("%s could not allocate disk info buffer\n", | |
993 | __func__); | |
994 | return 1; | |
995 | } | |
996 | ||
997 | load_section(fd, super, &dl->disk, | |
998 | super->active->data_section_offset, | |
999 | super->active->data_section_length, | |
1000 | 0); | |
1001 | dl->devname = devname ? xstrdup(devname) : NULL; | |
1002 | ||
1003 | fstat(fd, &stb); | |
1004 | dl->major = major(stb.st_rdev); | |
1005 | dl->minor = minor(stb.st_rdev); | |
1006 | dl->next = super->dlist; | |
1007 | dl->fd = keep ? fd : -1; | |
1008 | ||
1009 | dl->size = 0; | |
1010 | if (get_dev_size(fd, devname, &dsize)) | |
1011 | dl->size = dsize >> 9; | |
1012 | /* If the disks have different sizes, the LBAs will differ | |
1013 | * between phys disks. | |
1014 | * At this point here, the values in super->active must be valid | |
1015 | * for this phys disk. */ | |
1016 | dl->primary_lba = super->active->primary_lba; | |
1017 | dl->secondary_lba = super->active->secondary_lba; | |
1018 | dl->workspace_lba = super->active->workspace_lba; | |
1019 | dl->spare = NULL; | |
1020 | for (i = 0 ; i < super->max_part ; i++) | |
1021 | dl->vlist[i] = NULL; | |
1022 | super->dlist = dl; | |
1023 | dl->pdnum = -1; | |
1024 | for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++) | |
1025 | if (memcmp(super->phys->entries[i].guid, | |
1026 | dl->disk.guid, DDF_GUID_LEN) == 0) | |
1027 | dl->pdnum = i; | |
1028 | ||
1029 | /* Now the config list. */ | |
1030 | /* 'conf' is an array of config entries, some of which are | |
1031 | * probably invalid. Those which are good need to be copied into | |
1032 | * the conflist | |
1033 | */ | |
1034 | ||
1035 | conf = load_section(fd, super, super->conf, | |
1036 | super->active->config_section_offset, | |
1037 | super->active->config_section_length, | |
1038 | 0); | |
1039 | super->conf = conf; | |
1040 | vnum = 0; | |
1041 | for (confsec = 0; | |
1042 | confsec < be32_to_cpu(super->active->config_section_length); | |
1043 | confsec += super->conf_rec_len) { | |
1044 | struct vd_config *vd = | |
1045 | (struct vd_config *)((char*)conf + confsec*512); | |
1046 | struct vcl *vcl; | |
1047 | ||
1048 | if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) { | |
1049 | if (dl->spare) | |
1050 | continue; | |
1051 | if (posix_memalign((void**)&dl->spare, 512, | |
1052 | super->conf_rec_len*512) != 0) { | |
1053 | pr_err("%s could not allocate spare info buf\n", | |
1054 | __func__); | |
1055 | return 1; | |
1056 | } | |
1057 | ||
1058 | memcpy(dl->spare, vd, super->conf_rec_len*512); | |
1059 | continue; | |
1060 | } | |
1061 | if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC)) | |
1062 | continue; | |
1063 | for (vcl = super->conflist; vcl; vcl = vcl->next) { | |
1064 | if (memcmp(vcl->conf.guid, | |
1065 | vd->guid, DDF_GUID_LEN) == 0) | |
1066 | break; | |
1067 | } | |
1068 | ||
1069 | if (vcl) { | |
1070 | dl->vlist[vnum++] = vcl; | |
1071 | if (vcl->other_bvds != NULL && | |
1072 | vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) { | |
1073 | add_other_bvd(vcl, vd, super->conf_rec_len*512); | |
1074 | continue; | |
1075 | } | |
1076 | if (be32_to_cpu(vd->seqnum) <= | |
1077 | be32_to_cpu(vcl->conf.seqnum)) | |
1078 | continue; | |
1079 | } else { | |
1080 | if (posix_memalign((void**)&vcl, 512, | |
1081 | (super->conf_rec_len*512 + | |
1082 | offsetof(struct vcl, conf))) != 0) { | |
1083 | pr_err("%s could not allocate vcl buf\n", | |
1084 | __func__); | |
1085 | return 1; | |
1086 | } | |
1087 | vcl->next = super->conflist; | |
1088 | vcl->block_sizes = NULL; /* FIXME not for CONCAT */ | |
1089 | vcl->conf.sec_elmnt_count = vd->sec_elmnt_count; | |
1090 | if (alloc_other_bvds(super, vcl) != 0) { | |
1091 | pr_err("%s could not allocate other bvds\n", | |
1092 | __func__); | |
1093 | free(vcl); | |
1094 | return 1; | |
1095 | }; | |
1096 | super->conflist = vcl; | |
1097 | dl->vlist[vnum++] = vcl; | |
1098 | } | |
1099 | memcpy(&vcl->conf, vd, super->conf_rec_len*512); | |
1100 | for (i=0; i < max_virt_disks ; i++) | |
1101 | if (memcmp(super->virt->entries[i].guid, | |
1102 | vcl->conf.guid, DDF_GUID_LEN)==0) | |
1103 | break; | |
1104 | if (i < max_virt_disks) | |
1105 | vcl->vcnum = i; | |
1106 | } | |
1107 | ||
1108 | return 0; | |
1109 | } | |
1110 | ||
1111 | #ifndef MDASSEMBLE | |
1112 | static int load_super_ddf_all(struct supertype *st, int fd, | |
1113 | void **sbp, char *devname); | |
1114 | #endif | |
1115 | ||
1116 | static void free_super_ddf(struct supertype *st); | |
1117 | ||
1118 | static int load_super_ddf(struct supertype *st, int fd, | |
1119 | char *devname) | |
1120 | { | |
1121 | unsigned long long dsize; | |
1122 | struct ddf_super *super; | |
1123 | int rv; | |
1124 | ||
1125 | if (get_dev_size(fd, devname, &dsize) == 0) | |
1126 | return 1; | |
1127 | ||
1128 | if (!st->ignore_hw_compat && test_partition(fd)) | |
1129 | /* DDF is not allowed on partitions */ | |
1130 | return 1; | |
1131 | ||
1132 | /* 32M is a lower bound */ | |
1133 | if (dsize <= 32*1024*1024) { | |
1134 | if (devname) | |
1135 | pr_err("%s is too small for ddf: " | |
1136 | "size is %llu sectors.\n", | |
1137 | devname, dsize>>9); | |
1138 | return 1; | |
1139 | } | |
1140 | if (dsize & 511) { | |
1141 | if (devname) | |
1142 | pr_err("%s is an odd size for ddf: " | |
1143 | "size is %llu bytes.\n", | |
1144 | devname, dsize); | |
1145 | return 1; | |
1146 | } | |
1147 | ||
1148 | free_super_ddf(st); | |
1149 | ||
1150 | if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) { | |
1151 | pr_err("malloc of %zu failed.\n", | |
1152 | sizeof(*super)); | |
1153 | return 1; | |
1154 | } | |
1155 | memset(super, 0, sizeof(*super)); | |
1156 | ||
1157 | rv = load_ddf_headers(fd, super, devname); | |
1158 | if (rv) { | |
1159 | free(super); | |
1160 | return rv; | |
1161 | } | |
1162 | ||
1163 | /* Have valid headers and have chosen the best. Let's read in the rest*/ | |
1164 | ||
1165 | rv = load_ddf_global(fd, super, devname); | |
1166 | ||
1167 | if (rv) { | |
1168 | if (devname) | |
1169 | pr_err("Failed to load all information " | |
1170 | "sections on %s\n", devname); | |
1171 | free(super); | |
1172 | return rv; | |
1173 | } | |
1174 | ||
1175 | rv = load_ddf_local(fd, super, devname, 0); | |
1176 | ||
1177 | if (rv) { | |
1178 | if (devname) | |
1179 | pr_err("Failed to load all information " | |
1180 | "sections on %s\n", devname); | |
1181 | free(super); | |
1182 | return rv; | |
1183 | } | |
1184 | ||
1185 | /* Should possibly check the sections .... */ | |
1186 | ||
1187 | st->sb = super; | |
1188 | if (st->ss == NULL) { | |
1189 | st->ss = &super_ddf; | |
1190 | st->minor_version = 0; | |
1191 | st->max_devs = 512; | |
1192 | } | |
1193 | return 0; | |
1194 | ||
1195 | } | |
1196 | ||
1197 | static void free_super_ddf(struct supertype *st) | |
1198 | { | |
1199 | struct ddf_super *ddf = st->sb; | |
1200 | if (ddf == NULL) | |
1201 | return; | |
1202 | free(ddf->phys); | |
1203 | free(ddf->virt); | |
1204 | free(ddf->conf); | |
1205 | while (ddf->conflist) { | |
1206 | struct vcl *v = ddf->conflist; | |
1207 | ddf->conflist = v->next; | |
1208 | if (v->block_sizes) | |
1209 | free(v->block_sizes); | |
1210 | if (v->other_bvds) | |
1211 | /* | |
1212 | v->other_bvds[0] points to beginning of buffer, | |
1213 | see alloc_other_bvds() | |
1214 | */ | |
1215 | free(v->other_bvds[0]); | |
1216 | free(v); | |
1217 | } | |
1218 | while (ddf->dlist) { | |
1219 | struct dl *d = ddf->dlist; | |
1220 | ddf->dlist = d->next; | |
1221 | if (d->fd >= 0) | |
1222 | close(d->fd); | |
1223 | if (d->spare) | |
1224 | free(d->spare); | |
1225 | free(d); | |
1226 | } | |
1227 | while (ddf->add_list) { | |
1228 | struct dl *d = ddf->add_list; | |
1229 | ddf->add_list = d->next; | |
1230 | if (d->fd >= 0) | |
1231 | close(d->fd); | |
1232 | if (d->spare) | |
1233 | free(d->spare); | |
1234 | free(d); | |
1235 | } | |
1236 | free(ddf); | |
1237 | st->sb = NULL; | |
1238 | } | |
1239 | ||
1240 | static struct supertype *match_metadata_desc_ddf(char *arg) | |
1241 | { | |
1242 | /* 'ddf' only support containers */ | |
1243 | struct supertype *st; | |
1244 | if (strcmp(arg, "ddf") != 0 && | |
1245 | strcmp(arg, "default") != 0 | |
1246 | ) | |
1247 | return NULL; | |
1248 | ||
1249 | st = xcalloc(1, sizeof(*st)); | |
1250 | st->ss = &super_ddf; | |
1251 | st->max_devs = 512; | |
1252 | st->minor_version = 0; | |
1253 | st->sb = NULL; | |
1254 | return st; | |
1255 | } | |
1256 | ||
1257 | #ifndef MDASSEMBLE | |
1258 | ||
1259 | static mapping_t ddf_state[] = { | |
1260 | { "Optimal", 0}, | |
1261 | { "Degraded", 1}, | |
1262 | { "Deleted", 2}, | |
1263 | { "Missing", 3}, | |
1264 | { "Failed", 4}, | |
1265 | { "Partially Optimal", 5}, | |
1266 | { "-reserved-", 6}, | |
1267 | { "-reserved-", 7}, | |
1268 | { NULL, 0} | |
1269 | }; | |
1270 | ||
1271 | static mapping_t ddf_init_state[] = { | |
1272 | { "Not Initialised", 0}, | |
1273 | { "QuickInit in Progress", 1}, | |
1274 | { "Fully Initialised", 2}, | |
1275 | { "*UNKNOWN*", 3}, | |
1276 | { NULL, 0} | |
1277 | }; | |
1278 | static mapping_t ddf_access[] = { | |
1279 | { "Read/Write", 0}, | |
1280 | { "Reserved", 1}, | |
1281 | { "Read Only", 2}, | |
1282 | { "Blocked (no access)", 3}, | |
1283 | { NULL ,0} | |
1284 | }; | |
1285 | ||
1286 | static mapping_t ddf_level[] = { | |
1287 | { "RAID0", DDF_RAID0}, | |
1288 | { "RAID1", DDF_RAID1}, | |
1289 | { "RAID3", DDF_RAID3}, | |
1290 | { "RAID4", DDF_RAID4}, | |
1291 | { "RAID5", DDF_RAID5}, | |
1292 | { "RAID1E",DDF_RAID1E}, | |
1293 | { "JBOD", DDF_JBOD}, | |
1294 | { "CONCAT",DDF_CONCAT}, | |
1295 | { "RAID5E",DDF_RAID5E}, | |
1296 | { "RAID5EE",DDF_RAID5EE}, | |
1297 | { "RAID6", DDF_RAID6}, | |
1298 | { NULL, 0} | |
1299 | }; | |
1300 | static mapping_t ddf_sec_level[] = { | |
1301 | { "Striped", DDF_2STRIPED}, | |
1302 | { "Mirrored", DDF_2MIRRORED}, | |
1303 | { "Concat", DDF_2CONCAT}, | |
1304 | { "Spanned", DDF_2SPANNED}, | |
1305 | { NULL, 0} | |
1306 | }; | |
1307 | #endif | |
1308 | ||
1309 | static int all_ff(const char *guid) | |
1310 | { | |
1311 | int i; | |
1312 | for (i = 0; i < DDF_GUID_LEN; i++) | |
1313 | if (guid[i] != (char)0xff) | |
1314 | return 0; | |
1315 | return 1; | |
1316 | } | |
1317 | ||
1318 | static const char *guid_str(const char *guid) | |
1319 | { | |
1320 | static char buf[DDF_GUID_LEN*2+1]; | |
1321 | int i; | |
1322 | char *p = buf; | |
1323 | for (i = 0; i < DDF_GUID_LEN; i++) { | |
1324 | unsigned char c = guid[i]; | |
1325 | if (c >= 32 && c < 127) | |
1326 | p += sprintf(p, "%c", c); | |
1327 | else | |
1328 | p += sprintf(p, "%02x", c); | |
1329 | } | |
1330 | *p = '\0'; | |
1331 | return (const char *) buf; | |
1332 | } | |
1333 | ||
1334 | #ifndef MDASSEMBLE | |
1335 | static void print_guid(char *guid, int tstamp) | |
1336 | { | |
1337 | /* A GUIDs are part (or all) ASCII and part binary. | |
1338 | * They tend to be space padded. | |
1339 | * We print the GUID in HEX, then in parentheses add | |
1340 | * any initial ASCII sequence, and a possible | |
1341 | * time stamp from bytes 16-19 | |
1342 | */ | |
1343 | int l = DDF_GUID_LEN; | |
1344 | int i; | |
1345 | ||
1346 | for (i=0 ; i<DDF_GUID_LEN ; i++) { | |
1347 | if ((i&3)==0 && i != 0) printf(":"); | |
1348 | printf("%02X", guid[i]&255); | |
1349 | } | |
1350 | ||
1351 | printf("\n ("); | |
1352 | while (l && guid[l-1] == ' ') | |
1353 | l--; | |
1354 | for (i=0 ; i<l ; i++) { | |
1355 | if (guid[i] >= 0x20 && guid[i] < 0x7f) | |
1356 | fputc(guid[i], stdout); | |
1357 | else | |
1358 | break; | |
1359 | } | |
1360 | if (tstamp) { | |
1361 | time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE; | |
1362 | char tbuf[100]; | |
1363 | struct tm *tm; | |
1364 | tm = localtime(&then); | |
1365 | strftime(tbuf, 100, " %D %T",tm); | |
1366 | fputs(tbuf, stdout); | |
1367 | } | |
1368 | printf(")"); | |
1369 | } | |
1370 | ||
1371 | static void examine_vd(int n, struct ddf_super *sb, char *guid) | |
1372 | { | |
1373 | int crl = sb->conf_rec_len; | |
1374 | struct vcl *vcl; | |
1375 | ||
1376 | for (vcl = sb->conflist ; vcl ; vcl = vcl->next) { | |
1377 | unsigned int i; | |
1378 | struct vd_config *vc = &vcl->conf; | |
1379 | ||
1380 | if (!be32_eq(calc_crc(vc, crl*512), vc->crc)) | |
1381 | continue; | |
1382 | if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0) | |
1383 | continue; | |
1384 | ||
1385 | /* Ok, we know about this VD, let's give more details */ | |
1386 | printf(" Raid Devices[%d] : %d (", n, | |
1387 | be16_to_cpu(vc->prim_elmnt_count)); | |
1388 | for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) { | |
1389 | int j; | |
1390 | int cnt = be16_to_cpu(sb->phys->used_pdes); | |
1391 | for (j=0; j<cnt; j++) | |
1392 | if (be32_eq(vc->phys_refnum[i], | |
1393 | sb->phys->entries[j].refnum)) | |
1394 | break; | |
1395 | if (i) printf(" "); | |
1396 | if (j < cnt) | |
1397 | printf("%d", j); | |
1398 | else | |
1399 | printf("--"); | |
1400 | } | |
1401 | printf(")\n"); | |
1402 | if (vc->chunk_shift != 255) | |
1403 | printf(" Chunk Size[%d] : %d sectors\n", n, | |
1404 | 1 << vc->chunk_shift); | |
1405 | printf(" Raid Level[%d] : %s\n", n, | |
1406 | map_num(ddf_level, vc->prl)?:"-unknown-"); | |
1407 | if (vc->sec_elmnt_count != 1) { | |
1408 | printf(" Secondary Position[%d] : %d of %d\n", n, | |
1409 | vc->sec_elmnt_seq, vc->sec_elmnt_count); | |
1410 | printf(" Secondary Level[%d] : %s\n", n, | |
1411 | map_num(ddf_sec_level, vc->srl) ?: "-unknown-"); | |
1412 | } | |
1413 | printf(" Device Size[%d] : %llu\n", n, | |
1414 | be64_to_cpu(vc->blocks)/2); | |
1415 | printf(" Array Size[%d] : %llu\n", n, | |
1416 | be64_to_cpu(vc->array_blocks)/2); | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | static void examine_vds(struct ddf_super *sb) | |
1421 | { | |
1422 | int cnt = be16_to_cpu(sb->virt->populated_vdes); | |
1423 | unsigned int i; | |
1424 | printf(" Virtual Disks : %d\n", cnt); | |
1425 | ||
1426 | for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) { | |
1427 | struct virtual_entry *ve = &sb->virt->entries[i]; | |
1428 | if (all_ff(ve->guid)) | |
1429 | continue; | |
1430 | printf("\n"); | |
1431 | printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1); | |
1432 | printf("\n"); | |
1433 | printf(" unit[%d] : %d\n", i, be16_to_cpu(ve->unit)); | |
1434 | printf(" state[%d] : %s, %s%s\n", i, | |
1435 | map_num(ddf_state, ve->state & 7), | |
1436 | (ve->state & DDF_state_morphing) ? "Morphing, ": "", | |
1437 | (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent"); | |
1438 | printf(" init state[%d] : %s\n", i, | |
1439 | map_num(ddf_init_state, ve->init_state&DDF_initstate_mask)); | |
1440 | printf(" access[%d] : %s\n", i, | |
1441 | map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6)); | |
1442 | printf(" Name[%d] : %.16s\n", i, ve->name); | |
1443 | examine_vd(i, sb, ve->guid); | |
1444 | } | |
1445 | if (cnt) printf("\n"); | |
1446 | } | |
1447 | ||
1448 | static void examine_pds(struct ddf_super *sb) | |
1449 | { | |
1450 | int cnt = be16_to_cpu(sb->phys->used_pdes); | |
1451 | int i; | |
1452 | struct dl *dl; | |
1453 | printf(" Physical Disks : %d\n", cnt); | |
1454 | printf(" Number RefNo Size Device Type/State\n"); | |
1455 | ||
1456 | for (i=0 ; i<cnt ; i++) { | |
1457 | struct phys_disk_entry *pd = &sb->phys->entries[i]; | |
1458 | int type = be16_to_cpu(pd->type); | |
1459 | int state = be16_to_cpu(pd->state); | |
1460 | ||
1461 | //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0); | |
1462 | //printf("\n"); | |
1463 | printf(" %3d %08x ", i, | |
1464 | be32_to_cpu(pd->refnum)); | |
1465 | printf("%8lluK ", | |
1466 | be64_to_cpu(pd->config_size)>>1); | |
1467 | for (dl = sb->dlist; dl ; dl = dl->next) { | |
1468 | if (be32_eq(dl->disk.refnum, pd->refnum)) { | |
1469 | char *dv = map_dev(dl->major, dl->minor, 0); | |
1470 | if (dv) { | |
1471 | printf("%-15s", dv); | |
1472 | break; | |
1473 | } | |
1474 | } | |
1475 | } | |
1476 | if (!dl) | |
1477 | printf("%15s",""); | |
1478 | printf(" %s%s%s%s%s", | |
1479 | (type&2) ? "active":"", | |
1480 | (type&4) ? "Global-Spare":"", | |
1481 | (type&8) ? "spare" : "", | |
1482 | (type&16)? ", foreign" : "", | |
1483 | (type&32)? "pass-through" : ""); | |
1484 | if (state & DDF_Failed) | |
1485 | /* This over-rides these three */ | |
1486 | state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition); | |
1487 | printf("/%s%s%s%s%s%s%s", | |
1488 | (state&1)? "Online": "Offline", | |
1489 | (state&2)? ", Failed": "", | |
1490 | (state&4)? ", Rebuilding": "", | |
1491 | (state&8)? ", in-transition": "", | |
1492 | (state&16)? ", SMART-errors": "", | |
1493 | (state&32)? ", Unrecovered-Read-Errors": "", | |
1494 | (state&64)? ", Missing" : ""); | |
1495 | printf("\n"); | |
1496 | } | |
1497 | } | |
1498 | ||
1499 | static void examine_super_ddf(struct supertype *st, char *homehost) | |
1500 | { | |
1501 | struct ddf_super *sb = st->sb; | |
1502 | ||
1503 | printf(" Magic : %08x\n", be32_to_cpu(sb->anchor.magic)); | |
1504 | printf(" Version : %.8s\n", sb->anchor.revision); | |
1505 | printf("Controller GUID : "); print_guid(sb->controller.guid, 0); | |
1506 | printf("\n"); | |
1507 | printf(" Container GUID : "); print_guid(sb->anchor.guid, 1); | |
1508 | printf("\n"); | |
1509 | printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq)); | |
1510 | printf(" Redundant hdr : %s\n", be32_eq(sb->secondary.magic, | |
1511 | DDF_HEADER_MAGIC) | |
1512 | ?"yes" : "no"); | |
1513 | examine_vds(sb); | |
1514 | examine_pds(sb); | |
1515 | } | |
1516 | ||
1517 | static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map); | |
1518 | ||
1519 | static void uuid_from_ddf_guid(const char *guid, int uuid[4]); | |
1520 | static void uuid_from_super_ddf(struct supertype *st, int uuid[4]); | |
1521 | ||
1522 | static unsigned int get_vd_num_of_subarray(struct supertype *st) | |
1523 | { | |
1524 | /* | |
1525 | * Figure out the VD number for this supertype. | |
1526 | * Returns DDF_CONTAINER for the container itself, | |
1527 | * and DDF_NOTFOUND on error. | |
1528 | */ | |
1529 | struct ddf_super *ddf = st->sb; | |
1530 | struct mdinfo *sra; | |
1531 | char *sub, *end; | |
1532 | unsigned int vcnum; | |
1533 | ||
1534 | if (*st->container_devnm == '\0') | |
1535 | return DDF_CONTAINER; | |
1536 | ||
1537 | sra = sysfs_read(-1, st->devnm, GET_VERSION); | |
1538 | if (!sra || sra->array.major_version != -1 || | |
1539 | sra->array.minor_version != -2 || | |
1540 | !is_subarray(sra->text_version)) | |
1541 | return DDF_NOTFOUND; | |
1542 | ||
1543 | sub = strchr(sra->text_version + 1, '/'); | |
1544 | if (sub != NULL) | |
