]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
4 | * Copyright (C) 2006-2014 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 taken 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 | #include <stddef.h> | |
34 | ||
35 | /* a non-official T10 name for creation GUIDs */ | |
36 | static char T10[] = "Linux-MD"; | |
37 | ||
38 | /* DDF timestamps are 1980 based, so we need to add | |
39 | * second-in-decade-of-seventies to convert to linux timestamps. | |
40 | * 10 years with 2 leap years. | |
41 | */ | |
42 | #define DECADE (3600*24*(365*10+2)) | |
43 | unsigned long crc32( | |
44 | unsigned long crc, | |
45 | const unsigned char *buf, | |
46 | unsigned len); | |
47 | ||
48 | #define DDF_NOTFOUND (~0U) | |
49 | #define DDF_CONTAINER (DDF_NOTFOUND-1) | |
50 | ||
51 | /* Default for safe_mode_delay. Same value as for IMSM. | |
52 | */ | |
53 | static const int DDF_SAFE_MODE_DELAY = 4000; | |
54 | ||
55 | /* The DDF metadata handling. | |
56 | * DDF metadata lives at the end of the device. | |
57 | * The last 512 byte block provides an 'anchor' which is used to locate | |
58 | * the rest of the metadata which usually lives immediately behind the anchor. | |
59 | * | |
60 | * Note: | |
61 | * - all multibyte numeric fields are bigendian. | |
62 | * - all strings are space padded. | |
63 | * | |
64 | */ | |
65 | ||
66 | typedef struct __be16 { | |
67 | __u16 _v16; | |
68 | } be16; | |
69 | #define be16_eq(x, y) ((x)._v16 == (y)._v16) | |
70 | #define be16_and(x, y) ((x)._v16 & (y)._v16) | |
71 | #define be16_or(x, y) ((x)._v16 | (y)._v16) | |
72 | #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16) | |
73 | #define be16_set(x, y) ((x)._v16 |= (y)._v16) | |
74 | ||
75 | typedef struct __be32 { | |
76 | __u32 _v32; | |
77 | } be32; | |
78 | #define be32_eq(x, y) ((x)._v32 == (y)._v32) | |
79 | ||
80 | typedef struct __be64 { | |
81 | __u64 _v64; | |
82 | } be64; | |
83 | #define be64_eq(x, y) ((x)._v64 == (y)._v64) | |
84 | ||
85 | #define be16_to_cpu(be) __be16_to_cpu((be)._v16) | |
86 | static inline be16 cpu_to_be16(__u16 x) | |
87 | { | |
88 | be16 be = { ._v16 = __cpu_to_be16(x) }; | |
89 | return be; | |
90 | } | |
91 | ||
92 | #define be32_to_cpu(be) __be32_to_cpu((be)._v32) | |
93 | static inline be32 cpu_to_be32(__u32 x) | |
94 | { | |
95 | be32 be = { ._v32 = __cpu_to_be32(x) }; | |
96 | return be; | |
97 | } | |
98 | ||
99 | #define be64_to_cpu(be) __be64_to_cpu((be)._v64) | |
100 | static inline be64 cpu_to_be64(__u64 x) | |
101 | { | |
102 | be64 be = { ._v64 = __cpu_to_be64(x) }; | |
103 | return be; | |
104 | } | |
105 | ||
106 | /* Primary Raid Level (PRL) */ | |
107 | #define DDF_RAID0 0x00 | |
108 | #define DDF_RAID1 0x01 | |
109 | #define DDF_RAID3 0x03 | |
110 | #define DDF_RAID4 0x04 | |
111 | #define DDF_RAID5 0x05 | |
112 | #define DDF_RAID1E 0x11 | |
113 | #define DDF_JBOD 0x0f | |
114 | #define DDF_CONCAT 0x1f | |
115 | #define DDF_RAID5E 0x15 | |
116 | #define DDF_RAID5EE 0x25 | |
117 | #define DDF_RAID6 0x06 | |
118 | ||
119 | /* Raid Level Qualifier (RLQ) */ | |
120 | #define DDF_RAID0_SIMPLE 0x00 | |
121 | #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */ | |
122 | #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */ | |
123 | #define DDF_RAID3_0 0x00 /* parity in first extent */ | |
124 | #define DDF_RAID3_N 0x01 /* parity in last extent */ | |
125 | #define DDF_RAID4_0 0x00 /* parity in first extent */ | |
126 | #define DDF_RAID4_N 0x01 /* parity in last extent */ | |
127 | /* these apply to raid5e and raid5ee as well */ | |
128 | #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */ | |
129 | #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */ | |
130 | #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */ | |
131 | #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */ | |
132 | ||
133 | #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */ | |
134 | #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */ | |
135 | ||
136 | /* Secondary RAID Level (SRL) */ | |
137 | #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */ | |
138 | #define DDF_2MIRRORED 0x01 | |
139 | #define DDF_2CONCAT 0x02 | |
140 | #define DDF_2SPANNED 0x03 /* This is also weird - be careful */ | |
141 | ||
142 | /* Magic numbers */ | |
143 | #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11) | |
144 | #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111) | |
145 | #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222) | |
146 | #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333) | |
147 | #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD) | |
148 | #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE) | |
149 | #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555) | |
150 | #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888) | |
151 | #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0) | |
152 | #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C) | |
153 | ||
154 | #define DDF_GUID_LEN 24 | |
155 | #define DDF_REVISION_0 "01.00.00" | |
156 | #define DDF_REVISION_2 "01.02.00" | |
157 | ||
158 | struct ddf_header { | |
159 | be32 magic; /* DDF_HEADER_MAGIC */ | |
160 | be32 crc; | |
161 | char guid[DDF_GUID_LEN]; | |
162 | char revision[8]; /* 01.02.00 */ | |
163 | be32 seq; /* starts at '1' */ | |
164 | be32 timestamp; | |
165 | __u8 openflag; | |
166 | __u8 foreignflag; | |
167 | __u8 enforcegroups; | |
168 | __u8 pad0; /* 0xff */ | |
169 | __u8 pad1[12]; /* 12 * 0xff */ | |
170 | /* 64 bytes so far */ | |
171 | __u8 header_ext[32]; /* reserved: fill with 0xff */ | |
172 | be64 primary_lba; | |
173 | be64 secondary_lba; | |
174 | __u8 type; | |
175 | __u8 pad2[3]; /* 0xff */ | |
176 | be32 workspace_len; /* sectors for vendor space - | |
177 | * at least 32768(sectors) */ | |
178 | be64 workspace_lba; | |
179 | be16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */ | |
180 | be16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */ | |
181 | be16 max_partitions; /* i.e. max num of configuration | |
182 | record entries per disk */ | |
183 | be16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries | |
184 | *12/512) */ | |
185 | be16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */ | |
186 | __u8 pad3[54]; /* 0xff */ | |
187 | /* 192 bytes so far */ | |
188 | be32 controller_section_offset; | |
189 | be32 controller_section_length; | |
190 | be32 phys_section_offset; | |
191 | be32 phys_section_length; | |
192 | be32 virt_section_offset; | |
193 | be32 virt_section_length; | |
194 | be32 config_section_offset; | |
195 | be32 config_section_length; | |
196 | be32 data_section_offset; | |
197 | be32 data_section_length; | |
198 | be32 bbm_section_offset; | |
199 | be32 bbm_section_length; | |
200 | be32 diag_space_offset; | |
201 | be32 diag_space_length; | |
202 | be32 vendor_offset; | |
203 | be32 vendor_length; | |
204 | /* 256 bytes so far */ | |
205 | __u8 pad4[256]; /* 0xff */ | |
206 | }; | |
207 | ||
208 | /* type field */ | |
209 | #define DDF_HEADER_ANCHOR 0x00 | |
210 | #define DDF_HEADER_PRIMARY 0x01 | |
211 | #define DDF_HEADER_SECONDARY 0x02 | |
212 | ||
213 | /* The content of the 'controller section' - global scope */ | |
214 | struct ddf_controller_data { | |
215 | be32 magic; /* DDF_CONTROLLER_MAGIC */ | |
216 | be32 crc; | |
217 | char guid[DDF_GUID_LEN]; | |
218 | struct controller_type { | |
219 | be16 vendor_id; | |
220 | be16 device_id; | |
221 | be16 sub_vendor_id; | |
222 | be16 sub_device_id; | |
223 | } type; | |
224 | char product_id[16]; | |
225 | __u8 pad[8]; /* 0xff */ | |
226 | __u8 vendor_data[448]; | |
227 | }; | |
228 | ||
229 | /* The content of phys_section - global scope */ | |
230 | struct phys_disk { | |
231 | be32 magic; /* DDF_PHYS_RECORDS_MAGIC */ | |
232 | be32 crc; | |
233 | be16 used_pdes; /* This is a counter, not a max - the list | |
234 | * of used entries may not be dense */ | |
235 | be16 max_pdes; | |
236 | __u8 pad[52]; | |
237 | struct phys_disk_entry { | |
238 | char guid[DDF_GUID_LEN]; | |
239 | be32 refnum; | |
240 | be16 type; | |
241 | be16 state; | |
242 | be64 config_size; /* DDF structures must be after here */ | |
243 | char path[18]; /* Another horrible structure really | |
244 | * but is "used for information | |
245 | * purposes only" */ | |
246 | __u8 pad[6]; | |
247 | } entries[0]; | |
248 | }; | |
249 | ||
250 | /* phys_disk_entry.type is a bitmap - bigendian remember */ | |
251 | #define DDF_Forced_PD_GUID 1 | |
252 | #define DDF_Active_in_VD 2 | |
253 | #define DDF_Global_Spare 4 /* VD_CONF records are ignored */ | |
254 | #define DDF_Spare 8 /* overrides Global_spare */ | |
255 | #define DDF_Foreign 16 | |
256 | #define DDF_Legacy 32 /* no DDF on this device */ | |
257 | ||
258 | #define DDF_Interface_mask 0xf00 | |
259 | #define DDF_Interface_SCSI 0x100 | |
260 | #define DDF_Interface_SAS 0x200 | |
261 | #define DDF_Interface_SATA 0x300 | |
262 | #define DDF_Interface_FC 0x400 | |
263 | ||
264 | /* phys_disk_entry.state is a bigendian bitmap */ | |
265 | #define DDF_Online 1 | |
266 | #define DDF_Failed 2 /* overrides 1,4,8 */ | |
267 | #define DDF_Rebuilding 4 | |
268 | #define DDF_Transition 8 | |
269 | #define DDF_SMART 16 | |
270 | #define DDF_ReadErrors 32 | |
271 | #define DDF_Missing 64 | |
272 | ||
273 | /* The content of the virt_section global scope */ | |
274 | struct virtual_disk { | |
275 | be32 magic; /* DDF_VIRT_RECORDS_MAGIC */ | |
276 | be32 crc; | |
277 | be16 populated_vdes; | |
278 | be16 max_vdes; | |
279 | __u8 pad[52]; | |
280 | struct virtual_entry { | |
281 | char guid[DDF_GUID_LEN]; | |
282 | be16 unit; | |
283 | __u16 pad0; /* 0xffff */ | |
284 | be16 guid_crc; | |
285 | be16 type; | |
286 | __u8 state; | |
287 | __u8 init_state; | |
288 | __u8 pad1[14]; | |
289 | char name[16]; | |
290 | } entries[0]; | |
291 | }; | |
292 | ||
293 | /* virtual_entry.type is a bitmap - bigendian */ | |
294 | #define DDF_Shared 1 | |
295 | #define DDF_Enforce_Groups 2 | |
296 | #define DDF_Unicode 4 | |
297 | #define DDF_Owner_Valid 8 | |
298 | ||
299 | /* virtual_entry.state is a bigendian bitmap */ | |
300 | #define DDF_state_mask 0x7 | |
301 | #define DDF_state_optimal 0x0 | |
302 | #define DDF_state_degraded 0x1 | |
303 | #define DDF_state_deleted 0x2 | |
304 | #define DDF_state_missing 0x3 | |
305 | #define DDF_state_failed 0x4 | |
306 | #define DDF_state_part_optimal 0x5 | |
307 | ||
308 | #define DDF_state_morphing 0x8 | |
309 | #define DDF_state_inconsistent 0x10 | |
310 | ||
311 | /* virtual_entry.init_state is a bigendian bitmap */ | |
312 | #define DDF_initstate_mask 0x03 | |
313 | #define DDF_init_not 0x00 | |
314 | #define DDF_init_quick 0x01 /* initialisation is progress. | |
315 | * i.e. 'state_inconsistent' */ | |
316 | #define DDF_init_full 0x02 | |
317 | ||
318 | #define DDF_access_mask 0xc0 | |
319 | #define DDF_access_rw 0x00 | |
320 | #define DDF_access_ro 0x80 | |
321 | #define DDF_access_blocked 0xc0 | |
322 | ||
323 | /* The content of the config_section - local scope | |
324 | * It has multiple records each config_record_len sectors | |
325 | * They can be vd_config or spare_assign | |
326 | */ | |
327 | ||
328 | struct vd_config { | |
329 | be32 magic; /* DDF_VD_CONF_MAGIC */ | |
330 | be32 crc; | |
331 | char guid[DDF_GUID_LEN]; | |
332 | be32 timestamp; | |
333 | be32 seqnum; | |
334 | __u8 pad0[24]; | |
335 | be16 prim_elmnt_count; | |
336 | __u8 chunk_shift; /* 0 == 512, 1==1024 etc */ | |
337 | __u8 prl; | |
338 | __u8 rlq; | |
339 | __u8 sec_elmnt_count; | |
340 | __u8 sec_elmnt_seq; | |
341 | __u8 srl; | |
342 | be64 blocks; /* blocks per component could be different | |
343 | * on different component devices...(only | |
344 | * for concat I hope) */ | |
345 | be64 array_blocks; /* blocks in array */ | |
346 | __u8 pad1[8]; | |
347 | be32 spare_refs[8]; /* This is used to detect missing spares. | |
348 | * As we don't have an interface for that | |
349 | * the values are ignored. | |
350 | */ | |
351 | __u8 cache_pol[8]; | |
352 | __u8 bg_rate; | |
353 | __u8 pad2[3]; | |
354 | __u8 pad3[52]; | |
355 | __u8 pad4[192]; | |
356 | __u8 v0[32]; /* reserved- 0xff */ | |
357 | __u8 v1[32]; /* reserved- 0xff */ | |
358 | __u8 v2[16]; /* reserved- 0xff */ | |
359 | __u8 v3[16]; /* reserved- 0xff */ | |
360 | __u8 vendor[32]; | |
361 | be32 phys_refnum[0]; /* refnum of each disk in sequence */ | |
362 | /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a | |
363 | bvd are always the same size */ | |
364 | }; | |
365 | #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe]) | |
366 | ||
367 | /* vd_config.cache_pol[7] is a bitmap */ | |
368 | #define DDF_cache_writeback 1 /* else writethrough */ | |
369 | #define DDF_cache_wadaptive 2 /* only applies if writeback */ | |
370 | #define DDF_cache_readahead 4 | |
371 | #define DDF_cache_radaptive 8 /* only if doing read-ahead */ | |
372 | #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */ | |
373 | #define DDF_cache_wallowed 32 /* enable write caching */ | |
374 | #define DDF_cache_rallowed 64 /* enable read caching */ | |
375 | ||
376 | struct spare_assign { | |
377 | be32 magic; /* DDF_SPARE_ASSIGN_MAGIC */ | |
378 | be32 crc; | |
379 | be32 timestamp; | |
380 | __u8 reserved[7]; | |
381 | __u8 type; | |
382 | be16 populated; /* SAEs used */ | |
383 | be16 max; /* max SAEs */ | |
384 | __u8 pad[8]; | |
385 | struct spare_assign_entry { | |
386 | char guid[DDF_GUID_LEN]; | |
387 | be16 secondary_element; | |
388 | __u8 pad[6]; | |
389 | } spare_ents[0]; | |
390 | }; | |
391 | /* spare_assign.type is a bitmap */ | |
392 | #define DDF_spare_dedicated 0x1 /* else global */ | |
393 | #define DDF_spare_revertible 0x2 /* else committable */ | |
394 | #define DDF_spare_active 0x4 /* else not active */ | |
395 | #define DDF_spare_affinity 0x8 /* enclosure affinity */ | |
396 | ||
397 | /* The data_section contents - local scope */ | |
398 | struct disk_data { | |
399 | be32 magic; /* DDF_PHYS_DATA_MAGIC */ | |
400 | be32 crc; | |
401 | char guid[DDF_GUID_LEN]; | |
402 | be32 refnum; /* crc of some magic drive data ... */ | |
403 | __u8 forced_ref; /* set when above was not result of magic */ | |
404 | __u8 forced_guid; /* set if guid was forced rather than magic */ | |
405 | __u8 vendor[32]; | |
406 | __u8 pad[442]; | |
407 | }; | |
408 | ||
409 | /* bbm_section content */ | |
410 | struct bad_block_log { | |
411 | be32 magic; | |
412 | be32 crc; | |
413 | be16 entry_count; | |
414 | be32 spare_count; | |
415 | __u8 pad[10]; | |
416 | be64 first_spare; | |
417 | struct mapped_block { | |
418 | be64 defective_start; | |
419 | be32 replacement_start; | |
420 | be16 remap_count; | |
421 | __u8 pad[2]; | |
422 | } entries[0]; | |
423 | }; | |
424 | ||
425 | /* Struct for internally holding ddf structures */ | |
426 | /* The DDF structure stored on each device is potentially | |
427 | * quite different, as some data is global and some is local. | |
428 | * The global data is: | |
429 | * - ddf header | |
430 | * - controller_data | |
431 | * - Physical disk records | |
432 | * - Virtual disk records | |
433 | * The local data is: | |
434 | * - Configuration records | |
435 | * - Physical Disk data section | |
436 | * ( and Bad block and vendor which I don't care about yet). | |
437 | * | |
438 | * The local data is parsed into separate lists as it is read | |
439 | * and reconstructed for writing. This means that we only need | |
440 | * to make config changes once and they are automatically | |
441 | * propagated to all devices. | |
442 | * The global (config and disk data) records are each in a list | |
443 | * of separate data structures. When writing we find the entry | |
444 | * or entries applicable to the particular device. | |
445 | */ | |
446 | struct ddf_super { | |
447 | struct ddf_header anchor, primary, secondary; | |
448 | struct ddf_controller_data controller; | |
449 | struct ddf_header *active; | |
450 | struct phys_disk *phys; | |
451 | struct virtual_disk *virt; | |
452 | char *conf; | |
453 | int pdsize, vdsize; | |
454 | unsigned int max_part, mppe, conf_rec_len; | |
455 | int currentdev; | |
456 | int updates_pending; | |
457 | struct vcl { | |
458 | union { | |
459 | char space[512]; | |
460 | struct { | |
461 | struct vcl *next; | |
462 | unsigned int vcnum; /* index into ->virt */ | |
463 | /* For an array with a secondary level there are | |
464 | * multiple vd_config structures, all with the same | |
465 | * guid but with different sec_elmnt_seq. | |
466 | * One of these structures is in 'conf' below. | |
467 | * The others are in other_bvds, not in any | |
468 | * particular order. | |
469 | */ | |
470 | struct vd_config **other_bvds; | |
471 | __u64 *block_sizes; /* NULL if all the same */ | |
472 | }; | |
473 | }; | |
474 | struct vd_config conf; | |
475 | } *conflist, *currentconf; | |
476 | struct dl { | |
477 | union { | |
478 | char space[512]; | |
479 | struct { | |
480 | struct dl *next; | |
481 | int major, minor; | |
482 | char *devname; | |
483 | int fd; | |
484 | unsigned long long size; /* sectors */ | |
485 | be64 primary_lba; /* sectors */ | |
486 | be64 secondary_lba; /* sectors */ | |
487 | be64 workspace_lba; /* sectors */ | |
488 | int pdnum; /* index in ->phys */ | |
489 | struct spare_assign *spare; | |
490 | void *mdupdate; /* hold metadata update */ | |
491 | ||
492 | /* These fields used by auto-layout */ | |
493 | int raiddisk; /* slot to fill in autolayout */ | |
494 | __u64 esize; | |
495 | int displayed; | |
496 | }; | |
497 | }; | |
498 | struct disk_data disk; | |
499 | struct vcl *vlist[0]; /* max_part in size */ | |
500 | } *dlist, *add_list; | |
501 | }; | |
502 | ||
503 | static int load_super_ddf_all(struct supertype *st, int fd, | |
504 | void **sbp, char *devname); | |
505 | static int get_svd_state(const struct ddf_super *, const struct vcl *); | |
506 | static int | |
507 | validate_geometry_ddf_container(struct supertype *st, | |
508 | int level, int layout, int raiddisks, | |
509 | int chunk, unsigned long long size, | |
510 | unsigned long long data_offset, | |