1545 | vcnum = strtoul(sub + 1, &end, 10); | |
1546 | if (sub == NULL || *sub == '\0' || *end != '\0' || | |
1547 | vcnum >= be16_to_cpu(ddf->active->max_vd_entries)) | |
1548 | return DDF_NOTFOUND; | |
1549 | ||
1550 | return vcnum; | |
1551 | } | |
1552 | ||
1553 | static void brief_examine_super_ddf(struct supertype *st, int verbose) | |
1554 | { | |
1555 | /* We just write a generic DDF ARRAY entry | |
1556 | */ | |
1557 | struct mdinfo info; | |
1558 | char nbuf[64]; | |
1559 | getinfo_super_ddf(st, &info, NULL); | |
1560 | fname_from_uuid(st, &info, nbuf, ':'); | |
1561 | ||
1562 | printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5); | |
1563 | } | |
1564 | ||
1565 | static void brief_examine_subarrays_ddf(struct supertype *st, int verbose) | |
1566 | { | |
1567 | /* We just write a generic DDF ARRAY entry | |
1568 | */ | |
1569 | struct ddf_super *ddf = st->sb; | |
1570 | struct mdinfo info; | |
1571 | unsigned int i; | |
1572 | char nbuf[64]; | |
1573 | getinfo_super_ddf(st, &info, NULL); | |
1574 | fname_from_uuid(st, &info, nbuf, ':'); | |
1575 | ||
1576 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) { | |
1577 | struct virtual_entry *ve = &ddf->virt->entries[i]; | |
1578 | struct vcl vcl; | |
1579 | char nbuf1[64]; | |
1580 | if (all_ff(ve->guid)) | |
1581 | continue; | |
1582 | memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN); | |
1583 | ddf->currentconf =&vcl; | |
1584 | vcl.vcnum = i; | |
1585 | uuid_from_super_ddf(st, info.uuid); | |
1586 | fname_from_uuid(st, &info, nbuf1, ':'); | |
1587 | printf("ARRAY container=%s member=%d UUID=%s\n", | |
1588 | nbuf+5, i, nbuf1+5); | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | static void export_examine_super_ddf(struct supertype *st) | |
1593 | { | |
1594 | struct mdinfo info; | |
1595 | char nbuf[64]; | |
1596 | getinfo_super_ddf(st, &info, NULL); | |
1597 | fname_from_uuid(st, &info, nbuf, ':'); | |
1598 | printf("MD_METADATA=ddf\n"); | |
1599 | printf("MD_LEVEL=container\n"); | |
1600 | printf("MD_UUID=%s\n", nbuf+5); | |
1601 | printf("MD_DEVICES=%u\n", | |
1602 | be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes)); | |
1603 | } | |
1604 | ||
1605 | static int copy_metadata_ddf(struct supertype *st, int from, int to) | |
1606 | { | |
1607 | void *buf; | |
1608 | unsigned long long dsize, offset; | |
1609 | int bytes; | |
1610 | struct ddf_header *ddf; | |
1611 | int written = 0; | |
1612 | ||
1613 | /* The meta consists of an anchor, a primary, and a secondary. | |
1614 | * This all lives at the end of the device. | |
1615 | * So it is easiest to find the earliest of primary and | |
1616 | * secondary, and copy everything from there. | |
1617 | * | |
1618 | * Anchor is 512 from end It contains primary_lba and secondary_lba | |
1619 | * we choose one of those | |
1620 | */ | |
1621 | ||
1622 | if (posix_memalign(&buf, 4096, 4096) != 0) | |
1623 | return 1; | |
1624 | ||
1625 | if (!get_dev_size(from, NULL, &dsize)) | |
1626 | goto err; | |
1627 | ||
1628 | if (lseek64(from, dsize-512, 0) < 0) | |
1629 | goto err; | |
1630 | if (read(from, buf, 512) != 512) | |
1631 | goto err; | |
1632 | ddf = buf; | |
1633 | if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) || | |
1634 | !be32_eq(calc_crc(ddf, 512), ddf->crc) || | |
1635 | (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 && | |
1636 | memcmp(ddf->revision, DDF_REVISION_2, 8) != 0)) | |
1637 | goto err; | |
1638 | ||
1639 | offset = dsize - 512; | |
1640 | if ((be64_to_cpu(ddf->primary_lba) << 9) < offset) | |
1641 | offset = be64_to_cpu(ddf->primary_lba) << 9; | |
1642 | if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset) | |
1643 | offset = be64_to_cpu(ddf->secondary_lba) << 9; | |
1644 | ||
1645 | bytes = dsize - offset; | |
1646 | ||
1647 | if (lseek64(from, offset, 0) < 0 || | |
1648 | lseek64(to, offset, 0) < 0) | |
1649 | goto err; | |
1650 | while (written < bytes) { | |
1651 | int n = bytes - written; | |
1652 | if (n > 4096) | |
1653 | n = 4096; | |
1654 | if (read(from, buf, n) != n) | |
1655 | goto err; | |
1656 | if (write(to, buf, n) != n) | |
1657 | goto err; | |
1658 | written += n; | |
1659 | } | |
1660 | free(buf); | |
1661 | return 0; | |
1662 | err: | |
1663 | free(buf); | |
1664 | return 1; | |
1665 | } | |
1666 | ||
1667 | static void detail_super_ddf(struct supertype *st, char *homehost) | |
1668 | { | |
1669 | /* FIXME later | |
1670 | * Could print DDF GUID | |
1671 | * Need to find which array | |
1672 | * If whole, briefly list all arrays | |
1673 | * If one, give name | |
1674 | */ | |
1675 | } | |
1676 | ||
1677 | static const char *vendors_with_variable_volume_UUID[] = { | |
1678 | "LSI ", | |
1679 | }; | |
1680 | ||
1681 | static int volume_id_is_reliable(const struct ddf_super *ddf) | |
1682 | { | |
1683 | int n = ARRAY_SIZE(vendors_with_variable_volume_UUID); | |
1684 | int i; | |
1685 | for (i = 0; i < n; i++) | |
1686 | if (!memcmp(ddf->controller.guid, | |
1687 | vendors_with_variable_volume_UUID[i], 8)) | |
1688 | return 0; | |
1689 | return 1; | |
1690 | } | |
1691 | ||
1692 | static void uuid_of_ddf_subarray(const struct ddf_super *ddf, | |
1693 | unsigned int vcnum, int uuid[4]) | |
1694 | { | |
1695 | char buf[DDF_GUID_LEN+18], sha[20], *p; | |
1696 | struct sha1_ctx ctx; | |
1697 | if (volume_id_is_reliable(ddf)) { | |
1698 | uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid); | |
1699 | return; | |
1700 | } | |
1701 | /* | |
1702 | * Some fake RAID BIOSes (in particular, LSI ones) change the | |
1703 | * VD GUID at every boot. These GUIDs are not suitable for | |
1704 | * identifying an array. Luckily the header GUID appears to | |
1705 | * remain constant. | |
1706 | * We construct a pseudo-UUID from the header GUID and those | |
1707 | * properties of the subarray that we expect to remain constant. | |
1708 | */ | |
1709 | memset(buf, 0, sizeof(buf)); | |
1710 | p = buf; | |
1711 | memcpy(p, ddf->anchor.guid, DDF_GUID_LEN); | |
1712 | p += DDF_GUID_LEN; | |
1713 | memcpy(p, ddf->virt->entries[vcnum].name, 16); | |
1714 | p += 16; | |
1715 | *((__u16 *) p) = vcnum; | |
1716 | sha1_init_ctx(&ctx); | |
1717 | sha1_process_bytes(buf, sizeof(buf), &ctx); | |
1718 | sha1_finish_ctx(&ctx, sha); | |
1719 | memcpy(uuid, sha, 4*4); | |
1720 | } | |
1721 | ||
1722 | static void brief_detail_super_ddf(struct supertype *st) | |
1723 | { | |
1724 | struct mdinfo info; | |
1725 | char nbuf[64]; | |
1726 | struct ddf_super *ddf = st->sb; | |
1727 | unsigned int vcnum = get_vd_num_of_subarray(st); | |
1728 | if (vcnum == DDF_CONTAINER) | |
1729 | uuid_from_super_ddf(st, info.uuid); | |
1730 | else if (vcnum == DDF_NOTFOUND) | |
1731 | return; | |
1732 | else | |
1733 | uuid_of_ddf_subarray(ddf, vcnum, info.uuid); | |
1734 | fname_from_uuid(st, &info, nbuf,':'); | |
1735 | printf(" UUID=%s", nbuf + 5); | |
1736 | } | |
1737 | #endif | |
1738 | ||
1739 | static int match_home_ddf(struct supertype *st, char *homehost) | |
1740 | { | |
1741 | /* It matches 'this' host if the controller is a | |
1742 | * Linux-MD controller with vendor_data matching | |
1743 | * the hostname | |
1744 | */ | |
1745 | struct ddf_super *ddf = st->sb; | |
1746 | unsigned int len; | |
1747 | ||
1748 | if (!homehost) | |
1749 | return 0; | |
1750 | len = strlen(homehost); | |
1751 | ||
1752 | return (memcmp(ddf->controller.guid, T10, 8) == 0 && | |
1753 | len < sizeof(ddf->controller.vendor_data) && | |
1754 | memcmp(ddf->controller.vendor_data, homehost,len) == 0 && | |
1755 | ddf->controller.vendor_data[len] == 0); | |
1756 | } | |
1757 | ||
1758 | #ifndef MDASSEMBLE | |
1759 | static int find_index_in_bvd(const struct ddf_super *ddf, | |
1760 | const struct vd_config *conf, unsigned int n, | |
1761 | unsigned int *n_bvd) | |
1762 | { | |
1763 | /* | |
1764 | * Find the index of the n-th valid physical disk in this BVD | |
1765 | */ | |
1766 | unsigned int i, j; | |
1767 | for (i = 0, j = 0; i < ddf->mppe && | |
1768 | j < be16_to_cpu(conf->prim_elmnt_count); i++) { | |
1769 | if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) { | |
1770 | if (n == j) { | |
1771 | *n_bvd = i; | |
1772 | return 1; | |
1773 | } | |
1774 | j++; | |
1775 | } | |
1776 | } | |
1777 | dprintf("%s: couldn't find BVD member %u (total %u)\n", | |
1778 | __func__, n, be16_to_cpu(conf->prim_elmnt_count)); | |
1779 | return 0; | |
1780 | } | |
1781 | ||
1782 | static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst, | |
1783 | unsigned int n, | |
1784 | unsigned int *n_bvd, struct vcl **vcl) | |
1785 | { | |
1786 | struct vcl *v; | |
1787 | ||
1788 | for (v = ddf->conflist; v; v = v->next) { | |
1789 | unsigned int nsec, ibvd = 0; | |
1790 | struct vd_config *conf; | |
1791 | if (inst != v->vcnum) | |
1792 | continue; | |
1793 | conf = &v->conf; | |
1794 | if (conf->sec_elmnt_count == 1) { | |
1795 | if (find_index_in_bvd(ddf, conf, n, n_bvd)) { | |
1796 | *vcl = v; | |
1797 | return conf; | |
1798 | } else | |
1799 | goto bad; | |
1800 | } | |
1801 | if (v->other_bvds == NULL) { | |
1802 | pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n", | |
1803 | __func__, conf->sec_elmnt_count); | |
1804 | goto bad; | |
1805 | } | |
1806 | nsec = n / be16_to_cpu(conf->prim_elmnt_count); | |
1807 | if (conf->sec_elmnt_seq != nsec) { | |
1808 | for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) { | |
1809 | if (v->other_bvds[ibvd-1]->sec_elmnt_seq | |
1810 | == nsec) | |
1811 | break; | |
1812 | } | |
1813 | if (ibvd == conf->sec_elmnt_count) | |
1814 | goto bad; | |
1815 | conf = v->other_bvds[ibvd-1]; | |
1816 | } | |
1817 | if (!find_index_in_bvd(ddf, conf, | |
1818 | n - nsec*conf->sec_elmnt_count, n_bvd)) | |
1819 | goto bad; | |
1820 | dprintf("%s: found disk %u as member %u in bvd %d of array %u\n" | |
1821 | , __func__, n, *n_bvd, ibvd, inst); | |
1822 | *vcl = v; | |
1823 | return conf; | |
1824 | } | |
1825 | bad: | |
1826 | pr_err("%s: Could't find disk %d in array %u\n", __func__, n, inst); | |
1827 | return NULL; | |
1828 | } | |
1829 | #endif | |
1830 | ||
1831 | static int find_phys(const struct ddf_super *ddf, be32 phys_refnum) | |
1832 | { | |
1833 | /* Find the entry in phys_disk which has the given refnum | |
1834 | * and return it's index | |
1835 | */ | |
1836 | unsigned int i; | |
1837 | for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) | |
1838 | if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum)) | |
1839 | return i; | |
1840 | return -1; | |
1841 | } | |
1842 | ||
1843 | static void uuid_from_ddf_guid(const char *guid, int uuid[4]) | |
1844 | { | |
1845 | char buf[20]; | |
1846 | struct sha1_ctx ctx; | |
1847 | sha1_init_ctx(&ctx); | |
1848 | sha1_process_bytes(guid, DDF_GUID_LEN, &ctx); | |
1849 | sha1_finish_ctx(&ctx, buf); | |
1850 | memcpy(uuid, buf, 4*4); | |
1851 | } | |
1852 | ||
1853 | static void uuid_from_super_ddf(struct supertype *st, int uuid[4]) | |
1854 | { | |
1855 | /* The uuid returned here is used for: | |
1856 | * uuid to put into bitmap file (Create, Grow) | |
1857 | * uuid for backup header when saving critical section (Grow) | |
1858 | * comparing uuids when re-adding a device into an array | |
1859 | * In these cases the uuid required is that of the data-array, | |
1860 | * not the device-set. | |
1861 | * uuid to recognise same set when adding a missing device back | |
1862 | * to an array. This is a uuid for the device-set. | |
1863 | * | |
1864 | * For each of these we can make do with a truncated | |
1865 | * or hashed uuid rather than the original, as long as | |
1866 | * everyone agrees. | |
1867 | * In the case of SVD we assume the BVD is of interest, | |
1868 | * though that might be the case if a bitmap were made for | |
1869 | * a mirrored SVD - worry about that later. | |
1870 | * So we need to find the VD configuration record for the | |
1871 | * relevant BVD and extract the GUID and Secondary_Element_Seq. | |
1872 | * The first 16 bytes of the sha1 of these is used. | |
1873 | */ | |
1874 | struct ddf_super *ddf = st->sb; | |
1875 | struct vcl *vcl = ddf->currentconf; | |
1876 | ||
1877 | if (vcl) | |
1878 | uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid); | |
1879 | else | |
1880 | uuid_from_ddf_guid(ddf->anchor.guid, uuid); | |
1881 | } | |
1882 | ||
1883 | static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map); | |
1884 | ||
1885 | static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map) | |
1886 | { | |
1887 | struct ddf_super *ddf = st->sb; | |
1888 | int map_disks = info->array.raid_disks; | |
1889 | __u32 *cptr; | |
1890 | ||
1891 | if (ddf->currentconf) { | |
1892 | getinfo_super_ddf_bvd(st, info, map); | |
1893 | return; | |
1894 | } | |
1895 | memset(info, 0, sizeof(*info)); | |
1896 | ||
1897 | info->array.raid_disks = be16_to_cpu(ddf->phys->used_pdes); | |
1898 | info->array.level = LEVEL_CONTAINER; | |
1899 | info->array.layout = 0; | |
1900 | info->array.md_minor = -1; | |
1901 | cptr = (__u32 *)(ddf->anchor.guid + 16); | |
1902 | info->array.ctime = DECADE + __be32_to_cpu(*cptr); | |
1903 | ||
1904 | info->array.utime = 0; | |
1905 | info->array.chunk_size = 0; | |
1906 | info->container_enough = 1; | |
1907 | ||
1908 | info->disk.major = 0; | |
1909 | info->disk.minor = 0; | |
1910 | if (ddf->dlist) { | |
1911 | info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum); | |
1912 | info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum); | |
1913 | ||
1914 | info->data_offset = be64_to_cpu(ddf->phys-> | |
1915 | entries[info->disk.raid_disk]. | |
1916 | config_size); | |
1917 | info->component_size = ddf->dlist->size - info->data_offset; | |
1918 | } else { | |
1919 | info->disk.number = -1; | |
1920 | info->disk.raid_disk = -1; | |
1921 | // info->disk.raid_disk = find refnum in the table and use index; | |
1922 | } | |
1923 | info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
1924 | ||
1925 | info->recovery_start = MaxSector; | |
1926 | info->reshape_active = 0; | |
1927 | info->recovery_blocked = 0; | |
1928 | info->name[0] = 0; | |
1929 | ||
1930 | info->array.major_version = -1; | |
1931 | info->array.minor_version = -2; | |
1932 | strcpy(info->text_version, "ddf"); | |
1933 | info->safe_mode_delay = 0; | |
1934 | ||
1935 | uuid_from_super_ddf(st, info->uuid); | |
1936 | ||
1937 | if (map) { | |
1938 | int i; | |
1939 | for (i = 0 ; i < map_disks; i++) { | |
1940 | if (i < info->array.raid_disks && | |
1941 | (be16_to_cpu(ddf->phys->entries[i].state) | |
1942 | & DDF_Online) && | |
1943 | !(be16_to_cpu(ddf->phys->entries[i].state) | |
1944 | & DDF_Failed)) | |
1945 | map[i] = 1; | |
1946 | else | |
1947 | map[i] = 0; | |
1948 | } | |
1949 | } | |
1950 | } | |
1951 | ||
1952 | static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map) | |
1953 | { | |
1954 | struct ddf_super *ddf = st->sb; | |
1955 | struct vcl *vc = ddf->currentconf; | |
1956 | int cd = ddf->currentdev; | |
1957 | int n_prim; | |
1958 | int j; | |
1959 | struct dl *dl; | |
1960 | int map_disks = info->array.raid_disks; | |
1961 | __u32 *cptr; | |
1962 | struct vd_config *conf; | |
1963 | ||
1964 | memset(info, 0, sizeof(*info)); | |
1965 | if (layout_ddf2md(&vc->conf, &info->array) == -1) | |
1966 | return; | |
1967 | info->array.md_minor = -1; | |
1968 | cptr = (__u32 *)(vc->conf.guid + 16); | |
1969 | info->array.ctime = DECADE + __be32_to_cpu(*cptr); | |
1970 | info->array.utime = DECADE + be32_to_cpu(vc->conf.timestamp); | |
1971 | info->array.chunk_size = 512 << vc->conf.chunk_shift; | |
1972 | info->custom_array_size = 0; | |
1973 | ||
1974 | conf = &vc->conf; | |
1975 | n_prim = be16_to_cpu(conf->prim_elmnt_count); | |
1976 | if (conf->sec_elmnt_count > 1 && cd >= n_prim) { | |
1977 | int ibvd = cd / n_prim - 1; | |
1978 | cd %= n_prim; | |
1979 | conf = vc->other_bvds[ibvd]; | |
1980 | } | |
1981 | ||
1982 | if (cd >= 0 && (unsigned)cd < ddf->mppe) { | |
1983 | info->data_offset = | |
1984 | be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]); | |
1985 | if (vc->block_sizes) | |
1986 | info->component_size = vc->block_sizes[cd]; | |
1987 | else | |
1988 | info->component_size = be64_to_cpu(conf->blocks); | |
1989 | } | |
1990 | ||
1991 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
1992 | if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd])) | |
1993 | break; | |
1994 | ||
1995 | info->disk.major = 0; | |
1996 | info->disk.minor = 0; | |
1997 | info->disk.state = 0; | |
1998 | if (dl) { | |
1999 | info->disk.major = dl->major; | |
2000 | info->disk.minor = dl->minor; | |
2001 | info->disk.raid_disk = cd + conf->sec_elmnt_seq | |
2002 | * be16_to_cpu(conf->prim_elmnt_count); | |
2003 | info->disk.number = dl->pdnum; | |
2004 | info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE); | |
2005 | } | |
2006 | ||
2007 | info->container_member = ddf->currentconf->vcnum; | |
2008 | ||
2009 | info->recovery_start = MaxSector; | |
2010 | info->resync_start = 0; | |
2011 | info->reshape_active = 0; | |
2012 | info->recovery_blocked = 0; | |
2013 | if (!(ddf->virt->entries[info->container_member].state | |
2014 | & DDF_state_inconsistent) && | |
2015 | (ddf->virt->entries[info->container_member].init_state | |
2016 | & DDF_initstate_mask) | |
2017 | == DDF_init_full) | |
2018 | info->resync_start = MaxSector; | |
2019 | ||
2020 | uuid_from_super_ddf(st, info->uuid); | |
2021 | ||
2022 | info->array.major_version = -1; | |
2023 | info->array.minor_version = -2; | |
2024 | sprintf(info->text_version, "/%s/%d", | |
2025 | st->container_devnm, | |
2026 | info->container_member); | |
2027 | info->safe_mode_delay = DDF_SAFE_MODE_DELAY; | |
2028 | ||
2029 | memcpy(info->name, ddf->virt->entries[info->container_member].name, 16); | |
2030 | info->name[16]=0; | |
2031 | for(j=0; j<16; j++) | |
2032 | if (info->name[j] == ' ') | |
2033 | info->name[j] = 0; | |
2034 | ||
2035 | if (map) | |
2036 | for (j = 0; j < map_disks; j++) { | |
2037 | map[j] = 0; | |
2038 | if (j < info->array.raid_disks) { | |
2039 | int i = find_phys(ddf, vc->conf.phys_refnum[j]); | |
2040 | if (i >= 0 && | |
2041 | (be16_to_cpu(ddf->phys->entries[i].state) | |
2042 | & DDF_Online) && | |
2043 | !(be16_to_cpu(ddf->phys->entries[i].state) | |
2044 | & DDF_Failed)) | |
2045 | map[i] = 1; | |
2046 | } | |
2047 | } | |
2048 | } | |
2049 | ||
2050 | static int update_super_ddf(struct supertype *st, struct mdinfo *info, | |
2051 | char *update, | |
2052 | char *devname, int verbose, | |
2053 | int uuid_set, char *homehost) | |
2054 | { | |
2055 | /* For 'assemble' and 'force' we need to return non-zero if any | |
2056 | * change was made. For others, the return value is ignored. | |