511 | char *dev, unsigned long long *freesize, | |
512 | int verbose); | |
513 | ||
514 | static int validate_geometry_ddf_bvd(struct supertype *st, | |
515 | int level, int layout, int raiddisks, | |
516 | int *chunk, unsigned long long size, | |
517 | unsigned long long data_offset, | |
518 | char *dev, unsigned long long *freesize, | |
519 | int verbose); | |
520 | ||
521 | static void free_super_ddf(struct supertype *st); | |
522 | static int all_ff(const char *guid); | |
523 | static unsigned int get_pd_index_from_refnum(const struct vcl *vc, | |
524 | be32 refnum, unsigned int nmax, | |
525 | const struct vd_config **bvd, | |
526 | unsigned int *idx); | |
527 | static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map); | |
528 | static void uuid_from_ddf_guid(const char *guid, int uuid[4]); | |
529 | static void uuid_from_super_ddf(struct supertype *st, int uuid[4]); | |
530 | static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i); | |
531 | static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map); | |
532 | static int init_super_ddf_bvd(struct supertype *st, | |
533 | mdu_array_info_t *info, | |
534 | unsigned long long size, | |
535 | char *name, char *homehost, | |
536 | int *uuid, unsigned long long data_offset); | |
537 | ||
538 | #if DEBUG | |
539 | static void pr_state(struct ddf_super *ddf, const char *msg) | |
540 | { | |
541 | unsigned int i; | |
542 | dprintf("%s: ", msg); | |
543 | for (i = 0; i < be16_to_cpu(ddf->active->max_vd_entries); i++) { | |
544 | if (all_ff(ddf->virt->entries[i].guid)) | |
545 | continue; | |
546 | dprintf_cont("%u(s=%02x i=%02x) ", i, | |
547 | ddf->virt->entries[i].state, | |
548 | ddf->virt->entries[i].init_state); | |
549 | } | |
550 | dprintf_cont("\n"); | |
551 | } | |
552 | #else | |
553 | static void pr_state(const struct ddf_super *ddf, const char *msg) {} | |
554 | #endif | |
555 | ||
556 | static void _ddf_set_updates_pending(struct ddf_super *ddf, struct vd_config *vc, | |
557 | const char *func) | |
558 | { | |
559 | if (vc) { | |
560 | vc->timestamp = cpu_to_be32(time(0)-DECADE); | |
561 | vc->seqnum = cpu_to_be32(be32_to_cpu(vc->seqnum) + 1); | |
562 | } | |
563 | if (ddf->updates_pending) | |
564 | return; | |
565 | ddf->updates_pending = 1; | |
566 | ddf->active->seq = cpu_to_be32((be32_to_cpu(ddf->active->seq)+1)); | |
567 | pr_state(ddf, func); | |
568 | } | |
569 | ||
570 | #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__) | |
571 | ||
572 | static be32 calc_crc(void *buf, int len) | |
573 | { | |
574 | /* crcs are always at the same place as in the ddf_header */ | |
575 | struct ddf_header *ddf = buf; | |
576 | be32 oldcrc = ddf->crc; | |
577 | __u32 newcrc; | |
578 | ddf->crc = cpu_to_be32(0xffffffff); | |
579 | ||
580 | newcrc = crc32(0, buf, len); | |
581 | ddf->crc = oldcrc; | |
582 | /* The crc is stored (like everything) bigendian, so convert | |
583 | * here for simplicity | |
584 | */ | |
585 | return cpu_to_be32(newcrc); | |
586 | } | |
587 | ||
588 | #define DDF_INVALID_LEVEL 0xff | |
589 | #define DDF_NO_SECONDARY 0xff | |
590 | static int err_bad_md_layout(const mdu_array_info_t *array) | |
591 | { | |
592 | pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n", | |
593 | array->level, array->layout, array->raid_disks); | |
594 | return -1; | |
595 | } | |
596 | ||
597 | static int layout_md2ddf(const mdu_array_info_t *array, | |
598 | struct vd_config *conf) | |
599 | { | |
600 | be16 prim_elmnt_count = cpu_to_be16(array->raid_disks); | |
601 | __u8 prl = DDF_INVALID_LEVEL, rlq = 0; | |
602 | __u8 sec_elmnt_count = 1; | |
603 | __u8 srl = DDF_NO_SECONDARY; | |
604 | ||
605 | switch (array->level) { | |
606 | case LEVEL_LINEAR: | |
607 | prl = DDF_CONCAT; | |
608 | break; | |
609 | case 0: | |
610 | rlq = DDF_RAID0_SIMPLE; | |
611 | prl = DDF_RAID0; | |
612 | break; | |
613 | case 1: | |
614 | switch (array->raid_disks) { | |
615 | case 2: | |
616 | rlq = DDF_RAID1_SIMPLE; | |
617 | break; | |
618 | case 3: | |
619 | rlq = DDF_RAID1_MULTI; | |
620 | break; | |
621 | default: | |
622 | return err_bad_md_layout(array); | |
623 | } | |
624 | prl = DDF_RAID1; | |
625 | break; | |
626 | case 4: | |
627 | if (array->layout != 0) | |
628 | return err_bad_md_layout(array); | |
629 | rlq = DDF_RAID4_N; | |
630 | prl = DDF_RAID4; | |
631 | break; | |
632 | case 5: | |
633 | switch (array->layout) { | |
634 | case ALGORITHM_LEFT_ASYMMETRIC: | |
635 | rlq = DDF_RAID5_N_RESTART; | |
636 | break; | |
637 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
638 | rlq = DDF_RAID5_0_RESTART; | |
639 | break; | |
640 | case ALGORITHM_LEFT_SYMMETRIC: | |
641 | rlq = DDF_RAID5_N_CONTINUE; | |
642 | break; | |
643 | case ALGORITHM_RIGHT_SYMMETRIC: | |
644 | /* not mentioned in standard */ | |
645 | default: | |
646 | return err_bad_md_layout(array); | |
647 | } | |
648 | prl = DDF_RAID5; | |
649 | break; | |
650 | case 6: | |
651 | switch (array->layout) { | |
652 | case ALGORITHM_ROTATING_N_RESTART: | |
653 | rlq = DDF_RAID5_N_RESTART; | |
654 | break; | |
655 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
656 | rlq = DDF_RAID6_0_RESTART; | |
657 | break; | |
658 | case ALGORITHM_ROTATING_N_CONTINUE: | |
659 | rlq = DDF_RAID5_N_CONTINUE; | |
660 | break; | |
661 | default: | |
662 | return err_bad_md_layout(array); | |
663 | } | |
664 | prl = DDF_RAID6; | |
665 | break; | |
666 | case 10: | |
667 | if (array->raid_disks % 2 == 0 && array->layout == 0x102) { | |
668 | rlq = DDF_RAID1_SIMPLE; | |
669 | prim_elmnt_count = cpu_to_be16(2); | |
670 | sec_elmnt_count = array->raid_disks / 2; | |
671 | srl = DDF_2SPANNED; | |
672 | prl = DDF_RAID1; | |
673 | } else if (array->raid_disks % 3 == 0 && | |
674 | array->layout == 0x103) { | |
675 | rlq = DDF_RAID1_MULTI; | |
676 | prim_elmnt_count = cpu_to_be16(3); | |
677 | sec_elmnt_count = array->raid_disks / 3; | |
678 | srl = DDF_2SPANNED; | |
679 | prl = DDF_RAID1; | |
680 | } else if (array->layout == 0x201) { | |
681 | prl = DDF_RAID1E; | |
682 | rlq = DDF_RAID1E_OFFSET; | |
683 | } else if (array->layout == 0x102) { | |
684 | prl = DDF_RAID1E; | |
685 | rlq = DDF_RAID1E_ADJACENT; | |
686 | } else | |
687 | return err_bad_md_layout(array); | |
688 | break; | |
689 | default: | |
690 | return err_bad_md_layout(array); | |
691 | } | |
692 | conf->prl = prl; | |
693 | conf->prim_elmnt_count = prim_elmnt_count; | |
694 | conf->rlq = rlq; | |
695 | conf->srl = srl; | |
696 | conf->sec_elmnt_count = sec_elmnt_count; | |
697 | return 0; | |
698 | } | |
699 | ||
700 | static int err_bad_ddf_layout(const struct vd_config *conf) | |
701 | { | |
702 | pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n", | |
703 | conf->prl, conf->rlq, be16_to_cpu(conf->prim_elmnt_count)); | |
704 | return -1; | |
705 | } | |
706 | ||
707 | static int layout_ddf2md(const struct vd_config *conf, | |
708 | mdu_array_info_t *array) | |
709 | { | |
710 | int level = LEVEL_UNSUPPORTED; | |
711 | int layout = 0; | |
712 | int raiddisks = be16_to_cpu(conf->prim_elmnt_count); | |
713 | ||
714 | if (conf->sec_elmnt_count > 1) { | |
715 | /* see also check_secondary() */ | |
716 | if (conf->prl != DDF_RAID1 || | |
717 | (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED)) { | |
718 | pr_err("Unsupported secondary RAID level %u/%u\n", | |
719 | conf->prl, conf->srl); | |
720 | return -1; | |
721 | } | |
722 | if (raiddisks == 2 && conf->rlq == DDF_RAID1_SIMPLE) | |
723 | layout = 0x102; | |
724 | else if (raiddisks == 3 && conf->rlq == DDF_RAID1_MULTI) | |
725 | layout = 0x103; | |
726 | else | |
727 | return err_bad_ddf_layout(conf); | |
728 | raiddisks *= conf->sec_elmnt_count; | |
729 | level = 10; | |
730 | goto good; | |
731 | } | |
732 | ||
733 | switch (conf->prl) { | |
734 | case DDF_CONCAT: | |
735 | level = LEVEL_LINEAR; | |
736 | break; | |
737 | case DDF_RAID0: | |
738 | if (conf->rlq != DDF_RAID0_SIMPLE) | |
739 | return err_bad_ddf_layout(conf); | |
740 | level = 0; | |
741 | break; | |
742 | case DDF_RAID1: | |
743 | if (!((conf->rlq == DDF_RAID1_SIMPLE && raiddisks == 2) || | |
744 | (conf->rlq == DDF_RAID1_MULTI && raiddisks == 3))) | |
745 | return err_bad_ddf_layout(conf); | |
746 | level = 1; | |
747 | break; | |
748 | case DDF_RAID1E: | |
749 | if (conf->rlq == DDF_RAID1E_ADJACENT) | |
750 | layout = 0x102; | |
751 | else if (conf->rlq == DDF_RAID1E_OFFSET) | |
752 | layout = 0x201; | |
753 | else | |
754 | return err_bad_ddf_layout(conf); | |
755 | level = 10; | |
756 | break; | |
757 | case DDF_RAID4: | |
758 | if (conf->rlq != DDF_RAID4_N) | |
759 | return err_bad_ddf_layout(conf); | |
760 | level = 4; | |
761 | break; | |
762 | case DDF_RAID5: | |
763 | switch (conf->rlq) { | |
764 | case DDF_RAID5_N_RESTART: | |
765 | layout = ALGORITHM_LEFT_ASYMMETRIC; | |
766 | break; | |
767 | case DDF_RAID5_0_RESTART: | |
768 | layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
769 | break; | |
770 | case DDF_RAID5_N_CONTINUE: | |
771 | layout = ALGORITHM_LEFT_SYMMETRIC; | |
772 | break; | |
773 | default: | |
774 | return err_bad_ddf_layout(conf); | |
775 | } | |
776 | level = 5; | |
777 | break; | |
778 | case DDF_RAID6: | |
779 | switch (conf->rlq) { | |
780 | case DDF_RAID5_N_RESTART: | |
781 | layout = ALGORITHM_ROTATING_N_RESTART; | |
782 | break; | |
783 | case DDF_RAID6_0_RESTART: | |
784 | layout = ALGORITHM_ROTATING_ZERO_RESTART; | |
785 | break; | |
786 | case DDF_RAID5_N_CONTINUE: | |
787 | layout = ALGORITHM_ROTATING_N_CONTINUE; | |
788 | break; | |
789 | default: | |
790 | return err_bad_ddf_layout(conf); | |
791 | } | |
792 | level = 6; | |
793 | break; | |
794 | default: | |
795 | return err_bad_ddf_layout(conf); | |
796 | }; | |
797 | ||
798 | good: | |
799 | array->level = level; | |
800 | array->layout = layout; | |
801 | array->raid_disks = raiddisks; | |
802 | return 0; | |
803 | } | |
804 | ||
805 | static int load_ddf_header(int fd, unsigned long long lba, | |
806 | unsigned long long size, | |
807 | int type, | |
808 | struct ddf_header *hdr, struct ddf_header *anchor) | |
809 | { | |
810 | /* read a ddf header (primary or secondary) from fd/lba | |
811 | * and check that it is consistent with anchor | |
812 | * Need to check: | |
813 | * magic, crc, guid, rev, and LBA's header_type, and | |
814 | * everything after header_type must be the same | |
815 | */ | |
816 | if (lba >= size-1) | |
817 | return 0; | |
818 | ||
819 | if (lseek64(fd, lba<<9, 0) < 0) | |
820 | return 0; | |
821 | ||
822 | if (read(fd, hdr, 512) != 512) | |
823 | return 0; | |
824 | ||
825 | if (!be32_eq(hdr->magic, DDF_HEADER_MAGIC)) { | |
826 | pr_err("bad header magic\n"); | |
827 | return 0; | |
828 | } | |
829 | if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) { | |
830 | pr_err("bad CRC\n"); | |
831 | return 0; | |
832 | } | |
833 | if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 || | |
834 | memcmp(anchor->revision, hdr->revision, 8) != 0 || | |
835 | !be64_eq(anchor->primary_lba, hdr->primary_lba) || | |
836 | !be64_eq(anchor->secondary_lba, hdr->secondary_lba) || | |
837 | hdr->type != type || | |
838 | memcmp(anchor->pad2, hdr->pad2, 512 - | |
839 | offsetof(struct ddf_header, pad2)) != 0) { | |
840 | pr_err("header mismatch\n"); | |
841 | return 0; | |
842 | } | |
843 | ||
844 | /* Looks good enough to me... */ | |
845 | return 1; | |
846 | } | |
847 | ||
848 | static void *load_section(int fd, struct ddf_super *super, void *buf, | |
849 | be32 offset_be, be32 len_be, int check) | |
850 | { | |
851 | unsigned long long offset = be32_to_cpu(offset_be); | |
852 | unsigned long long len = be32_to_cpu(len_be); | |
853 | int dofree = (buf == NULL); | |
854 | ||
855 | if (check) | |
856 | if (len != 2 && len != 8 && len != 32 && | |
857 | len != 128 && len != 512) | |
858 | return NULL; | |
859 | ||
860 | if (len > 1024) | |
861 | return NULL; | |
862 | if (!buf && posix_memalign(&buf, 512, len<<9) != 0) | |
863 | buf = NULL; | |
864 | ||
865 | if (!buf) | |
866 | return NULL; | |
867 | ||
868 | if (super->active->type == 1) | |
869 | offset += be64_to_cpu(super->active->primary_lba); | |
870 | else | |
871 | offset += be64_to_cpu(super->active->secondary_lba); | |
872 | ||
873 | if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) { | |
874 | if (dofree) | |
875 | free(buf); | |
876 | return NULL; | |
877 | } | |
878 | if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) { | |
879 | if (dofree) | |
880 | free(buf); | |
881 | return NULL; | |
882 | } | |
883 | return buf; | |
884 | } | |
885 | ||
886 | static int load_ddf_headers(int fd, struct ddf_super *super, char *devname) | |
887 | { | |
888 | unsigned long long dsize; | |
889 | ||
890 | get_dev_size(fd, NULL, &dsize); | |
891 | ||
892 | if (lseek64(fd, dsize-512, 0) < 0) { | |
893 | if (devname) | |
894 | pr_err("Cannot seek to anchor block on %s: %s\n", | |
895 | devname, strerror(errno)); | |
896 | return 1; | |
897 | } | |
898 | if (read(fd, &super->anchor, 512) != 512) { | |
899 | if (devname) | |
900 | pr_err("Cannot read anchor block on %s: %s\n", | |
901 | devname, strerror(errno)); | |
902 | return 1; | |
903 | } | |
904 | if (!be32_eq(super->anchor.magic, DDF_HEADER_MAGIC)) { | |
905 | if (devname) | |
906 | pr_err("no DDF anchor found on %s\n", | |
907 | devname); | |
908 | return 2; | |
909 | } | |
910 | if (!be32_eq(calc_crc(&super->anchor, 512), super->anchor.crc)) { | |
911 | if (devname) | |
912 | pr_err("bad CRC on anchor on %s\n", | |
913 | devname); | |
914 | return 2; | |
915 | } | |
916 | if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 && | |
917 | memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) { | |
918 | if (devname) | |
919 | pr_err("can only support super revision %.8s and earlier, not %.8s on %s\n", | |
920 | DDF_REVISION_2, super->anchor.revision,devname); | |
921 | return 2; | |
922 | } | |
923 | super->active = NULL; | |
924 | if (load_ddf_header(fd, be64_to_cpu(super->anchor.primary_lba), | |
925 | dsize >> 9, 1, | |
926 | &super->primary, &super->anchor) == 0) { | |
927 | if (devname) | |
928 | pr_err("Failed to load primary DDF header on %s\n", devname); | |
929 | } else | |
930 | super->active = &super->primary; | |
931 | ||
932 | if (load_ddf_header(fd, be64_to_cpu(super->anchor.secondary_lba), | |
933 | dsize >> 9, 2, | |
934 | &super->secondary, &super->anchor)) { | |
935 | if (super->active == NULL || | |
936 | (be32_to_cpu(super->primary.seq) | |
937 | < be32_to_cpu(super->secondary.seq) && | |
938 | !super->secondary.openflag) || | |
939 | (be32_to_cpu(super->primary.seq) == | |
940 | be32_to_cpu(super->secondary.seq) && | |
941 | super->primary.openflag && !super->secondary.openflag)) | |
942 | super->active = &super->secondary; | |
943 | } else if (devname && | |
944 | be64_to_cpu(super->anchor.secondary_lba) != ~(__u64)0) | |
945 | pr_err("Failed to load secondary DDF header on %s\n", | |
946 | devname); | |
947 | if (super->active == NULL) | |
948 | return 2; | |
949 | return 0; | |
950 | } | |
951 | ||
952 | static int load_ddf_global(int fd, struct ddf_super *super, char *devname) | |
953 | { | |
954 | void *ok; | |
955 | ok = load_section(fd, super, &super->controller, | |
956 | super->active->controller_section_offset, | |
957 | super->active->controller_section_length, | |
958 | 0); | |
959 | super->phys = load_section(fd, super, NULL, | |
960 | super->active->phys_section_offset, | |
961 | super->active->phys_section_length, | |
962 | 1); | |
963 | super->pdsize = be32_to_cpu(super->active->phys_section_length) * 512; | |
964 | ||
965 | super->virt = load_section(fd, super, NULL, | |
966 | super->active->virt_section_offset, | |
967 | super->active->virt_section_length, | |
968 | 1); | |
969 | super->vdsize = be32_to_cpu(super->active->virt_section_length) * 512; | |
970 | if (!ok || | |
971 | !super->phys || | |
972 | !super->virt) { | |
973 | free(super->phys); | |
974 | free(super->virt); | |
975 | super->phys = NULL; | |
976 | super->virt = NULL; | |
977 | return 2; | |
978 | } | |
979 | super->conflist = NULL; | |
980 | super->dlist = NULL; | |
981 | ||
982 | super->max_part = be16_to_cpu(super->active->max_partitions); | |
983 | super->mppe = be16_to_cpu(super->active->max_primary_element_entries); | |
984 | super->conf_rec_len = be16_to_cpu(super->active->config_record_len); | |
985 | return 0; | |
986 | } | |
987 | ||
988 | #define DDF_UNUSED_BVD 0xff | |
989 | static int alloc_other_bvds(const struct ddf_super *ddf, struct vcl *vcl) | |
990 | { | |
991 | unsigned int n_vds = vcl->conf.sec_elmnt_count - 1; | |
992 | unsigned int i, vdsize; | |
993 | void *p; | |
994 | if (n_vds == 0) { | |
995 | vcl->other_bvds = NULL; | |
996 | return 0; | |
997 | } | |
998 | vdsize = ddf->conf_rec_len * 512; | |
999 | if (posix_memalign(&p, 512, n_vds * | |
1000 | (vdsize + sizeof(struct vd_config *))) != 0) | |
1001 | return -1; | |
1002 | vcl->other_bvds = (struct vd_config **) (p + n_vds * vdsize); | |
1003 | for (i = 0; i < n_vds; i++) { | |
1004 | vcl->other_bvds[i] = p + i * vdsize; | |
1005 | memset(vcl->other_bvds[i], 0, vdsize); | |
1006 | vcl->other_bvds[i]->sec_elmnt_seq = DDF_UNUSED_BVD; | |
1007 | } | |
1008 | return 0; | |
1009 | } | |
1010 | ||
1011 | static void add_other_bvd(struct vcl *vcl, struct vd_config *vd, | |
1012 | unsigned int len) | |
1013 | { | |
1014 | int i; | |
1015 | for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++) | |
1016 | if (vcl->other_bvds[i]->sec_elmnt_seq == vd->sec_elmnt_seq) | |
1017 | break; | |
1018 | ||
1019 | if (i < vcl->conf.sec_elmnt_count-1) { | |
1020 | if (be32_to_cpu(vd->seqnum) <= | |
1021 | be32_to_cpu(vcl->other_bvds[i]->seqnum)) | |
1022 | return; | |
1023 | } else { | |
1024 | for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++) | |
1025 | if (vcl->other_bvds[i]->sec_elmnt_seq == DDF_UNUSED_BVD) | |
1026 | break; | |
1027 | if (i == vcl->conf.sec_elmnt_count-1) { | |
1028 | pr_err("no space for sec level config %u, count is %u\n", | |
1029 | vd->sec_elmnt_seq, vcl->conf.sec_elmnt_count); | |
1030 | return; | |
1031 | } | |