2057 | * Update options are: | |
2058 | * force-one : This device looks a bit old but needs to be included, | |
2059 | * update age info appropriately. | |
2060 | * assemble: clear any 'faulty' flag to allow this device to | |
2061 | * be assembled. | |
2062 | * force-array: Array is degraded but being forced, mark it clean | |
2063 | * if that will be needed to assemble it. | |
2064 | * | |
2065 | * newdev: not used ???? | |
2066 | * grow: Array has gained a new device - this is currently for | |
2067 | * linear only | |
2068 | * resync: mark as dirty so a resync will happen. | |
2069 | * uuid: Change the uuid of the array to match what is given | |
2070 | * homehost: update the recorded homehost | |
2071 | * name: update the name - preserving the homehost | |
2072 | * _reshape_progress: record new reshape_progress position. | |
2073 | * | |
2074 | * Following are not relevant for this version: | |
2075 | * sparc2.2 : update from old dodgey metadata | |
2076 | * super-minor: change the preferred_minor number | |
2077 | * summaries: update redundant counters. | |
2078 | */ | |
2079 | int rv = 0; | |
2080 | // struct ddf_super *ddf = st->sb; | |
2081 | // struct vd_config *vd = find_vdcr(ddf, info->container_member); | |
2082 | // struct virtual_entry *ve = find_ve(ddf); | |
2083 | ||
2084 | /* we don't need to handle "force-*" or "assemble" as | |
2085 | * there is no need to 'trick' the kernel. We the metadata is | |
2086 | * first updated to activate the array, all the implied modifications | |
2087 | * will just happen. | |
2088 | */ | |
2089 | ||
2090 | if (strcmp(update, "grow") == 0) { | |
2091 | /* FIXME */ | |
2092 | } else if (strcmp(update, "resync") == 0) { | |
2093 | // info->resync_checkpoint = 0; | |
2094 | } else if (strcmp(update, "homehost") == 0) { | |
2095 | /* homehost is stored in controller->vendor_data, | |
2096 | * or it is when we are the vendor | |
2097 | */ | |
2098 | // if (info->vendor_is_local) | |
2099 | // strcpy(ddf->controller.vendor_data, homehost); | |
2100 | rv = -1; | |
2101 | } else if (strcmp(update, "name") == 0) { | |
2102 | /* name is stored in virtual_entry->name */ | |
2103 | // memset(ve->name, ' ', 16); | |
2104 | // strncpy(ve->name, info->name, 16); | |
2105 | rv = -1; | |
2106 | } else if (strcmp(update, "_reshape_progress") == 0) { | |
2107 | /* We don't support reshape yet */ | |
2108 | } else if (strcmp(update, "assemble") == 0 ) { | |
2109 | /* Do nothing, just succeed */ | |
2110 | rv = 0; | |
2111 | } else | |
2112 | rv = -1; | |
2113 | ||
2114 | // update_all_csum(ddf); | |
2115 | ||
2116 | return rv; | |
2117 | } | |
2118 | ||
2119 | static void make_header_guid(char *guid) | |
2120 | { | |
2121 | be32 stamp; | |
2122 | /* Create a DDF Header of Virtual Disk GUID */ | |
2123 | ||
2124 | /* 24 bytes of fiction required. | |
2125 | * first 8 are a 'vendor-id' - "Linux-MD" | |
2126 | * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000 | |
2127 | * Remaining 8 random number plus timestamp | |
2128 | */ | |
2129 | memcpy(guid, T10, sizeof(T10)); | |
2130 | stamp = cpu_to_be32(0xdeadbeef); | |
2131 | memcpy(guid+8, &stamp, 4); | |
2132 | stamp = cpu_to_be32(0); | |
2133 | memcpy(guid+12, &stamp, 4); | |
2134 | stamp = cpu_to_be32(time(0) - DECADE); | |
2135 | memcpy(guid+16, &stamp, 4); | |
2136 | stamp._v32 = random32(); | |
2137 | memcpy(guid+20, &stamp, 4); | |
2138 | } | |
2139 | ||
2140 | static unsigned int find_unused_vde(const struct ddf_super *ddf) | |
2141 | { | |
2142 | unsigned int i; | |
2143 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) { | |
2144 | if (all_ff(ddf->virt->entries[i].guid)) | |
2145 | return i; | |
2146 | } | |
2147 | return DDF_NOTFOUND; | |
2148 | } | |
2149 | ||
2150 | static unsigned int find_vde_by_name(const struct ddf_super *ddf, | |
2151 | const char *name) | |
2152 | { | |
2153 | unsigned int i; | |
2154 | if (name == NULL) | |
2155 | return DDF_NOTFOUND; | |
2156 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) { | |
2157 | if (all_ff(ddf->virt->entries[i].guid)) | |
2158 | continue; | |
2159 | if (!strncmp(name, ddf->virt->entries[i].name, | |
2160 | sizeof(ddf->virt->entries[i].name))) | |
2161 | return i; | |
2162 | } | |
2163 | return DDF_NOTFOUND; | |
2164 | } | |
2165 | ||
2166 | #ifndef MDASSEMBLE | |
2167 | static unsigned int find_vde_by_guid(const struct ddf_super *ddf, | |
2168 | const char *guid) | |
2169 | { | |
2170 | unsigned int i; | |
2171 | if (guid == NULL || all_ff(guid)) | |
2172 | return DDF_NOTFOUND; | |
2173 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) | |
2174 | if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN)) | |
2175 | return i; | |
2176 | return DDF_NOTFOUND; | |
2177 | } | |
2178 | #endif | |
2179 | ||
2180 | static int init_super_ddf_bvd(struct supertype *st, | |
2181 | mdu_array_info_t *info, | |
2182 | unsigned long long size, | |
2183 | char *name, char *homehost, | |
2184 | int *uuid, unsigned long long data_offset); | |
2185 | ||
2186 | static int init_super_ddf(struct supertype *st, | |
2187 | mdu_array_info_t *info, | |
2188 | unsigned long long size, char *name, char *homehost, | |
2189 | int *uuid, unsigned long long data_offset) | |
2190 | { | |
2191 | /* This is primarily called by Create when creating a new array. | |
2192 | * We will then get add_to_super called for each component, and then | |
2193 | * write_init_super called to write it out to each device. | |
2194 | * For DDF, Create can create on fresh devices or on a pre-existing | |
2195 | * array. | |
2196 | * To create on a pre-existing array a different method will be called. | |
2197 | * This one is just for fresh drives. | |
2198 | * | |
2199 | * We need to create the entire 'ddf' structure which includes: | |
2200 | * DDF headers - these are easy. | |
2201 | * Controller data - a Sector describing this controller .. not that | |
2202 | * this is a controller exactly. | |
2203 | * Physical Disk Record - one entry per device, so | |
2204 | * leave plenty of space. | |
2205 | * Virtual Disk Records - again, just leave plenty of space. | |
2206 | * This just lists VDs, doesn't give details | |
2207 | * Config records - describes the VDs that use this disk | |
2208 | * DiskData - describes 'this' device. | |
2209 | * BadBlockManagement - empty | |
2210 | * Diag Space - empty | |
2211 | * Vendor Logs - Could we put bitmaps here? | |
2212 | * | |
2213 | */ | |
2214 | struct ddf_super *ddf; | |
2215 | char hostname[17]; | |
2216 | int hostlen; | |
2217 | int max_phys_disks, max_virt_disks; | |
2218 | unsigned long long sector; | |
2219 | int clen; | |
2220 | int i; | |
2221 | int pdsize, vdsize; | |
2222 | struct phys_disk *pd; | |
2223 | struct virtual_disk *vd; | |
2224 | ||
2225 | if (data_offset != INVALID_SECTORS) { | |
2226 | pr_err("data-offset not supported by DDF\n"); | |
2227 | return 0; | |
2228 | } | |
2229 | ||
2230 | if (st->sb) | |
2231 | return init_super_ddf_bvd(st, info, size, name, homehost, uuid, | |
2232 | data_offset); | |
2233 | ||
2234 | if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) { | |
2235 | pr_err("%s could not allocate superblock\n", __func__); | |
2236 | return 0; | |
2237 | } | |
2238 | memset(ddf, 0, sizeof(*ddf)); | |
2239 | ddf->dlist = NULL; /* no physical disks yet */ | |
2240 | ddf->conflist = NULL; /* No virtual disks yet */ | |
2241 | st->sb = ddf; | |
2242 | ||
2243 | if (info == NULL) { | |
2244 | /* zeroing superblock */ | |
2245 | return 0; | |
2246 | } | |
2247 | ||
2248 | /* At least 32MB *must* be reserved for the ddf. So let's just | |
2249 | * start 32MB from the end, and put the primary header there. | |
2250 | * Don't do secondary for now. | |
2251 | * We don't know exactly where that will be yet as it could be | |
2252 | * different on each device. To just set up the lengths. | |
2253 | * | |
2254 | */ | |
2255 | ||
2256 | ddf->anchor.magic = DDF_HEADER_MAGIC; | |
2257 | make_header_guid(ddf->anchor.guid); | |
2258 | ||
2259 | memcpy(ddf->anchor.revision, DDF_REVISION_2, 8); | |
2260 | ddf->anchor.seq = cpu_to_be32(1); | |
2261 | ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE); | |
2262 | ddf->anchor.openflag = 0xFF; | |
2263 | ddf->anchor.foreignflag = 0; | |
2264 | ddf->anchor.enforcegroups = 0; /* Is this best?? */ | |
2265 | ddf->anchor.pad0 = 0xff; | |
2266 | memset(ddf->anchor.pad1, 0xff, 12); | |
2267 | memset(ddf->anchor.header_ext, 0xff, 32); | |
2268 | ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0); | |
2269 | ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0); | |
2270 | ddf->anchor.type = DDF_HEADER_ANCHOR; | |
2271 | memset(ddf->anchor.pad2, 0xff, 3); | |
2272 | ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */ | |
2273 | /* Put this at bottom of 32M reserved.. */ | |
2274 | ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0); | |
2275 | max_phys_disks = 1023; /* Should be enough */ | |
2276 | ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks); | |
2277 | max_virt_disks = 255; | |
2278 | ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks); /* ?? */ | |
2279 | ddf->anchor.max_partitions = cpu_to_be16(64); /* ?? */ | |
2280 | ddf->max_part = 64; | |
2281 | ddf->mppe = 256; | |
2282 | ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512; | |
2283 | ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len); | |
2284 | ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe); | |
2285 | memset(ddf->anchor.pad3, 0xff, 54); | |
2286 | /* controller sections is one sector long immediately | |
2287 | * after the ddf header */ | |
2288 | sector = 1; | |
2289 | ddf->anchor.controller_section_offset = cpu_to_be32(sector); | |
2290 | ddf->anchor.controller_section_length = cpu_to_be32(1); | |
2291 | sector += 1; | |
2292 | ||
2293 | /* phys is 8 sectors after that */ | |
2294 | pdsize = ROUND_UP(sizeof(struct phys_disk) + | |
2295 | sizeof(struct phys_disk_entry)*max_phys_disks, | |
2296 | 512); | |
2297 | switch(pdsize/512) { | |
2298 | case 2: case 8: case 32: case 128: case 512: break; | |
2299 | default: abort(); | |
2300 | } | |
2301 | ddf->anchor.phys_section_offset = cpu_to_be32(sector); | |
2302 | ddf->anchor.phys_section_length = | |
2303 | cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */ | |
2304 | sector += pdsize/512; | |
2305 | ||
2306 | /* virt is another 32 sectors */ | |
2307 | vdsize = ROUND_UP(sizeof(struct virtual_disk) + | |
2308 | sizeof(struct virtual_entry) * max_virt_disks, | |
2309 | 512); | |
2310 | switch(vdsize/512) { | |
2311 | case 2: case 8: case 32: case 128: case 512: break; | |
2312 | default: abort(); | |
2313 | } | |
2314 | ddf->anchor.virt_section_offset = cpu_to_be32(sector); | |
2315 | ddf->anchor.virt_section_length = | |
2316 | cpu_to_be32(vdsize/512); /* max_vd_entries/8 */ | |
2317 | sector += vdsize/512; | |
2318 | ||
2319 | clen = ddf->conf_rec_len * (ddf->max_part+1); | |
2320 | ddf->anchor.config_section_offset = cpu_to_be32(sector); | |
2321 | ddf->anchor.config_section_length = cpu_to_be32(clen); | |
2322 | sector += clen; | |
2323 | ||
2324 | ddf->anchor.data_section_offset = cpu_to_be32(sector); | |
2325 | ddf->anchor.data_section_length = cpu_to_be32(1); | |
2326 | sector += 1; | |
2327 | ||
2328 | ddf->anchor.bbm_section_length = cpu_to_be32(0); | |
2329 | ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF); | |
2330 | ddf->anchor.diag_space_length = cpu_to_be32(0); | |
2331 | ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF); | |
2332 | ddf->anchor.vendor_length = cpu_to_be32(0); | |
2333 | ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF); | |
2334 | ||
2335 | memset(ddf->anchor.pad4, 0xff, 256); | |
2336 | ||
2337 | memcpy(&ddf->primary, &ddf->anchor, 512); | |
2338 | memcpy(&ddf->secondary, &ddf->anchor, 512); | |
2339 | ||
2340 | ddf->primary.openflag = 1; /* I guess.. */ | |
2341 | ddf->primary.type = DDF_HEADER_PRIMARY; | |
2342 | ||
2343 | ddf->secondary.openflag = 1; /* I guess.. */ | |
2344 | ddf->secondary.type = DDF_HEADER_SECONDARY; | |
2345 | ||
2346 | ddf->active = &ddf->primary; | |
2347 | ||
2348 | ddf->controller.magic = DDF_CONTROLLER_MAGIC; | |
2349 | ||
2350 | /* 24 more bytes of fiction required. | |
2351 | * first 8 are a 'vendor-id' - "Linux-MD" | |
2352 | * Remaining 16 are serial number.... maybe a hostname would do? | |
2353 | */ | |
2354 | memcpy(ddf->controller.guid, T10, sizeof(T10)); | |
2355 | gethostname(hostname, sizeof(hostname)); | |
2356 | hostname[sizeof(hostname) - 1] = 0; | |
2357 | hostlen = strlen(hostname); | |
2358 | memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen); | |
2359 | for (i = strlen(T10) ; i+hostlen < 24; i++) | |
2360 | ddf->controller.guid[i] = ' '; | |
2361 | ||
2362 | ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD); | |
2363 | ddf->controller.type.device_id = cpu_to_be16(0xBEEF); | |
2364 | ddf->controller.type.sub_vendor_id = cpu_to_be16(0); | |
2365 | ddf->controller.type.sub_device_id = cpu_to_be16(0); | |
2366 | memcpy(ddf->controller.product_id, "What Is My PID??", 16); | |
2367 | memset(ddf->controller.pad, 0xff, 8); | |
2368 | memset(ddf->controller.vendor_data, 0xff, 448); | |
2369 | if (homehost && strlen(homehost) < 440) | |
2370 | strcpy((char*)ddf->controller.vendor_data, homehost); | |
2371 | ||
2372 | if (posix_memalign((void**)&pd, 512, pdsize) != 0) { | |
2373 | pr_err("%s could not allocate pd\n", __func__); | |
2374 | return 0; | |
2375 | } | |
2376 | ddf->phys = pd; | |
2377 | ddf->pdsize = pdsize; | |
2378 | ||
2379 | memset(pd, 0xff, pdsize); | |
2380 | memset(pd, 0, sizeof(*pd)); | |
2381 | pd->magic = DDF_PHYS_RECORDS_MAGIC; | |
2382 | pd->used_pdes = cpu_to_be16(0); | |
2383 | pd->max_pdes = cpu_to_be16(max_phys_disks); | |
2384 | memset(pd->pad, 0xff, 52); | |
2385 | for (i = 0; i < max_phys_disks; i++) | |
2386 | memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN); | |
2387 | ||
2388 | if (posix_memalign((void**)&vd, 512, vdsize) != 0) { | |
2389 | pr_err("%s could not allocate vd\n", __func__); | |
2390 | return 0; | |
2391 | } | |
2392 | ddf->virt = vd; | |
2393 | ddf->vdsize = vdsize; | |
2394 | memset(vd, 0, vdsize); | |
2395 | vd->magic = DDF_VIRT_RECORDS_MAGIC; | |
2396 | vd->populated_vdes = cpu_to_be16(0); | |
2397 | vd->max_vdes = cpu_to_be16(max_virt_disks); | |
2398 | memset(vd->pad, 0xff, 52); | |
2399 | ||
2400 | for (i=0; i<max_virt_disks; i++) | |
2401 | memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry)); | |
2402 | ||
2403 | st->sb = ddf; | |
2404 | ddf_set_updates_pending(ddf); | |
2405 | return 1; | |
2406 | } | |
2407 | ||
2408 | static int chunk_to_shift(int chunksize) | |
2409 | { | |
2410 | return ffs(chunksize/512)-1; | |
2411 | } | |
2412 | ||
2413 | #ifndef MDASSEMBLE | |
2414 | struct extent { | |
2415 | unsigned long long start, size; | |
2416 | }; | |
2417 | static int cmp_extent(const void *av, const void *bv) | |
2418 | { | |
2419 | const struct extent *a = av; | |
2420 | const struct extent *b = bv; | |
2421 | if (a->start < b->start) | |
2422 | return -1; | |
2423 | if (a->start > b->start) | |
2424 | return 1; | |
2425 | return 0; | |
2426 | } | |
2427 | ||
2428 | static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl) | |
2429 | { | |
2430 | /* find a list of used extents on the give physical device | |
2431 | * (dnum) of the given ddf. | |
2432 | * Return a malloced array of 'struct extent' | |
2433 | ||
2434 | * FIXME ignore DDF_Legacy devices? | |
2435 | ||
2436 | */ | |
2437 | struct extent *rv; | |
2438 | int n = 0; | |
2439 | unsigned int i; | |
2440 | __u16 state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state); | |
2441 | ||
2442 | if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online) | |
2443 | return NULL; | |
2444 | ||
2445 | rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2)); | |
2446 | ||
2447 | for (i = 0; i < ddf->max_part; i++) { | |
2448 | const struct vd_config *bvd; | |
2449 | unsigned int ibvd; | |
2450 | struct vcl *v = dl->vlist[i]; | |
2451 | if (v == NULL || | |
2452 | get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe, | |
2453 | &bvd, &ibvd) == DDF_NOTFOUND) | |
2454 | continue; | |
2455 | rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]); | |
2456 | rv[n].size = be64_to_cpu(bvd->blocks); | |
2457 | n++; | |
2458 | } | |
2459 | qsort(rv, n, sizeof(*rv), cmp_extent); | |
2460 | ||
2461 | rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size); | |
2462 | rv[n].size = 0; | |
2463 | return rv; | |
2464 | } | |
2465 | #endif | |
2466 | ||
2467 | static int init_super_ddf_bvd(struct supertype *st, | |
2468 | mdu_array_info_t *info, | |
2469 | unsigned long long size, | |
2470 | char *name, char *homehost, | |
2471 | int *uuid, unsigned long long data_offset) | |
2472 | { | |
2473 | /* We are creating a BVD inside a pre-existing container. | |
2474 | * so st->sb is already set. | |
2475 | * We need to create a new vd_config and a new virtual_entry | |
2476 | */ | |
2477 | struct ddf_super *ddf = st->sb; | |
2478 | unsigned int venum, i; | |
2479 | struct virtual_entry *ve; | |
2480 | struct vcl *vcl; | |
2481 | struct vd_config *vc; | |
2482 | ||
2483 | if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) { | |
2484 | pr_err("This ddf already has an array called %s\n", name); | |
2485 | return 0; | |
2486 | } | |
2487 | venum = find_unused_vde(ddf); | |
2488 | if (venum == DDF_NOTFOUND) { | |
2489 | pr_err("Cannot find spare slot for virtual disk\n"); | |
2490 | return 0; | |
2491 | } | |
2492 | ve = &ddf->virt->entries[venum]; | |
2493 | ||
2494 | /* A Virtual Disk GUID contains the T10 Vendor ID, controller type, | |
2495 | * timestamp, random number | |
2496 | */ | |
2497 | make_header_guid(ve->guid); | |
2498 | ve->unit = cpu_to_be16(info->md_minor); | |
2499 | ve->pad0 = 0xFFFF; | |
2500 | ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid, | |
2501 | DDF_GUID_LEN); | |
2502 | ve->type = cpu_to_be16(0); | |
2503 | ve->state = DDF_state_degraded; /* Will be modified as devices are added */ | |
2504 | if (info->state & 1) /* clean */ | |
2505 | ve->init_state = DDF_init_full; | |
2506 | else | |
2507 | ve->init_state = DDF_init_not; | |
2508 | ||
2509 | memset(ve->pad1, 0xff, 14); | |
2510 | memset(ve->name, ' ', 16); | |
2511 | if (name) | |
2512 | strncpy(ve->name, name, 16); | |
2513 | ddf->virt->populated_vdes = | |
2514 | cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1); | |
2515 | ||
2516 | /* Now create a new vd_config */ | |
2517 | if (posix_memalign((void**)&vcl, 512, | |
2518 | (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) { | |
2519 | pr_err("%s could not allocate vd_config\n", __func__); | |
2520 | return 0; | |
2521 | } | |
2522 | vcl->vcnum = venum; | |
2523 | vcl->block_sizes = NULL; /* FIXME not for CONCAT */ | |
2524 | vc = &vcl->conf; | |
2525 | ||
2526 | vc->magic = DDF_VD_CONF_MAGIC; | |
2527 | memcpy(vc->guid, ve->guid, DDF_GUID_LEN); | |
2528 | vc->timestamp = cpu_to_be32(time(0)-DECADE); | |
2529 | vc->seqnum = cpu_to_be32(1); | |
2530 | memset(vc->pad0, 0xff, 24); | |
2531 | vc->chunk_shift = chunk_to_shift(info->chunk_size); | |
2532 | if (layout_md2ddf(info, vc) == -1 || | |
2533 | be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) { | |
2534 | pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n", | |
2535 | __func__, info->level, info->layout, info->raid_disks); | |
2536 | free(vcl); | |
2537 | return 0; | |
2538 | } | |
2539 | vc->sec_elmnt_seq = 0; | |
2540 | if (alloc_other_bvds(ddf, vcl) != 0) { | |
2541 | pr_err("%s could not allocate other bvds\n", | |
2542 | __func__); | |
2543 | free(vcl); | |
2544 | return 0; | |
2545 | } | |
2546 | vc->blocks = cpu_to_be64(info->size * 2); | |
2547 | vc->array_blocks = cpu_to_be64( | |
2548 | calc_array_size(info->level, info->raid_disks, info->layout, | |
2549 | info->chunk_size, info->size*2)); | |
2550 | memset(vc->pad1, 0xff, 8); | |
2551 | vc->spare_refs[0] = cpu_to_be32(0xffffffff); | |
2552 | vc->spare_refs[1] = cpu_to_be32(0xffffffff); | |
2553 | vc->spare_refs[2] = cpu_to_be32(0xffffffff); | |
2554 | vc->spare_refs[3] = cpu_to_be32(0xffffffff); | |
2555 | vc->spare_refs[4] = cpu_to_be32(0xffffffff); | |
2556 | vc->spare_refs[5] = cpu_to_be32(0xffffffff); | |
2557 | vc->spare_refs[6] = cpu_to_be32(0xffffffff); | |
2558 | vc->spare_refs[7] = cpu_to_be32(0xffffffff); | |
2559 | memset(vc->cache_pol, 0, 8); | |
2560 | vc->bg_rate = 0x80; | |
2561 | memset(vc->pad2, 0xff, 3); | |
2562 | memset(vc->pad3, 0xff, 52); | |
2563 | memset(vc->pad4, 0xff, 192); | |
2564 | memset(vc->v0, 0xff, 32); | |
2565 | memset(vc->v1, 0xff, 32); | |
2566 | memset(vc->v2, 0xff, 16); | |
2567 | memset(vc->v3, 0xff, 16); | |
2568 | memset(vc->vendor, 0xff, 32); | |
2569 | ||
2570 | memset(vc->phys_refnum, 0xff, 4*ddf->mppe); | |
2571 | memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe); | |
2572 | ||
2573 | for (i = 1; i < vc->sec_elmnt_count; i++) { | |
2574 | memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512); | |
2575 | vcl->other_bvds[i-1]->sec_elmnt_seq = i; | |
2576 | } | |
2577 | ||
2578 | vcl->next = ddf->conflist; | |
2579 | ddf->conflist = vcl; | |
2580 | ddf->currentconf = vcl; | |
2581 | ddf_set_updates_pending(ddf); | |
2582 | return 1; | |
2583 | } | |
2584 | ||
2585 | ||
2586 | #ifndef MDASSEMBLE | |
2587 | static int get_svd_state(const struct ddf_super *, const struct vcl *); | |
2588 | ||
2589 | static void add_to_super_ddf_bvd(struct supertype *st, | |
2590 | mdu_disk_info_t *dk, int fd, char *devname) | |
2591 | { | |
2592 | /* fd and devname identify a device with-in the ddf container (st). | |
2593 | * dk identifies a location in the new BVD. | |
2594 | * We need to find suitable free space in that device and update | |
2595 | * the phys_refnum and lba_offset for the newly created vd_config. | |
2596 | * We might also want to update the type in the phys_disk | |
2597 | * section. | |
2598 | * | |
2599 | * Alternately: fd == -1 and we have already chosen which device to | |
2600 | * use and recorded in dlist->raid_disk; | |
2601 | */ | |
2602 | struct dl *dl; | |
2603 | struct ddf_super *ddf = st->sb; | |
2604 | struct vd_config *vc; | |
2605 | unsigned int i; | |
2606 | unsigned long long blocks, pos, esize; | |
2607 | struct extent *ex; | |
2608 | unsigned int raid_disk = dk->raid_disk; | |
2609 | ||
2610 | if (fd == -1) { | |
2611 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2612 | if (dl->raiddisk == dk->raid_disk) | |
2613 | break; | |
2614 | } else { | |
2615 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2616 | if (dl->major == dk->major && | |
2617 | dl->minor == dk->minor) | |
2618 | break; | |
2619 | } | |
2620 | if (!dl || ! (dk->state & (1<<MD_DISK_SYNC))) | |
2621 | return; | |
2622 | ||
2623 | vc = &ddf->currentconf->conf; | |
2624 | if (vc->sec_elmnt_count > 1) { | |
2625 | unsigned int n = be16_to_cpu(vc->prim_elmnt_count); | |
2626 | if (raid_disk >= n) | |
2627 | vc = ddf->currentconf->other_bvds[raid_disk / n - 1]; | |
2628 | raid_disk %= n; | |
2629 | } | |
2630 | ||
2631 | ex = get_extents(ddf, dl); | |
2632 | if (!ex) | |
2633 | return; | |
2634 | ||
2635 | i = 0; pos = 0; | |
2636 | blocks = be64_to_cpu(vc->blocks); | |
2637 | if (ddf->currentconf->block_sizes) | |
2638 | blocks = ddf->currentconf->block_sizes[dk->raid_disk]; | |
2639 | ||
2640 | do { | |
2641 | esize = ex[i].start - pos; | |
2642 | if (esize >= blocks) | |
2643 | break; | |
2644 | pos = ex[i].start + ex[i].size; | |
2645 | i++; | |
2646 | } while (ex[i-1].size); | |
2647 | ||
2648 | free(ex); | |
2649 | if (esize < blocks) | |
2650 | return; | |
2651 | ||
2652 | ddf->currentdev = dk->raid_disk; | |
2653 | vc->phys_refnum[raid_disk] = dl->disk.refnum; | |
2654 | LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos); | |
2655 | ||
2656 | for (i = 0; i < ddf->max_part ; i++) | |
2657 | if (dl->vlist[i] == NULL) | |
2658 | break; | |
2659 | if (i == ddf->max_part) | |
2660 | return; | |
2661 | dl->vlist[i] = ddf->currentconf; | |
2662 | ||
2663 | if (fd >= 0) | |
2664 | dl->fd = fd; | |
2665 | if (devname) | |
2666 | dl->devname = devname; | |
2667 | ||
2668 | /* Check if we can mark array as optimal yet */ | |
2669 | i = ddf->currentconf->vcnum; | |
2670 | ddf->virt->entries[i].state = | |
2671 | (ddf->virt->entries[i].state & ~DDF_state_mask) | |
2672 | | get_svd_state(ddf, ddf->currentconf); | |
2673 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
2674 | cpu_to_be16(DDF_Global_Spare)); | |
2675 | be16_set(ddf->phys->entries[dl->pdnum].type, | |
2676 | cpu_to_be16(DDF_Active_in_VD)); | |
2677 | dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n", | |
2678 | __func__, dl->pdnum, be32_to_cpu(dl->disk.refnum), | |
2679 | ddf->currentconf->vcnum, guid_str(vc->guid), | |
2680 | dk->raid_disk); | |
2681 | ddf_set_updates_pending(ddf); | |
2682 | } | |
2683 | ||
2684 | static unsigned int find_unused_pde(const struct ddf_super *ddf) | |
2685 | { | |
2686 | unsigned int i; | |
2687 | for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) { | |
2688 | if (all_ff(ddf->phys->entries[i].guid)) | |
2689 | return i; | |
2690 | } | |
2691 | return DDF_NOTFOUND; | |
2692 | } | |
2693 | ||
2694 | /* add a device to a container, either while creating it or while | |
2695 | * expanding a pre-existing container | |
2696 | */ | |
2697 | static int add_to_super_ddf(struct supertype *st, | |
2698 | mdu_disk_info_t *dk, int fd, char *devname, | |
2699 | unsigned long long data_offset) | |
2700 | { | |
2701 | struct ddf_super *ddf = st->sb; | |
2702 | struct dl *dd; | |
2703 | time_t now; | |
2704 | struct tm *tm; | |
2705 | unsigned long long size; | |
2706 | struct phys_disk_entry *pde; | |
2707 | unsigned int n, i; | |
2708 | struct stat stb; | |
2709 | __u32 *tptr; | |
2710 | ||
2711 | if (ddf->currentconf) { | |
2712 | add_to_super_ddf_bvd(st, dk, fd, devname); | |
2713 | return 0; | |
2714 | } | |
2715 | ||
2716 | /* This is device numbered dk->number. We need to create | |
2717 | * a phys_disk entry and a more detailed disk_data entry. | |
2718 | */ | |
2719 | fstat(fd, &stb); | |
2720 | n = find_unused_pde(ddf); | |
2721 | if (n == DDF_NOTFOUND) { | |
2722 | pr_err("%s: No free slot in array, cannot add disk\n", | |
2723 | __func__); | |
2724 | return 1; | |
2725 | } | |
2726 | pde = &ddf->phys->entries[n]; | |
2727 | get_dev_size(fd, NULL, &size); | |
2728 | if (size <= 32*1024*1024) { | |
2729 | pr_err("%s: device size must be at least 32MB\n", | |
2730 | __func__); | |
2731 | return 1; | |
2732 | } | |
2733 | size >>= 9; | |
2734 | ||
2735 | if (posix_memalign((void**)&dd, 512, | |
2736 | sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) { | |
2737 | pr_err("%s could allocate buffer for new disk, aborting\n", | |
2738 | __func__); | |
2739 | return 1; | |
2740 | } | |
2741 | dd->major = major(stb.st_rdev); | |
2742 | dd->minor = minor(stb.st_rdev); | |
2743 | dd->devname = devname; | |
2744 | dd->fd = fd; | |
2745 | dd->spare = NULL; | |
2746 | ||
2747 | dd->disk.magic = DDF_PHYS_DATA_MAGIC; | |
2748 | now = time(0); | |
2749 | tm = localtime(&now); | |
2750 | sprintf(dd->disk.guid, "%8s%04d%02d%02d", | |
2751 | T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday); | |
2752 | tptr = (__u32 *)(dd->disk.guid + 16); | |
2753 | *tptr++ = random32(); | |
2754 | *tptr = random32(); | |
2755 | ||
2756 | do { | |
2757 | /* Cannot be bothered finding a CRC of some irrelevant details*/ | |
2758 | dd->disk.refnum._v32 = random32(); | |
2759 | for (i = be16_to_cpu(ddf->active->max_pd_entries); | |
2760 | i > 0; i--) | |
2761 | if (be32_eq(ddf->phys->entries[i-1].refnum, | |
2762 | dd->disk.refnum)) | |
2763 | break; | |
2764 | } while (i > 0); | |
2765 | ||
2766 | dd->disk.forced_ref = 1; | |
2767 | dd->disk.forced_guid = 1; | |
2768 | memset(dd->disk.vendor, ' ', 32); | |
2769 | memcpy(dd->disk.vendor, "Linux", 5); | |
2770 | memset(dd->disk.pad, 0xff, 442); | |
2771 | for (i = 0; i < ddf->max_part ; i++) | |
2772 | dd->vlist[i] = NULL; | |
2773 | ||
2774 | dd->pdnum = n; | |
2775 | ||
2776 | if (st->update_tail) { | |
2777 | int len = (sizeof(struct phys_disk) + | |
2778 | sizeof(struct phys_disk_entry)); | |
2779 | struct phys_disk *pd; | |
2780 | ||
2781 | pd = xmalloc(len); | |
2782 | pd->magic = DDF_PHYS_RECORDS_MAGIC; | |
2783 | pd->used_pdes = cpu_to_be16(n); | |
2784 | pde = &pd->entries[0]; | |
2785 | dd->mdupdate = pd; | |
2786 | } else | |
2787 | ddf->phys->used_pdes = cpu_to_be16( | |
2788 | 1 + be16_to_cpu(ddf->phys->used_pdes)); | |
2789 | ||
2790 | memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN); | |
2791 | pde->refnum = dd->disk.refnum; | |
2792 | pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare); | |
2793 | pde->state = cpu_to_be16(DDF_Online); | |
2794 | dd->size = size; | |
2795 | /* | |
2796 | * If there is already a device in dlist, try to reserve the same | |
2797 | * amount of workspace. Otherwise, use 32MB. | |
2798 | * We checked disk size above already. | |
2799 | */ | |
2800 | #define __calc_lba(new, old, lba, mb) do { \ | |
2801 | unsigned long long dif; \ | |
2802 | if ((old) != NULL) \ | |
2803 | dif = (old)->size - be64_to_cpu((old)->lba); \ | |
2804 | else \ | |
2805 | dif = (new)->size; \ | |
2806 | if ((new)->size > dif) \ | |
2807 | (new)->lba = cpu_to_be64((new)->size - dif); \ | |
2808 | else \ | |
2809 | (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \ | |
2810 | } while (0) | |
2811 | __calc_lba(dd, ddf->dlist, workspace_lba, 32); | |
2812 | __calc_lba(dd, ddf->dlist, primary_lba, 16); | |
2813 | __calc_lba(dd, ddf->dlist, secondary_lba, 32); | |
2814 | pde->config_size = dd->workspace_lba; | |
2815 | ||
2816 | sprintf(pde->path, "%17.17s","Information: nil") ; | |
2817 | memset(pde->pad, 0xff, 6); | |
2818 | ||
2819 | if (st->update_tail) { | |
2820 | dd->next = ddf->add_list; | |
2821 | ddf->add_list = dd; | |
2822 | } else { | |
2823 | dd->next = ddf->dlist; | |
2824 | ddf->dlist = dd; | |
2825 | ddf_set_updates_pending(ddf); | |
2826 | } | |
2827 | ||
2828 | return 0; | |
2829 | } | |
2830 | ||
2831 | static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk) | |
2832 | { | |
2833 | struct ddf_super *ddf = st->sb; | |
2834 | struct dl *dl; | |
2835 | ||
2836 | /* mdmon has noticed that this disk (dk->major/dk->minor) has | |
2837 | * disappeared from the container. | |
2838 | * We need to arrange that it disappears from the metadata and | |
2839 | * internal data structures too. | |
2840 | * Most of the work is done by ddf_process_update which edits | |
2841 | * the metadata and closes the file handle and attaches the memory | |
2842 | * where free_updates will free it. | |
2843 | */ | |
2844 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2845 | if (dl->major == dk->major && | |
2846 | dl->minor == dk->minor) | |
2847 | break; | |
2848 | if (!dl) | |
2849 | return -1; | |
2850 | ||
2851 | if (st->update_tail) { | |
2852 | int len = (sizeof(struct phys_disk) + | |
2853 | sizeof(struct phys_disk_entry)); | |
2854 | struct phys_disk *pd; | |
2855 | ||
2856 | pd = xmalloc(len); | |
2857 | pd->magic = DDF_PHYS_RECORDS_MAGIC; | |
2858 | pd->used_pdes = cpu_to_be16(dl->pdnum); | |
2859 | pd->entries[0].state = cpu_to_be16(DDF_Missing); | |
2860 | append_metadata_update(st, pd, len); | |
2861 | } | |
2862 | return 0; | |
2863 | } | |
2864 | #endif | |
2865 | ||
2866 | /* | |
2867 | * This is the write_init_super method for a ddf container. It is | |
2868 | * called when creating a container or adding another device to a | |
2869 | * container. | |
2870 | */ | |
2871 | #define NULL_CONF_SZ 4096 | |
2872 | ||
2873 | static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type) | |
2874 | { | |
2875 | unsigned long long sector; | |
2876 | struct ddf_header *header; | |
2877 | int fd, i, n_config, conf_size, buf_size; | |
2878 | int ret = 0; | |
2879 | char *conf; | |
2880 | ||
2881 | fd = d->fd; | |
2882 | ||
2883 | switch (type) { | |
2884 | case DDF_HEADER_PRIMARY: | |
2885 | header = &ddf->primary; | |
2886 | sector = be64_to_cpu(header->primary_lba); | |
2887 | break; | |
2888 | case DDF_HEADER_SECONDARY: | |
2889 | header = &ddf->secondary; | |
2890 | sector = be64_to_cpu(header->secondary_lba); | |
2891 | break; | |
2892 | default: | |
2893 | return 0; | |
2894 | } | |
2895 | ||
2896 | header->type = type; | |
2897 | header->openflag = 1; | |
2898 | header->crc = calc_crc(header, 512); | |
2899 | ||
2900 | lseek64(fd, sector<<9, 0); | |
2901 | if (write(fd, header, 512) < 0) | |
2902 | goto out; | |
2903 | ||
2904 | ddf->controller.crc = calc_crc(&ddf->controller, 512); | |
2905 | if (write(fd, &ddf->controller, 512) < 0) | |
2906 | goto out; | |
2907 | ||
2908 | ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize); | |
2909 | if (write(fd, ddf->phys, ddf->pdsize) < 0) | |
2910 | goto out; | |
2911 | ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize); | |
2912 | if (write(fd, ddf->virt, ddf->vdsize) < 0) | |
2913 | goto out; | |
2914 | ||
2915 | /* Now write lots of config records. */ | |
2916 | n_config = ddf->max_part; | |
2917 | conf_size = ddf->conf_rec_len * 512; | |
2918 | conf = ddf->conf; | |
2919 | buf_size = conf_size * (n_config + 1); | |
2920 | if (!conf) { | |
2921 | if (posix_memalign((void**)&conf, 512, buf_size) != 0) | |
2922 | goto out; | |
2923 | ddf->conf = conf; | |
2924 | } | |
2925 | for (i = 0 ; i <= n_config ; i++) { | |
2926 | struct vcl *c; | |
2927 | struct vd_config *vdc = NULL; | |
2928 | if (i == n_config) { | |
2929 | c = (struct vcl *)d->spare; | |
2930 | if (c) | |
2931 | vdc = &c->conf; | |
2932 | } else { | |
2933 | unsigned int dummy; | |
2934 | c = d->vlist[i]; | |
2935 | if (c) | |
2936 | get_pd_index_from_refnum( | |
2937 | c, d->disk.refnum, | |
2938 | ddf->mppe, | |
2939 | (const struct vd_config **)&vdc, | |
2940 | &dummy); | |
2941 | } | |
2942 | if (c) { | |
2943 | dprintf("writing conf record %i on disk %08x for %s/%u\n", | |
2944 | i, be32_to_cpu(d->disk.refnum), | |
2945 | guid_str(vdc->guid), | |
2946 | vdc->sec_elmnt_seq); | |
2947 | vdc->seqnum = header->seq; | |
2948 | vdc->crc = calc_crc(vdc, conf_size); | |
2949 | memcpy(conf + i*conf_size, vdc, conf_size); | |
2950 | } else | |
2951 | memset(conf + i*conf_size, 0xff, conf_size); | |
2952 | } | |
2953 | if (write(fd, conf, buf_size) != buf_size) | |
2954 | goto out; | |
2955 | ||
2956 | d->disk.crc = calc_crc(&d->disk, 512); | |
2957 | if (write(fd, &d->disk, 512) < 0) | |
2958 | goto out; | |
2959 | ||
2960 | ret = 1; | |
2961 | out: | |
2962 | header->openflag = 0; | |
2963 | header->crc = calc_crc(header, 512); | |
2964 | ||
2965 | lseek64(fd, sector<<9, 0); | |
2966 | if (write(fd, header, 512) < 0) | |
2967 | ret = 0; | |
2968 | ||
2969 | return ret; | |
2970 | } | |
2971 | ||
2972 | static int _write_super_to_disk(struct ddf_super *ddf, struct dl *d) | |
2973 | { | |
2974 | unsigned long long size; | |
2975 | int fd = d->fd; | |
2976 | if (fd < 0) | |
2977 | return 0; | |
2978 | ||
2979 | /* We need to fill in the primary, (secondary) and workspace | |
2980 | * lba's in the headers, set their checksums, | |
2981 | * Also checksum phys, virt.... | |
2982 | * | |
2983 | * Then write everything out, finally the anchor is written. | |
2984 | */ | |
2985 | get_dev_size(fd, NULL, &size); | |
2986 | size /= 512; | |
2987 | if (be64_to_cpu(d->workspace_lba) != 0ULL) | |
2988 | ddf->anchor.workspace_lba = d->workspace_lba; | |
2989 | else | |
2990 | ddf->anchor.workspace_lba = | |
2991 | cpu_to_be64(size - 32*1024*2); | |
2992 | if (be64_to_cpu(d->primary_lba) != 0ULL) | |
2993 | ddf->anchor.primary_lba = d->primary_lba; | |
2994 | else | |
2995 | ddf->anchor.primary_lba = | |
2996 | cpu_to_be64(size - 16*1024*2); | |
2997 | if (be64_to_cpu(d->secondary_lba) != 0ULL) | |
2998 | ddf->anchor.secondary_lba = d->secondary_lba; | |
2999 | else | |
3000 | ddf->anchor.secondary_lba = | |
3001 | cpu_to_be64(size - 32*1024*2); | |
3002 | ddf->anchor.seq = ddf->active->seq; | |
3003 | memcpy(&ddf->primary, &ddf->anchor, 512); | |
3004 | memcpy(&ddf->secondary, &ddf->anchor, 512); | |
3005 | ||
3006 | ddf->anchor.openflag = 0xFF; /* 'open' means nothing */ | |
3007 | ddf->anchor.seq = cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */ | |
3008 | ddf->anchor.crc = calc_crc(&ddf->anchor, 512); | |
3009 | ||
3010 | if (!__write_ddf_structure(d, ddf, DDF_HEADER_PRIMARY)) | |
3011 | return 0; | |
3012 | ||
3013 | if (!__write_ddf_structure(d, ddf, DDF_HEADER_SECONDARY)) | |
3014 | return 0; | |
3015 | ||
3016 | lseek64(fd, (size-1)*512, SEEK_SET); | |
3017 | if (write(fd, &ddf->anchor, 512) < 0) | |
3018 | return 0; | |
3019 | ||
3020 | return 1; | |
3021 | } | |
3022 | ||
3023 | #ifndef MDASSEMBLE | |
3024 | static int __write_init_super_ddf(struct supertype *st) | |
3025 | { | |
3026 | struct ddf_super *ddf = st->sb; | |
3027 | struct dl *d; | |
3028 | int attempts = 0; | |
3029 | int successes = 0; | |
3030 | ||
3031 | pr_state(ddf, __func__); | |
3032 | ||
3033 | /* try to write updated metadata, | |
3034 | * if we catch a failure move on to the next disk | |
3035 | */ | |
3036 | for (d = ddf->dlist; d; d=d->next) { | |
3037 | attempts++; | |
3038 | successes += _write_super_to_disk(ddf, d); | |
3039 | } | |
3040 | ||
3041 | return attempts != successes; | |