1032 | } | |
1033 | memcpy(vcl->other_bvds[i], vd, len); | |
1034 | } | |
1035 | ||
1036 | static int load_ddf_local(int fd, struct ddf_super *super, | |
1037 | char *devname, int keep) | |
1038 | { | |
1039 | struct dl *dl; | |
1040 | struct stat stb; | |
1041 | char *conf; | |
1042 | unsigned int i; | |
1043 | unsigned int confsec; | |
1044 | int vnum; | |
1045 | unsigned int max_virt_disks = | |
1046 | be16_to_cpu(super->active->max_vd_entries); | |
1047 | unsigned long long dsize; | |
1048 | ||
1049 | /* First the local disk info */ | |
1050 | if (posix_memalign((void**)&dl, 512, | |
1051 | sizeof(*dl) + | |
1052 | (super->max_part) * sizeof(dl->vlist[0])) != 0) { | |
1053 | pr_err("could not allocate disk info buffer\n"); | |
1054 | return 1; | |
1055 | } | |
1056 | ||
1057 | load_section(fd, super, &dl->disk, | |
1058 | super->active->data_section_offset, | |
1059 | super->active->data_section_length, | |
1060 | 0); | |
1061 | dl->devname = devname ? xstrdup(devname) : NULL; | |
1062 | ||
1063 | fstat(fd, &stb); | |
1064 | dl->major = major(stb.st_rdev); | |
1065 | dl->minor = minor(stb.st_rdev); | |
1066 | dl->next = super->dlist; | |
1067 | dl->fd = keep ? fd : -1; | |
1068 | ||
1069 | dl->size = 0; | |
1070 | if (get_dev_size(fd, devname, &dsize)) | |
1071 | dl->size = dsize >> 9; | |
1072 | /* If the disks have different sizes, the LBAs will differ | |
1073 | * between phys disks. | |
1074 | * At this point here, the values in super->active must be valid | |
1075 | * for this phys disk. */ | |
1076 | dl->primary_lba = super->active->primary_lba; | |
1077 | dl->secondary_lba = super->active->secondary_lba; | |
1078 | dl->workspace_lba = super->active->workspace_lba; | |
1079 | dl->spare = NULL; | |
1080 | for (i = 0 ; i < super->max_part ; i++) | |
1081 | dl->vlist[i] = NULL; | |
1082 | super->dlist = dl; | |
1083 | dl->pdnum = -1; | |
1084 | for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++) | |
1085 | if (memcmp(super->phys->entries[i].guid, | |
1086 | dl->disk.guid, DDF_GUID_LEN) == 0) | |
1087 | dl->pdnum = i; | |
1088 | ||
1089 | /* Now the config list. */ | |
1090 | /* 'conf' is an array of config entries, some of which are | |
1091 | * probably invalid. Those which are good need to be copied into | |
1092 | * the conflist | |
1093 | */ | |
1094 | ||
1095 | conf = load_section(fd, super, super->conf, | |
1096 | super->active->config_section_offset, | |
1097 | super->active->config_section_length, | |
1098 | 0); | |
1099 | super->conf = conf; | |
1100 | vnum = 0; | |
1101 | for (confsec = 0; | |
1102 | confsec < be32_to_cpu(super->active->config_section_length); | |
1103 | confsec += super->conf_rec_len) { | |
1104 | struct vd_config *vd = | |
1105 | (struct vd_config *)((char*)conf + confsec*512); | |
1106 | struct vcl *vcl; | |
1107 | ||
1108 | if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) { | |
1109 | if (dl->spare) | |
1110 | continue; | |
1111 | if (posix_memalign((void**)&dl->spare, 512, | |
1112 | super->conf_rec_len*512) != 0) { | |
1113 | pr_err("could not allocate spare info buf\n"); | |
1114 | return 1; | |
1115 | } | |
1116 | ||
1117 | memcpy(dl->spare, vd, super->conf_rec_len*512); | |
1118 | continue; | |
1119 | } | |
1120 | if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC)) | |
1121 | /* Must be vendor-unique - I cannot handle those */ | |
1122 | continue; | |
1123 | ||
1124 | for (vcl = super->conflist; vcl; vcl = vcl->next) { | |
1125 | if (memcmp(vcl->conf.guid, | |
1126 | vd->guid, DDF_GUID_LEN) == 0) | |
1127 | break; | |
1128 | } | |
1129 | ||
1130 | if (vcl) { | |
1131 | dl->vlist[vnum++] = vcl; | |
1132 | if (vcl->other_bvds != NULL && | |
1133 | vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) { | |
1134 | add_other_bvd(vcl, vd, super->conf_rec_len*512); | |
1135 | continue; | |
1136 | } | |
1137 | if (be32_to_cpu(vd->seqnum) <= | |
1138 | be32_to_cpu(vcl->conf.seqnum)) | |
1139 | continue; | |
1140 | } else { | |
1141 | if (posix_memalign((void**)&vcl, 512, | |
1142 | (super->conf_rec_len*512 + | |
1143 | offsetof(struct vcl, conf))) != 0) { | |
1144 | pr_err("could not allocate vcl buf\n"); | |
1145 | return 1; | |
1146 | } | |
1147 | vcl->next = super->conflist; | |
1148 | vcl->block_sizes = NULL; /* FIXME not for CONCAT */ | |
1149 | vcl->conf.sec_elmnt_count = vd->sec_elmnt_count; | |
1150 | if (alloc_other_bvds(super, vcl) != 0) { | |
1151 | pr_err("could not allocate other bvds\n"); | |
1152 | free(vcl); | |
1153 | return 1; | |
1154 | }; | |
1155 | super->conflist = vcl; | |
1156 | dl->vlist[vnum++] = vcl; | |
1157 | } | |
1158 | memcpy(&vcl->conf, vd, super->conf_rec_len*512); | |
1159 | for (i=0; i < max_virt_disks ; i++) | |
1160 | if (memcmp(super->virt->entries[i].guid, | |
1161 | vcl->conf.guid, DDF_GUID_LEN)==0) | |
1162 | break; | |
1163 | if (i < max_virt_disks) | |
1164 | vcl->vcnum = i; | |
1165 | } | |
1166 | ||
1167 | return 0; | |
1168 | } | |
1169 | ||
1170 | static int load_super_ddf(struct supertype *st, int fd, | |
1171 | char *devname) | |
1172 | { | |
1173 | unsigned long long dsize; | |
1174 | struct ddf_super *super; | |
1175 | int rv; | |
1176 | ||
1177 | if (get_dev_size(fd, devname, &dsize) == 0) | |
1178 | return 1; | |
1179 | ||
1180 | if (test_partition(fd)) | |
1181 | /* DDF is not allowed on partitions */ | |
1182 | return 1; | |
1183 | ||
1184 | /* 32M is a lower bound */ | |
1185 | if (dsize <= 32*1024*1024) { | |
1186 | if (devname) | |
1187 | pr_err("%s is too small for ddf: size is %llu sectors.\n", | |
1188 | devname, dsize>>9); | |
1189 | return 1; | |
1190 | } | |
1191 | if (dsize & 511) { | |
1192 | if (devname) | |
1193 | pr_err("%s is an odd size for ddf: size is %llu bytes.\n", | |
1194 | devname, dsize); | |
1195 | return 1; | |
1196 | } | |
1197 | ||
1198 | free_super_ddf(st); | |
1199 | ||
1200 | if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) { | |
1201 | pr_err("malloc of %zu failed.\n", | |
1202 | sizeof(*super)); | |
1203 | return 1; | |
1204 | } | |
1205 | memset(super, 0, sizeof(*super)); | |
1206 | ||
1207 | rv = load_ddf_headers(fd, super, devname); | |
1208 | if (rv) { | |
1209 | free(super); | |
1210 | return rv; | |
1211 | } | |
1212 | ||
1213 | /* Have valid headers and have chosen the best. Let's read in the rest*/ | |
1214 | ||
1215 | rv = load_ddf_global(fd, super, devname); | |
1216 | ||
1217 | if (rv) { | |
1218 | if (devname) | |
1219 | pr_err("Failed to load all information sections on %s\n", devname); | |
1220 | free(super); | |
1221 | return rv; | |
1222 | } | |
1223 | ||
1224 | rv = load_ddf_local(fd, super, devname, 0); | |
1225 | ||
1226 | if (rv) { | |
1227 | if (devname) | |
1228 | pr_err("Failed to load all information sections on %s\n", devname); | |
1229 | free(super); | |
1230 | return rv; | |
1231 | } | |
1232 | ||
1233 | /* Should possibly check the sections .... */ | |
1234 | ||
1235 | st->sb = super; | |
1236 | if (st->ss == NULL) { | |
1237 | st->ss = &super_ddf; | |
1238 | st->minor_version = 0; | |
1239 | st->max_devs = 512; | |
1240 | } | |
1241 | return 0; | |
1242 | ||
1243 | } | |
1244 | ||
1245 | static void free_super_ddf(struct supertype *st) | |
1246 | { | |
1247 | struct ddf_super *ddf = st->sb; | |
1248 | if (ddf == NULL) | |
1249 | return; | |
1250 | free(ddf->phys); | |
1251 | free(ddf->virt); | |
1252 | free(ddf->conf); | |
1253 | while (ddf->conflist) { | |
1254 | struct vcl *v = ddf->conflist; | |
1255 | ddf->conflist = v->next; | |
1256 | if (v->block_sizes) | |
1257 | free(v->block_sizes); | |
1258 | if (v->other_bvds) | |
1259 | /* | |
1260 | v->other_bvds[0] points to beginning of buffer, | |
1261 | see alloc_other_bvds() | |
1262 | */ | |
1263 | free(v->other_bvds[0]); | |
1264 | free(v); | |
1265 | } | |
1266 | while (ddf->dlist) { | |
1267 | struct dl *d = ddf->dlist; | |
1268 | ddf->dlist = d->next; | |
1269 | if (d->fd >= 0) | |
1270 | close(d->fd); | |
1271 | if (d->spare) | |
1272 | free(d->spare); | |
1273 | free(d); | |
1274 | } | |
1275 | while (ddf->add_list) { | |
1276 | struct dl *d = ddf->add_list; | |
1277 | ddf->add_list = d->next; | |
1278 | if (d->fd >= 0) | |
1279 | close(d->fd); | |
1280 | if (d->spare) | |
1281 | free(d->spare); | |
1282 | free(d); | |
1283 | } | |
1284 | free(ddf); | |
1285 | st->sb = NULL; | |
1286 | } | |
1287 | ||
1288 | static struct supertype *match_metadata_desc_ddf(char *arg) | |
1289 | { | |
1290 | /* 'ddf' only supports containers */ | |
1291 | struct supertype *st; | |
1292 | if (strcmp(arg, "ddf") != 0 && | |
1293 | strcmp(arg, "default") != 0 | |
1294 | ) | |
1295 | return NULL; | |
1296 | ||
1297 | st = xcalloc(1, sizeof(*st)); | |
1298 | st->ss = &super_ddf; | |
1299 | st->max_devs = 512; | |
1300 | st->minor_version = 0; | |
1301 | st->sb = NULL; | |
1302 | return st; | |
1303 | } | |
1304 | ||
1305 | static mapping_t ddf_state[] = { | |
1306 | { "Optimal", 0}, | |
1307 | { "Degraded", 1}, | |
1308 | { "Deleted", 2}, | |
1309 | { "Missing", 3}, | |
1310 | { "Failed", 4}, | |
1311 | { "Partially Optimal", 5}, | |
1312 | { "-reserved-", 6}, | |
1313 | { "-reserved-", 7}, | |
1314 | { NULL, 0} | |
1315 | }; | |
1316 | ||
1317 | static mapping_t ddf_init_state[] = { | |
1318 | { "Not Initialised", 0}, | |
1319 | { "QuickInit in Progress", 1}, | |
1320 | { "Fully Initialised", 2}, | |
1321 | { "*UNKNOWN*", 3}, | |
1322 | { NULL, 0} | |
1323 | }; | |
1324 | static mapping_t ddf_access[] = { | |
1325 | { "Read/Write", 0}, | |
1326 | { "Reserved", 1}, | |
1327 | { "Read Only", 2}, | |
1328 | { "Blocked (no access)", 3}, | |
1329 | { NULL ,0} | |
1330 | }; | |
1331 | ||
1332 | static mapping_t ddf_level[] = { | |
1333 | { "RAID0", DDF_RAID0}, | |
1334 | { "RAID1", DDF_RAID1}, | |
1335 | { "RAID3", DDF_RAID3}, | |
1336 | { "RAID4", DDF_RAID4}, | |
1337 | { "RAID5", DDF_RAID5}, | |
1338 | { "RAID1E",DDF_RAID1E}, | |
1339 | { "JBOD", DDF_JBOD}, | |
1340 | { "CONCAT",DDF_CONCAT}, | |
1341 | { "RAID5E",DDF_RAID5E}, | |
1342 | { "RAID5EE",DDF_RAID5EE}, | |
1343 | { "RAID6", DDF_RAID6}, | |
1344 | { NULL, 0} | |
1345 | }; | |
1346 | static mapping_t ddf_sec_level[] = { | |
1347 | { "Striped", DDF_2STRIPED}, | |
1348 | { "Mirrored", DDF_2MIRRORED}, | |
1349 | { "Concat", DDF_2CONCAT}, | |
1350 | { "Spanned", DDF_2SPANNED}, | |
1351 | { NULL, 0} | |
1352 | }; | |
1353 | ||
1354 | static int all_ff(const char *guid) | |
1355 | { | |
1356 | int i; | |
1357 | for (i = 0; i < DDF_GUID_LEN; i++) | |
1358 | if (guid[i] != (char)0xff) | |
1359 | return 0; | |
1360 | return 1; | |
1361 | } | |
1362 | ||
1363 | static const char *guid_str(const char *guid) | |
1364 | { | |
1365 | static char buf[DDF_GUID_LEN*2+1]; | |
1366 | int i; | |
1367 | char *p = buf; | |
1368 | for (i = 0; i < DDF_GUID_LEN; i++) { | |
1369 | unsigned char c = guid[i]; | |
1370 | if (c >= 32 && c < 127) | |
1371 | p += sprintf(p, "%c", c); | |
1372 | else | |
1373 | p += sprintf(p, "%02x", c); | |
1374 | } | |
1375 | *p = '\0'; | |
1376 | return (const char *) buf; | |
1377 | } | |
1378 | ||
1379 | static void print_guid(char *guid, int tstamp) | |
1380 | { | |
1381 | /* A GUIDs are part (or all) ASCII and part binary. | |
1382 | * They tend to be space padded. | |
1383 | * We print the GUID in HEX, then in parentheses add | |
1384 | * any initial ASCII sequence, and a possible | |
1385 | * time stamp from bytes 16-19 | |
1386 | */ | |
1387 | int l = DDF_GUID_LEN; | |
1388 | int i; | |
1389 | ||
1390 | for (i=0 ; i<DDF_GUID_LEN ; i++) { | |
1391 | if ((i&3)==0 && i != 0) printf(":"); | |
1392 | printf("%02X", guid[i]&255); | |
1393 | } | |
1394 | ||
1395 | printf("\n ("); | |
1396 | while (l && guid[l-1] == ' ') | |
1397 | l--; | |
1398 | for (i=0 ; i<l ; i++) { | |
1399 | if (guid[i] >= 0x20 && guid[i] < 0x7f) | |
1400 | fputc(guid[i], stdout); | |
1401 | else | |
1402 | break; | |
1403 | } | |
1404 | if (tstamp) { | |
1405 | time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE; | |
1406 | char tbuf[100]; | |
1407 | struct tm *tm; | |
1408 | tm = localtime(&then); | |
1409 | strftime(tbuf, 100, " %D %T",tm); | |
1410 | fputs(tbuf, stdout); | |
1411 | } | |
1412 | printf(")"); | |
1413 | } | |
1414 | ||
1415 | static void examine_vd(int n, struct ddf_super *sb, char *guid) | |
1416 | { | |
1417 | int crl = sb->conf_rec_len; | |
1418 | struct vcl *vcl; | |
1419 | ||
1420 | for (vcl = sb->conflist ; vcl ; vcl = vcl->next) { | |
1421 | unsigned int i; | |
1422 | struct vd_config *vc = &vcl->conf; | |
1423 | ||
1424 | if (!be32_eq(calc_crc(vc, crl*512), vc->crc)) | |
1425 | continue; | |
1426 | if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0) | |
1427 | continue; | |
1428 | ||
1429 | /* Ok, we know about this VD, let's give more details */ | |
1430 | printf(" Raid Devices[%d] : %d (", n, | |
1431 | be16_to_cpu(vc->prim_elmnt_count)); | |
1432 | for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) { | |
1433 | int j; | |
1434 | int cnt = be16_to_cpu(sb->phys->max_pdes); | |
1435 | for (j=0; j<cnt; j++) | |
1436 | if (be32_eq(vc->phys_refnum[i], | |
1437 | sb->phys->entries[j].refnum)) | |
1438 | break; | |
1439 | if (i) printf(" "); | |
1440 | if (j < cnt) | |
1441 | printf("%d", j); | |
1442 | else | |
1443 | printf("--"); | |
1444 | printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb, vc)[i])/2); | |
1445 | } | |
1446 | printf(")\n"); | |
1447 | if (vc->chunk_shift != 255) | |
1448 | printf(" Chunk Size[%d] : %d sectors\n", n, | |
1449 | 1 << vc->chunk_shift); | |
1450 | printf(" Raid Level[%d] : %s\n", n, | |
1451 | map_num(ddf_level, vc->prl)?:"-unknown-"); | |
1452 | if (vc->sec_elmnt_count != 1) { | |
1453 | printf(" Secondary Position[%d] : %d of %d\n", n, | |
1454 | vc->sec_elmnt_seq, vc->sec_elmnt_count); | |
1455 | printf(" Secondary Level[%d] : %s\n", n, | |
1456 | map_num(ddf_sec_level, vc->srl) ?: "-unknown-"); | |
1457 | } | |
1458 | printf(" Device Size[%d] : %llu\n", n, | |
1459 | be64_to_cpu(vc->blocks)/2); | |
1460 | printf(" Array Size[%d] : %llu\n", n, | |
1461 | be64_to_cpu(vc->array_blocks)/2); | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | static void examine_vds(struct ddf_super *sb) | |
1466 | { | |
1467 | int cnt = be16_to_cpu(sb->virt->populated_vdes); | |
1468 | unsigned int i; | |
1469 | printf(" Virtual Disks : %d\n", cnt); | |
1470 | ||
1471 | for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) { | |
1472 | struct virtual_entry *ve = &sb->virt->entries[i]; | |
1473 | if (all_ff(ve->guid)) | |
1474 | continue; | |
1475 | printf("\n"); | |
1476 | printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1); | |
1477 | printf("\n"); | |
1478 | printf(" unit[%d] : %d\n", i, be16_to_cpu(ve->unit)); | |
1479 | printf(" state[%d] : %s, %s%s\n", i, | |
1480 | map_num(ddf_state, ve->state & 7), | |
1481 | (ve->state & DDF_state_morphing) ? "Morphing, ": "", | |
1482 | (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent"); | |
1483 | printf(" init state[%d] : %s\n", i, | |
1484 | map_num(ddf_init_state, ve->init_state&DDF_initstate_mask)); | |
1485 | printf(" access[%d] : %s\n", i, | |
1486 | map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6)); | |
1487 | printf(" Name[%d] : %.16s\n", i, ve->name); | |
1488 | examine_vd(i, sb, ve->guid); | |
1489 | } | |
1490 | if (cnt) printf("\n"); | |
1491 | } | |
1492 | ||
1493 | static void examine_pds(struct ddf_super *sb) | |
1494 | { | |
1495 | int cnt = be16_to_cpu(sb->phys->max_pdes); | |
1496 | int i; | |
1497 | struct dl *dl; | |
1498 | int unlisted = 0; | |
1499 | printf(" Physical Disks : %d\n", cnt); | |
1500 | printf(" Number RefNo Size Device Type/State\n"); | |
1501 | ||
1502 | for (dl = sb->dlist; dl; dl = dl->next) | |
1503 | dl->displayed = 0; | |
1504 | ||
1505 | for (i=0 ; i<cnt ; i++) { | |
1506 | struct phys_disk_entry *pd = &sb->phys->entries[i]; | |
1507 | int type = be16_to_cpu(pd->type); | |
1508 | int state = be16_to_cpu(pd->state); | |
1509 | ||
1510 | if (be32_to_cpu(pd->refnum) == 0xffffffff) | |
1511 | /* Not in use */ | |
1512 | continue; | |
1513 | //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0); | |
1514 | //printf("\n"); | |
1515 | printf(" %3d %08x ", i, | |
1516 | be32_to_cpu(pd->refnum)); | |
1517 | printf("%8lluK ", | |
1518 | be64_to_cpu(pd->config_size)>>1); | |
1519 | for (dl = sb->dlist; dl ; dl = dl->next) { | |
1520 | if (be32_eq(dl->disk.refnum, pd->refnum)) { | |
1521 | char *dv = map_dev(dl->major, dl->minor, 0); | |
1522 | if (dv) { | |
1523 | printf("%-15s", dv); | |
1524 | break; | |
1525 | } | |
1526 | } | |
1527 | } | |
1528 | if (!dl) | |
1529 | printf("%15s",""); | |
1530 | else | |
1531 | dl->displayed = 1; | |
1532 | printf(" %s%s%s%s%s", | |
1533 | (type&2) ? "active":"", | |
1534 | (type&4) ? "Global-Spare":"", | |
1535 | (type&8) ? "spare" : "", | |
1536 | (type&16)? ", foreign" : "", | |
1537 | (type&32)? "pass-through" : ""); | |
1538 | if (state & DDF_Failed) | |
1539 | /* This over-rides these three */ | |
1540 | state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition); | |
1541 | printf("/%s%s%s%s%s%s%s", | |
1542 | (state&1)? "Online": "Offline", | |
1543 | (state&2)? ", Failed": "", | |
1544 | (state&4)? ", Rebuilding": "", | |
1545 | (state&8)? ", in-transition": "", | |
1546 | (state&16)? ", SMART-errors": "", | |
1547 | (state&32)? ", Unrecovered-Read-Errors": "", | |
1548 | (state&64)? ", Missing" : ""); | |
1549 | printf("\n"); | |
1550 | } | |
1551 | for (dl = sb->dlist; dl; dl = dl->next) { | |
1552 | char *dv; | |
1553 | if (dl->displayed) | |
1554 | continue; | |
1555 | if (!unlisted) | |
1556 | printf(" Physical disks not in metadata!:\n"); | |
1557 | unlisted = 1; | |
1558 | dv = map_dev(dl->major, dl->minor, 0); | |
1559 | printf(" %08x %s\n", be32_to_cpu(dl->disk.refnum), | |
1560 | dv ? dv : "-unknown-"); | |
1561 | } | |
1562 | if (unlisted) | |
1563 | printf("\n"); | |
1564 | } | |
1565 | ||
1566 | static void examine_super_ddf(struct supertype *st, char *homehost) | |
1567 | { | |
1568 | struct ddf_super *sb = st->sb; | |
1569 | ||
1570 | printf(" Magic : %08x\n", be32_to_cpu(sb->anchor.magic)); | |
1571 | printf(" Version : %.8s\n", sb->anchor.revision); | |
1572 | printf("Controller GUID : "); print_guid(sb->controller.guid, 0); | |
1573 | printf("\n"); | |
1574 | printf(" Container GUID : "); print_guid(sb->anchor.guid, 1); | |
1575 | printf("\n"); | |
1576 | printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq)); | |
1577 | printf(" Redundant hdr : %s\n", (be32_eq(sb->secondary.magic, | |
1578 | DDF_HEADER_MAGIC) | |
1579 | ?"yes" : "no")); | |
1580 | examine_vds(sb); | |
1581 | examine_pds(sb); | |
1582 | } | |
1583 | ||
1584 | static unsigned int get_vd_num_of_subarray(struct supertype *st) | |
1585 | { | |
1586 | /* | |
1587 | * Figure out the VD number for this supertype. | |
1588 | * Returns DDF_CONTAINER for the container itself, | |
1589 | * and DDF_NOTFOUND on error. | |
1590 | */ | |
1591 | struct ddf_super *ddf = st->sb; | |
1592 | struct mdinfo *sra; | |
1593 | char *sub, *end; | |
1594 | unsigned int vcnum; | |
1595 | ||
1596 | if (*st->container_devnm == '\0') | |
1597 | return DDF_CONTAINER; | |
1598 | ||
1599 | sra = sysfs_read(-1, st->devnm, GET_VERSION); | |
1600 | if (!sra || sra->array.major_version != -1 || | |
1601 | sra->array.minor_version != -2 || | |
1602 | !is_subarray(sra->text_version)) | |
1603 | return DDF_NOTFOUND; | |
1604 | ||
1605 | sub = strchr(sra->text_version + 1, '/'); | |
1606 | if (sub != NULL) | |
1607 | vcnum = strtoul(sub + 1, &end, 10); | |
1608 | if (sub == NULL || *sub == '\0' || *end != '\0' || | |
1609 | vcnum >= be16_to_cpu(ddf->active->max_vd_entries)) | |
1610 | return DDF_NOTFOUND; | |
1611 | ||
1612 | return vcnum; | |
1613 | } | |
1614 | ||
1615 | static void brief_examine_super_ddf(struct supertype *st, int verbose) | |
1616 | { | |
1617 | /* We just write a generic DDF ARRAY entry | |
1618 | */ | |
1619 | struct mdinfo info; | |
1620 | char nbuf[64]; | |
1621 | getinfo_super_ddf(st, &info, NULL); | |
1622 | fname_from_uuid(st, &info, nbuf, ':'); | |
1623 | ||
1624 | printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5); | |
1625 | } | |
1626 | ||
1627 | static void brief_examine_subarrays_ddf(struct supertype *st, int verbose) | |
1628 | { | |
1629 | /* We write a DDF ARRAY member entry for each vd, identifying container | |
1630 | * by uuid and member by unit number and uuid. | |
1631 | */ | |
1632 | struct ddf_super *ddf = st->sb; | |
1633 | struct mdinfo info; | |
1634 | unsigned int i; | |
1635 | char nbuf[64]; | |
1636 | getinfo_super_ddf(st, &info, NULL); | |
1637 | fname_from_uuid(st, &info, nbuf, ':'); | |
1638 | ||
1639 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) { | |
1640 | struct virtual_entry *ve = &ddf->virt->entries[i]; | |
1641 | struct vcl vcl; | |
1642 | char nbuf1[64]; | |
1643 | char namebuf[17]; | |
1644 | if (all_ff(ve->guid)) | |
1645 | continue; | |
1646 | memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN); | |
1647 | ddf->currentconf =&vcl; | |
1648 | vcl.vcnum = i; | |
1649 | uuid_from_super_ddf(st, info.uuid); | |
1650 | fname_from_uuid(st, &info, nbuf1, ':'); | |
1651 | _ddf_array_name(namebuf, ddf, i); | |
1652 | printf("ARRAY%s%s container=%s member=%d UUID=%s\n", | |
1653 | namebuf[0] == '\0' ? "" : " /dev/md/", namebuf, | |
1654 | nbuf+5, i, nbuf1+5); | |
1655 | } | |
1656 | } | |
1657 | ||
1658 | static void export_examine_super_ddf(struct supertype *st) | |
1659 | { | |
1660 | struct mdinfo info; | |
1661 | char nbuf[64]; | |
1662 | getinfo_super_ddf(st, &info, NULL); | |
1663 | fname_from_uuid(st, &info, nbuf, ':'); | |
1664 | printf("MD_METADATA=ddf\n"); | |
1665 | printf("MD_LEVEL=container\n"); | |
1666 | printf("MD_UUID=%s\n", nbuf+5); | |
1667 | printf("MD_DEVICES=%u\n", | |
1668 | be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes)); | |
1669 | } | |
1670 | ||
1671 | static int copy_metadata_ddf(struct supertype *st, int from, int to) | |
1672 | { | |
1673 | void *buf; | |
1674 | unsigned long long dsize, offset; | |
1675 | int bytes; | |
1676 | struct ddf_header *ddf; | |
1677 | int written = 0; | |
1678 | ||
1679 | /* The meta consists of an anchor, a primary, and a secondary. | |
1680 | * This all lives at the end of the device. | |
1681 | * So it is easiest to find the earliest of primary and | |
1682 | * secondary, and copy everything from there. | |
1683 | * | |
1684 | * Anchor is 512 from end. It contains primary_lba and secondary_lba | |
1685 | * we choose one of those | |
1686 | */ | |
1687 | ||
1688 | if (posix_memalign(&buf, 4096, 4096) != 0) | |
1689 | return 1; | |
1690 | ||
1691 | if (!get_dev_size(from, NULL, &dsize)) | |
1692 | goto err; | |
1693 | ||
1694 | if (lseek64(from, dsize-512, 0) < 0) | |
1695 | goto err; | |
1696 | if (read(from, buf, 512) != 512) | |
1697 | goto err; | |
1698 | ddf = buf; | |
1699 | if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) || | |
1700 | !be32_eq(calc_crc(ddf, 512), ddf->crc) || | |
1701 | (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 && | |
1702 | memcmp(ddf->revision, DDF_REVISION_2, 8) != 0)) | |
1703 | goto err; | |
1704 | ||
1705 | offset = dsize - 512; | |
1706 | if ((be64_to_cpu(ddf->primary_lba) << 9) < offset) | |
1707 | offset = be64_to_cpu(ddf->primary_lba) << 9; | |
1708 | if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset) | |
1709 | offset = be64_to_cpu(ddf->secondary_lba) << 9; | |
1710 | ||
1711 | bytes = dsize - offset; | |
1712 | ||
1713 | if (lseek64(from, offset, 0) < 0 || | |
1714 | lseek64(to, offset, 0) < 0) | |
1715 | goto err; | |
1716 | while (written < bytes) { | |
1717 | int n = bytes - written; | |
1718 | if (n > 4096) | |
1719 | n = 4096; | |
1720 | if (read(from, buf, n) != n) | |
1721 | goto err; | |
1722 | if (write(to, buf, n) != n) | |
1723 | goto err; | |
1724 | written += n; | |
1725 | } | |
1726 | free(buf); | |
1727 | return 0; | |
1728 | err: | |
1729 | free(buf); | |
1730 | return 1; | |
1731 | } | |
1732 | ||
1733 | static void detail_super_ddf(struct supertype *st, char *homehost) | |
1734 | { | |
1735 | struct ddf_super *sb = st->sb; | |
1736 | int cnt = be16_to_cpu(sb->virt->populated_vdes); | |
1737 | ||
1738 | printf(" Container GUID : "); print_guid(sb->anchor.guid, 1); | |
1739 | printf("\n"); | |
1740 | printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq)); | |
1741 | printf(" Virtual Disks : %d\n", cnt); | |
1742 | printf("\n"); | |
1743 | } | |
1744 | ||
1745 | static const char *vendors_with_variable_volume_UUID[] = { | |
1746 | "LSI ", | |
1747 | }; | |
1748 | ||
1749 | static int volume_id_is_reliable(const struct ddf_super *ddf) | |
1750 | { | |
1751 | int n = ARRAY_SIZE(vendors_with_variable_volume_UUID); | |
1752 | int i; | |
1753 | for (i = 0; i < n; i++) | |
1754 | if (!memcmp(ddf->controller.guid, | |
1755 | vendors_with_variable_volume_UUID[i], 8)) | |
1756 | return 0; | |
1757 | return 1; | |
1758 | } | |
1759 | ||
1760 | static void uuid_of_ddf_subarray(const struct ddf_super *ddf, | |
1761 | unsigned int vcnum, int uuid[4]) | |
1762 | { | |
1763 | char buf[DDF_GUID_LEN+18], sha[20], *p; | |
1764 | struct sha1_ctx ctx; | |
1765 | if (volume_id_is_reliable(ddf)) { | |
1766 | uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid); | |
1767 | return; | |
1768 | } | |
1769 | /* | |
1770 | * Some fake RAID BIOSes (in particular, LSI ones) change the | |
1771 | * VD GUID at every boot. These GUIDs are not suitable for | |
1772 | * identifying an array. Luckily the header GUID appears to | |
1773 | * remain constant. | |
1774 | * We construct a pseudo-UUID from the header GUID and those | |
1775 | * properties of the subarray that we expect to remain constant. | |
1776 | */ | |
1777 | memset(buf, 0, sizeof(buf)); | |
1778 | p = buf; | |
1779 | memcpy(p, ddf->anchor.guid, DDF_GUID_LEN); | |
1780 | p += DDF_GUID_LEN; | |
1781 | memcpy(p, ddf->virt->entries[vcnum].name, 16); | |
1782 | p += 16; | |
1783 | *((__u16 *) p) = vcnum; | |
1784 | sha1_init_ctx(&ctx); | |
1785 | sha1_process_bytes(buf, sizeof(buf), &ctx); | |
1786 | sha1_finish_ctx(&ctx, sha); | |
1787 | memcpy(uuid, sha, 4*4); | |
1788 | } | |
1789 | ||
1790 | static void brief_detail_super_ddf(struct supertype *st) | |
1791 | { | |
1792 | struct mdinfo info; | |
1793 | char nbuf[64]; | |
1794 | struct ddf_super *ddf = st->sb; | |
1795 | unsigned int vcnum = get_vd_num_of_subarray(st); | |
1796 | if (vcnum == DDF_CONTAINER) | |
1797 | uuid_from_super_ddf(st, info.uuid); | |
1798 | else if (vcnum == DDF_NOTFOUND) | |
1799 | return; | |
1800 | else | |
1801 | uuid_of_ddf_subarray(ddf, vcnum, info.uuid); | |
1802 | fname_from_uuid(st, &info, nbuf,':'); | |
1803 | printf(" UUID=%s", nbuf + 5); | |
1804 | } | |
1805 | ||
1806 | static int match_home_ddf(struct supertype *st, char *homehost) | |
1807 | { | |
1808 | /* It matches 'this' host if the controller is a | |
1809 | * Linux-MD controller with vendor_data matching | |
1810 | * the hostname. It would be nice if we could | |
1811 | * test against controller found in /sys or somewhere... | |
1812 | */ | |
1813 | struct ddf_super *ddf = st->sb; | |
1814 | unsigned int len; | |
1815 | ||
1816 | if (!homehost) | |
1817 | return 0; | |
1818 | len = strlen(homehost); | |
1819 | ||
1820 | return (memcmp(ddf->controller.guid, T10, 8) == 0 && | |
1821 | len < sizeof(ddf->controller.vendor_data) && | |
1822 | memcmp(ddf->controller.vendor_data, homehost,len) == 0 && | |
1823 | ddf->controller.vendor_data[len] == 0); | |
1824 | } | |
1825 | ||
1826 | static int find_index_in_bvd(const struct ddf_super *ddf, | |
1827 | const struct vd_config *conf, unsigned int n, | |
1828 | unsigned int *n_bvd) | |
1829 | { | |
1830 | /* | |
1831 | * Find the index of the n-th valid physical disk in this BVD. | |
1832 | * Unused entries can be sprinkled in with the used entries, | |
1833 | * but don't count. | |
1834 | */ | |
1835 | unsigned int i, j; | |
1836 | for (i = 0, j = 0; | |
1837 | i < ddf->mppe && j < be16_to_cpu(conf->prim_elmnt_count); | |
1838 | i++) { | |
1839 | if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) { | |
1840 | if (n == j) { | |
1841 | *n_bvd = i; | |
1842 | return 1; | |
1843 | } | |
1844 | j++; | |
1845 | } | |
1846 | } | |
1847 | dprintf("couldn't find BVD member %u (total %u)\n", | |
1848 | n, be16_to_cpu(conf->prim_elmnt_count)); | |
1849 | return 0; | |
1850 | } | |
1851 | ||
1852 | /* Given a member array instance number, and a raid disk within that instance, | |
1853 | * find the vd_config structure. The offset of the given disk in the phys_refnum | |
1854 | * table is returned in n_bvd. | |
1855 | * For two-level members with a secondary raid level the vd_config for | |
1856 | * the appropriate BVD is returned. | |
1857 | * The return value is always &vlc->conf, where vlc is returned in last pointer. | |
1858 | */ | |
1859 | static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst, | |
1860 | unsigned int n, | |
1861 | unsigned int *n_bvd, struct vcl **vcl) | |
1862 | { | |
1863 | struct vcl *v; | |
1864 | ||
1865 | for (v = ddf->conflist; v; v = v->next) { | |
1866 | unsigned int nsec, ibvd = 0; | |
1867 | struct vd_config *conf; | |
1868 | if (inst != v->vcnum) | |
1869 | continue; | |
1870 | conf = &v->conf; | |
1871 | if (conf->sec_elmnt_count == 1) { | |
1872 | if (find_index_in_bvd(ddf, conf, n, n_bvd)) { | |
1873 | *vcl = v; | |
1874 | return conf; | |
1875 | } else | |
1876 | goto bad; | |
1877 | } | |
1878 | if (v->other_bvds == NULL) { | |
1879 | pr_err("BUG: other_bvds is NULL, nsec=%u\n", | |
1880 | conf->sec_elmnt_count); | |
1881 | goto bad; | |
1882 | } | |
1883 | nsec = n / be16_to_cpu(conf->prim_elmnt_count); | |
1884 | if (conf->sec_elmnt_seq != nsec) { | |
1885 | for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) { | |
1886 | if (v->other_bvds[ibvd-1]->sec_elmnt_seq == | |
1887 | nsec) | |
1888 | break; | |
1889 | } | |
1890 | if (ibvd == conf->sec_elmnt_count) | |
1891 | goto bad; | |
1892 | conf = v->other_bvds[ibvd-1]; | |
1893 | } | |
1894 | if (!find_index_in_bvd(ddf, conf, | |
1895 | n - nsec*conf->sec_elmnt_count, n_bvd)) | |
1896 | goto bad; | |
1897 | dprintf("found disk %u as member %u in bvd %d of array %u\n", | |
1898 | n, *n_bvd, ibvd, inst); | |
1899 | *vcl = v; | |
1900 | return conf; | |
1901 | } | |
1902 | bad: | |
1903 | pr_err("Could't find disk %d in array %u\n", n, inst); | |
1904 | return NULL; | |
1905 | } | |
1906 | ||
1907 | static int find_phys(const struct ddf_super *ddf, be32 phys_refnum) | |
1908 | { | |
1909 | /* Find the entry in phys_disk which has the given refnum | |
1910 | * and return it's index | |
1911 | */ | |
1912 | unsigned int i; | |
1913 | for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) | |
1914 | if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum)) | |
1915 | return i; | |
1916 | return -1; | |
1917 | } | |
1918 | ||
1919 | static void uuid_from_ddf_guid(const char *guid, int uuid[4]) | |
1920 | { | |
1921 | char buf[20]; | |
1922 | struct sha1_ctx ctx; | |
1923 | sha1_init_ctx(&ctx); | |
1924 | sha1_process_bytes(guid, DDF_GUID_LEN, &ctx); | |
1925 | sha1_finish_ctx(&ctx, buf); | |
1926 | memcpy(uuid, buf, 4*4); | |
1927 | } | |
1928 | ||
1929 | static void uuid_from_super_ddf(struct supertype *st, int uuid[4]) | |
1930 | { | |
1931 | /* The uuid returned here is used for: | |
1932 | * uuid to put into bitmap file (Create, Grow) | |
1933 | * uuid for backup header when saving critical section (Grow) | |
1934 | * comparing uuids when re-adding a device into an array | |
1935 | * In these cases the uuid required is that of the data-array, | |
1936 | * not the device-set. | |
1937 | * uuid to recognise same set when adding a missing device back | |
1938 | * to an array. This is a uuid for the device-set. | |
1939 | * | |
1940 | * For each of these we can make do with a truncated | |
1941 | * or hashed uuid rather than the original, as long as | |
1942 | * everyone agrees. | |
1943 | * In the case of SVD we assume the BVD is of interest, | |
1944 | * though that might be the case if a bitmap were made for | |
1945 | * a mirrored SVD - worry about that later. | |
1946 | * So we need to find the VD configuration record for the | |
1947 | * relevant BVD and extract the GUID and Secondary_Element_Seq. | |
1948 | * The first 16 bytes of the sha1 of these is used. | |
1949 | */ | |
1950 | struct ddf_super *ddf = st->sb; | |
1951 | struct vcl *vcl = ddf->currentconf; | |
1952 | ||
1953 | if (vcl) | |
1954 | uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid); | |
1955 | else | |
1956 | uuid_from_ddf_guid(ddf->anchor.guid, uuid); | |
1957 | } | |
1958 | ||
1959 | static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map) | |
1960 | { | |
1961 | struct ddf_super *ddf = st->sb; | |
1962 | int map_disks = info->array.raid_disks; | |
1963 | __u32 *cptr; | |
1964 | ||
1965 | if (ddf->currentconf) { | |
1966 | getinfo_super_ddf_bvd(st, info, map); | |
1967 | return; | |
1968 | } | |
1969 | memset(info, 0, sizeof(*info)); | |
1970 | ||
1971 | info->array.raid_disks = be16_to_cpu(ddf->phys->used_pdes); | |
1972 | info->array.level = LEVEL_CONTAINER; | |
1973 | info->array.layout = 0; | |
1974 | info->array.md_minor = -1; | |
1975 | cptr = (__u32 *)(ddf->anchor.guid + 16); | |
1976 | info->array.ctime = DECADE + __be32_to_cpu(*cptr); | |
1977 | ||
1978 | info->array.chunk_size = 0; | |
1979 | info->container_enough = 1; | |
1980 | ||
1981 | info->disk.major = 0; | |
1982 | info->disk.minor = 0; | |
1983 | if (ddf->dlist) { | |
1984 | struct phys_disk_entry *pde = NULL; | |
1985 | info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum); | |
1986 | info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum); | |
1987 | ||
1988 | info->data_offset = be64_to_cpu(ddf->phys-> | |
1989 | entries[info->disk.raid_disk]. | |
1990 | config_size); | |
1991 | info->component_size = ddf->dlist->size - info->data_offset; | |
1992 | if (info->disk.raid_disk >= 0) | |
1993 | pde = ddf->phys->entries + info->disk.raid_disk; | |
1994 | if (pde && | |
1995 | !(be16_to_cpu(pde->state) & DDF_Failed) && | |
1996 | !(be16_to_cpu(pde->state) & DDF_Missing)) | |
1997 | info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
1998 | else | |
1999 | info->disk.state = 1 << MD_DISK_FAULTY; | |
2000 | ||
2001 | } else { | |
2002 | /* There should always be a dlist, but just in case...*/ | |
2003 | info->disk.number = -1; | |
2004 | info->disk.raid_disk = -1; | |
2005 | info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE); | |
2006 | } | |
2007 | info->events = be32_to_cpu(ddf->active->seq); | |
2008 | info->array.utime = DECADE + be32_to_cpu(ddf->active->timestamp); | |
2009 | ||
2010 | info->recovery_start = MaxSector; | |
2011 | info->reshape_active = 0; | |
2012 | info->recovery_blocked = 0; | |
2013 | info->name[0] = 0; | |
2014 | ||
2015 | info->array.major_version = -1; | |
2016 | info->array.minor_version = -2; | |
2017 | strcpy(info->text_version, "ddf"); | |
2018 | info->safe_mode_delay = 0; | |
2019 | ||
2020 | uuid_from_super_ddf(st, info->uuid); | |
2021 | ||
2022 | if (map) { | |
2023 | int i, e = 0; | |
2024 | int max = be16_to_cpu(ddf->phys->max_pdes); | |
2025 | for (i = e = 0 ; i < map_disks ; i++, e++) { | |
2026 | while (e < max && | |
2027 | be32_to_cpu(ddf->phys->entries[e].refnum) == 0xffffffff) | |
2028 | e++; | |
2029 | if (i < info->array.raid_disks && e < max && | |
2030 | !(be16_to_cpu(ddf->phys->entries[e].state) & | |
2031 | DDF_Failed)) | |
2032 | map[i] = 1; | |
2033 | else | |
2034 | map[i] = 0; | |
2035 | } | |
2036 | } | |
2037 | } | |
2038 | ||
2039 | /* size of name must be at least 17 bytes! */ | |
2040 | static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i) | |
2041 | { | |
2042 | int j; | |
2043 | memcpy(name, ddf->virt->entries[i].name, 16); | |
2044 | name[16] = 0; | |
2045 | for(j = 0; j < 16; j++) | |
2046 | if (name[j] == ' ') | |
2047 | name[j] = 0; | |
2048 | } | |
2049 | ||
2050 | static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map) | |
2051 | { | |
2052 | struct ddf_super *ddf = st->sb; | |
2053 | struct vcl *vc = ddf->currentconf; | |