3042 | } | |
3043 | ||
3044 | static int write_init_super_ddf(struct supertype *st) | |
3045 | { | |
3046 | struct ddf_super *ddf = st->sb; | |
3047 | struct vcl *currentconf = ddf->currentconf; | |
3048 | ||
3049 | /* we are done with currentconf reset it to point st at the container */ | |
3050 | ddf->currentconf = NULL; | |
3051 | ||
3052 | if (st->update_tail) { | |
3053 | /* queue the virtual_disk and vd_config as metadata updates */ | |
3054 | struct virtual_disk *vd; | |
3055 | struct vd_config *vc; | |
3056 | int len, tlen; | |
3057 | unsigned int i; | |
3058 | ||
3059 | if (!currentconf) { | |
3060 | int len = (sizeof(struct phys_disk) + | |
3061 | sizeof(struct phys_disk_entry)); | |
3062 | ||
3063 | /* adding a disk to the container. */ | |
3064 | if (!ddf->add_list) | |
3065 | return 0; | |
3066 | ||
3067 | append_metadata_update(st, ddf->add_list->mdupdate, len); | |
3068 | ddf->add_list->mdupdate = NULL; | |
3069 | return 0; | |
3070 | } | |
3071 | ||
3072 | /* Newly created VD */ | |
3073 | ||
3074 | /* First the virtual disk. We have a slightly fake header */ | |
3075 | len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry); | |
3076 | vd = xmalloc(len); | |
3077 | *vd = *ddf->virt; | |
3078 | vd->entries[0] = ddf->virt->entries[currentconf->vcnum]; | |
3079 | vd->populated_vdes = cpu_to_be16(currentconf->vcnum); | |
3080 | append_metadata_update(st, vd, len); | |
3081 | ||
3082 | /* Then the vd_config */ | |
3083 | len = ddf->conf_rec_len * 512; | |
3084 | tlen = len * currentconf->conf.sec_elmnt_count; | |
3085 | vc = xmalloc(tlen); | |
3086 | memcpy(vc, ¤tconf->conf, len); | |
3087 | for (i = 1; i < currentconf->conf.sec_elmnt_count; i++) | |
3088 | memcpy((char *)vc + i*len, currentconf->other_bvds[i-1], | |
3089 | len); | |
3090 | append_metadata_update(st, vc, tlen); | |
3091 | ||
3092 | /* FIXME I need to close the fds! */ | |
3093 | return 0; | |
3094 | } else { | |
3095 | struct dl *d; | |
3096 | if (!currentconf) | |
3097 | for (d = ddf->dlist; d; d=d->next) | |
3098 | while (Kill(d->devname, NULL, 0, -1, 1) == 0); | |
3099 | return __write_init_super_ddf(st); | |
3100 | } | |
3101 | } | |
3102 | ||
3103 | #endif | |
3104 | ||
3105 | static __u64 avail_size_ddf(struct supertype *st, __u64 devsize, | |
3106 | unsigned long long data_offset) | |
3107 | { | |
3108 | /* We must reserve the last 32Meg */ | |
3109 | if (devsize <= 32*1024*2) | |
3110 | return 0; | |
3111 | return devsize - 32*1024*2; | |
3112 | } | |
3113 | ||
3114 | #ifndef MDASSEMBLE | |
3115 | ||
3116 | static int reserve_space(struct supertype *st, int raiddisks, | |
3117 | unsigned long long size, int chunk, | |
3118 | unsigned long long *freesize) | |
3119 | { | |
3120 | /* Find 'raiddisks' spare extents at least 'size' big (but | |
3121 | * only caring about multiples of 'chunk') and remember | |
3122 | * them. | |
3123 | * If the cannot be found, fail. | |
3124 | */ | |
3125 | struct dl *dl; | |
3126 | struct ddf_super *ddf = st->sb; | |
3127 | int cnt = 0; | |
3128 | ||
3129 | for (dl = ddf->dlist; dl ; dl=dl->next) { | |
3130 | dl->raiddisk = -1; | |
3131 | dl->esize = 0; | |
3132 | } | |
3133 | /* Now find largest extent on each device */ | |
3134 | for (dl = ddf->dlist ; dl ; dl=dl->next) { | |
3135 | struct extent *e = get_extents(ddf, dl); | |
3136 | unsigned long long pos = 0; | |
3137 | int i = 0; | |
3138 | int found = 0; | |
3139 | unsigned long long minsize = size; | |
3140 | ||
3141 | if (size == 0) | |
3142 | minsize = chunk; | |
3143 | ||
3144 | if (!e) | |
3145 | continue; | |
3146 | do { | |
3147 | unsigned long long esize; | |
3148 | esize = e[i].start - pos; | |
3149 | if (esize >= minsize) { | |
3150 | found = 1; | |
3151 | minsize = esize; | |
3152 | } | |
3153 | pos = e[i].start + e[i].size; | |
3154 | i++; | |
3155 | } while (e[i-1].size); | |
3156 | if (found) { | |
3157 | cnt++; | |
3158 | dl->esize = minsize; | |
3159 | } | |
3160 | free(e); | |
3161 | } | |
3162 | if (cnt < raiddisks) { | |
3163 | pr_err("not enough devices with space to create array.\n"); | |
3164 | return 0; /* No enough free spaces large enough */ | |
3165 | } | |
3166 | if (size == 0) { | |
3167 | /* choose the largest size of which there are at least 'raiddisk' */ | |
3168 | for (dl = ddf->dlist ; dl ; dl=dl->next) { | |
3169 | struct dl *dl2; | |
3170 | if (dl->esize <= size) | |
3171 | continue; | |
3172 | /* This is bigger than 'size', see if there are enough */ | |
3173 | cnt = 0; | |
3174 | for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next) | |
3175 | if (dl2->esize >= dl->esize) | |
3176 | cnt++; | |
3177 | if (cnt >= raiddisks) | |
3178 | size = dl->esize; | |
3179 | } | |
3180 | if (chunk) { | |
3181 | size = size / chunk; | |
3182 | size *= chunk; | |
3183 | } | |
3184 | *freesize = size; | |
3185 | if (size < 32) { | |
3186 | pr_err("not enough spare devices to create array.\n"); | |
3187 | return 0; | |
3188 | } | |
3189 | } | |
3190 | /* We have a 'size' of which there are enough spaces. | |
3191 | * We simply do a first-fit */ | |
3192 | cnt = 0; | |
3193 | for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) { | |
3194 | if (dl->esize < size) | |
3195 | continue; | |
3196 | ||
3197 | dl->raiddisk = cnt; | |
3198 | cnt++; | |
3199 | } | |
3200 | return 1; | |
3201 | } | |
3202 | ||
3203 | static int | |
3204 | validate_geometry_ddf_container(struct supertype *st, | |
3205 | int level, int layout, int raiddisks, | |
3206 | int chunk, unsigned long long size, | |
3207 | unsigned long long data_offset, | |
3208 | char *dev, unsigned long long *freesize, | |
3209 | int verbose); | |
3210 | ||
3211 | static int validate_geometry_ddf_bvd(struct supertype *st, | |
3212 | int level, int layout, int raiddisks, | |
3213 | int *chunk, unsigned long long size, | |
3214 | unsigned long long data_offset, | |
3215 | char *dev, unsigned long long *freesize, | |
3216 | int verbose); | |
3217 | ||
3218 | static int validate_geometry_ddf(struct supertype *st, | |
3219 | int level, int layout, int raiddisks, | |
3220 | int *chunk, unsigned long long size, | |
3221 | unsigned long long data_offset, | |
3222 | char *dev, unsigned long long *freesize, | |
3223 | int verbose) | |
3224 | { | |
3225 | int fd; | |
3226 | struct mdinfo *sra; | |
3227 | int cfd; | |
3228 | ||
3229 | /* ddf potentially supports lots of things, but it depends on | |
3230 | * what devices are offered (and maybe kernel version?) | |
3231 | * If given unused devices, we will make a container. | |
3232 | * If given devices in a container, we will make a BVD. | |
3233 | * If given BVDs, we make an SVD, changing all the GUIDs in the process. | |
3234 | */ | |
3235 | ||
3236 | if (*chunk == UnSet) | |
3237 | *chunk = DEFAULT_CHUNK; | |
3238 | ||
3239 | if (level == -1000000) level = LEVEL_CONTAINER; | |
3240 | if (level == LEVEL_CONTAINER) { | |
3241 | /* Must be a fresh device to add to a container */ | |
3242 | return validate_geometry_ddf_container(st, level, layout, | |
3243 | raiddisks, *chunk, | |
3244 | size, data_offset, dev, | |
3245 | freesize, | |
3246 | verbose); | |
3247 | } | |
3248 | ||
3249 | if (!dev) { | |
3250 | mdu_array_info_t array = { | |
3251 | .level = level, .layout = layout, | |
3252 | .raid_disks = raiddisks | |
3253 | }; | |
3254 | struct vd_config conf; | |
3255 | if (layout_md2ddf(&array, &conf) == -1) { | |
3256 | if (verbose) | |
3257 | pr_err("DDF does not support level %d /layout %d arrays with %d disks\n", | |
3258 | level, layout, raiddisks); | |
3259 | return 0; | |
3260 | } | |
3261 | /* Should check layout? etc */ | |
3262 | ||
3263 | if (st->sb && freesize) { | |
3264 | /* --create was given a container to create in. | |
3265 | * So we need to check that there are enough | |
3266 | * free spaces and return the amount of space. | |
3267 | * We may as well remember which drives were | |
3268 | * chosen so that add_to_super/getinfo_super | |
3269 | * can return them. | |
3270 | */ | |
3271 | return reserve_space(st, raiddisks, size, *chunk, freesize); | |
3272 | } | |
3273 | return 1; | |
3274 | } | |
3275 | ||
3276 | if (st->sb) { | |
3277 | /* A container has already been opened, so we are | |
3278 | * creating in there. Maybe a BVD, maybe an SVD. | |
3279 | * Should make a distinction one day. | |
3280 | */ | |
3281 | return validate_geometry_ddf_bvd(st, level, layout, raiddisks, | |
3282 | chunk, size, data_offset, dev, | |
3283 | freesize, | |
3284 | verbose); | |
3285 | } | |
3286 | /* This is the first device for the array. | |
3287 | * If it is a container, we read it in and do automagic allocations, | |
3288 | * no other devices should be given. | |
3289 | * Otherwise it must be a member device of a container, and we | |
3290 | * do manual allocation. | |
3291 | * Later we should check for a BVD and make an SVD. | |
3292 | */ | |
3293 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
3294 | if (fd >= 0) { | |
3295 | sra = sysfs_read(fd, NULL, GET_VERSION); | |
3296 | close(fd); | |
3297 | if (sra && sra->array.major_version == -1 && | |
3298 | strcmp(sra->text_version, "ddf") == 0) { | |
3299 | ||
3300 | /* load super */ | |
3301 | /* find space for 'n' devices. */ | |
3302 | /* remember the devices */ | |
3303 | /* Somehow return the fact that we have enough */ | |
3304 | } | |
3305 | ||
3306 | if (verbose) | |
3307 | pr_err("ddf: Cannot create this array " | |
3308 | "on device %s - a container is required.\n", | |
3309 | dev); | |
3310 | return 0; | |
3311 | } | |
3312 | if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) { | |
3313 | if (verbose) | |
3314 | pr_err("ddf: Cannot open %s: %s\n", | |
3315 | dev, strerror(errno)); | |
3316 | return 0; | |
3317 | } | |
3318 | /* Well, it is in use by someone, maybe a 'ddf' container. */ | |
3319 | cfd = open_container(fd); | |
3320 | if (cfd < 0) { | |
3321 | close(fd); | |
3322 | if (verbose) | |
3323 | pr_err("ddf: Cannot use %s: %s\n", | |
3324 | dev, strerror(EBUSY)); | |
3325 | return 0; | |
3326 | } | |
3327 | sra = sysfs_read(cfd, NULL, GET_VERSION); | |
3328 | close(fd); | |
3329 | if (sra && sra->array.major_version == -1 && | |
3330 | strcmp(sra->text_version, "ddf") == 0) { | |
3331 | /* This is a member of a ddf container. Load the container | |
3332 | * and try to create a bvd | |
3333 | */ | |
3334 | struct ddf_super *ddf; | |
3335 | if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) { | |
3336 | st->sb = ddf; | |
3337 | strcpy(st->container_devnm, fd2devnm(cfd)); | |
3338 | close(cfd); | |
3339 | return validate_geometry_ddf_bvd(st, level, layout, | |
3340 | raiddisks, chunk, size, | |
3341 | data_offset, | |
3342 | dev, freesize, | |
3343 | verbose); | |
3344 | } | |
3345 | close(cfd); | |
3346 | } else /* device may belong to a different container */ | |
3347 | return 0; | |
3348 | ||
3349 | return 1; | |
3350 | } | |
3351 | ||
3352 | static int | |
3353 | validate_geometry_ddf_container(struct supertype *st, | |
3354 | int level, int layout, int raiddisks, | |
3355 | int chunk, unsigned long long size, | |
3356 | unsigned long long data_offset, | |
3357 | char *dev, unsigned long long *freesize, | |
3358 | int verbose) | |
3359 | { | |
3360 | int fd; | |
3361 | unsigned long long ldsize; | |
3362 | ||
3363 | if (level != LEVEL_CONTAINER) | |
3364 | return 0; | |
3365 | if (!dev) | |
3366 | return 1; | |
3367 | ||
3368 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
3369 | if (fd < 0) { | |
3370 | if (verbose) | |
3371 | pr_err("ddf: Cannot open %s: %s\n", | |
3372 | dev, strerror(errno)); | |
3373 | return 0; | |
3374 | } | |
3375 | if (!get_dev_size(fd, dev, &ldsize)) { | |
3376 | close(fd); | |
3377 | return 0; | |
3378 | } | |
3379 | close(fd); | |
3380 | ||
3381 | *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS); | |
3382 | if (*freesize == 0) | |
3383 | return 0; | |
3384 | ||
3385 | return 1; | |
3386 | } | |
3387 | ||
3388 | static int validate_geometry_ddf_bvd(struct supertype *st, | |
3389 | int level, int layout, int raiddisks, | |
3390 | int *chunk, unsigned long long size, | |
3391 | unsigned long long data_offset, | |
3392 | char *dev, unsigned long long *freesize, | |
3393 | int verbose) | |
3394 | { | |
3395 | struct stat stb; | |
3396 | struct ddf_super *ddf = st->sb; | |
3397 | struct dl *dl; | |
3398 | unsigned long long pos = 0; | |
3399 | unsigned long long maxsize; | |
3400 | struct extent *e; | |
3401 | int i; | |
3402 | /* ddf/bvd supports lots of things, but not containers */ | |
3403 | if (level == LEVEL_CONTAINER) { | |
3404 | if (verbose) | |
3405 | pr_err("DDF cannot create a container within an container\n"); | |
3406 | return 0; | |
3407 | } | |
3408 | /* We must have the container info already read in. */ | |
3409 | if (!ddf) | |
3410 | return 0; | |
3411 | ||
3412 | if (!dev) { | |
3413 | /* General test: make sure there is space for | |
3414 | * 'raiddisks' device extents of size 'size'. | |
3415 | */ | |
3416 | unsigned long long minsize = size; | |
3417 | int dcnt = 0; | |
3418 | if (minsize == 0) | |
3419 | minsize = 8; | |
3420 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
3421 | { | |
3422 | int found = 0; | |
3423 | pos = 0; | |
3424 | ||
3425 | i = 0; | |
3426 | e = get_extents(ddf, dl); | |
3427 | if (!e) continue; | |
3428 | do { | |
3429 | unsigned long long esize; | |
3430 | esize = e[i].start - pos; | |
3431 | if (esize >= minsize) | |
3432 | found = 1; | |
3433 | pos = e[i].start + e[i].size; | |
3434 | i++; | |
3435 | } while (e[i-1].size); | |
3436 | if (found) | |
3437 | dcnt++; | |
3438 | free(e); | |
3439 | } | |
3440 | if (dcnt < raiddisks) { | |
3441 | if (verbose) | |
3442 | pr_err("ddf: Not enough devices with " | |
3443 | "space for this array (%d < %d)\n", | |
3444 | dcnt, raiddisks); | |
3445 | return 0; | |
3446 | } | |
3447 | return 1; | |
3448 | } | |
3449 | /* This device must be a member of the set */ | |
3450 | if (stat(dev, &stb) < 0) | |
3451 | return 0; | |
3452 | if ((S_IFMT & stb.st_mode) != S_IFBLK) | |
3453 | return 0; | |
3454 | for (dl = ddf->dlist ; dl ; dl = dl->next) { | |
3455 | if (dl->major == (int)major(stb.st_rdev) && | |
3456 | dl->minor == (int)minor(stb.st_rdev)) | |
3457 | break; | |
3458 | } | |
3459 | if (!dl) { | |
3460 | if (verbose) | |
3461 | pr_err("ddf: %s is not in the " | |
3462 | "same DDF set\n", | |
3463 | dev); | |
3464 | return 0; | |
3465 | } | |
3466 | e = get_extents(ddf, dl); | |
3467 | maxsize = 0; | |
3468 | i = 0; | |
3469 | if (e) do { | |
3470 | unsigned long long esize; | |
3471 | esize = e[i].start - pos; | |
3472 | if (esize >= maxsize) | |
3473 | maxsize = esize; | |
3474 | pos = e[i].start + e[i].size; | |
3475 | i++; | |
3476 | } while (e[i-1].size); | |
3477 | *freesize = maxsize; | |
3478 | // FIXME here I am | |
3479 | ||
3480 | return 1; | |
3481 | } | |
3482 | ||
3483 | static int load_super_ddf_all(struct supertype *st, int fd, | |
3484 | void **sbp, char *devname) | |
3485 | { | |
3486 | struct mdinfo *sra; | |
3487 | struct ddf_super *super; | |
3488 | struct mdinfo *sd, *best = NULL; | |
3489 | int bestseq = 0; | |
3490 | int seq; | |
3491 | char nm[20]; | |
3492 | int dfd; | |
3493 | ||
3494 | sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE); | |
3495 | if (!sra) | |
3496 | return 1; | |
3497 | if (sra->array.major_version != -1 || | |
3498 | sra->array.minor_version != -2 || | |
3499 | strcmp(sra->text_version, "ddf") != 0) | |
3500 | return 1; | |
3501 | ||
3502 | if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0) | |
3503 | return 1; | |
3504 | memset(super, 0, sizeof(*super)); | |
3505 | ||
3506 | /* first, try each device, and choose the best ddf */ | |
3507 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
3508 | int rv; | |
3509 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
3510 | dfd = dev_open(nm, O_RDONLY); | |
3511 | if (dfd < 0) | |
3512 | return 2; | |
3513 | rv = load_ddf_headers(dfd, super, NULL); | |
3514 | close(dfd); | |
3515 | if (rv == 0) { | |
3516 | seq = be32_to_cpu(super->active->seq); | |
3517 | if (super->active->openflag) | |
3518 | seq--; | |
3519 | if (!best || seq > bestseq) { | |
3520 | bestseq = seq; | |
3521 | best = sd; | |
3522 | } | |
3523 | } | |
3524 | } | |
3525 | if (!best) | |
3526 | return 1; | |
3527 | /* OK, load this ddf */ | |
3528 | sprintf(nm, "%d:%d", best->disk.major, best->disk.minor); | |
3529 | dfd = dev_open(nm, O_RDONLY); | |
3530 | if (dfd < 0) | |
3531 | return 1; | |
3532 | load_ddf_headers(dfd, super, NULL); | |
3533 | load_ddf_global(dfd, super, NULL); | |
3534 | close(dfd); | |
3535 | /* Now we need the device-local bits */ | |
3536 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
3537 | int rv; | |
3538 | ||
3539 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
3540 | dfd = dev_open(nm, O_RDWR); | |
3541 | if (dfd < 0) | |
3542 | return 2; | |
3543 | rv = load_ddf_headers(dfd, super, NULL); | |
3544 | if (rv == 0) | |
3545 | rv = load_ddf_local(dfd, super, NULL, 1); | |
3546 | if (rv) | |
3547 | return 1; | |
3548 | } | |
3549 | ||
3550 | *sbp = super; | |
3551 | if (st->ss == NULL) { | |
3552 | st->ss = &super_ddf; | |
3553 | st->minor_version = 0; | |
3554 | st->max_devs = 512; | |
3555 | } | |
3556 | strcpy(st->container_devnm, fd2devnm(fd)); | |
3557 | return 0; | |
3558 | } | |
3559 | ||
3560 | static int load_container_ddf(struct supertype *st, int fd, | |
3561 | char *devname) | |
3562 | { | |
3563 | return load_super_ddf_all(st, fd, &st->sb, devname); | |
3564 | } | |
3565 | ||
3566 | #endif /* MDASSEMBLE */ | |
3567 | ||
3568 | static int check_secondary(const struct vcl *vc) | |
3569 | { | |
3570 | const struct vd_config *conf = &vc->conf; | |
3571 | int i; | |
3572 | ||
3573 | /* The only DDF secondary RAID level md can support is | |
3574 | * RAID 10, if the stripe sizes and Basic volume sizes | |
3575 | * are all equal. | |
3576 | * Other configurations could in theory be supported by exposing | |
3577 | * the BVDs to user space and using device mapper for the secondary | |
3578 | * mapping. So far we don't support that. | |
3579 | */ | |
3580 | ||
3581 | __u64 sec_elements[4] = {0, 0, 0, 0}; | |
3582 | #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63))) | |
3583 | #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0) | |
3584 | ||
3585 | if (vc->other_bvds == NULL) { | |
3586 | pr_err("No BVDs for secondary RAID found\n"); | |
3587 | return -1; | |
3588 | } | |
3589 | if (conf->prl != DDF_RAID1) { | |
3590 | pr_err("Secondary RAID level only supported for mirrored BVD\n"); | |
3591 | return -1; | |
3592 | } | |
3593 | if (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED) { | |
3594 | pr_err("Secondary RAID level %d is unsupported\n", | |
3595 | conf->srl); | |
3596 | return -1; | |
3597 | } | |
3598 | __set_sec_seen(conf->sec_elmnt_seq); | |
3599 | for (i = 0; i < conf->sec_elmnt_count-1; i++) { | |
3600 | const struct vd_config *bvd = vc->other_bvds[i]; | |
3601 | if (bvd->sec_elmnt_seq == DDF_UNUSED_BVD) | |
3602 | continue; | |
3603 | if (bvd->srl != conf->srl) { | |
3604 | pr_err("Inconsistent secondary RAID level across BVDs\n"); | |
3605 | return -1; | |
3606 | } | |
3607 | if (bvd->prl != conf->prl) { | |
3608 | pr_err("Different RAID levels for BVDs are unsupported\n"); | |
3609 | return -1; | |
3610 | } | |
3611 | if (!be16_eq(bvd->prim_elmnt_count, conf->prim_elmnt_count)) { | |
3612 | pr_err("All BVDs must have the same number of primary elements\n"); | |
3613 | return -1; | |
3614 | } | |
3615 | if (bvd->chunk_shift != conf->chunk_shift) { | |
3616 | pr_err("Different strip sizes for BVDs are unsupported\n"); | |
3617 | return -1; | |
3618 | } | |
3619 | if (!be64_eq(bvd->array_blocks, conf->array_blocks)) { | |
3620 | pr_err("Different BVD sizes are unsupported\n"); | |
3621 | return -1; | |
3622 | } | |
3623 | __set_sec_seen(bvd->sec_elmnt_seq); | |
3624 | } | |
3625 | for (i = 0; i < conf->sec_elmnt_count; i++) { | |
3626 | if (!__was_sec_seen(i)) { | |
3627 | pr_err("BVD %d is missing\n", i); | |
3628 | return -1; | |
3629 | } | |
3630 | } | |
3631 | return 0; | |
3632 | } | |
3633 | ||
3634 | static unsigned int get_pd_index_from_refnum(const struct vcl *vc, | |
3635 | be32 refnum, unsigned int nmax, | |
3636 | const struct vd_config **bvd, | |
3637 | unsigned int *idx) | |
3638 | { | |
3639 | unsigned int i, j, n, sec, cnt; | |
3640 | ||
3641 | cnt = be16_to_cpu(vc->conf.prim_elmnt_count); | |
3642 | sec = (vc->conf.sec_elmnt_count == 1 ? 0 : vc->conf.sec_elmnt_seq); | |
3643 | ||
3644 | for (i = 0, j = 0 ; i < nmax ; i++) { | |
3645 | /* j counts valid entries for this BVD */ | |
3646 | if (be32_to_cpu(vc->conf.phys_refnum[i]) != 0xffffffff) | |
3647 | j++; | |
3648 | if (be32_eq(vc->conf.phys_refnum[i], refnum)) { | |
3649 | *bvd = &vc->conf; | |
3650 | *idx = i; | |
3651 | return sec * cnt + j - 1; | |
3652 | } | |
3653 | } | |
3654 | if (vc->other_bvds == NULL) | |
3655 | goto bad; | |
3656 | ||
3657 | for (n = 1; n < vc->conf.sec_elmnt_count; n++) { | |
3658 | struct vd_config *vd = vc->other_bvds[n-1]; | |
3659 | sec = vd->sec_elmnt_seq; | |
3660 | if (sec == DDF_UNUSED_BVD) | |
3661 | continue; | |
3662 | for (i = 0, j = 0 ; i < nmax ; i++) { | |
3663 | if (be32_to_cpu(vd->phys_refnum[i]) != 0xffffffff) | |
3664 | j++; | |
3665 | if (be32_eq(vd->phys_refnum[i], refnum)) { | |
3666 | *bvd = vd; | |
3667 | *idx = i; | |
3668 | return sec * cnt + j - 1; | |
3669 | } | |
3670 | } | |
3671 | } | |
3672 | bad: | |
3673 | *bvd = NULL; | |
3674 | return DDF_NOTFOUND; | |
3675 | } | |
3676 | ||
3677 | static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray) | |
3678 | { | |
3679 | /* Given a container loaded by load_super_ddf_all, | |
3680 | * extract information about all the arrays into | |
3681 | * an mdinfo tree. | |
3682 | * | |
3683 | * For each vcl in conflist: create an mdinfo, fill it in, | |
3684 | * then look for matching devices (phys_refnum) in dlist | |
3685 | * and create appropriate device mdinfo. | |
3686 | */ | |
3687 | struct ddf_super *ddf = st->sb; | |
3688 | struct mdinfo *rest = NULL; | |
3689 | struct vcl *vc; | |
3690 | ||
3691 | for (vc = ddf->conflist ; vc ; vc=vc->next) | |
3692 | { | |
3693 | unsigned int i; | |
3694 | unsigned int j; | |
3695 | struct mdinfo *this; | |
3696 | char *ep; | |
3697 | __u32 *cptr; | |
3698 | unsigned int pd; | |
3699 | ||
3700 | if (subarray && | |
3701 | (strtoul(subarray, &ep, 10) != vc->vcnum || | |
3702 | *ep != '\0')) | |
3703 | continue; | |
3704 | ||
3705 | if (vc->conf.sec_elmnt_count > 1) { | |
3706 | if (check_secondary(vc) != 0) | |
3707 | continue; | |
3708 | } | |
3709 | ||
3710 | this = xcalloc(1, sizeof(*this)); | |
3711 | this->next = rest; | |
3712 | rest = this; | |
3713 | ||
3714 | if (layout_ddf2md(&vc->conf, &this->array)) | |
3715 | continue; | |
3716 | this->array.md_minor = -1; | |
3717 | this->array.major_version = -1; | |
3718 | this->array.minor_version = -2; | |
3719 | this->safe_mode_delay = DDF_SAFE_MODE_DELAY; | |
3720 | cptr = (__u32 *)(vc->conf.guid + 16); | |
3721 | this->array.ctime = DECADE + __be32_to_cpu(*cptr); | |
3722 | this->array.utime = DECADE + | |
3723 | be32_to_cpu(vc->conf.timestamp); | |
3724 | this->array.chunk_size = 512 << vc->conf.chunk_shift; | |
3725 | ||
3726 | i = vc->vcnum; | |
3727 | if ((ddf->virt->entries[i].state & DDF_state_inconsistent) || | |
3728 | (ddf->virt->entries[i].init_state & DDF_initstate_mask) != | |
3729 | DDF_init_full) { | |
3730 | this->array.state = 0; | |
3731 | this->resync_start = 0; | |
3732 | } else { | |
3733 | this->array.state = 1; | |
3734 | this->resync_start = MaxSector; | |
3735 | } | |
3736 | memcpy(this->name, ddf->virt->entries[i].name, 16); | |
3737 | this->name[16]=0; | |
3738 | for(j=0; j<16; j++) | |
3739 | if (this->name[j] == ' ') | |
3740 | this->name[j] = 0; | |
3741 | ||
3742 | memset(this->uuid, 0, sizeof(this->uuid)); | |
3743 | this->component_size = be64_to_cpu(vc->conf.blocks); | |
3744 | this->array.size = this->component_size / 2; | |
3745 | this->container_member = i; | |
3746 | ||
3747 | ddf->currentconf = vc; | |
3748 | uuid_from_super_ddf(st, this->uuid); | |
3749 | if (!subarray) | |
3750 | ddf->currentconf = NULL; | |
3751 | ||
3752 | sprintf(this->text_version, "/%s/%d", | |
3753 | st->container_devnm, this->container_member); | |
3754 | ||
3755 | for (pd = 0; pd < be16_to_cpu(ddf->phys->used_pdes); pd++) { | |
3756 | struct mdinfo *dev; | |
3757 | struct dl *d; | |
3758 | const struct vd_config *bvd; | |
3759 | unsigned int iphys; | |
3760 | int stt; | |
3761 | ||
3762 | if (be32_to_cpu(ddf->phys->entries[pd].refnum) | |
3763 | == 0xFFFFFFFF) | |
3764 | continue; | |
3765 | ||
3766 | stt = be16_to_cpu(ddf->phys->entries[pd].state); | |
3767 | if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding)) | |
3768 | != DDF_Online) | |
3769 | continue; | |
3770 | ||
3771 | i = get_pd_index_from_refnum( | |
3772 | vc, ddf->phys->entries[pd].refnum, | |
3773 | ddf->mppe, &bvd, &iphys); | |
3774 | if (i == DDF_NOTFOUND) | |
3775 | continue; | |
3776 | ||
3777 | this->array.working_disks++; | |
3778 | ||
3779 | for (d = ddf->dlist; d ; d=d->next) | |
3780 | if (be32_eq(d->disk.refnum, | |
3781 | ddf->phys->entries[pd].refnum)) | |
3782 | break; | |
3783 | if (d == NULL) | |
3784 | /* Haven't found that one yet, maybe there are others */ | |
3785 | continue; | |
3786 | ||
3787 | dev = xcalloc(1, sizeof(*dev)); | |
3788 | dev->next = this->devs; | |
3789 | this->devs = dev; | |
3790 | ||
3791 | dev->disk.number = be32_to_cpu(d->disk.refnum); | |
3792 | dev->disk.major = d->major; | |
3793 | dev->disk.minor = d->minor; | |
3794 | dev->disk.raid_disk = i; | |
3795 | dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE); | |
3796 | dev->recovery_start = MaxSector; | |
3797 | ||
3798 | dev->events = be32_to_cpu(ddf->primary.seq); | |
3799 | dev->data_offset = | |
3800 | be64_to_cpu(LBA_OFFSET(ddf, bvd)[iphys]); | |
3801 | dev->component_size = be64_to_cpu(bvd->blocks); | |
3802 | if (d->devname) | |
3803 | strcpy(dev->name, d->devname); | |
3804 | } | |
3805 | } | |
3806 | return rest; | |
3807 | } | |
3808 | ||
3809 | static int store_super_ddf(struct supertype *st, int fd) | |
3810 | { | |
3811 | struct ddf_super *ddf = st->sb; | |
3812 | unsigned long long dsize; | |
3813 | void *buf; | |
3814 | int rc; | |
3815 | ||
3816 | if (!ddf) | |
3817 | return 1; | |
3818 | ||
3819 | if (!get_dev_size(fd, NULL, &dsize)) | |
3820 | return 1; | |
3821 | ||
3822 | if (ddf->dlist || ddf->conflist) { | |
3823 | struct stat sta; | |
3824 | struct dl *dl; | |
3825 | int ofd, ret; | |
3826 | ||
3827 | if (fstat(fd, &sta) == -1 || !S_ISBLK(sta.st_mode)) { | |
3828 | pr_err("%s: file descriptor for invalid device\n", | |
3829 | __func__); | |
3830 | return 1; | |
3831 | } | |
3832 | for (dl = ddf->dlist; dl; dl = dl->next) | |
3833 | if (dl->major == (int)major(sta.st_rdev) && | |
3834 | dl->minor == (int)minor(sta.st_rdev)) | |
3835 | break; | |
3836 | if (!dl) { | |
3837 | pr_err("%s: couldn't find disk %d/%d\n", __func__, | |
3838 | (int)major(sta.st_rdev), | |
3839 | (int)minor(sta.st_rdev)); | |
3840 | return 1; | |
3841 | } | |
3842 | ofd = dl->fd; | |
3843 | dl->fd = fd; | |
3844 | ret = (_write_super_to_disk(ddf, dl) != 1); | |
3845 | dl->fd = ofd; | |
3846 | return ret; | |
3847 | } | |
3848 | ||
3849 | if (posix_memalign(&buf, 512, 512) != 0) | |
3850 | return 1; | |
3851 | memset(buf, 0, 512); | |
3852 | ||
3853 | lseek64(fd, dsize-512, 0); | |
3854 | rc = write(fd, buf, 512); | |
3855 | free(buf); | |
3856 | if (rc < 0) | |
3857 | return 1; | |
3858 | return 0; | |
3859 | } | |
3860 | ||
3861 | static int compare_super_ddf(struct supertype *st, struct supertype *tst) | |
3862 | { | |
3863 | /* | |
3864 | * return: | |
3865 | * 0 same, or first was empty, and second was copied | |
3866 | * 1 second had wrong number | |
3867 | * 2 wrong uuid | |
3868 | * 3 wrong other info | |
3869 | */ | |
3870 | struct ddf_super *first = st->sb; | |
3871 | struct ddf_super *second = tst->sb; | |
3872 | struct dl *dl1, *dl2; | |
3873 | struct vcl *vl1, *vl2; | |
3874 | unsigned int max_vds, max_pds, pd, vd; | |
3875 | ||
3876 | if (!first) { | |
3877 | st->sb = tst->sb; | |
3878 | tst->sb = NULL; | |
3879 | return 0; | |
3880 | } | |
3881 | ||
3882 | if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0) | |
3883 | return 2; | |
3884 | ||
3885 | if (!be32_eq(first->anchor.seq, second->anchor.seq)) { | |
3886 | dprintf("%s: sequence number mismatch %u/%u\n", __func__, | |
3887 | be32_to_cpu(first->anchor.seq), | |
3888 | be32_to_cpu(second->anchor.seq)); | |
3889 | return 3; | |
3890 | } | |
3891 | if (first->max_part != second->max_part || | |
3892 | !be16_eq(first->phys->used_pdes, second->phys->used_pdes) || | |
3893 | !be16_eq(first->virt->populated_vdes, | |
3894 | second->virt->populated_vdes)) { | |
3895 | dprintf("%s: PD/VD number mismatch\n", __func__); | |
3896 | return 3; | |
3897 | } | |
3898 | ||
3899 | max_pds = be16_to_cpu(first->phys->used_pdes); | |
3900 | for (dl2 = second->dlist; dl2; dl2 = dl2->next) { | |
3901 | for (pd = 0; pd < max_pds; pd++) | |
3902 | if (be32_eq(first->phys->entries[pd].refnum, | |
3903 | dl2->disk.refnum)) | |
3904 | break; | |
3905 | if (pd == max_pds) { | |
3906 | dprintf("%s: no match for disk %08x\n", __func__, | |
3907 | be32_to_cpu(dl2->disk.refnum)); | |
3908 | return 3; | |
3909 | } | |
3910 | } | |
3911 | ||
3912 | max_vds = be16_to_cpu(first->active->max_vd_entries); | |
3913 | for (vl2 = second->conflist; vl2; vl2 = vl2->next) { | |
3914 | if (!be32_eq(vl2->conf.magic, DDF_VD_CONF_MAGIC)) | |
3915 | continue; | |
3916 | for (vd = 0; vd < max_vds; vd++) | |
3917 | if (!memcmp(first->virt->entries[vd].guid, | |
3918 | vl2->conf.guid, DDF_GUID_LEN)) | |
3919 | break; | |
3920 | if (vd == max_vds) { | |
3921 | dprintf("%s: no match for VD config\n", __func__); | |
3922 | return 3; | |
3923 | } | |
3924 | } | |
3925 | /* FIXME should I look at anything else? */ | |
3926 | ||
3927 | /* | |
3928 | At this point we are fairly sure that the meta data matches. | |
3929 | But the new disk may contain additional local data. | |
3930 | Add it to the super block. | |
3931 | */ | |
3932 | for (vl2 = second->conflist; vl2; vl2 = vl2->next) { | |
3933 | for (vl1 = first->conflist; vl1; vl1 = vl1->next) | |
3934 | if (!memcmp(vl1->conf.guid, vl2->conf.guid, | |
3935 | DDF_GUID_LEN)) | |
3936 | break; | |
3937 | if (vl1) { | |
3938 | if (vl1->other_bvds != NULL && | |
3939 | vl1->conf.sec_elmnt_seq != | |
3940 | vl2->conf.sec_elmnt_seq) { | |
3941 | dprintf("%s: adding BVD %u\n", __func__, | |
3942 | vl2->conf.sec_elmnt_seq); | |
3943 | add_other_bvd(vl1, &vl2->conf, | |
3944 | first->conf_rec_len*512); | |
3945 | } | |
3946 | continue; | |
3947 | } | |
3948 | ||
3949 | if (posix_memalign((void **)&vl1, 512, | |
3950 | (first->conf_rec_len*512 + | |
3951 | offsetof(struct vcl, conf))) != 0) { | |
3952 | pr_err("%s could not allocate vcl buf\n", | |
3953 | __func__); | |
3954 | return 3; | |
3955 | } | |
3956 | ||
3957 | vl1->next = first->conflist; | |
3958 | vl1->block_sizes = NULL; | |
3959 | memcpy(&vl1->conf, &vl2->conf, first->conf_rec_len*512); | |
3960 | if (alloc_other_bvds(first, vl1) != 0) { | |
3961 | pr_err("%s could not allocate other bvds\n", | |
3962 | __func__); | |
3963 | free(vl1); | |
3964 | return 3; | |
3965 | } | |
3966 | for (vd = 0; vd < max_vds; vd++) | |
3967 | if (!memcmp(first->virt->entries[vd].guid, | |
3968 | vl1->conf.guid, DDF_GUID_LEN)) | |
3969 | break; | |
3970 | vl1->vcnum = vd; | |
3971 | dprintf("%s: added config for VD %u\n", __func__, vl1->vcnum); | |
3972 | first->conflist = vl1; | |
3973 | } | |
3974 | ||
3975 | for (dl2 = second->dlist; dl2; dl2 = dl2->next) { | |
3976 | for (dl1 = first->dlist; dl1; dl1 = dl1->next) | |
3977 | if (be32_eq(dl1->disk.refnum, dl2->disk.refnum)) | |
3978 | break; | |
3979 | if (dl1) | |
3980 | continue; | |
3981 | ||
3982 | if (posix_memalign((void **)&dl1, 512, | |
3983 | sizeof(*dl1) + (first->max_part) * sizeof(dl1->vlist[0])) | |
3984 | != 0) { | |
3985 | pr_err("%s could not allocate disk info buffer\n", | |
3986 | __func__); | |
3987 | return 3; | |
3988 | } | |
3989 | memcpy(dl1, dl2, sizeof(*dl1)); | |
3990 | dl1->mdupdate = NULL; | |
3991 | dl1->next = first->dlist; | |
3992 | dl1->fd = -1; | |
3993 | for (pd = 0; pd < max_pds; pd++) | |
3994 | if (be32_eq(first->phys->entries[pd].refnum, | |
3995 | dl1->disk.refnum)) | |
3996 | break; | |
3997 | dl1->pdnum = pd; | |
3998 | if (dl2->spare) { | |
3999 | if (posix_memalign((void **)&dl1->spare, 512, | |
4000 | first->conf_rec_len*512) != 0) { | |
4001 | pr_err("%s could not allocate spare info buf\n", | |
4002 | __func__); | |
4003 | return 3; | |
4004 | } | |
4005 | memcpy(dl1->spare, dl2->spare, first->conf_rec_len*512); | |
4006 | } | |
4007 | for (vd = 0 ; vd < first->max_part ; vd++) { | |
4008 | if (!dl2->vlist[vd]) { | |
4009 | dl1->vlist[vd] = NULL; | |
4010 | continue; | |
4011 | } | |
4012 | for (vl1 = first->conflist; vl1; vl1 = vl1->next) { | |
4013 | if (!memcmp(vl1->conf.guid, | |
4014 | dl2->vlist[vd]->conf.guid, | |
4015 | DDF_GUID_LEN)) | |
4016 | break; | |
4017 | dl1->vlist[vd] = vl1; | |
4018 | } | |
4019 | } | |
4020 | first->dlist = dl1; | |
4021 | dprintf("%s: added disk %d: %08x\n", __func__, dl1->pdnum, | |
4022 | be32_to_cpu(dl1->disk.refnum)); | |
4023 | } | |
4024 | ||
4025 | return 0; | |
4026 | } | |
4027 | ||
4028 | #ifndef MDASSEMBLE | |
4029 | /* | |
4030 | * A new array 'a' has been started which claims to be instance 'inst' | |
4031 | * within container 'c'. | |
4032 | * We need to confirm that the array matches the metadata in 'c' so | |
4033 | * that we don't corrupt any metadata. | |
4034 | */ | |
4035 | static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst) | |
4036 | { | |
4037 | struct ddf_super *ddf = c->sb; | |
4038 | int n = atoi(inst); | |
4039 | struct mdinfo *dev; | |
4040 | struct dl *dl; | |
4041 | static const char faulty[] = "faulty"; | |
4042 | ||
4043 | if (all_ff(ddf->virt->entries[n].guid)) { | |
4044 | pr_err("%s: subarray %d doesn't exist\n", __func__, n); | |
4045 | return -ENODEV; | |
4046 | } | |
4047 | dprintf("%s: new subarray %d, GUID: %s\n", __func__, n, | |
4048 | guid_str(ddf->virt->entries[n].guid)); | |
4049 | for (dev = a->info.devs; dev; dev = dev->next) { | |
4050 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4051 | if (dl->major == dev->disk.major && | |
4052 | dl->minor == dev->disk.minor) | |
4053 | break; | |
4054 | if (!dl) { | |
4055 | pr_err("%s: device %d/%d of subarray %d not found in meta data\n", | |
4056 | __func__, dev->disk.major, dev->disk.minor, n); | |
4057 | return -1; | |
4058 | } | |
4059 | if ((be16_to_cpu(ddf->phys->entries[dl->pdnum].state) & | |
4060 | (DDF_Online|DDF_Missing|DDF_Failed)) != DDF_Online) { | |
4061 | pr_err("%s: new subarray %d contains broken device %d/%d (%02x)\n", | |
4062 | __func__, n, dl->major, dl->minor, | |
4063 | be16_to_cpu( | |
4064 | ddf->phys->entries[dl->pdnum].state)); | |
4065 | if (write(dev->state_fd, faulty, sizeof(faulty)-1) != | |
4066 | sizeof(faulty) - 1) | |
4067 | pr_err("Write to state_fd failed\n"); | |
4068 | dev->curr_state = DS_FAULTY; | |
4069 | } | |
4070 | } | |
4071 | a->info.container_member = n; | |
4072 | return 0; | |
4073 | } | |
4074 | ||
4075 | /* | |
4076 | * The array 'a' is to be marked clean in the metadata. | |
4077 | * If '->resync_start' is not ~(unsigned long long)0, then the array is only | |
4078 | * clean up to the point (in sectors). If that cannot be recorded in the | |
4079 | * metadata, then leave it as dirty. | |
4080 | * | |
4081 | * For DDF, we need to clear the DDF_state_inconsistent bit in the | |
4082 | * !global! virtual_disk.virtual_entry structure. | |
4083 | */ | |
4084 | static int ddf_set_array_state(struct active_array *a, int consistent) | |
4085 | { | |
4086 | struct ddf_super *ddf = a->container->sb; | |
4087 | int inst = a->info.container_member; | |
4088 | int old = ddf->virt->entries[inst].state; | |
4089 | if (consistent == 2) { | |
4090 | /* Should check if a recovery should be started FIXME */ | |
4091 | consistent = 1; | |
4092 | if (!is_resync_complete(&a->info)) | |
4093 | consistent = 0; | |
4094 | } | |
4095 | if (consistent) | |
4096 | ddf->virt->entries[inst].state &= ~DDF_state_inconsistent; | |
4097 | else | |
4098 | ddf->virt->entries[inst].state |= DDF_state_inconsistent; | |
4099 | if (old != ddf->virt->entries[inst].state) | |
4100 | ddf_set_updates_pending(ddf); | |
4101 | ||
4102 | old = ddf->virt->entries[inst].init_state; | |
4103 | ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask; | |
4104 | if (is_resync_complete(&a->info)) | |
4105 | ddf->virt->entries[inst].init_state |= DDF_init_full; | |
4106 | else if (a->info.resync_start == 0) | |
4107 | ddf->virt->entries[inst].init_state |= DDF_init_not; | |
4108 | else | |
4109 | ddf->virt->entries[inst].init_state |= DDF_init_quick; | |
4110 | if (old != ddf->virt->entries[inst].init_state) | |
4111 | ddf_set_updates_pending(ddf); | |
4112 | ||
4113 | dprintf("ddf mark %d/%s (%d) %s %llu\n", inst, | |
4114 | guid_str(ddf->virt->entries[inst].guid), a->curr_state, | |
4115 | consistent?"clean":"dirty", | |