2054 | int cd = ddf->currentdev; | |
2055 | int n_prim; | |
2056 | int j; | |
2057 | struct dl *dl = NULL; | |
2058 | int map_disks = info->array.raid_disks; | |
2059 | __u32 *cptr; | |
2060 | struct vd_config *conf; | |
2061 | ||
2062 | memset(info, 0, sizeof(*info)); | |
2063 | if (layout_ddf2md(&vc->conf, &info->array) == -1) | |
2064 | return; | |
2065 | info->array.md_minor = -1; | |
2066 | cptr = (__u32 *)(vc->conf.guid + 16); | |
2067 | info->array.ctime = DECADE + __be32_to_cpu(*cptr); | |
2068 | info->array.utime = DECADE + be32_to_cpu(vc->conf.timestamp); | |
2069 | info->array.chunk_size = 512 << vc->conf.chunk_shift; | |
2070 | info->custom_array_size = be64_to_cpu(vc->conf.array_blocks); | |
2071 | ||
2072 | conf = &vc->conf; | |
2073 | n_prim = be16_to_cpu(conf->prim_elmnt_count); | |
2074 | if (conf->sec_elmnt_count > 1 && cd >= n_prim) { | |
2075 | int ibvd = cd / n_prim - 1; | |
2076 | cd %= n_prim; | |
2077 | conf = vc->other_bvds[ibvd]; | |
2078 | } | |
2079 | ||
2080 | if (cd >= 0 && (unsigned)cd < ddf->mppe) { | |
2081 | info->data_offset = | |
2082 | be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]); | |
2083 | if (vc->block_sizes) | |
2084 | info->component_size = vc->block_sizes[cd]; | |
2085 | else | |
2086 | info->component_size = be64_to_cpu(conf->blocks); | |
2087 | ||
2088 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2089 | if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd])) | |
2090 | break; | |
2091 | } | |
2092 | ||
2093 | info->disk.major = 0; | |
2094 | info->disk.minor = 0; | |
2095 | info->disk.state = 0; | |
2096 | if (dl && dl->pdnum >= 0) { | |
2097 | info->disk.major = dl->major; | |
2098 | info->disk.minor = dl->minor; | |
2099 | info->disk.raid_disk = cd + conf->sec_elmnt_seq | |
2100 | * be16_to_cpu(conf->prim_elmnt_count); | |
2101 | info->disk.number = dl->pdnum; | |
2102 | info->disk.state = 0; | |
2103 | if (info->disk.number >= 0 && | |
2104 | (be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Online) && | |
2105 | !(be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Failed)) | |
2106 | info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE); | |
2107 | info->events = be32_to_cpu(ddf->active->seq); | |
2108 | } | |
2109 | ||
2110 | info->container_member = ddf->currentconf->vcnum; | |
2111 | ||
2112 | info->recovery_start = MaxSector; | |
2113 | info->resync_start = 0; | |
2114 | info->reshape_active = 0; | |
2115 | info->recovery_blocked = 0; | |
2116 | if (!(ddf->virt->entries[info->container_member].state & | |
2117 | DDF_state_inconsistent) && | |
2118 | (ddf->virt->entries[info->container_member].init_state & | |
2119 | DDF_initstate_mask) == DDF_init_full) | |
2120 | info->resync_start = MaxSector; | |
2121 | ||
2122 | uuid_from_super_ddf(st, info->uuid); | |
2123 | ||
2124 | info->array.major_version = -1; | |
2125 | info->array.minor_version = -2; | |
2126 | sprintf(info->text_version, "/%s/%d", | |
2127 | st->container_devnm, | |
2128 | info->container_member); | |
2129 | info->safe_mode_delay = DDF_SAFE_MODE_DELAY; | |
2130 | ||
2131 | _ddf_array_name(info->name, ddf, info->container_member); | |
2132 | ||
2133 | if (map) | |
2134 | for (j = 0; j < map_disks; j++) { | |
2135 | map[j] = 0; | |
2136 | if (j < info->array.raid_disks) { | |
2137 | int i = find_phys(ddf, vc->conf.phys_refnum[j]); | |
2138 | if (i >= 0 && | |
2139 | (be16_to_cpu(ddf->phys->entries[i].state) | |
2140 | & DDF_Online) && | |
2141 | !(be16_to_cpu(ddf->phys->entries[i].state) | |
2142 | & DDF_Failed)) | |
2143 | map[i] = 1; | |
2144 | } | |
2145 | } | |
2146 | } | |
2147 | ||
2148 | static int update_super_ddf(struct supertype *st, struct mdinfo *info, | |
2149 | char *update, | |
2150 | char *devname, int verbose, | |
2151 | int uuid_set, char *homehost) | |
2152 | { | |
2153 | /* For 'assemble' and 'force' we need to return non-zero if any | |
2154 | * change was made. For others, the return value is ignored. | |
2155 | * Update options are: | |
2156 | * force-one : This device looks a bit old but needs to be included, | |
2157 | * update age info appropriately. | |
2158 | * assemble: clear any 'faulty' flag to allow this device to | |
2159 | * be assembled. | |
2160 | * force-array: Array is degraded but being forced, mark it clean | |
2161 | * if that will be needed to assemble it. | |
2162 | * | |
2163 | * newdev: not used ???? | |
2164 | * grow: Array has gained a new device - this is currently for | |
2165 | * linear only | |
2166 | * resync: mark as dirty so a resync will happen. | |
2167 | * uuid: Change the uuid of the array to match what is given | |
2168 | * homehost: update the recorded homehost | |
2169 | * name: update the name - preserving the homehost | |
2170 | * _reshape_progress: record new reshape_progress position. | |
2171 | * | |
2172 | * Following are not relevant for this version: | |
2173 | * sparc2.2 : update from old dodgey metadata | |
2174 | * super-minor: change the preferred_minor number | |
2175 | * summaries: update redundant counters. | |
2176 | */ | |
2177 | int rv = 0; | |
2178 | // struct ddf_super *ddf = st->sb; | |
2179 | // struct vd_config *vd = find_vdcr(ddf, info->container_member); | |
2180 | // struct virtual_entry *ve = find_ve(ddf); | |
2181 | ||
2182 | /* we don't need to handle "force-*" or "assemble" as | |
2183 | * there is no need to 'trick' the kernel. When the metadata is | |
2184 | * first updated to activate the array, all the implied modifications | |
2185 | * will just happen. | |
2186 | */ | |
2187 | ||
2188 | if (strcmp(update, "grow") == 0) { | |
2189 | /* FIXME */ | |
2190 | } else if (strcmp(update, "resync") == 0) { | |
2191 | // info->resync_checkpoint = 0; | |
2192 | } else if (strcmp(update, "homehost") == 0) { | |
2193 | /* homehost is stored in controller->vendor_data, | |
2194 | * or it is when we are the vendor | |
2195 | */ | |
2196 | // if (info->vendor_is_local) | |
2197 | // strcpy(ddf->controller.vendor_data, homehost); | |
2198 | rv = -1; | |
2199 | } else if (strcmp(update, "name") == 0) { | |
2200 | /* name is stored in virtual_entry->name */ | |
2201 | // memset(ve->name, ' ', 16); | |
2202 | // strncpy(ve->name, info->name, 16); | |
2203 | rv = -1; | |
2204 | } else if (strcmp(update, "_reshape_progress") == 0) { | |
2205 | /* We don't support reshape yet */ | |
2206 | } else if (strcmp(update, "assemble") == 0 ) { | |
2207 | /* Do nothing, just succeed */ | |
2208 | rv = 0; | |
2209 | } else | |
2210 | rv = -1; | |
2211 | ||
2212 | // update_all_csum(ddf); | |
2213 | ||
2214 | return rv; | |
2215 | } | |
2216 | ||
2217 | static void make_header_guid(char *guid) | |
2218 | { | |
2219 | be32 stamp; | |
2220 | /* Create a DDF Header of Virtual Disk GUID */ | |
2221 | ||
2222 | /* 24 bytes of fiction required. | |
2223 | * first 8 are a 'vendor-id' - "Linux-MD" | |
2224 | * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000 | |
2225 | * Remaining 8 random number plus timestamp | |
2226 | */ | |
2227 | memcpy(guid, T10, sizeof(T10)); | |
2228 | stamp = cpu_to_be32(0xdeadbeef); | |
2229 | memcpy(guid+8, &stamp, 4); | |
2230 | stamp = cpu_to_be32(0); | |
2231 | memcpy(guid+12, &stamp, 4); | |
2232 | stamp = cpu_to_be32(time(0) - DECADE); | |
2233 | memcpy(guid+16, &stamp, 4); | |
2234 | stamp._v32 = random32(); | |
2235 | memcpy(guid+20, &stamp, 4); | |
2236 | } | |
2237 | ||
2238 | static unsigned int find_unused_vde(const struct ddf_super *ddf) | |
2239 | { | |
2240 | unsigned int i; | |
2241 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) { | |
2242 | if (all_ff(ddf->virt->entries[i].guid)) | |
2243 | return i; | |
2244 | } | |
2245 | return DDF_NOTFOUND; | |
2246 | } | |
2247 | ||
2248 | static unsigned int find_vde_by_name(const struct ddf_super *ddf, | |
2249 | const char *name) | |
2250 | { | |
2251 | unsigned int i; | |
2252 | if (name == NULL) | |
2253 | return DDF_NOTFOUND; | |
2254 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) { | |
2255 | if (all_ff(ddf->virt->entries[i].guid)) | |
2256 | continue; | |
2257 | if (!strncmp(name, ddf->virt->entries[i].name, | |
2258 | sizeof(ddf->virt->entries[i].name))) | |
2259 | return i; | |
2260 | } | |
2261 | return DDF_NOTFOUND; | |
2262 | } | |
2263 | ||
2264 | static unsigned int find_vde_by_guid(const struct ddf_super *ddf, | |
2265 | const char *guid) | |
2266 | { | |
2267 | unsigned int i; | |
2268 | if (guid == NULL || all_ff(guid)) | |
2269 | return DDF_NOTFOUND; | |
2270 | for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) | |
2271 | if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN)) | |
2272 | return i; | |
2273 | return DDF_NOTFOUND; | |
2274 | } | |
2275 | ||
2276 | static int init_super_ddf(struct supertype *st, | |
2277 | mdu_array_info_t *info, | |
2278 | struct shape *s, char *name, char *homehost, | |
2279 | int *uuid, unsigned long long data_offset) | |
2280 | { | |
2281 | /* This is primarily called by Create when creating a new array. | |
2282 | * We will then get add_to_super called for each component, and then | |
2283 | * write_init_super called to write it out to each device. | |
2284 | * For DDF, Create can create on fresh devices or on a pre-existing | |
2285 | * array. | |
2286 | * To create on a pre-existing array a different method will be called. | |
2287 | * This one is just for fresh drives. | |
2288 | * | |
2289 | * We need to create the entire 'ddf' structure which includes: | |
2290 | * DDF headers - these are easy. | |
2291 | * Controller data - a Sector describing this controller .. not that | |
2292 | * this is a controller exactly. | |
2293 | * Physical Disk Record - one entry per device, so | |
2294 | * leave plenty of space. | |
2295 | * Virtual Disk Records - again, just leave plenty of space. | |
2296 | * This just lists VDs, doesn't give details. | |
2297 | * Config records - describe the VDs that use this disk | |
2298 | * DiskData - describes 'this' device. | |
2299 | * BadBlockManagement - empty | |
2300 | * Diag Space - empty | |
2301 | * Vendor Logs - Could we put bitmaps here? | |
2302 | * | |
2303 | */ | |
2304 | struct ddf_super *ddf; | |
2305 | char hostname[17]; | |
2306 | int hostlen; | |
2307 | int max_phys_disks, max_virt_disks; | |
2308 | unsigned long long sector; | |
2309 | int clen; | |
2310 | int i; | |
2311 | int pdsize, vdsize; | |
2312 | struct phys_disk *pd; | |
2313 | struct virtual_disk *vd; | |
2314 | ||
2315 | if (st->sb) | |
2316 | return init_super_ddf_bvd(st, info, s->size, name, homehost, uuid, | |
2317 | data_offset); | |
2318 | ||
2319 | if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) { | |
2320 | pr_err("could not allocate superblock\n"); | |
2321 | return 0; | |
2322 | } | |
2323 | memset(ddf, 0, sizeof(*ddf)); | |
2324 | st->sb = ddf; | |
2325 | ||
2326 | if (info == NULL) { | |
2327 | /* zeroing superblock */ | |
2328 | return 0; | |
2329 | } | |
2330 | ||
2331 | /* At least 32MB *must* be reserved for the ddf. So let's just | |
2332 | * start 32MB from the end, and put the primary header there. | |
2333 | * Don't do secondary for now. | |
2334 | * We don't know exactly where that will be yet as it could be | |
2335 | * different on each device. So just set up the lengths. | |
2336 | */ | |
2337 | ||
2338 | ddf->anchor.magic = DDF_HEADER_MAGIC; | |
2339 | make_header_guid(ddf->anchor.guid); | |
2340 | ||
2341 | memcpy(ddf->anchor.revision, DDF_REVISION_2, 8); | |
2342 | ddf->anchor.seq = cpu_to_be32(1); | |
2343 | ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE); | |
2344 | ddf->anchor.openflag = 0xFF; | |
2345 | ddf->anchor.foreignflag = 0; | |
2346 | ddf->anchor.enforcegroups = 0; /* Is this best?? */ | |
2347 | ddf->anchor.pad0 = 0xff; | |
2348 | memset(ddf->anchor.pad1, 0xff, 12); | |
2349 | memset(ddf->anchor.header_ext, 0xff, 32); | |
2350 | ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0); | |
2351 | ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0); | |
2352 | ddf->anchor.type = DDF_HEADER_ANCHOR; | |
2353 | memset(ddf->anchor.pad2, 0xff, 3); | |
2354 | ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */ | |
2355 | /* Put this at bottom of 32M reserved.. */ | |
2356 | ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0); | |
2357 | max_phys_disks = 1023; /* Should be enough, 4095 is also allowed */ | |
2358 | ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks); | |
2359 | max_virt_disks = 255; /* 15, 63, 255, 1024, 4095 are all allowed */ | |
2360 | ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks); | |
2361 | ddf->max_part = 64; | |
2362 | ddf->anchor.max_partitions = cpu_to_be16(ddf->max_part); | |
2363 | ddf->mppe = 256; /* 16, 64, 256, 1024, 4096 are all allowed */ | |
2364 | ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512; | |
2365 | ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len); | |
2366 | ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe); | |
2367 | memset(ddf->anchor.pad3, 0xff, 54); | |
2368 | /* Controller section is one sector long immediately | |
2369 | * after the ddf header */ | |
2370 | sector = 1; | |
2371 | ddf->anchor.controller_section_offset = cpu_to_be32(sector); | |
2372 | ddf->anchor.controller_section_length = cpu_to_be32(1); | |
2373 | sector += 1; | |
2374 | ||
2375 | /* phys is 8 sectors after that */ | |
2376 | pdsize = ROUND_UP(sizeof(struct phys_disk) + | |
2377 | sizeof(struct phys_disk_entry)*max_phys_disks, | |
2378 | 512); | |
2379 | switch(pdsize/512) { | |
2380 | case 2: case 8: case 32: case 128: case 512: break; | |
2381 | default: abort(); | |
2382 | } | |
2383 | ddf->anchor.phys_section_offset = cpu_to_be32(sector); | |
2384 | ddf->anchor.phys_section_length = | |
2385 | cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */ | |
2386 | sector += pdsize/512; | |
2387 | ||
2388 | /* virt is another 32 sectors */ | |
2389 | vdsize = ROUND_UP(sizeof(struct virtual_disk) + | |
2390 | sizeof(struct virtual_entry) * max_virt_disks, | |
2391 | 512); | |
2392 | switch(vdsize/512) { | |
2393 | case 2: case 8: case 32: case 128: case 512: break; | |
2394 | default: abort(); | |
2395 | } | |
2396 | ddf->anchor.virt_section_offset = cpu_to_be32(sector); | |
2397 | ddf->anchor.virt_section_length = | |
2398 | cpu_to_be32(vdsize/512); /* max_vd_entries/8 */ | |
2399 | sector += vdsize/512; | |
2400 | ||
2401 | clen = ddf->conf_rec_len * (ddf->max_part+1); | |
2402 | ddf->anchor.config_section_offset = cpu_to_be32(sector); | |
2403 | ddf->anchor.config_section_length = cpu_to_be32(clen); | |
2404 | sector += clen; | |
2405 | ||
2406 | ddf->anchor.data_section_offset = cpu_to_be32(sector); | |
2407 | ddf->anchor.data_section_length = cpu_to_be32(1); | |
2408 | sector += 1; | |
2409 | ||
2410 | ddf->anchor.bbm_section_length = cpu_to_be32(0); | |
2411 | ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF); | |
2412 | ddf->anchor.diag_space_length = cpu_to_be32(0); | |
2413 | ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF); | |
2414 | ddf->anchor.vendor_length = cpu_to_be32(0); | |
2415 | ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF); | |
2416 | ||
2417 | memset(ddf->anchor.pad4, 0xff, 256); | |
2418 | ||
2419 | memcpy(&ddf->primary, &ddf->anchor, 512); | |
2420 | memcpy(&ddf->secondary, &ddf->anchor, 512); | |
2421 | ||
2422 | ddf->primary.openflag = 1; /* I guess.. */ | |
2423 | ddf->primary.type = DDF_HEADER_PRIMARY; | |
2424 | ||
2425 | ddf->secondary.openflag = 1; /* I guess.. */ | |
2426 | ddf->secondary.type = DDF_HEADER_SECONDARY; | |
2427 | ||
2428 | ddf->active = &ddf->primary; | |
2429 | ||
2430 | ddf->controller.magic = DDF_CONTROLLER_MAGIC; | |
2431 | ||
2432 | /* 24 more bytes of fiction required. | |
2433 | * first 8 are a 'vendor-id' - "Linux-MD" | |
2434 | * Remaining 16 are serial number.... maybe a hostname would do? | |
2435 | */ | |
2436 | memcpy(ddf->controller.guid, T10, sizeof(T10)); | |
2437 | gethostname(hostname, sizeof(hostname)); | |
2438 | hostname[sizeof(hostname) - 1] = 0; | |
2439 | hostlen = strlen(hostname); | |
2440 | memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen); | |
2441 | for (i = strlen(T10) ; i+hostlen < 24; i++) | |
2442 | ddf->controller.guid[i] = ' '; | |
2443 | ||
2444 | ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD); | |
2445 | ddf->controller.type.device_id = cpu_to_be16(0xBEEF); | |
2446 | ddf->controller.type.sub_vendor_id = cpu_to_be16(0); | |
2447 | ddf->controller.type.sub_device_id = cpu_to_be16(0); | |
2448 | memcpy(ddf->controller.product_id, "What Is My PID??", 16); | |
2449 | memset(ddf->controller.pad, 0xff, 8); | |
2450 | memset(ddf->controller.vendor_data, 0xff, 448); | |
2451 | if (homehost && strlen(homehost) < 440) | |
2452 | strcpy((char*)ddf->controller.vendor_data, homehost); | |
2453 | ||
2454 | if (posix_memalign((void**)&pd, 512, pdsize) != 0) { | |
2455 | pr_err("could not allocate pd\n"); | |
2456 | return 0; | |
2457 | } | |
2458 | ddf->phys = pd; | |
2459 | ddf->pdsize = pdsize; | |
2460 | ||
2461 | memset(pd, 0xff, pdsize); | |
2462 | memset(pd, 0, sizeof(*pd)); | |
2463 | pd->magic = DDF_PHYS_RECORDS_MAGIC; | |
2464 | pd->used_pdes = cpu_to_be16(0); | |
2465 | pd->max_pdes = cpu_to_be16(max_phys_disks); | |
2466 | memset(pd->pad, 0xff, 52); | |
2467 | for (i = 0; i < max_phys_disks; i++) | |
2468 | memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN); | |
2469 | ||
2470 | if (posix_memalign((void**)&vd, 512, vdsize) != 0) { | |
2471 | pr_err("could not allocate vd\n"); | |
2472 | return 0; | |
2473 | } | |
2474 | ddf->virt = vd; | |
2475 | ddf->vdsize = vdsize; | |
2476 | memset(vd, 0, vdsize); | |
2477 | vd->magic = DDF_VIRT_RECORDS_MAGIC; | |
2478 | vd->populated_vdes = cpu_to_be16(0); | |
2479 | vd->max_vdes = cpu_to_be16(max_virt_disks); | |
2480 | memset(vd->pad, 0xff, 52); | |
2481 | ||
2482 | for (i=0; i<max_virt_disks; i++) | |
2483 | memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry)); | |
2484 | ||
2485 | st->sb = ddf; | |
2486 | ddf_set_updates_pending(ddf, NULL); | |
2487 | return 1; | |
2488 | } | |
2489 | ||
2490 | static int chunk_to_shift(int chunksize) | |
2491 | { | |
2492 | return ffs(chunksize/512)-1; | |
2493 | } | |
2494 | ||
2495 | struct extent { | |
2496 | unsigned long long start, size; | |
2497 | }; | |
2498 | static int cmp_extent(const void *av, const void *bv) | |
2499 | { | |
2500 | const struct extent *a = av; | |
2501 | const struct extent *b = bv; | |
2502 | if (a->start < b->start) | |
2503 | return -1; | |
2504 | if (a->start > b->start) | |
2505 | return 1; | |
2506 | return 0; | |
2507 | } | |
2508 | ||
2509 | static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl) | |
2510 | { | |
2511 | /* Find a list of used extents on the given physical device | |
2512 | * (dnum) of the given ddf. | |
2513 | * Return a malloced array of 'struct extent' | |
2514 | */ | |
2515 | struct extent *rv; | |
2516 | int n = 0; | |
2517 | unsigned int i; | |
2518 | __u16 state; | |
2519 | ||
2520 | if (dl->pdnum < 0) | |
2521 | return NULL; | |
2522 | state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state); | |
2523 | ||
2524 | if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online) | |
2525 | return NULL; | |
2526 | ||
2527 | rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2)); | |
2528 | ||
2529 | for (i = 0; i < ddf->max_part; i++) { | |
2530 | const struct vd_config *bvd; | |
2531 | unsigned int ibvd; | |
2532 | struct vcl *v = dl->vlist[i]; | |
2533 | if (v == NULL || | |
2534 | get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe, | |
2535 | &bvd, &ibvd) == DDF_NOTFOUND) | |
2536 | continue; | |
2537 | rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]); | |
2538 | rv[n].size = be64_to_cpu(bvd->blocks); | |
2539 | n++; | |
2540 | } | |
2541 | qsort(rv, n, sizeof(*rv), cmp_extent); | |
2542 | ||
2543 | rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size); | |
2544 | rv[n].size = 0; | |
2545 | return rv; | |
2546 | } | |
2547 | ||
2548 | static unsigned long long find_space( | |
2549 | struct ddf_super *ddf, struct dl *dl, | |
2550 | unsigned long long data_offset, | |
2551 | unsigned long long *size) | |
2552 | { | |
2553 | /* Find if the requested amount of space is available. | |
2554 | * If it is, return start. | |
2555 | * If not, set *size to largest space. | |
2556 | * If data_offset != INVALID_SECTORS, then the space must start | |
2557 | * at this location. | |
2558 | */ | |
2559 | struct extent *e = get_extents(ddf, dl); | |
2560 | int i = 0; | |
2561 | unsigned long long pos = 0; | |
2562 | unsigned long long max_size = 0; | |
2563 | ||
2564 | if (!e) { | |
2565 | *size = 0; | |
2566 | return INVALID_SECTORS; | |
2567 | } | |
2568 | do { | |
2569 | unsigned long long esize = e[i].start - pos; | |
2570 | if (data_offset != INVALID_SECTORS && | |
2571 | pos <= data_offset && | |
2572 | e[i].start > data_offset) { | |
2573 | pos = data_offset; | |
2574 | esize = e[i].start - pos; | |
2575 | } | |
2576 | if (data_offset != INVALID_SECTORS && | |
2577 | pos != data_offset) { | |
2578 | i++; | |
2579 | continue; | |
2580 | } | |
2581 | if (esize >= *size) { | |
2582 | /* Found! */ | |
2583 | free(e); | |
2584 | return pos; | |
2585 | } | |
2586 | if (esize > max_size) | |
2587 | max_size = esize; | |
2588 | pos = e[i].start + e[i].size; | |
2589 | i++; | |
2590 | } while (e[i-1].size); | |
2591 | *size = max_size; | |
2592 | free(e); | |
2593 | return INVALID_SECTORS; | |
2594 | } | |
2595 | ||
2596 | static int init_super_ddf_bvd(struct supertype *st, | |
2597 | mdu_array_info_t *info, | |
2598 | unsigned long long size, | |
2599 | char *name, char *homehost, | |
2600 | int *uuid, unsigned long long data_offset) | |
2601 | { | |
2602 | /* We are creating a BVD inside a pre-existing container. | |
2603 | * so st->sb is already set. | |
2604 | * We need to create a new vd_config and a new virtual_entry | |
2605 | */ | |
2606 | struct ddf_super *ddf = st->sb; | |
2607 | unsigned int venum, i; | |
2608 | struct virtual_entry *ve; | |
2609 | struct vcl *vcl; | |
2610 | struct vd_config *vc; | |
2611 | ||
2612 | if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) { | |
2613 | pr_err("This ddf already has an array called %s\n", name); | |
2614 | return 0; | |
2615 | } | |
2616 | venum = find_unused_vde(ddf); | |
2617 | if (venum == DDF_NOTFOUND) { | |
2618 | pr_err("Cannot find spare slot for virtual disk\n"); | |
2619 | return 0; | |
2620 | } | |
2621 | ve = &ddf->virt->entries[venum]; | |
2622 | ||
2623 | /* A Virtual Disk GUID contains the T10 Vendor ID, controller type, | |
2624 | * timestamp, random number | |
2625 | */ | |
2626 | make_header_guid(ve->guid); | |
2627 | ve->unit = cpu_to_be16(info->md_minor); | |
2628 | ve->pad0 = 0xFFFF; | |
2629 | ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid, | |
2630 | DDF_GUID_LEN); | |
2631 | ve->type = cpu_to_be16(0); | |
2632 | ve->state = DDF_state_degraded; /* Will be modified as devices are added */ | |
2633 | if (info->state & 1) /* clean */ | |
2634 | ve->init_state = DDF_init_full; | |
2635 | else | |
2636 | ve->init_state = DDF_init_not; | |
2637 | ||
2638 | memset(ve->pad1, 0xff, 14); | |
2639 | memset(ve->name, ' ', 16); | |
2640 | if (name) | |
2641 | strncpy(ve->name, name, 16); | |
2642 | ddf->virt->populated_vdes = | |
2643 | cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1); | |
2644 | ||
2645 | /* Now create a new vd_config */ | |
2646 | if (posix_memalign((void**)&vcl, 512, | |
2647 | (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) { | |
2648 | pr_err("could not allocate vd_config\n"); | |
2649 | return 0; | |
2650 | } | |
2651 | vcl->vcnum = venum; | |
2652 | vcl->block_sizes = NULL; /* FIXME not for CONCAT */ | |
2653 | vc = &vcl->conf; | |
2654 | ||
2655 | vc->magic = DDF_VD_CONF_MAGIC; | |
2656 | memcpy(vc->guid, ve->guid, DDF_GUID_LEN); | |
2657 | vc->timestamp = cpu_to_be32(time(0)-DECADE); | |
2658 | vc->seqnum = cpu_to_be32(1); | |
2659 | memset(vc->pad0, 0xff, 24); | |
2660 | vc->chunk_shift = chunk_to_shift(info->chunk_size); | |
2661 | if (layout_md2ddf(info, vc) == -1 || | |
2662 | be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) { | |
2663 | pr_err("unsupported RAID level/layout %d/%d with %d disks\n", | |
2664 | info->level, info->layout, info->raid_disks); | |
2665 | free(vcl); | |
2666 | return 0; | |
2667 | } | |
2668 | vc->sec_elmnt_seq = 0; | |
2669 | if (alloc_other_bvds(ddf, vcl) != 0) { | |
2670 | pr_err("could not allocate other bvds\n"); | |
2671 | free(vcl); | |
2672 | return 0; | |
2673 | } | |
2674 | vc->blocks = cpu_to_be64(size * 2); | |
2675 | vc->array_blocks = cpu_to_be64( | |
2676 | calc_array_size(info->level, info->raid_disks, info->layout, | |
2677 | info->chunk_size, size * 2)); | |
2678 | memset(vc->pad1, 0xff, 8); | |
2679 | vc->spare_refs[0] = cpu_to_be32(0xffffffff); | |
2680 | vc->spare_refs[1] = cpu_to_be32(0xffffffff); | |
2681 | vc->spare_refs[2] = cpu_to_be32(0xffffffff); | |
2682 | vc->spare_refs[3] = cpu_to_be32(0xffffffff); | |
2683 | vc->spare_refs[4] = cpu_to_be32(0xffffffff); | |
2684 | vc->spare_refs[5] = cpu_to_be32(0xffffffff); | |
2685 | vc->spare_refs[6] = cpu_to_be32(0xffffffff); | |
2686 | vc->spare_refs[7] = cpu_to_be32(0xffffffff); | |
2687 | memset(vc->cache_pol, 0, 8); | |
2688 | vc->bg_rate = 0x80; | |
2689 | memset(vc->pad2, 0xff, 3); | |
2690 | memset(vc->pad3, 0xff, 52); | |
2691 | memset(vc->pad4, 0xff, 192); | |
2692 | memset(vc->v0, 0xff, 32); | |
2693 | memset(vc->v1, 0xff, 32); | |
2694 | memset(vc->v2, 0xff, 16); | |
2695 | memset(vc->v3, 0xff, 16); | |
2696 | memset(vc->vendor, 0xff, 32); | |
2697 | ||
2698 | memset(vc->phys_refnum, 0xff, 4*ddf->mppe); | |
2699 | memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe); | |
2700 | ||
2701 | for (i = 1; i < vc->sec_elmnt_count; i++) { | |
2702 | memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512); | |
2703 | vcl->other_bvds[i-1]->sec_elmnt_seq = i; | |
2704 | } | |
2705 | ||
2706 | vcl->next = ddf->conflist; | |
2707 | ddf->conflist = vcl; | |
2708 | ddf->currentconf = vcl; | |
2709 | ddf_set_updates_pending(ddf, NULL); | |
2710 | return 1; | |
2711 | } | |
2712 | ||
2713 | static void add_to_super_ddf_bvd(struct supertype *st, | |
2714 | mdu_disk_info_t *dk, int fd, char *devname, | |
2715 | unsigned long long data_offset) | |
2716 | { | |
2717 | /* fd and devname identify a device within the ddf container (st). | |
2718 | * dk identifies a location in the new BVD. | |
2719 | * We need to find suitable free space in that device and update | |
2720 | * the phys_refnum and lba_offset for the newly created vd_config. | |
2721 | * We might also want to update the type in the phys_disk | |
2722 | * section. | |
2723 | * | |
2724 | * Alternately: fd == -1 and we have already chosen which device to | |
2725 | * use and recorded in dlist->raid_disk; | |
2726 | */ | |
2727 | struct dl *dl; | |
2728 | struct ddf_super *ddf = st->sb; | |
2729 | struct vd_config *vc; | |
2730 | unsigned int i; | |
2731 | unsigned long long blocks, pos; | |
2732 | unsigned int raid_disk = dk->raid_disk; | |
2733 | ||
2734 | if (fd == -1) { | |
2735 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2736 | if (dl->raiddisk == dk->raid_disk) | |
2737 | break; | |
2738 | } else { | |
2739 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2740 | if (dl->major == dk->major && | |
2741 | dl->minor == dk->minor) | |
2742 | break; | |
2743 | } | |
2744 | if (!dl || dl->pdnum < 0 || ! (dk->state & (1<<MD_DISK_SYNC))) | |
2745 | return; | |
2746 | ||
2747 | vc = &ddf->currentconf->conf; | |
2748 | if (vc->sec_elmnt_count > 1) { | |
2749 | unsigned int n = be16_to_cpu(vc->prim_elmnt_count); | |
2750 | if (raid_disk >= n) | |
2751 | vc = ddf->currentconf->other_bvds[raid_disk / n - 1]; | |
2752 | raid_disk %= n; | |
2753 | } | |
2754 | ||
2755 | blocks = be64_to_cpu(vc->blocks); | |
2756 | if (ddf->currentconf->block_sizes) | |
2757 | blocks = ddf->currentconf->block_sizes[dk->raid_disk]; | |
2758 | ||
2759 | pos = find_space(ddf, dl, data_offset, &blocks); | |
2760 | if (pos == INVALID_SECTORS) | |
2761 | return; | |
2762 | ||
2763 | ddf->currentdev = dk->raid_disk; | |
2764 | vc->phys_refnum[raid_disk] = dl->disk.refnum; | |
2765 | LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos); | |
2766 | ||
2767 | for (i = 0; i < ddf->max_part ; i++) | |
2768 | if (dl->vlist[i] == NULL) | |
2769 | break; | |
2770 | if (i == ddf->max_part) | |
2771 | return; | |
2772 | dl->vlist[i] = ddf->currentconf; | |
2773 | ||
2774 | if (fd >= 0) | |
2775 | dl->fd = fd; | |
2776 | if (devname) | |
2777 | dl->devname = devname; | |
2778 | ||
2779 | /* Check if we can mark array as optimal yet */ | |
2780 | i = ddf->currentconf->vcnum; | |
2781 | ddf->virt->entries[i].state = | |
2782 | (ddf->virt->entries[i].state & ~DDF_state_mask) | |
2783 | | get_svd_state(ddf, ddf->currentconf); | |
2784 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
2785 | cpu_to_be16(DDF_Global_Spare)); | |
2786 | be16_set(ddf->phys->entries[dl->pdnum].type, | |
2787 | cpu_to_be16(DDF_Active_in_VD)); | |
2788 | dprintf("added disk %d/%08x to VD %d/%s as disk %d\n", | |
2789 | dl->pdnum, be32_to_cpu(dl->disk.refnum), | |
2790 | ddf->currentconf->vcnum, guid_str(vc->guid), | |
2791 | dk->raid_disk); | |
2792 | ddf_set_updates_pending(ddf, vc); | |
2793 | } | |
2794 | ||
2795 | static unsigned int find_unused_pde(const struct ddf_super *ddf) | |
2796 | { | |
2797 | unsigned int i; | |
2798 | for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) { | |
2799 | if (all_ff(ddf->phys->entries[i].guid)) | |
2800 | return i; | |
2801 | } | |
2802 | return DDF_NOTFOUND; | |
2803 | } | |
2804 | ||
2805 | static void _set_config_size(struct phys_disk_entry *pde, const struct dl *dl) | |
2806 | { | |
2807 | __u64 cfs, t; | |
2808 | cfs = min(dl->size - 32*1024*2ULL, be64_to_cpu(dl->primary_lba)); | |
2809 | t = be64_to_cpu(dl->secondary_lba); | |
2810 | if (t != ~(__u64)0) | |
2811 | cfs = min(cfs, t); | |
2812 | /* | |
2813 | * Some vendor DDF structures interpret workspace_lba | |
2814 | * very differently than we do: Make a sanity check on the value. | |
2815 | */ | |
2816 | t = be64_to_cpu(dl->workspace_lba); | |
2817 | if (t < cfs) { | |
2818 | __u64 wsp = cfs - t; | |
2819 | if (wsp > 1024*1024*2ULL && wsp > dl->size / 16) { | |
2820 | pr_err("%x:%x: workspace size 0x%llx too big, ignoring\n", | |
2821 | dl->major, dl->minor, (unsigned long long)wsp); | |
2822 | } else | |
2823 | cfs = t; | |
2824 | } | |
2825 | pde->config_size = cpu_to_be64(cfs); | |
2826 | dprintf("%x:%x config_size %llx, DDF structure is %llx blocks\n", | |
2827 | dl->major, dl->minor, | |
2828 | (unsigned long long)cfs, (unsigned long long)(dl->size-cfs)); | |
2829 | } | |
2830 | ||
2831 | /* Add a device to a container, either while creating it or while | |
2832 | * expanding a pre-existing container | |
2833 | */ | |
2834 | static int add_to_super_ddf(struct supertype *st, | |
2835 | mdu_disk_info_t *dk, int fd, char *devname, | |
2836 | unsigned long long data_offset) | |
2837 | { | |
2838 | struct ddf_super *ddf = st->sb; | |
2839 | struct dl *dd; | |
2840 | time_t now; | |
2841 | struct tm *tm; | |
2842 | unsigned long long size; | |
2843 | struct phys_disk_entry *pde; | |
2844 | unsigned int n, i; | |
2845 | struct stat stb; | |
2846 | __u32 *tptr; | |
2847 | ||
2848 | if (ddf->currentconf) { | |
2849 | add_to_super_ddf_bvd(st, dk, fd, devname, data_offset); | |
2850 | return 0; | |
2851 | } | |
2852 | ||
2853 | /* This is device numbered dk->number. We need to create | |
2854 | * a phys_disk entry and a more detailed disk_data entry. | |
2855 | */ | |
2856 | fstat(fd, &stb); | |
2857 | n = find_unused_pde(ddf); | |
2858 | if (n == DDF_NOTFOUND) { | |
2859 | pr_err("No free slot in array, cannot add disk\n"); | |
2860 | return 1; | |
2861 | } | |
2862 | pde = &ddf->phys->entries[n]; | |
2863 | get_dev_size(fd, NULL, &size); | |
2864 | if (size <= 32*1024*1024) { | |
2865 | pr_err("device size must be at least 32MB\n"); | |
2866 | return 1; | |
2867 | } | |
2868 | size >>= 9; | |
2869 | ||
2870 | if (posix_memalign((void**)&dd, 512, | |
2871 | sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) { | |
2872 | pr_err("could allocate buffer for new disk, aborting\n"); | |
2873 | return 1; | |
2874 | } | |
2875 | dd->major = major(stb.st_rdev); | |
2876 | dd->minor = minor(stb.st_rdev); | |
2877 | dd->devname = devname; | |
2878 | dd->fd = fd; | |
2879 | dd->spare = NULL; | |
2880 | ||
2881 | dd->disk.magic = DDF_PHYS_DATA_MAGIC; | |
2882 | now = time(0); | |
2883 | tm = localtime(&now); | |
2884 | sprintf(dd->disk.guid, "%8s%04d%02d%02d", T10, | |
2885 | (__u16)tm->tm_year+1900, | |
2886 | (__u8)tm->tm_mon+1, (__u8)tm->tm_mday); | |
2887 | tptr = (__u32 *)(dd->disk.guid + 16); | |
2888 | *tptr++ = random32(); | |
2889 | *tptr = random32(); | |
2890 | ||
2891 | do { | |
2892 | /* Cannot be bothered finding a CRC of some irrelevant details*/ | |
2893 | dd->disk.refnum._v32 = random32(); | |
2894 | for (i = be16_to_cpu(ddf->active->max_pd_entries); | |
2895 | i > 0; i--) | |
2896 | if (be32_eq(ddf->phys->entries[i-1].refnum, | |
2897 | dd->disk.refnum)) | |
2898 | break; | |
2899 | } while (i > 0); | |
2900 | ||
2901 | dd->disk.forced_ref = 1; | |
2902 | dd->disk.forced_guid = 1; | |
2903 | memset(dd->disk.vendor, ' ', 32); | |
2904 | memcpy(dd->disk.vendor, "Linux", 5); | |
2905 | memset(dd->disk.pad, 0xff, 442); | |
2906 | for (i = 0; i < ddf->max_part ; i++) | |
2907 | dd->vlist[i] = NULL; | |
2908 | ||
2909 | dd->pdnum = n; | |
2910 | ||
2911 | if (st->update_tail) { | |
2912 | int len = (sizeof(struct phys_disk) + | |
2913 | sizeof(struct phys_disk_entry)); | |
2914 | struct phys_disk *pd; | |
2915 | ||
2916 | pd = xmalloc(len); | |
2917 | pd->magic = DDF_PHYS_RECORDS_MAGIC; | |
2918 | pd->used_pdes = cpu_to_be16(n); | |
2919 | pde = &pd->entries[0]; | |
2920 | dd->mdupdate = pd; | |
2921 | } else | |
2922 | ddf->phys->used_pdes = cpu_to_be16( | |
2923 | 1 + be16_to_cpu(ddf->phys->used_pdes)); | |
2924 | ||
2925 | memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN); | |
2926 | pde->refnum = dd->disk.refnum; | |
2927 | pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare); | |
2928 | pde->state = cpu_to_be16(DDF_Online); | |
2929 | dd->size = size; | |
2930 | /* | |
2931 | * If there is already a device in dlist, try to reserve the same | |
2932 | * amount of workspace. Otherwise, use 32MB. | |
2933 | * We checked disk size above already. | |
2934 | */ | |
2935 | #define __calc_lba(new, old, lba, mb) do { \ | |
2936 | unsigned long long dif; \ | |
2937 | if ((old) != NULL) \ | |
2938 | dif = (old)->size - be64_to_cpu((old)->lba); \ | |
2939 | else \ | |
2940 | dif = (new)->size; \ | |
2941 | if ((new)->size > dif) \ | |
2942 | (new)->lba = cpu_to_be64((new)->size - dif); \ | |
2943 | else \ | |
2944 | (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \ | |
2945 | } while (0) | |
2946 | __calc_lba(dd, ddf->dlist, workspace_lba, 32); | |
2947 | __calc_lba(dd, ddf->dlist, primary_lba, 16); | |
2948 | if (ddf->dlist == NULL || | |
2949 | be64_to_cpu(ddf->dlist->secondary_lba) != ~(__u64)0) | |
2950 | __calc_lba(dd, ddf->dlist, secondary_lba, 32); | |
2951 | _set_config_size(pde, dd); | |
2952 | ||
2953 | sprintf(pde->path, "%17.17s","Information: nil") ; | |
2954 | memset(pde->pad, 0xff, 6); | |
2955 | ||
2956 | if (st->update_tail) { | |
2957 | dd->next = ddf->add_list; | |
2958 | ddf->add_list = dd; | |
2959 | } else { | |
2960 | dd->next = ddf->dlist; | |
2961 | ddf->dlist = dd; | |
2962 | ddf_set_updates_pending(ddf, NULL); | |
2963 | } | |
2964 | ||
2965 | return 0; | |
2966 | } | |
2967 | ||
2968 | static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk) | |
2969 | { | |
2970 | struct ddf_super *ddf = st->sb; | |
2971 | struct dl *dl; | |
2972 | ||
2973 | /* mdmon has noticed that this disk (dk->major/dk->minor) has | |
2974 | * disappeared from the container. | |
2975 | * We need to arrange that it disappears from the metadata and | |
2976 | * internal data structures too. | |