4116 | a->info.resync_start); | |
4117 | return consistent; | |
4118 | } | |
4119 | ||
4120 | static int get_bvd_state(const struct ddf_super *ddf, | |
4121 | const struct vd_config *vc) | |
4122 | { | |
4123 | unsigned int i, n_bvd, working = 0; | |
4124 | unsigned int n_prim = be16_to_cpu(vc->prim_elmnt_count); | |
4125 | int pd, st, state; | |
4126 | for (i = 0; i < n_prim; i++) { | |
4127 | if (!find_index_in_bvd(ddf, vc, i, &n_bvd)) | |
4128 | continue; | |
4129 | pd = find_phys(ddf, vc->phys_refnum[n_bvd]); | |
4130 | if (pd < 0) | |
4131 | continue; | |
4132 | st = be16_to_cpu(ddf->phys->entries[pd].state); | |
4133 | if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding)) | |
4134 | == DDF_Online) | |
4135 | working++; | |
4136 | } | |
4137 | ||
4138 | state = DDF_state_degraded; | |
4139 | if (working == n_prim) | |
4140 | state = DDF_state_optimal; | |
4141 | else | |
4142 | switch (vc->prl) { | |
4143 | case DDF_RAID0: | |
4144 | case DDF_CONCAT: | |
4145 | case DDF_JBOD: | |
4146 | state = DDF_state_failed; | |
4147 | break; | |
4148 | case DDF_RAID1: | |
4149 | if (working == 0) | |
4150 | state = DDF_state_failed; | |
4151 | else if (working >= 2) | |
4152 | state = DDF_state_part_optimal; | |
4153 | break; | |
4154 | case DDF_RAID4: | |
4155 | case DDF_RAID5: | |
4156 | if (working < n_prim - 1) | |
4157 | state = DDF_state_failed; | |
4158 | break; | |
4159 | case DDF_RAID6: | |
4160 | if (working < n_prim - 2) | |
4161 | state = DDF_state_failed; | |
4162 | else if (working == n_prim - 1) | |
4163 | state = DDF_state_part_optimal; | |
4164 | break; | |
4165 | } | |
4166 | return state; | |
4167 | } | |
4168 | ||
4169 | static int secondary_state(int state, int other, int seclevel) | |
4170 | { | |
4171 | if (state == DDF_state_optimal && other == DDF_state_optimal) | |
4172 | return DDF_state_optimal; | |
4173 | if (seclevel == DDF_2MIRRORED) { | |
4174 | if (state == DDF_state_optimal || other == DDF_state_optimal) | |
4175 | return DDF_state_part_optimal; | |
4176 | if (state == DDF_state_failed && other == DDF_state_failed) | |
4177 | return DDF_state_failed; | |
4178 | return DDF_state_degraded; | |
4179 | } else { | |
4180 | if (state == DDF_state_failed || other == DDF_state_failed) | |
4181 | return DDF_state_failed; | |
4182 | if (state == DDF_state_degraded || other == DDF_state_degraded) | |
4183 | return DDF_state_degraded; | |
4184 | return DDF_state_part_optimal; | |
4185 | } | |
4186 | } | |
4187 | ||
4188 | static int get_svd_state(const struct ddf_super *ddf, const struct vcl *vcl) | |
4189 | { | |
4190 | int state = get_bvd_state(ddf, &vcl->conf); | |
4191 | unsigned int i; | |
4192 | for (i = 1; i < vcl->conf.sec_elmnt_count; i++) { | |
4193 | state = secondary_state( | |
4194 | state, | |
4195 | get_bvd_state(ddf, vcl->other_bvds[i-1]), | |
4196 | vcl->conf.srl); | |
4197 | } | |
4198 | return state; | |
4199 | } | |
4200 | ||
4201 | /* | |
4202 | * The state of each disk is stored in the global phys_disk structure | |
4203 | * in phys_disk.entries[n].state. | |
4204 | * This makes various combinations awkward. | |
4205 | * - When a device fails in any array, it must be failed in all arrays | |
4206 | * that include a part of this device. | |
4207 | * - When a component is rebuilding, we cannot include it officially in the | |
4208 | * array unless this is the only array that uses the device. | |
4209 | * | |
4210 | * So: when transitioning: | |
4211 | * Online -> failed, just set failed flag. monitor will propagate | |
4212 | * spare -> online, the device might need to be added to the array. | |
4213 | * spare -> failed, just set failed. Don't worry if in array or not. | |
4214 | */ | |
4215 | static void ddf_set_disk(struct active_array *a, int n, int state) | |
4216 | { | |
4217 | struct ddf_super *ddf = a->container->sb; | |
4218 | unsigned int inst = a->info.container_member, n_bvd; | |
4219 | struct vcl *vcl; | |
4220 | struct vd_config *vc = find_vdcr(ddf, inst, (unsigned int)n, | |
4221 | &n_bvd, &vcl); | |
4222 | int pd; | |
4223 | struct mdinfo *mdi; | |
4224 | struct dl *dl; | |
4225 | ||
4226 | dprintf("%s: %d to %x\n", __func__, n, state); | |
4227 | if (vc == NULL) { | |
4228 | dprintf("ddf: cannot find instance %d!!\n", inst); | |
4229 | return; | |
4230 | } | |
4231 | /* Find the matching slot in 'info'. */ | |
4232 | for (mdi = a->info.devs; mdi; mdi = mdi->next) | |
4233 | if (mdi->disk.raid_disk == n) | |
4234 | break; | |
4235 | if (!mdi) { | |
4236 | pr_err("%s: cannot find raid disk %d\n", | |
4237 | __func__, n); | |
4238 | return; | |
4239 | } | |
4240 | ||
4241 | /* and find the 'dl' entry corresponding to that. */ | |
4242 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4243 | if (mdi->state_fd >= 0 && | |
4244 | mdi->disk.major == dl->major && | |
4245 | mdi->disk.minor == dl->minor) | |
4246 | break; | |
4247 | if (!dl) { | |
4248 | pr_err("%s: cannot find raid disk %d (%d/%d)\n", | |
4249 | __func__, n, | |
4250 | mdi->disk.major, mdi->disk.minor); | |
4251 | return; | |
4252 | } | |
4253 | ||
4254 | pd = find_phys(ddf, vc->phys_refnum[n_bvd]); | |
4255 | if (pd < 0 || pd != dl->pdnum) { | |
4256 | /* disk doesn't currently exist or has changed. | |
4257 | * If it is now in_sync, insert it. */ | |
4258 | dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n", | |
4259 | __func__, dl->pdnum, dl->major, dl->minor, | |
4260 | be32_to_cpu(dl->disk.refnum)); | |
4261 | dprintf("%s: array %u disk %u ref %08x pd %d\n", | |
4262 | __func__, inst, n_bvd, | |
4263 | be32_to_cpu(vc->phys_refnum[n_bvd]), pd); | |
4264 | if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) { | |
4265 | pd = dl->pdnum; /* FIXME: is this really correct ? */ | |
4266 | vc->phys_refnum[n_bvd] = dl->disk.refnum; | |
4267 | LBA_OFFSET(ddf, vc)[n_bvd] = | |
4268 | cpu_to_be64(mdi->data_offset); | |
4269 | be16_clear(ddf->phys->entries[pd].type, | |
4270 | cpu_to_be16(DDF_Global_Spare)); | |
4271 | be16_set(ddf->phys->entries[pd].type, | |
4272 | cpu_to_be16(DDF_Active_in_VD)); | |
4273 | ddf_set_updates_pending(ddf); | |
4274 | } | |
4275 | } else { | |
4276 | be16 old = ddf->phys->entries[pd].state; | |
4277 | if (state & DS_FAULTY) | |
4278 | be16_set(ddf->phys->entries[pd].state, | |
4279 | cpu_to_be16(DDF_Failed)); | |
4280 | if (state & DS_INSYNC) { | |
4281 | be16_set(ddf->phys->entries[pd].state, | |
4282 | cpu_to_be16(DDF_Online)); | |
4283 | be16_clear(ddf->phys->entries[pd].state, | |
4284 | cpu_to_be16(DDF_Rebuilding)); | |
4285 | } | |
4286 | if (!be16_eq(old, ddf->phys->entries[pd].state)) | |
4287 | ddf_set_updates_pending(ddf); | |
4288 | } | |
4289 | ||
4290 | dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n, | |
4291 | be32_to_cpu(dl->disk.refnum), state, | |
4292 | be16_to_cpu(ddf->phys->entries[pd].state)); | |
4293 | ||
4294 | /* Now we need to check the state of the array and update | |
4295 | * virtual_disk.entries[n].state. | |
4296 | * It needs to be one of "optimal", "degraded", "failed". | |
4297 | * I don't understand 'deleted' or 'missing'. | |
4298 | */ | |
4299 | state = get_svd_state(ddf, vcl); | |
4300 | ||
4301 | if (ddf->virt->entries[inst].state != | |
4302 | ((ddf->virt->entries[inst].state & ~DDF_state_mask) | |
4303 | | state)) { | |
4304 | ||
4305 | ddf->virt->entries[inst].state = | |
4306 | (ddf->virt->entries[inst].state & ~DDF_state_mask) | |
4307 | | state; | |
4308 | ddf_set_updates_pending(ddf); | |
4309 | } | |
4310 | ||
4311 | } | |
4312 | ||
4313 | static void ddf_sync_metadata(struct supertype *st) | |
4314 | { | |
4315 | ||
4316 | /* | |
4317 | * Write all data to all devices. | |
4318 | * Later, we might be able to track whether only local changes | |
4319 | * have been made, or whether any global data has been changed, | |
4320 | * but ddf is sufficiently weird that it probably always | |
4321 | * changes global data .... | |
4322 | */ | |
4323 | struct ddf_super *ddf = st->sb; | |
4324 | if (!ddf->updates_pending) | |
4325 | return; | |
4326 | ddf->updates_pending = 0; | |
4327 | __write_init_super_ddf(st); | |
4328 | dprintf("ddf: sync_metadata\n"); | |
4329 | } | |
4330 | ||
4331 | static int del_from_conflist(struct vcl **list, const char *guid) | |
4332 | { | |
4333 | struct vcl **p; | |
4334 | int found = 0; | |
4335 | for (p = list; p && *p; p = &((*p)->next)) | |
4336 | if (!memcmp((*p)->conf.guid, guid, DDF_GUID_LEN)) { | |
4337 | found = 1; | |
4338 | *p = (*p)->next; | |
4339 | } | |
4340 | return found; | |
4341 | } | |
4342 | ||
4343 | static int _kill_subarray_ddf(struct ddf_super *ddf, const char *guid) | |
4344 | { | |
4345 | struct dl *dl; | |
4346 | unsigned int vdnum, i; | |
4347 | vdnum = find_vde_by_guid(ddf, guid); | |
4348 | if (vdnum == DDF_NOTFOUND) { | |
4349 | pr_err("%s: could not find VD %s\n", __func__, | |
4350 | guid_str(guid)); | |
4351 | return -1; | |
4352 | } | |
4353 | if (del_from_conflist(&ddf->conflist, guid) == 0) { | |
4354 | pr_err("%s: could not find conf %s\n", __func__, | |
4355 | guid_str(guid)); | |
4356 | return -1; | |
4357 | } | |
4358 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4359 | for (i = 0; i < ddf->max_part; i++) | |
4360 | if (dl->vlist[i] != NULL && | |
4361 | !memcmp(dl->vlist[i]->conf.guid, guid, | |
4362 | DDF_GUID_LEN)) | |
4363 | dl->vlist[i] = NULL; | |
4364 | memset(ddf->virt->entries[vdnum].guid, 0xff, DDF_GUID_LEN); | |
4365 | dprintf("%s: deleted %s\n", __func__, guid_str(guid)); | |
4366 | return 0; | |
4367 | } | |
4368 | ||
4369 | static int kill_subarray_ddf(struct supertype *st) | |
4370 | { | |
4371 | struct ddf_super *ddf = st->sb; | |
4372 | /* | |
4373 | * currentconf is set in container_content_ddf, | |
4374 | * called with subarray arg | |
4375 | */ | |
4376 | struct vcl *victim = ddf->currentconf; | |
4377 | struct vd_config *conf; | |
4378 | ddf->currentconf = NULL; | |
4379 | unsigned int vdnum; | |
4380 | if (!victim) { | |
4381 | pr_err("%s: nothing to kill\n", __func__); | |
4382 | return -1; | |
4383 | } | |
4384 | conf = &victim->conf; | |
4385 | vdnum = find_vde_by_guid(ddf, conf->guid); | |
4386 | if (vdnum == DDF_NOTFOUND) { | |
4387 | pr_err("%s: could not find VD %s\n", __func__, | |
4388 | guid_str(conf->guid)); | |
4389 | return -1; | |
4390 | } | |
4391 | if (st->update_tail) { | |
4392 | struct virtual_disk *vd; | |
4393 | int len = sizeof(struct virtual_disk) | |
4394 | + sizeof(struct virtual_entry); | |
4395 | vd = xmalloc(len); | |
4396 | if (vd == NULL) { | |
4397 | pr_err("%s: failed to allocate %d bytes\n", __func__, | |
4398 | len); | |
4399 | return -1; | |
4400 | } | |
4401 | memset(vd, 0 , len); | |
4402 | vd->magic = DDF_VIRT_RECORDS_MAGIC; | |
4403 | vd->populated_vdes = cpu_to_be16(0); | |
4404 | memcpy(vd->entries[0].guid, conf->guid, DDF_GUID_LEN); | |
4405 | /* we use DDF_state_deleted as marker */ | |
4406 | vd->entries[0].state = DDF_state_deleted; | |
4407 | append_metadata_update(st, vd, len); | |
4408 | } else { | |
4409 | _kill_subarray_ddf(ddf, conf->guid); | |
4410 | ddf_set_updates_pending(ddf); | |
4411 | ddf_sync_metadata(st); | |
4412 | } | |
4413 | return 0; | |
4414 | } | |
4415 | ||
4416 | static void copy_matching_bvd(struct ddf_super *ddf, | |
4417 | struct vd_config *conf, | |
4418 | const struct metadata_update *update) | |
4419 | { | |
4420 | unsigned int mppe = | |
4421 | be16_to_cpu(ddf->anchor.max_primary_element_entries); | |
4422 | unsigned int len = ddf->conf_rec_len * 512; | |
4423 | char *p; | |
4424 | struct vd_config *vc; | |
4425 | for (p = update->buf; p < update->buf + update->len; p += len) { | |
4426 | vc = (struct vd_config *) p; | |
4427 | if (vc->sec_elmnt_seq == conf->sec_elmnt_seq) { | |
4428 | memcpy(conf->phys_refnum, vc->phys_refnum, | |
4429 | mppe * (sizeof(__u32) + sizeof(__u64))); | |
4430 | return; | |
4431 | } | |
4432 | } | |
4433 | pr_err("%s: no match for BVD %d of %s in update\n", __func__, | |
4434 | conf->sec_elmnt_seq, guid_str(conf->guid)); | |
4435 | } | |
4436 | ||
4437 | static void ddf_process_update(struct supertype *st, | |
4438 | struct metadata_update *update) | |
4439 | { | |
4440 | /* Apply this update to the metadata. | |
4441 | * The first 4 bytes are a DDF_*_MAGIC which guides | |
4442 | * our actions. | |
4443 | * Possible update are: | |
4444 | * DDF_PHYS_RECORDS_MAGIC | |
4445 | * Add a new physical device or remove an old one. | |
4446 | * Changes to this record only happen implicitly. | |
4447 | * used_pdes is the device number. | |
4448 | * DDF_VIRT_RECORDS_MAGIC | |
4449 | * Add a new VD. Possibly also change the 'access' bits. | |
4450 | * populated_vdes is the entry number. | |
4451 | * DDF_VD_CONF_MAGIC | |
4452 | * New or updated VD. the VIRT_RECORD must already | |
4453 | * exist. For an update, phys_refnum and lba_offset | |
4454 | * (at least) are updated, and the VD_CONF must | |
4455 | * be written to precisely those devices listed with | |
4456 | * a phys_refnum. | |
4457 | * DDF_SPARE_ASSIGN_MAGIC | |
4458 | * replacement Spare Assignment Record... but for which device? | |
4459 | * | |
4460 | * So, e.g.: | |
4461 | * - to create a new array, we send a VIRT_RECORD and | |
4462 | * a VD_CONF. Then assemble and start the array. | |
4463 | * - to activate a spare we send a VD_CONF to add the phys_refnum | |
4464 | * and offset. This will also mark the spare as active with | |
4465 | * a spare-assignment record. | |
4466 | */ | |
4467 | struct ddf_super *ddf = st->sb; | |
4468 | be32 *magic = (be32 *)update->buf; | |
4469 | struct phys_disk *pd; | |
4470 | struct virtual_disk *vd; | |
4471 | struct vd_config *vc; | |
4472 | struct vcl *vcl; | |
4473 | struct dl *dl; | |
4474 | unsigned int ent; | |
4475 | unsigned int pdnum, pd2, len; | |
4476 | ||
4477 | dprintf("Process update %x\n", be32_to_cpu(*magic)); | |
4478 | ||
4479 | if (be32_eq(*magic, DDF_PHYS_RECORDS_MAGIC)) { | |
4480 | ||
4481 | if (update->len != (sizeof(struct phys_disk) + | |
4482 | sizeof(struct phys_disk_entry))) | |
4483 | return; | |
4484 | pd = (struct phys_disk*)update->buf; | |
4485 | ||
4486 | ent = be16_to_cpu(pd->used_pdes); | |
4487 | if (ent >= be16_to_cpu(ddf->phys->max_pdes)) | |
4488 | return; | |
4489 | if (be16_and(pd->entries[0].state, cpu_to_be16(DDF_Missing))) { | |
4490 | struct dl **dlp; | |
4491 | /* removing this disk. */ | |
4492 | be16_set(ddf->phys->entries[ent].state, | |
4493 | cpu_to_be16(DDF_Missing)); | |
4494 | for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) { | |
4495 | struct dl *dl = *dlp; | |
4496 | if (dl->pdnum == (signed)ent) { | |
4497 | close(dl->fd); | |
4498 | dl->fd = -1; | |
4499 | /* FIXME this doesn't free | |
4500 | * dl->devname */ | |
4501 | update->space = dl; | |
4502 | *dlp = dl->next; | |
4503 | break; | |
4504 | } | |
4505 | } | |
4506 | ddf_set_updates_pending(ddf); | |
4507 | return; | |
4508 | } | |
4509 | if (!all_ff(ddf->phys->entries[ent].guid)) | |
4510 | return; | |
4511 | ddf->phys->entries[ent] = pd->entries[0]; | |
4512 | ddf->phys->used_pdes = cpu_to_be16 | |
4513 | (1 + be16_to_cpu(ddf->phys->used_pdes)); | |
4514 | ddf_set_updates_pending(ddf); | |
4515 | if (ddf->add_list) { | |
4516 | struct active_array *a; | |
4517 | struct dl *al = ddf->add_list; | |
4518 | ddf->add_list = al->next; | |
4519 | ||
4520 | al->next = ddf->dlist; | |
4521 | ddf->dlist = al; | |
4522 | ||
4523 | /* As a device has been added, we should check | |
4524 | * for any degraded devices that might make | |
4525 | * use of this spare */ | |
4526 | for (a = st->arrays ; a; a=a->next) | |
4527 | a->check_degraded = 1; | |
4528 | } | |
4529 | } else if (be32_eq(*magic, DDF_VIRT_RECORDS_MAGIC)) { | |
4530 | ||
4531 | if (update->len != (sizeof(struct virtual_disk) + | |
4532 | sizeof(struct virtual_entry))) | |
4533 | return; | |
4534 | vd = (struct virtual_disk*)update->buf; | |
4535 | ||
4536 | if (vd->entries[0].state == DDF_state_deleted) { | |
4537 | if (_kill_subarray_ddf(ddf, vd->entries[0].guid)) | |
4538 | return; | |
4539 | } else { | |
4540 | ||
4541 | ent = find_vde_by_guid(ddf, vd->entries[0].guid); | |
4542 | if (ent != DDF_NOTFOUND) { | |
4543 | dprintf("%s: VD %s exists already in slot %d\n", | |
4544 | __func__, guid_str(vd->entries[0].guid), | |
4545 | ent); | |
4546 | return; | |
4547 | } | |
4548 | ent = find_unused_vde(ddf); | |
4549 | if (ent == DDF_NOTFOUND) | |
4550 | return; | |
4551 | ddf->virt->entries[ent] = vd->entries[0]; | |
4552 | ddf->virt->populated_vdes = | |
4553 | cpu_to_be16( | |
4554 | 1 + be16_to_cpu( | |
4555 | ddf->virt->populated_vdes)); | |
4556 | dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n", | |
4557 | __func__, guid_str(vd->entries[0].guid), ent, | |
4558 | ddf->virt->entries[ent].state, | |
4559 | ddf->virt->entries[ent].init_state); | |
4560 | } | |
4561 | ddf_set_updates_pending(ddf); | |
4562 | } | |
4563 | ||
4564 | else if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) { | |
4565 | vc = (struct vd_config*)update->buf; | |
4566 | len = ddf->conf_rec_len * 512; | |
4567 | if ((unsigned int)update->len != len * vc->sec_elmnt_count) { | |
4568 | pr_err("%s: %s: insufficient data (%d) for %u BVDs\n", | |
4569 | __func__, guid_str(vc->guid), update->len, | |
4570 | vc->sec_elmnt_count); | |
4571 | return; | |
4572 | } | |
4573 | for (vcl = ddf->conflist; vcl ; vcl = vcl->next) | |
4574 | if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0) | |
4575 | break; | |
4576 | dprintf("%s: conf update for %s (%s)\n", __func__, | |
4577 | guid_str(vc->guid), (vcl ? "old" : "new")); | |
4578 | if (vcl) { | |
4579 | /* An update, just copy the phys_refnum and lba_offset | |
4580 | * fields | |
4581 | */ | |
4582 | unsigned int i; | |
4583 | unsigned int k; | |
4584 | copy_matching_bvd(ddf, &vcl->conf, update); | |
4585 | for (k = 0; k < be16_to_cpu(vc->prim_elmnt_count); k++) | |
4586 | dprintf("BVD %u has %08x at %llu\n", 0, | |
4587 | be32_to_cpu(vcl->conf.phys_refnum[k]), | |
4588 | be64_to_cpu(LBA_OFFSET(ddf, | |
4589 | &vcl->conf)[k])); | |
4590 | for (i = 1; i < vc->sec_elmnt_count; i++) { | |
4591 | copy_matching_bvd(ddf, vcl->other_bvds[i-1], | |
4592 | update); | |
4593 | for (k = 0; k < be16_to_cpu( | |
4594 | vc->prim_elmnt_count); k++) | |
4595 | dprintf("BVD %u has %08x at %llu\n", i, | |
4596 | be32_to_cpu | |
4597 | (vcl->other_bvds[i-1]-> | |
4598 | phys_refnum[k]), | |
4599 | be64_to_cpu | |
4600 | (LBA_OFFSET | |
4601 | (ddf, | |
4602 | vcl->other_bvds[i-1])[k])); | |
4603 | } | |
4604 | } else { | |
4605 | /* A new VD_CONF */ | |
4606 | unsigned int i; | |
4607 | if (!update->space) | |
4608 | return; | |
4609 | vcl = update->space; | |