2977 | * Most of the work is done by ddf_process_update which edits | |
2978 | * the metadata and closes the file handle and attaches the memory | |
2979 | * where free_updates will free it. | |
2980 | */ | |
2981 | for (dl = ddf->dlist; dl ; dl = dl->next) | |
2982 | if (dl->major == dk->major && | |
2983 | dl->minor == dk->minor) | |
2984 | break; | |
2985 | if (!dl || dl->pdnum < 0) | |
2986 | return -1; | |
2987 | ||
2988 | if (st->update_tail) { | |
2989 | int len = (sizeof(struct phys_disk) + | |
2990 | sizeof(struct phys_disk_entry)); | |
2991 | struct phys_disk *pd; | |
2992 | ||
2993 | pd = xmalloc(len); | |
2994 | pd->magic = DDF_PHYS_RECORDS_MAGIC; | |
2995 | pd->used_pdes = cpu_to_be16(dl->pdnum); | |
2996 | pd->entries[0].state = cpu_to_be16(DDF_Missing); | |
2997 | append_metadata_update(st, pd, len); | |
2998 | } | |
2999 | return 0; | |
3000 | } | |
3001 | ||
3002 | /* | |
3003 | * This is the write_init_super method for a ddf container. It is | |
3004 | * called when creating a container or adding another device to a | |
3005 | * container. | |
3006 | */ | |
3007 | ||
3008 | static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type) | |
3009 | { | |
3010 | unsigned long long sector; | |
3011 | struct ddf_header *header; | |
3012 | int fd, i, n_config, conf_size, buf_size; | |
3013 | int ret = 0; | |
3014 | char *conf; | |
3015 | ||
3016 | fd = d->fd; | |
3017 | ||
3018 | switch (type) { | |
3019 | case DDF_HEADER_PRIMARY: | |
3020 | header = &ddf->primary; | |
3021 | sector = be64_to_cpu(header->primary_lba); | |
3022 | break; | |
3023 | case DDF_HEADER_SECONDARY: | |
3024 | header = &ddf->secondary; | |
3025 | sector = be64_to_cpu(header->secondary_lba); | |
3026 | break; | |
3027 | default: | |
3028 | return 0; | |
3029 | } | |
3030 | if (sector == ~(__u64)0) | |
3031 | return 0; | |
3032 | ||
3033 | header->type = type; | |
3034 | header->openflag = 1; | |
3035 | header->crc = calc_crc(header, 512); | |
3036 | ||
3037 | lseek64(fd, sector<<9, 0); | |
3038 | if (write(fd, header, 512) < 0) | |
3039 | goto out; | |
3040 | ||
3041 | ddf->controller.crc = calc_crc(&ddf->controller, 512); | |
3042 | if (write(fd, &ddf->controller, 512) < 0) | |
3043 | goto out; | |
3044 | ||
3045 | ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize); | |
3046 | if (write(fd, ddf->phys, ddf->pdsize) < 0) | |
3047 | goto out; | |
3048 | ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize); | |
3049 | if (write(fd, ddf->virt, ddf->vdsize) < 0) | |
3050 | goto out; | |
3051 | ||
3052 | /* Now write lots of config records. */ | |
3053 | n_config = ddf->max_part; | |
3054 | conf_size = ddf->conf_rec_len * 512; | |
3055 | conf = ddf->conf; | |
3056 | buf_size = conf_size * (n_config + 1); | |
3057 | if (!conf) { | |
3058 | if (posix_memalign((void**)&conf, 512, buf_size) != 0) | |
3059 | goto out; | |
3060 | ddf->conf = conf; | |
3061 | } | |
3062 | for (i = 0 ; i <= n_config ; i++) { | |
3063 | struct vcl *c; | |
3064 | struct vd_config *vdc = NULL; | |
3065 | if (i == n_config) { | |
3066 | c = (struct vcl *)d->spare; | |
3067 | if (c) | |
3068 | vdc = &c->conf; | |
3069 | } else { | |
3070 | unsigned int dummy; | |
3071 | c = d->vlist[i]; | |
3072 | if (c) | |
3073 | get_pd_index_from_refnum( | |
3074 | c, d->disk.refnum, | |
3075 | ddf->mppe, | |
3076 | (const struct vd_config **)&vdc, | |
3077 | &dummy); | |
3078 | } | |
3079 | if (vdc) { | |
3080 | dprintf("writing conf record %i on disk %08x for %s/%u\n", | |
3081 | i, be32_to_cpu(d->disk.refnum), | |
3082 | guid_str(vdc->guid), | |
3083 | vdc->sec_elmnt_seq); | |
3084 | vdc->crc = calc_crc(vdc, conf_size); | |
3085 | memcpy(conf + i*conf_size, vdc, conf_size); | |
3086 | } else | |
3087 | memset(conf + i*conf_size, 0xff, conf_size); | |
3088 | } | |
3089 | if (write(fd, conf, buf_size) != buf_size) | |
3090 | goto out; | |
3091 | ||
3092 | d->disk.crc = calc_crc(&d->disk, 512); | |
3093 | if (write(fd, &d->disk, 512) < 0) | |
3094 | goto out; | |
3095 | ||
3096 | ret = 1; | |
3097 | out: | |
3098 | header->openflag = 0; | |
3099 | header->crc = calc_crc(header, 512); | |
3100 | ||
3101 | lseek64(fd, sector<<9, 0); | |
3102 | if (write(fd, header, 512) < 0) | |
3103 | ret = 0; | |
3104 | ||
3105 | return ret; | |
3106 | } | |
3107 | ||
3108 | static int _write_super_to_disk(struct ddf_super *ddf, struct dl *d) | |
3109 | { | |
3110 | unsigned long long size; | |
3111 | int fd = d->fd; | |
3112 | if (fd < 0) | |
3113 | return 0; | |
3114 | ||
3115 | /* We need to fill in the primary, (secondary) and workspace | |
3116 | * lba's in the headers, set their checksums, | |
3117 | * Also checksum phys, virt.... | |
3118 | * | |
3119 | * Then write everything out, finally the anchor is written. | |
3120 | */ | |
3121 | get_dev_size(fd, NULL, &size); | |
3122 | size /= 512; | |
3123 | memcpy(&ddf->anchor, ddf->active, 512); | |
3124 | if (be64_to_cpu(d->workspace_lba) != 0ULL) | |
3125 | ddf->anchor.workspace_lba = d->workspace_lba; | |
3126 | else | |
3127 | ddf->anchor.workspace_lba = | |
3128 | cpu_to_be64(size - 32*1024*2); | |
3129 | if (be64_to_cpu(d->primary_lba) != 0ULL) | |
3130 | ddf->anchor.primary_lba = d->primary_lba; | |
3131 | else | |
3132 | ddf->anchor.primary_lba = | |
3133 | cpu_to_be64(size - 16*1024*2); | |
3134 | if (be64_to_cpu(d->secondary_lba) != 0ULL) | |
3135 | ddf->anchor.secondary_lba = d->secondary_lba; | |
3136 | else | |
3137 | ddf->anchor.secondary_lba = | |
3138 | cpu_to_be64(size - 32*1024*2); | |
3139 | ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE); | |
3140 | memcpy(&ddf->primary, &ddf->anchor, 512); | |
3141 | memcpy(&ddf->secondary, &ddf->anchor, 512); | |
3142 | ||
3143 | ddf->anchor.type = DDF_HEADER_ANCHOR; | |
3144 | ddf->anchor.openflag = 0xFF; /* 'open' means nothing */ | |
3145 | ddf->anchor.seq = cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */ | |
3146 | ddf->anchor.crc = calc_crc(&ddf->anchor, 512); | |
3147 | ||
3148 | if (!__write_ddf_structure(d, ddf, DDF_HEADER_PRIMARY)) | |
3149 | return 0; | |
3150 | ||
3151 | if (!__write_ddf_structure(d, ddf, DDF_HEADER_SECONDARY)) | |
3152 | return 0; | |
3153 | ||
3154 | lseek64(fd, (size-1)*512, SEEK_SET); | |
3155 | if (write(fd, &ddf->anchor, 512) < 0) | |
3156 | return 0; | |
3157 | ||
3158 | return 1; | |
3159 | } | |
3160 | ||
3161 | static int __write_init_super_ddf(struct supertype *st) | |
3162 | { | |
3163 | struct ddf_super *ddf = st->sb; | |
3164 | struct dl *d; | |
3165 | int attempts = 0; | |
3166 | int successes = 0; | |
3167 | ||
3168 | pr_state(ddf, __func__); | |
3169 | ||
3170 | /* try to write updated metadata, | |
3171 | * if we catch a failure move on to the next disk | |
3172 | */ | |
3173 | for (d = ddf->dlist; d; d=d->next) { | |
3174 | attempts++; | |
3175 | successes += _write_super_to_disk(ddf, d); | |
3176 | } | |
3177 | ||
3178 | return attempts != successes; | |
3179 | } | |
3180 | ||
3181 | static int write_init_super_ddf(struct supertype *st) | |
3182 | { | |
3183 | struct ddf_super *ddf = st->sb; | |
3184 | struct vcl *currentconf = ddf->currentconf; | |
3185 | ||
3186 | /* We are done with currentconf - reset it so st refers to the container */ | |
3187 | ddf->currentconf = NULL; | |
3188 | ||
3189 | if (st->update_tail) { | |
3190 | /* queue the virtual_disk and vd_config as metadata updates */ | |
3191 | struct virtual_disk *vd; | |
3192 | struct vd_config *vc; | |
3193 | int len, tlen; | |
3194 | unsigned int i; | |
3195 | ||
3196 | if (!currentconf) { | |
3197 | /* Must be adding a physical disk to the container */ | |
3198 | int len = (sizeof(struct phys_disk) + | |
3199 | sizeof(struct phys_disk_entry)); | |
3200 | ||
3201 | /* adding a disk to the container. */ | |
3202 | if (!ddf->add_list) | |
3203 | return 0; | |
3204 | ||
3205 | append_metadata_update(st, ddf->add_list->mdupdate, len); | |
3206 | ddf->add_list->mdupdate = NULL; | |
3207 | return 0; | |
3208 | } | |
3209 | ||
3210 | /* Newly created VD */ | |
3211 | ||
3212 | /* First the virtual disk. We have a slightly fake header */ | |
3213 | len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry); | |
3214 | vd = xmalloc(len); | |
3215 | *vd = *ddf->virt; | |
3216 | vd->entries[0] = ddf->virt->entries[currentconf->vcnum]; | |
3217 | vd->populated_vdes = cpu_to_be16(currentconf->vcnum); | |
3218 | append_metadata_update(st, vd, len); | |
3219 | ||
3220 | /* Then the vd_config */ | |
3221 | len = ddf->conf_rec_len * 512; | |
3222 | tlen = len * currentconf->conf.sec_elmnt_count; | |
3223 | vc = xmalloc(tlen); | |
3224 | memcpy(vc, ¤tconf->conf, len); | |
3225 | for (i = 1; i < currentconf->conf.sec_elmnt_count; i++) | |
3226 | memcpy((char *)vc + i*len, currentconf->other_bvds[i-1], | |
3227 | len); | |
3228 | append_metadata_update(st, vc, tlen); | |
3229 | ||
3230 | return 0; | |
3231 | } else { | |
3232 | struct dl *d; | |
3233 | if (!currentconf) | |
3234 | for (d = ddf->dlist; d; d=d->next) | |
3235 | while (Kill(d->devname, NULL, 0, -1, 1) == 0); | |
3236 | /* Note: we don't close the fd's now, but a subsequent | |
3237 | * ->free_super() will | |
3238 | */ | |
3239 | return __write_init_super_ddf(st); | |
3240 | } | |
3241 | } | |
3242 | ||
3243 | static __u64 avail_size_ddf(struct supertype *st, __u64 devsize, | |
3244 | unsigned long long data_offset) | |
3245 | { | |
3246 | /* We must reserve the last 32Meg */ | |
3247 | if (devsize <= 32*1024*2) | |
3248 | return 0; | |
3249 | return devsize - 32*1024*2; | |
3250 | } | |
3251 | ||
3252 | static int reserve_space(struct supertype *st, int raiddisks, | |
3253 | unsigned long long size, int chunk, | |
3254 | unsigned long long data_offset, | |
3255 | unsigned long long *freesize) | |
3256 | { | |
3257 | /* Find 'raiddisks' spare extents at least 'size' big (but | |
3258 | * only caring about multiples of 'chunk') and remember | |
3259 | * them. If size==0, find the largest size possible. | |
3260 | * Report available size in *freesize | |
3261 | * If space cannot be found, fail. | |
3262 | */ | |
3263 | struct dl *dl; | |
3264 | struct ddf_super *ddf = st->sb; | |
3265 | int cnt = 0; | |
3266 | ||
3267 | for (dl = ddf->dlist; dl ; dl=dl->next) { | |
3268 | dl->raiddisk = -1; | |
3269 | dl->esize = 0; | |
3270 | } | |
3271 | /* Now find largest extent on each device */ | |
3272 | for (dl = ddf->dlist ; dl ; dl=dl->next) { | |
3273 | unsigned long long minsize = ULLONG_MAX; | |
3274 | ||
3275 | find_space(ddf, dl, data_offset, &minsize); | |
3276 | if (minsize >= size && minsize >= (unsigned)chunk) { | |
3277 | cnt++; | |
3278 | dl->esize = minsize; | |
3279 | } | |
3280 | } | |
3281 | if (cnt < raiddisks) { | |
3282 | pr_err("not enough devices with space to create array.\n"); | |
3283 | return 0; /* No enough free spaces large enough */ | |
3284 | } | |
3285 | if (size == 0) { | |
3286 | /* choose the largest size of which there are at least 'raiddisk' */ | |
3287 | for (dl = ddf->dlist ; dl ; dl=dl->next) { | |
3288 | struct dl *dl2; | |
3289 | if (dl->esize <= size) | |
3290 | continue; | |
3291 | /* This is bigger than 'size', see if there are enough */ | |
3292 | cnt = 0; | |
3293 | for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next) | |
3294 | if (dl2->esize >= dl->esize) | |
3295 | cnt++; | |
3296 | if (cnt >= raiddisks) | |
3297 | size = dl->esize; | |
3298 | } | |
3299 | if (chunk) { | |
3300 | size = size / chunk; | |
3301 | size *= chunk; | |
3302 | } | |
3303 | *freesize = size; | |
3304 | if (size < 32) { | |
3305 | pr_err("not enough spare devices to create array.\n"); | |
3306 | return 0; | |
3307 | } | |
3308 | } | |
3309 | /* We have a 'size' of which there are enough spaces. | |
3310 | * We simply do a first-fit */ | |
3311 | cnt = 0; | |
3312 | for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) { | |
3313 | if (dl->esize < size) | |
3314 | continue; | |
3315 | ||
3316 | dl->raiddisk = cnt; | |
3317 | cnt++; | |
3318 | } | |
3319 | return 1; | |
3320 | } | |
3321 | ||
3322 | static int validate_geometry_ddf(struct supertype *st, | |
3323 | int level, int layout, int raiddisks, | |
3324 | int *chunk, unsigned long long size, | |
3325 | unsigned long long data_offset, | |
3326 | char *dev, unsigned long long *freesize, | |
3327 | int consistency_policy, int verbose) | |
3328 | { | |
3329 | int fd; | |
3330 | struct mdinfo *sra; | |
3331 | int cfd; | |
3332 | ||
3333 | /* ddf potentially supports lots of things, but it depends on | |
3334 | * what devices are offered (and maybe kernel version?) | |
3335 | * If given unused devices, we will make a container. | |
3336 | * If given devices in a container, we will make a BVD. | |
3337 | * If given BVDs, we make an SVD, changing all the GUIDs in the process. | |
3338 | */ | |
3339 | ||
3340 | if (*chunk == UnSet) | |
3341 | *chunk = DEFAULT_CHUNK; | |
3342 | ||
3343 | if (level == LEVEL_NONE) | |
3344 | level = LEVEL_CONTAINER; | |
3345 | if (level == LEVEL_CONTAINER) { | |
3346 | /* Must be a fresh device to add to a container */ | |
3347 | return validate_geometry_ddf_container(st, level, layout, | |
3348 | raiddisks, *chunk, | |
3349 | size, data_offset, dev, | |
3350 | freesize, | |
3351 | verbose); | |
3352 | } | |
3353 | ||
3354 | if (!dev) { | |
3355 | mdu_array_info_t array = { | |
3356 | .level = level, | |
3357 | .layout = layout, | |
3358 | .raid_disks = raiddisks | |
3359 | }; | |
3360 | struct vd_config conf; | |
3361 | if (layout_md2ddf(&array, &conf) == -1) { | |
3362 | if (verbose) | |
3363 | pr_err("DDF does not support level %d /layout %d arrays with %d disks\n", | |
3364 | level, layout, raiddisks); | |
3365 | return 0; | |
3366 | } | |
3367 | /* Should check layout? etc */ | |
3368 | ||
3369 | if (st->sb && freesize) { | |
3370 | /* --create was given a container to create in. | |
3371 | * So we need to check that there are enough | |
3372 | * free spaces and return the amount of space. | |
3373 | * We may as well remember which drives were | |
3374 | * chosen so that add_to_super/getinfo_super | |
3375 | * can return them. | |
3376 | */ | |
3377 | return reserve_space(st, raiddisks, size, *chunk, | |
3378 | data_offset, freesize); | |
3379 | } | |
3380 | return 1; | |
3381 | } | |
3382 | ||
3383 | if (st->sb) { | |
3384 | /* A container has already been opened, so we are | |
3385 | * creating in there. Maybe a BVD, maybe an SVD. | |
3386 | * Should make a distinction one day. | |
3387 | */ | |
3388 | return validate_geometry_ddf_bvd(st, level, layout, raiddisks, | |
3389 | chunk, size, data_offset, dev, | |
3390 | freesize, | |
3391 | verbose); | |
3392 | } | |
3393 | /* This is the first device for the array. | |
3394 | * If it is a container, we read it in and do automagic allocations, | |
3395 | * no other devices should be given. | |
3396 | * Otherwise it must be a member device of a container, and we | |
3397 | * do manual allocation. | |
3398 | * Later we should check for a BVD and make an SVD. | |
3399 | */ | |
3400 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
3401 | if (fd >= 0) { | |
3402 | close(fd); | |
3403 | /* Just a bare device, no good to us */ | |
3404 | if (verbose) | |
3405 | pr_err("ddf: Cannot create this array on device %s - a container is required.\n", | |
3406 | dev); | |
3407 | return 0; | |
3408 | } | |
3409 | if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) { | |
3410 | if (verbose) | |
3411 | pr_err("ddf: Cannot open %s: %s\n", | |
3412 | dev, strerror(errno)); | |
3413 | return 0; | |
3414 | } | |
3415 | /* Well, it is in use by someone, maybe a 'ddf' container. */ | |
3416 | cfd = open_container(fd); | |
3417 | if (cfd < 0) { | |
3418 | close(fd); | |
3419 | if (verbose) | |
3420 | pr_err("ddf: Cannot use %s: %s\n", | |
3421 | dev, strerror(EBUSY)); | |
3422 | return 0; | |
3423 | } | |
3424 | sra = sysfs_read(cfd, NULL, GET_VERSION); | |
3425 | close(fd); | |
3426 | if (sra && sra->array.major_version == -1 && | |
3427 | strcmp(sra->text_version, "ddf") == 0) { | |
3428 | /* This is a member of a ddf container. Load the container | |
3429 | * and try to create a bvd | |
3430 | */ | |
3431 | struct ddf_super *ddf; | |
3432 | if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) { | |
3433 | st->sb = ddf; | |
3434 | strcpy(st->container_devnm, fd2devnm(cfd)); | |
3435 | close(cfd); | |
3436 | return validate_geometry_ddf_bvd(st, level, layout, | |
3437 | raiddisks, chunk, size, | |
3438 | data_offset, | |
3439 | dev, freesize, | |
3440 | verbose); | |
3441 | } | |
3442 | close(cfd); | |
3443 | } else /* device may belong to a different container */ | |
3444 | return 0; | |
3445 | ||
3446 | return 1; | |
3447 | } | |
3448 | ||
3449 | static int | |
3450 | validate_geometry_ddf_container(struct supertype *st, | |
3451 | int level, int layout, int raiddisks, | |
3452 | int chunk, unsigned long long size, | |
3453 | unsigned long long data_offset, | |
3454 | char *dev, unsigned long long *freesize, | |
3455 | int verbose) | |
3456 | { | |
3457 | int fd; | |
3458 | unsigned long long ldsize; | |
3459 | ||
3460 | if (level != LEVEL_CONTAINER) | |
3461 | return 0; | |
3462 | if (!dev) | |
3463 | return 1; | |
3464 | ||
3465 | fd = open(dev, O_RDONLY|O_EXCL, 0); | |
3466 | if (fd < 0) { | |
3467 | if (verbose) | |
3468 | pr_err("ddf: Cannot open %s: %s\n", | |
3469 | dev, strerror(errno)); | |
3470 | return 0; | |
3471 | } | |
3472 | if (!get_dev_size(fd, dev, &ldsize)) { | |
3473 | close(fd); | |
3474 | return 0; | |
3475 | } | |
3476 | close(fd); | |
3477 | ||
3478 | *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS); | |
3479 | if (*freesize == 0) | |
3480 | return 0; | |
3481 | ||
3482 | return 1; | |
3483 | } | |
3484 | ||
3485 | static int validate_geometry_ddf_bvd(struct supertype *st, | |
3486 | int level, int layout, int raiddisks, | |
3487 | int *chunk, unsigned long long size, | |
3488 | unsigned long long data_offset, | |
3489 | char *dev, unsigned long long *freesize, | |
3490 | int verbose) | |
3491 | { | |
3492 | dev_t rdev; | |
3493 | struct ddf_super *ddf = st->sb; | |
3494 | struct dl *dl; | |