4610 | update->space = NULL; | |
4611 | vcl->next = ddf->conflist; | |
4612 | memcpy(&vcl->conf, vc, len); | |
4613 | ent = find_vde_by_guid(ddf, vc->guid); | |
4614 | if (ent == DDF_NOTFOUND) | |
4615 | return; | |
4616 | vcl->vcnum = ent; | |
4617 | ddf->conflist = vcl; | |
4618 | for (i = 1; i < vc->sec_elmnt_count; i++) | |
4619 | memcpy(vcl->other_bvds[i-1], | |
4620 | update->buf + len * i, len); | |
4621 | } | |
4622 | /* Set DDF_Transition on all Failed devices - to help | |
4623 | * us detect those that are no longer in use | |
4624 | */ | |
4625 | for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->used_pdes); | |
4626 | pdnum++) | |
4627 | if (be16_and(ddf->phys->entries[pdnum].state, | |
4628 | cpu_to_be16(DDF_Failed))) | |
4629 | be16_set(ddf->phys->entries[pdnum].state, | |
4630 | cpu_to_be16(DDF_Transition)); | |
4631 | /* Now make sure vlist is correct for each dl. */ | |
4632 | for (dl = ddf->dlist; dl; dl = dl->next) { | |
4633 | unsigned int vn = 0; | |
4634 | int in_degraded = 0; | |
4635 | for (vcl = ddf->conflist; vcl ; vcl = vcl->next) { | |
4636 | unsigned int dn, ibvd; | |
4637 | const struct vd_config *conf; | |
4638 | int vstate; | |
4639 | dn = get_pd_index_from_refnum(vcl, | |
4640 | dl->disk.refnum, | |
4641 | ddf->mppe, | |
4642 | &conf, &ibvd); | |
4643 | if (dn == DDF_NOTFOUND) | |
4644 | continue; | |
4645 | dprintf("dev %d/%08x has %s (sec=%u) at %d\n", | |
4646 | dl->pdnum, | |
4647 | be32_to_cpu(dl->disk.refnum), | |
4648 | guid_str(conf->guid), | |
4649 | conf->sec_elmnt_seq, vn); | |
4650 | /* Clear the Transition flag */ | |
4651 | if (be16_and | |
4652 | (ddf->phys->entries[dl->pdnum].state, | |
4653 | cpu_to_be16(DDF_Failed))) | |
4654 | be16_clear(ddf->phys | |
4655 | ->entries[dl->pdnum].state, | |
4656 | cpu_to_be16(DDF_Transition)); | |
4657 | dl->vlist[vn++] = vcl; | |
4658 | vstate = ddf->virt->entries[vcl->vcnum].state | |
4659 | & DDF_state_mask; | |
4660 | if (vstate == DDF_state_degraded || | |
4661 | vstate == DDF_state_part_optimal) | |
4662 | in_degraded = 1; | |
4663 | } | |
4664 | while (vn < ddf->max_part) | |
4665 | dl->vlist[vn++] = NULL; | |
4666 | if (dl->vlist[0]) { | |
4667 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4668 | cpu_to_be16(DDF_Global_Spare)); | |
4669 | if (!be16_and(ddf->phys | |
4670 | ->entries[dl->pdnum].type, | |
4671 | cpu_to_be16(DDF_Active_in_VD))) { | |
4672 | be16_set(ddf->phys | |
4673 | ->entries[dl->pdnum].type, | |
4674 | cpu_to_be16(DDF_Active_in_VD)); | |
4675 | if (in_degraded) | |
4676 | be16_set(ddf->phys | |
4677 | ->entries[dl->pdnum] | |
4678 | .state, | |
4679 | cpu_to_be16 | |
4680 | (DDF_Rebuilding)); | |
4681 | } | |
4682 | } | |
4683 | if (dl->spare) { | |
4684 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4685 | cpu_to_be16(DDF_Global_Spare)); | |
4686 | be16_set(ddf->phys->entries[dl->pdnum].type, | |
4687 | cpu_to_be16(DDF_Spare)); | |
4688 | } | |
4689 | if (!dl->vlist[0] && !dl->spare) { | |
4690 | be16_set(ddf->phys->entries[dl->pdnum].type, | |
4691 | cpu_to_be16(DDF_Global_Spare)); | |
4692 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4693 | cpu_to_be16(DDF_Spare)); | |
4694 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4695 | cpu_to_be16(DDF_Active_in_VD)); | |
4696 | } | |
4697 | } | |
4698 | ||
4699 | /* Now remove any 'Failed' devices that are not part | |
4700 | * of any VD. They will have the Transition flag set. | |
4701 | * Once done, we need to update all dl->pdnum numbers. | |
4702 | */ | |
4703 | pd2 = 0; | |
4704 | for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->used_pdes); | |
4705 | pdnum++) { | |
4706 | if (be16_and(ddf->phys->entries[pdnum].state, | |
4707 | cpu_to_be16(DDF_Failed)) | |
4708 | && be16_and(ddf->phys->entries[pdnum].state, | |
4709 | cpu_to_be16(DDF_Transition))) { | |
4710 | /* skip this one unless in dlist*/ | |
4711 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4712 | if (dl->pdnum == (int)pdnum) | |
4713 | break; | |
4714 | if (!dl) | |
4715 | continue; | |
4716 | } | |
4717 | if (pdnum == pd2) | |
4718 | pd2++; | |
4719 | else { | |
4720 | ddf->phys->entries[pd2] = | |
4721 | ddf->phys->entries[pdnum]; | |
4722 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4723 | if (dl->pdnum == (int)pdnum) | |
4724 | dl->pdnum = pd2; | |
4725 | pd2++; | |
4726 | } | |
4727 | } | |
4728 | ddf->phys->used_pdes = cpu_to_be16(pd2); | |
4729 | while (pd2 < pdnum) { | |
4730 | memset(ddf->phys->entries[pd2].guid, 0xff, | |
4731 | DDF_GUID_LEN); | |
4732 | pd2++; | |
4733 | } | |
4734 | ||
4735 | ddf_set_updates_pending(ddf); | |
4736 | } | |
4737 | /* case DDF_SPARE_ASSIGN_MAGIC */ | |
4738 | } | |
4739 | ||
4740 | static void ddf_prepare_update(struct supertype *st, | |
4741 | struct metadata_update *update) | |
4742 | { | |
4743 | /* This update arrived at managemon. | |
4744 | * We are about to pass it to monitor. | |
4745 | * If a malloc is needed, do it here. | |
4746 | */ | |
4747 | struct ddf_super *ddf = st->sb; | |
4748 | be32 *magic = (be32 *)update->buf; | |
4749 | if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) { | |
4750 | struct vcl *vcl; | |
4751 | struct vd_config *conf = (struct vd_config *) update->buf; | |
4752 | if (posix_memalign(&update->space, 512, | |
4753 | offsetof(struct vcl, conf) | |
4754 | + ddf->conf_rec_len * 512) != 0) { | |
4755 | update->space = NULL; | |
4756 | return; | |
4757 | } | |
4758 | vcl = update->space; | |
4759 | vcl->conf.sec_elmnt_count = conf->sec_elmnt_count; | |
4760 | if (alloc_other_bvds(ddf, vcl) != 0) { | |
4761 | free(update->space); | |
4762 | update->space = NULL; | |
4763 | } | |
4764 | } | |
4765 | } | |
4766 | ||
4767 | /* | |
4768 | * Check degraded state of a RAID10. | |
4769 | * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error | |
4770 | */ | |
4771 | static int raid10_degraded(struct mdinfo *info) | |
4772 | { | |
4773 | int n_prim, n_bvds; | |
4774 | int i; | |
4775 | struct mdinfo *d; | |
4776 | char *found; | |
4777 | int ret = -1; | |
4778 | ||
4779 | n_prim = info->array.layout & ~0x100; | |
4780 | n_bvds = info->array.raid_disks / n_prim; | |
4781 | found = xmalloc(n_bvds); | |
4782 | if (found == NULL) | |
4783 | return ret; | |
4784 | memset(found, 0, n_bvds); | |
4785 | for (d = info->devs; d; d = d->next) { | |
4786 | i = d->disk.raid_disk / n_prim; | |
4787 | if (i >= n_bvds) { | |
4788 | pr_err("%s: BUG: invalid raid disk\n", __func__); | |
4789 | goto out; | |
4790 | } | |
4791 | if (d->state_fd > 0) | |
4792 | found[i]++; | |
4793 | } | |
4794 | ret = 2; | |
4795 | for (i = 0; i < n_bvds; i++) | |
4796 | if (!found[i]) { | |
4797 | dprintf("%s: BVD %d/%d failed\n", __func__, i, n_bvds); | |
4798 | ret = 0; | |
4799 | goto out; | |
4800 | } else if (found[i] < n_prim) { | |
4801 | dprintf("%s: BVD %d/%d degraded\n", __func__, i, | |
4802 | n_bvds); | |
4803 | ret = 1; | |
4804 | } | |
4805 | out: | |
4806 | free(found); | |
4807 | return ret; | |
4808 | } | |
4809 | ||
4810 | /* | |
4811 | * Check if the array 'a' is degraded but not failed. | |
4812 | * If it is, find as many spares as are available and needed and | |
4813 | * arrange for their inclusion. | |
4814 | * We only choose devices which are not already in the array, | |
4815 | * and prefer those with a spare-assignment to this array. | |
4816 | * otherwise we choose global spares - assuming always that | |
4817 | * there is enough room. | |
4818 | * For each spare that we assign, we return an 'mdinfo' which | |
4819 | * describes the position for the device in the array. | |
4820 | * We also add to 'updates' a DDF_VD_CONF_MAGIC update with | |
4821 | * the new phys_refnum and lba_offset values. | |
4822 | * | |
4823 | * Only worry about BVDs at the moment. | |
4824 | */ | |
4825 | static struct mdinfo *ddf_activate_spare(struct active_array *a, | |
4826 | struct metadata_update **updates) | |
4827 | { | |
4828 | int working = 0; | |
4829 | struct mdinfo *d; | |
4830 | struct ddf_super *ddf = a->container->sb; | |
4831 | int global_ok = 0; | |
4832 | struct mdinfo *rv = NULL; | |
4833 | struct mdinfo *di; | |
4834 | struct metadata_update *mu; | |
4835 | struct dl *dl; | |
4836 | int i; | |
4837 | unsigned int j; | |
4838 | struct vcl *vcl; | |
4839 | struct vd_config *vc; | |
4840 | unsigned int n_bvd; | |
4841 | ||
4842 | for (d = a->info.devs ; d ; d = d->next) { | |
4843 | if ((d->curr_state & DS_FAULTY) && | |
4844 | d->state_fd >= 0) | |
4845 | /* wait for Removal to happen */ | |
4846 | return NULL; | |
4847 | if (d->state_fd >= 0) | |
4848 | working ++; | |
4849 | } | |
4850 | ||
4851 | dprintf("%s: working=%d (%d) level=%d\n", __func__, working, | |
4852 | a->info.array.raid_disks, | |
4853 | a->info.array.level); | |
4854 | if (working == a->info.array.raid_disks) | |
4855 | return NULL; /* array not degraded */ | |
4856 | switch (a->info.array.level) { | |
4857 | case 1: | |
4858 | if (working == 0) | |
4859 | return NULL; /* failed */ | |
4860 | break; | |
4861 | case 4: | |
4862 | case 5: | |
4863 | if (working < a->info.array.raid_disks - 1) | |
4864 | return NULL; /* failed */ | |
4865 | break; | |
4866 | case 6: | |
4867 | if (working < a->info.array.raid_disks - 2) | |
4868 | return NULL; /* failed */ | |
4869 | break; | |
4870 | case 10: | |
4871 | if (raid10_degraded(&a->info) < 1) | |
4872 | return NULL; | |
4873 | break; | |
4874 | default: /* concat or stripe */ | |
4875 | return NULL; /* failed */ | |
4876 | } | |
4877 | ||
4878 | /* For each slot, if it is not working, find a spare */ | |
4879 | dl = ddf->dlist; | |
4880 | for (i = 0; i < a->info.array.raid_disks; i++) { | |
4881 | for (d = a->info.devs ; d ; d = d->next) | |
4882 | if (d->disk.raid_disk == i) | |
4883 | break; | |
4884 | dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0); | |
4885 | if (d && (d->state_fd >= 0)) | |
4886 | continue; | |
4887 | ||
4888 | /* OK, this device needs recovery. Find a spare */ | |
4889 | again: | |
4890 | for ( ; dl ; dl = dl->next) { | |
4891 | unsigned long long esize; | |
4892 | unsigned long long pos; | |
4893 | struct mdinfo *d2; | |
4894 | int is_global = 0; | |
4895 | int is_dedicated = 0; | |
4896 | struct extent *ex; | |
4897 | unsigned int j; | |
4898 | be16 state = ddf->phys->entries[dl->pdnum].state; | |
4899 | if (be16_and(state, | |
4900 | cpu_to_be16(DDF_Failed|DDF_Missing)) || | |
4901 | !be16_and(state, | |
4902 | cpu_to_be16(DDF_Online))) | |
4903 | continue; | |
4904 | ||
4905 | /* If in this array, skip */ | |
4906 | for (d2 = a->info.devs ; d2 ; d2 = d2->next) | |
4907 | if (d2->state_fd >= 0 && | |
4908 | d2->disk.major == dl->major && | |
4909 | d2->disk.minor == dl->minor) { | |
4910 | dprintf("%x:%x (%08x) already in array\n", | |
4911 | dl->major, dl->minor, | |
4912 | be32_to_cpu(dl->disk.refnum)); | |
4913 | break; | |
4914 | } | |
4915 | if (d2) | |
4916 | continue; | |
4917 | if (be16_and(ddf->phys->entries[dl->pdnum].type, | |
4918 | cpu_to_be16(DDF_Spare))) { | |
4919 | /* Check spare assign record */ | |
4920 | if (dl->spare) { | |
4921 | if (dl->spare->type & DDF_spare_dedicated) { | |
4922 | /* check spare_ents for guid */ | |
4923 | for (j = 0 ; | |
4924 | j < be16_to_cpu | |
4925 | (dl->spare | |
4926 | ->populated); | |
4927 | j++) { | |
4928 | if (memcmp(dl->spare->spare_ents[j].guid, | |
4929 | ddf->virt->entries[a->info.container_member].guid, | |
4930 | DDF_GUID_LEN) == 0) | |
4931 | is_dedicated = 1; | |
4932 | } | |
4933 | } else | |
4934 | is_global = 1; | |
4935 | } | |
4936 | } else if (be16_and(ddf->phys->entries[dl->pdnum].type, | |
4937 | cpu_to_be16(DDF_Global_Spare))) { | |
4938 | is_global = 1; | |
4939 | } else if (!be16_and(ddf->phys | |
4940 | ->entries[dl->pdnum].state, | |
4941 | cpu_to_be16(DDF_Failed))) { | |
4942 | /* we can possibly use some of this */ | |
4943 | is_global = 1; | |
4944 | } | |
4945 | if ( ! (is_dedicated || | |
4946 | (is_global && global_ok))) { | |
4947 | dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor, | |
4948 | is_dedicated, is_global); | |
4949 | continue; | |
4950 | } | |
4951 | ||
4952 | /* We are allowed to use this device - is there space? | |
4953 | * We need a->info.component_size sectors */ | |
4954 | ex = get_extents(ddf, dl); | |
4955 | if (!ex) { | |
4956 | dprintf("cannot get extents\n"); | |
4957 | continue; | |
4958 | } | |
4959 | j = 0; pos = 0; | |
4960 | esize = 0; | |
4961 | ||
4962 | do { | |
4963 | esize = ex[j].start - pos; | |
4964 | if (esize >= a->info.component_size) | |
4965 | break; | |
4966 | pos = ex[j].start + ex[j].size; | |
4967 | j++; | |
4968 | } while (ex[j-1].size); | |
4969 | ||
4970 | free(ex); | |
4971 | if (esize < a->info.component_size) { | |
4972 | dprintf("%x:%x has no room: %llu %llu\n", | |
4973 | dl->major, dl->minor, | |
4974 | esize, a->info.component_size); | |
4975 | /* No room */ | |
4976 | continue; | |
4977 | } | |
4978 | ||
4979 | /* Cool, we have a device with some space at pos */ | |
4980 | di = xcalloc(1, sizeof(*di)); | |
4981 | di->disk.number = i; | |
4982 | di->disk.raid_disk = i; | |
4983 | di->disk.major = dl->major; | |
4984 | di->disk.minor = dl->minor; | |
4985 | di->disk.state = 0; | |
4986 | di->recovery_start = 0; | |
4987 | di->data_offset = pos; | |
4988 | di->component_size = a->info.component_size; | |
4989 | di->container_member = dl->pdnum; | |
4990 | di->next = rv; | |
4991 | rv = di; | |
4992 | dprintf("%x:%x (%08x) to be %d at %llu\n", | |
4993 | dl->major, dl->minor, | |
4994 | be32_to_cpu(dl->disk.refnum), i, pos); | |
4995 | ||
4996 | break; | |
4997 | } | |
4998 | if (!dl && ! global_ok) { | |
4999 | /* not enough dedicated spares, try global */ | |
5000 | global_ok = 1; | |
5001 | dl = ddf->dlist; | |
5002 | goto again; | |
5003 | } | |
5004 | } | |
5005 | ||
5006 | if (!rv) | |
5007 | /* No spares found */ | |
5008 | return rv; | |
5009 | /* Now 'rv' has a list of devices to return. | |
5010 | * Create a metadata_update record to update the | |
5011 | * phys_refnum and lba_offset values | |
5012 | */ | |
5013 | vc = find_vdcr(ddf, a->info.container_member, rv->disk.raid_disk, | |
5014 | &n_bvd, &vcl); | |
5015 | if (vc == NULL) | |
5016 | return NULL; | |
5017 | ||
5018 | mu = xmalloc(sizeof(*mu)); | |
5019 | if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) { | |
5020 | free(mu); | |
5021 | mu = NULL; | |
5022 | } | |
5023 | ||
5024 | mu->len = ddf->conf_rec_len * 512 * vcl->conf.sec_elmnt_count; | |
5025 | mu->buf = xmalloc(mu->len); | |
5026 | mu->space = NULL; | |
5027 | mu->space_list = NULL; | |
5028 | mu->next = *updates; | |
5029 | memcpy(mu->buf, &vcl->conf, ddf->conf_rec_len * 512); | |
5030 | for (j = 1; j < vcl->conf.sec_elmnt_count; j++) | |
5031 | memcpy(mu->buf + j * ddf->conf_rec_len * 512, | |
5032 | vcl->other_bvds[j-1], ddf->conf_rec_len * 512); | |
5033 | ||
5034 | vc = (struct vd_config*)mu->buf; | |
5035 | for (di = rv ; di ; di = di->next) { | |
5036 | unsigned int i_sec, i_prim; | |
5037 | i_sec = di->disk.raid_disk | |
5038 | / be16_to_cpu(vcl->conf.prim_elmnt_count); | |
5039 | i_prim = di->disk.raid_disk | |
5040 | % be16_to_cpu(vcl->conf.prim_elmnt_count); | |
5041 | vc = (struct vd_config *)(mu->buf | |
5042 | + i_sec * ddf->conf_rec_len * 512); | |
5043 | for (dl = ddf->dlist; dl; dl = dl->next) | |
5044 | if (dl->major == di->disk.major | |
5045 | && dl->minor == di->disk.minor) | |
5046 | break; | |
5047 | if (!dl) { | |
5048 | pr_err("%s: BUG: can't find disk %d (%d/%d)\n", | |
5049 | __func__, di->disk.raid_disk, | |
5050 | di->disk.major, di->disk.minor); | |
5051 | return NULL; | |
5052 | } | |
5053 | vc->phys_refnum[i_prim] = ddf->phys->entries[dl->pdnum].refnum; | |
5054 | LBA_OFFSET(ddf, vc)[i_prim] = cpu_to_be64(di->data_offset); | |
5055 | dprintf("BVD %u gets %u: %08x at %llu\n", i_sec, i_prim, | |
5056 | be32_to_cpu(vc->phys_refnum[i_prim]), | |
5057 | be64_to_cpu(LBA_OFFSET(ddf, vc)[i_prim])); | |
5058 | } | |
5059 | *updates = mu; | |
5060 | return rv; | |
5061 | } | |
5062 | #endif /* MDASSEMBLE */ | |
5063 | ||
5064 | static int ddf_level_to_layout(int level) | |
5065 | { | |
5066 | switch(level) { | |
5067 | case 0: | |
5068 | case 1: | |
5069 | return 0; | |
5070 | case 5: | |
5071 | return ALGORITHM_LEFT_SYMMETRIC; | |
5072 | case 6: | |
5073 | return ALGORITHM_ROTATING_N_CONTINUE; | |
5074 | case 10: | |
5075 | return 0x102; | |
5076 | default: | |
5077 | return UnSet; | |
5078 | } | |
5079 | } | |
5080 | ||
5081 | static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk) | |
5082 | { | |
5083 | if (level && *level == UnSet) | |
5084 | *level = LEVEL_CONTAINER; | |
5085 | ||
5086 | if (level && layout && *layout == UnSet) | |
5087 | *layout = ddf_level_to_layout(*level); | |
5088 | } | |
5089 | ||
5090 | struct superswitch super_ddf = { | |
5091 | #ifndef MDASSEMBLE | |
5092 | .examine_super = examine_super_ddf, | |
5093 | .brief_examine_super = brief_examine_super_ddf, | |
5094 | .brief_examine_subarrays = brief_examine_subarrays_ddf, | |
5095 | .export_examine_super = export_examine_super_ddf, | |
5096 | .detail_super = detail_super_ddf, | |
5097 | .brief_detail_super = brief_detail_super_ddf, | |
5098 | .validate_geometry = validate_geometry_ddf, | |
5099 | .write_init_super = write_init_super_ddf, | |
5100 | .add_to_super = add_to_super_ddf, | |
5101 | .remove_from_super = remove_from_super_ddf, | |
5102 | .load_container = load_container_ddf, | |
5103 | .copy_metadata = copy_metadata_ddf, | |
5104 | .kill_subarray = kill_subarray_ddf, | |
5105 | #endif | |
5106 | .match_home = match_home_ddf, | |
5107 | .uuid_from_super= uuid_from_super_ddf, | |
5108 | .getinfo_super = getinfo_super_ddf, | |
5109 | .update_super = update_super_ddf, | |
5110 | ||
5111 | .avail_size = avail_size_ddf, | |
5112 | ||
5113 | .compare_super = compare_super_ddf, | |
5114 | ||
5115 | .load_super = load_super_ddf, | |
5116 | .init_super = init_super_ddf, | |
5117 | .store_super = store_super_ddf, | |
5118 | .free_super = free_super_ddf, | |
5119 | .match_metadata_desc = match_metadata_desc_ddf, | |
5120 | .container_content = container_content_ddf, | |
5121 | .default_geometry = default_geometry_ddf, | |
5122 | ||
5123 | .external = 1, | |
5124 | ||
5125 | #ifndef MDASSEMBLE | |
5126 | /* for mdmon */ | |
5127 | .open_new = ddf_open_new, | |
5128 | .set_array_state= ddf_set_array_state, | |
5129 | .set_disk = ddf_set_disk, | |
5130 | .sync_metadata = ddf_sync_metadata, | |
5131 | .process_update = ddf_process_update, | |
5132 | .prepare_update = ddf_prepare_update, | |
5133 | .activate_spare = ddf_activate_spare, | |
5134 | #endif | |
5135 | .name = "ddf", | |
5136 | }; |