3495 | unsigned long long maxsize; | |
3496 | /* ddf/bvd supports lots of things, but not containers */ | |
3497 | if (level == LEVEL_CONTAINER) { | |
3498 | if (verbose) | |
3499 | pr_err("DDF cannot create a container within an container\n"); | |
3500 | return 0; | |
3501 | } | |
3502 | /* We must have the container info already read in. */ | |
3503 | if (!ddf) | |
3504 | return 0; | |
3505 | ||
3506 | if (!dev) { | |
3507 | /* General test: make sure there is space for | |
3508 | * 'raiddisks' device extents of size 'size'. | |
3509 | */ | |
3510 | unsigned long long minsize = size; | |
3511 | int dcnt = 0; | |
3512 | if (minsize == 0) | |
3513 | minsize = 8; | |
3514 | for (dl = ddf->dlist; dl ; dl = dl->next) { | |
3515 | if (find_space(ddf, dl, data_offset, &minsize) != | |
3516 | INVALID_SECTORS) | |
3517 | dcnt++; | |
3518 | } | |
3519 | if (dcnt < raiddisks) { | |
3520 | if (verbose) | |
3521 | pr_err("ddf: Not enough devices with space for this array (%d < %d)\n", | |
3522 | dcnt, raiddisks); | |
3523 | return 0; | |
3524 | } | |
3525 | return 1; | |
3526 | } | |
3527 | /* This device must be a member of the set */ | |
3528 | if (!stat_is_blkdev(dev, &rdev)) | |
3529 | return 0; | |
3530 | for (dl = ddf->dlist ; dl ; dl = dl->next) { | |
3531 | if (dl->major == (int)major(rdev) && | |
3532 | dl->minor == (int)minor(rdev)) | |
3533 | break; | |
3534 | } | |
3535 | if (!dl) { | |
3536 | if (verbose) | |
3537 | pr_err("ddf: %s is not in the same DDF set\n", | |
3538 | dev); | |
3539 | return 0; | |
3540 | } | |
3541 | maxsize = ULLONG_MAX; | |
3542 | find_space(ddf, dl, data_offset, &maxsize); | |
3543 | *freesize = maxsize; | |
3544 | ||
3545 | return 1; | |
3546 | } | |
3547 | ||
3548 | static int load_super_ddf_all(struct supertype *st, int fd, | |
3549 | void **sbp, char *devname) | |
3550 | { | |
3551 | struct mdinfo *sra; | |
3552 | struct ddf_super *super; | |
3553 | struct mdinfo *sd, *best = NULL; | |
3554 | int bestseq = 0; | |
3555 | int seq; | |
3556 | char nm[20]; | |
3557 | int dfd; | |
3558 | ||
3559 | sra = sysfs_read(fd, NULL, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE); | |
3560 | if (!sra) | |
3561 | return 1; | |
3562 | if (sra->array.major_version != -1 || | |
3563 | sra->array.minor_version != -2 || | |
3564 | strcmp(sra->text_version, "ddf") != 0) | |
3565 | return 1; | |
3566 | ||
3567 | if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0) | |
3568 | return 1; | |
3569 | memset(super, 0, sizeof(*super)); | |
3570 | ||
3571 | /* first, try each device, and choose the best ddf */ | |
3572 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
3573 | int rv; | |
3574 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
3575 | dfd = dev_open(nm, O_RDONLY); | |
3576 | if (dfd < 0) | |
3577 | return 2; | |
3578 | rv = load_ddf_headers(dfd, super, NULL); | |
3579 | close(dfd); | |
3580 | if (rv == 0) { | |
3581 | seq = be32_to_cpu(super->active->seq); | |
3582 | if (super->active->openflag) | |
3583 | seq--; | |
3584 | if (!best || seq > bestseq) { | |
3585 | bestseq = seq; | |
3586 | best = sd; | |
3587 | } | |
3588 | } | |
3589 | } | |
3590 | if (!best) | |
3591 | return 1; | |
3592 | /* OK, load this ddf */ | |
3593 | sprintf(nm, "%d:%d", best->disk.major, best->disk.minor); | |
3594 | dfd = dev_open(nm, O_RDONLY); | |
3595 | if (dfd < 0) | |
3596 | return 1; | |
3597 | load_ddf_headers(dfd, super, NULL); | |
3598 | load_ddf_global(dfd, super, NULL); | |
3599 | close(dfd); | |
3600 | /* Now we need the device-local bits */ | |
3601 | for (sd = sra->devs ; sd ; sd = sd->next) { | |
3602 | int rv; | |
3603 | ||
3604 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
3605 | dfd = dev_open(nm, O_RDWR); | |
3606 | if (dfd < 0) | |
3607 | return 2; | |
3608 | rv = load_ddf_headers(dfd, super, NULL); | |
3609 | if (rv == 0) | |
3610 | rv = load_ddf_local(dfd, super, NULL, 1); | |
3611 | if (rv) | |
3612 | return 1; | |
3613 | } | |
3614 | ||
3615 | *sbp = super; | |
3616 | if (st->ss == NULL) { | |
3617 | st->ss = &super_ddf; | |
3618 | st->minor_version = 0; | |
3619 | st->max_devs = 512; | |
3620 | } | |
3621 | strcpy(st->container_devnm, fd2devnm(fd)); | |
3622 | return 0; | |
3623 | } | |
3624 | ||
3625 | static int load_container_ddf(struct supertype *st, int fd, | |
3626 | char *devname) | |
3627 | { | |
3628 | return load_super_ddf_all(st, fd, &st->sb, devname); | |
3629 | } | |
3630 | ||
3631 | static int check_secondary(const struct vcl *vc) | |
3632 | { | |
3633 | const struct vd_config *conf = &vc->conf; | |
3634 | int i; | |
3635 | ||
3636 | /* The only DDF secondary RAID level md can support is | |
3637 | * RAID 10, if the stripe sizes and Basic volume sizes | |
3638 | * are all equal. | |
3639 | * Other configurations could in theory be supported by exposing | |
3640 | * the BVDs to user space and using device mapper for the secondary | |
3641 | * mapping. So far we don't support that. | |
3642 | */ | |
3643 | ||
3644 | __u64 sec_elements[4] = {0, 0, 0, 0}; | |
3645 | #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63))) | |
3646 | #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0) | |
3647 | ||
3648 | if (vc->other_bvds == NULL) { | |
3649 | pr_err("No BVDs for secondary RAID found\n"); | |
3650 | return -1; | |
3651 | } | |
3652 | if (conf->prl != DDF_RAID1) { | |
3653 | pr_err("Secondary RAID level only supported for mirrored BVD\n"); | |
3654 | return -1; | |
3655 | } | |
3656 | if (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED) { | |
3657 | pr_err("Secondary RAID level %d is unsupported\n", | |
3658 | conf->srl); | |
3659 | return -1; | |
3660 | } | |
3661 | __set_sec_seen(conf->sec_elmnt_seq); | |
3662 | for (i = 0; i < conf->sec_elmnt_count-1; i++) { | |
3663 | const struct vd_config *bvd = vc->other_bvds[i]; | |
3664 | if (bvd->sec_elmnt_seq == DDF_UNUSED_BVD) | |
3665 | continue; | |
3666 | if (bvd->srl != conf->srl) { | |
3667 | pr_err("Inconsistent secondary RAID level across BVDs\n"); | |
3668 | return -1; | |
3669 | } | |
3670 | if (bvd->prl != conf->prl) { | |
3671 | pr_err("Different RAID levels for BVDs are unsupported\n"); | |
3672 | return -1; | |
3673 | } | |
3674 | if (!be16_eq(bvd->prim_elmnt_count, conf->prim_elmnt_count)) { | |
3675 | pr_err("All BVDs must have the same number of primary elements\n"); | |
3676 | return -1; | |
3677 | } | |
3678 | if (bvd->chunk_shift != conf->chunk_shift) { | |
3679 | pr_err("Different strip sizes for BVDs are unsupported\n"); | |
3680 | return -1; | |
3681 | } | |
3682 | if (!be64_eq(bvd->array_blocks, conf->array_blocks)) { | |
3683 | pr_err("Different BVD sizes are unsupported\n"); | |
3684 | return -1; | |
3685 | } | |
3686 | __set_sec_seen(bvd->sec_elmnt_seq); | |
3687 | } | |
3688 | for (i = 0; i < conf->sec_elmnt_count; i++) { | |
3689 | if (!__was_sec_seen(i)) { | |
3690 | /* pr_err("BVD %d is missing\n", i); */ | |
3691 | return -1; | |
3692 | } | |
3693 | } | |
3694 | return 0; | |
3695 | } | |
3696 | ||
3697 | static unsigned int get_pd_index_from_refnum(const struct vcl *vc, | |
3698 | be32 refnum, unsigned int nmax, | |
3699 | const struct vd_config **bvd, | |
3700 | unsigned int *idx) | |
3701 | { | |
3702 | unsigned int i, j, n, sec, cnt; | |
3703 | ||
3704 | cnt = be16_to_cpu(vc->conf.prim_elmnt_count); | |
3705 | sec = (vc->conf.sec_elmnt_count == 1 ? 0 : vc->conf.sec_elmnt_seq); | |
3706 | ||
3707 | for (i = 0, j = 0 ; i < nmax ; i++) { | |
3708 | /* j counts valid entries for this BVD */ | |
3709 | if (be32_eq(vc->conf.phys_refnum[i], refnum)) { | |
3710 | *bvd = &vc->conf; | |
3711 | *idx = i; | |
3712 | return sec * cnt + j; | |
3713 | } | |
3714 | if (be32_to_cpu(vc->conf.phys_refnum[i]) != 0xffffffff) | |
3715 | j++; | |
3716 | } | |
3717 | if (vc->other_bvds == NULL) | |
3718 | goto bad; | |
3719 | ||
3720 | for (n = 1; n < vc->conf.sec_elmnt_count; n++) { | |
3721 | struct vd_config *vd = vc->other_bvds[n-1]; | |
3722 | sec = vd->sec_elmnt_seq; | |
3723 | if (sec == DDF_UNUSED_BVD) | |
3724 | continue; | |
3725 | for (i = 0, j = 0 ; i < nmax ; i++) { | |
3726 | if (be32_eq(vd->phys_refnum[i], refnum)) { | |
3727 | *bvd = vd; | |
3728 | *idx = i; | |
3729 | return sec * cnt + j; | |
3730 | } | |
3731 | if (be32_to_cpu(vd->phys_refnum[i]) != 0xffffffff) | |
3732 | j++; | |
3733 | } | |
3734 | } | |
3735 | bad: | |
3736 | *bvd = NULL; | |
3737 | return DDF_NOTFOUND; | |
3738 | } | |
3739 | ||
3740 | static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray) | |
3741 | { | |
3742 | /* Given a container loaded by load_super_ddf_all, | |
3743 | * extract information about all the arrays into | |
3744 | * an mdinfo tree. | |
3745 | * | |
3746 | * For each vcl in conflist: create an mdinfo, fill it in, | |
3747 | * then look for matching devices (phys_refnum) in dlist | |
3748 | * and create appropriate device mdinfo. | |
3749 | */ | |
3750 | struct ddf_super *ddf = st->sb; | |
3751 | struct mdinfo *rest = NULL; | |
3752 | struct vcl *vc; | |
3753 | ||
3754 | for (vc = ddf->conflist ; vc ; vc=vc->next) { | |
3755 | unsigned int i; | |
3756 | struct mdinfo *this; | |
3757 | char *ep; | |
3758 | __u32 *cptr; | |
3759 | unsigned int pd; | |
3760 | ||
3761 | if (subarray && | |
3762 | (strtoul(subarray, &ep, 10) != vc->vcnum || | |
3763 | *ep != '\0')) | |
3764 | continue; | |
3765 | ||
3766 | if (vc->conf.sec_elmnt_count > 1) { | |
3767 | if (check_secondary(vc) != 0) | |
3768 | continue; | |
3769 | } | |
3770 | ||
3771 | this = xcalloc(1, sizeof(*this)); | |
3772 | this->next = rest; | |
3773 | rest = this; | |
3774 | ||
3775 | if (layout_ddf2md(&vc->conf, &this->array)) | |
3776 | continue; | |
3777 | this->array.md_minor = -1; | |
3778 | this->array.major_version = -1; | |
3779 | this->array.minor_version = -2; | |
3780 | this->safe_mode_delay = DDF_SAFE_MODE_DELAY; | |
3781 | cptr = (__u32 *)(vc->conf.guid + 16); | |
3782 | this->array.ctime = DECADE + __be32_to_cpu(*cptr); | |
3783 | this->array.utime = DECADE + | |
3784 | be32_to_cpu(vc->conf.timestamp); | |
3785 | this->array.chunk_size = 512 << vc->conf.chunk_shift; | |
3786 | ||
3787 | i = vc->vcnum; | |
3788 | if ((ddf->virt->entries[i].state & DDF_state_inconsistent) || | |
3789 | (ddf->virt->entries[i].init_state & DDF_initstate_mask) != | |
3790 | DDF_init_full) { | |
3791 | this->array.state = 0; | |
3792 | this->resync_start = 0; | |
3793 | } else { | |
3794 | this->array.state = 1; | |
3795 | this->resync_start = MaxSector; | |
3796 | } | |
3797 | _ddf_array_name(this->name, ddf, i); | |
3798 | memset(this->uuid, 0, sizeof(this->uuid)); | |
3799 | this->component_size = be64_to_cpu(vc->conf.blocks); | |
3800 | this->array.size = this->component_size / 2; | |
3801 | this->container_member = i; | |
3802 | ||
3803 | ddf->currentconf = vc; | |
3804 | uuid_from_super_ddf(st, this->uuid); | |
3805 | if (!subarray) | |
3806 | ddf->currentconf = NULL; | |
3807 | ||
3808 | sprintf(this->text_version, "/%s/%d", | |
3809 | st->container_devnm, this->container_member); | |
3810 | ||
3811 | for (pd = 0; pd < be16_to_cpu(ddf->phys->max_pdes); pd++) { | |
3812 | struct mdinfo *dev; | |
3813 | struct dl *d; | |
3814 | const struct vd_config *bvd; | |
3815 | unsigned int iphys; | |
3816 | int stt; | |
3817 | ||
3818 | if (be32_to_cpu(ddf->phys->entries[pd].refnum) == | |
3819 | 0xffffffff) | |
3820 | continue; | |
3821 | ||
3822 | stt = be16_to_cpu(ddf->phys->entries[pd].state); | |
3823 | if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding)) != | |
3824 | DDF_Online) | |
3825 | continue; | |
3826 | ||
3827 | i = get_pd_index_from_refnum( | |
3828 | vc, ddf->phys->entries[pd].refnum, | |
3829 | ddf->mppe, &bvd, &iphys); | |
3830 | if (i == DDF_NOTFOUND) | |
3831 | continue; | |
3832 | ||
3833 | this->array.working_disks++; | |
3834 | ||
3835 | for (d = ddf->dlist; d ; d=d->next) | |
3836 | if (be32_eq(d->disk.refnum, | |
3837 | ddf->phys->entries[pd].refnum)) | |
3838 | break; | |
3839 | if (d == NULL) | |
3840 | /* Haven't found that one yet, maybe there are others */ | |
3841 | continue; | |
3842 | ||
3843 | dev = xcalloc(1, sizeof(*dev)); | |
3844 | dev->next = this->devs; | |
3845 | this->devs = dev; | |
3846 | ||
3847 | dev->disk.number = be32_to_cpu(d->disk.refnum); | |
3848 | dev->disk.major = d->major; | |
3849 | dev->disk.minor = d->minor; | |
3850 | dev->disk.raid_disk = i; | |
3851 | dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE); | |
3852 | dev->recovery_start = MaxSector; | |
3853 | ||
3854 | dev->events = be32_to_cpu(ddf->active->seq); | |
3855 | dev->data_offset = | |
3856 | be64_to_cpu(LBA_OFFSET(ddf, bvd)[iphys]); | |
3857 | dev->component_size = be64_to_cpu(bvd->blocks); | |
3858 | if (d->devname) | |
3859 | strcpy(dev->name, d->devname); | |
3860 | } | |
3861 | } | |
3862 | return rest; | |
3863 | } | |
3864 | ||
3865 | static int store_super_ddf(struct supertype *st, int fd) | |
3866 | { | |
3867 | struct ddf_super *ddf = st->sb; | |
3868 | unsigned long long dsize; | |
3869 | void *buf; | |
3870 | int rc; | |
3871 | ||
3872 | if (!ddf) | |
3873 | return 1; | |
3874 | ||
3875 | if (!get_dev_size(fd, NULL, &dsize)) | |
3876 | return 1; | |
3877 | ||
3878 | if (ddf->dlist || ddf->conflist) { | |
3879 | struct stat sta; | |
3880 | struct dl *dl; | |
3881 | int ofd, ret; | |
3882 | ||
3883 | if (fstat(fd, &sta) == -1 || !S_ISBLK(sta.st_mode)) { | |
3884 | pr_err("file descriptor for invalid device\n"); | |
3885 | return 1; | |
3886 | } | |
3887 | for (dl = ddf->dlist; dl; dl = dl->next) | |
3888 | if (dl->major == (int)major(sta.st_rdev) && | |
3889 | dl->minor == (int)minor(sta.st_rdev)) | |
3890 | break; | |
3891 | if (!dl) { | |
3892 | pr_err("couldn't find disk %d/%d\n", | |
3893 | (int)major(sta.st_rdev), | |
3894 | (int)minor(sta.st_rdev)); | |
3895 | return 1; | |
3896 | } | |
3897 | ofd = dl->fd; | |
3898 | dl->fd = fd; | |
3899 | ret = (_write_super_to_disk(ddf, dl) != 1); | |
3900 | dl->fd = ofd; | |
3901 | return ret; | |
3902 | } | |
3903 | ||
3904 | if (posix_memalign(&buf, 512, 512) != 0) | |
3905 | return 1; | |
3906 | memset(buf, 0, 512); | |
3907 | ||
3908 | lseek64(fd, dsize-512, 0); | |
3909 | rc = write(fd, buf, 512); | |
3910 | free(buf); | |
3911 | if (rc < 0) | |
3912 | return 1; | |
3913 | return 0; | |
3914 | } | |
3915 | ||
3916 | static int compare_super_ddf(struct supertype *st, struct supertype *tst) | |
3917 | { | |
3918 | /* | |
3919 | * return: | |
3920 | * 0 same, or first was empty, and second was copied | |
3921 | * 1 second had wrong magic number - but that isn't possible | |
3922 | * 2 wrong uuid | |
3923 | * 3 wrong other info | |
3924 | */ | |
3925 | struct ddf_super *first = st->sb; | |
3926 | struct ddf_super *second = tst->sb; | |
3927 | struct dl *dl1, *dl2; | |
3928 | struct vcl *vl1, *vl2; | |
3929 | unsigned int max_vds, max_pds, pd, vd; | |
3930 | ||
3931 | if (!first) { | |
3932 | st->sb = tst->sb; | |
3933 | tst->sb = NULL; | |
3934 | return 0; | |
3935 | } | |
3936 | ||
3937 | if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0) | |
3938 | return 2; | |
3939 | ||
3940 | /* It is only OK to compare info in the anchor. Anything else | |
3941 | * could be changing due to a reconfig so must be ignored. | |
3942 | * guid really should be enough anyway. | |
3943 | */ | |
3944 | ||
3945 | if (!be32_eq(first->active->seq, second->active->seq)) { | |
3946 | dprintf("sequence number mismatch %u<->%u\n", | |
3947 | be32_to_cpu(first->active->seq), | |
3948 | be32_to_cpu(second->active->seq)); | |
3949 | return 0; | |
3950 | } | |
3951 | ||
3952 | /* | |
3953 | * At this point we are fairly sure that the meta data matches. | |
3954 | * But the new disk may contain additional local data. | |
3955 | * Add it to the super block. | |
3956 | */ | |
3957 | max_vds = be16_to_cpu(first->active->max_vd_entries); | |
3958 | max_pds = be16_to_cpu(first->phys->max_pdes); | |
3959 | for (vl2 = second->conflist; vl2; vl2 = vl2->next) { | |
3960 | for (vl1 = first->conflist; vl1; vl1 = vl1->next) | |
3961 | if (!memcmp(vl1->conf.guid, vl2->conf.guid, | |
3962 | DDF_GUID_LEN)) | |
3963 | break; | |
3964 | if (vl1) { | |
3965 | if (vl1->other_bvds != NULL && | |
3966 | vl1->conf.sec_elmnt_seq != | |
3967 | vl2->conf.sec_elmnt_seq) { | |
3968 | dprintf("adding BVD %u\n", | |
3969 | vl2->conf.sec_elmnt_seq); | |
3970 | add_other_bvd(vl1, &vl2->conf, | |
3971 | first->conf_rec_len*512); | |
3972 | } | |
3973 | continue; | |
3974 | } | |
3975 | ||
3976 | if (posix_memalign((void **)&vl1, 512, | |
3977 | (first->conf_rec_len*512 + | |
3978 | offsetof(struct vcl, conf))) != 0) { | |
3979 | pr_err("could not allocate vcl buf\n"); | |
3980 | return 3; | |
3981 | } | |
3982 | ||
3983 | vl1->next = first->conflist; | |
3984 | vl1->block_sizes = NULL; | |
3985 | memcpy(&vl1->conf, &vl2->conf, first->conf_rec_len*512); | |
3986 | if (alloc_other_bvds(first, vl1) != 0) { | |
3987 | pr_err("could not allocate other bvds\n"); | |
3988 | free(vl1); | |
3989 | return 3; | |
3990 | } | |
3991 | for (vd = 0; vd < max_vds; vd++) | |
3992 | if (!memcmp(first->virt->entries[vd].guid, | |
3993 | vl1->conf.guid, DDF_GUID_LEN)) | |
3994 | break; | |
3995 | vl1->vcnum = vd; | |
3996 | dprintf("added config for VD %u\n", vl1->vcnum); | |
3997 | first->conflist = vl1; | |
3998 | } | |
3999 | ||
4000 | for (dl2 = second->dlist; dl2; dl2 = dl2->next) { | |
4001 | for (dl1 = first->dlist; dl1; dl1 = dl1->next) | |
4002 | if (be32_eq(dl1->disk.refnum, dl2->disk.refnum)) | |
4003 | break; | |
4004 | if (dl1) | |
4005 | continue; | |
4006 | ||
4007 | if (posix_memalign((void **)&dl1, 512, | |
4008 | sizeof(*dl1) + (first->max_part) * | |
4009 | sizeof(dl1->vlist[0])) != 0) { | |
4010 | pr_err("could not allocate disk info buffer\n"); | |
4011 | return 3; | |
4012 | } | |
4013 | memcpy(dl1, dl2, sizeof(*dl1)); | |
4014 | dl1->mdupdate = NULL; | |
4015 | dl1->next = first->dlist; | |
4016 | dl1->fd = -1; | |
4017 | for (pd = 0; pd < max_pds; pd++) | |
4018 | if (be32_eq(first->phys->entries[pd].refnum, | |
4019 | dl1->disk.refnum)) | |
4020 | break; | |
4021 | dl1->pdnum = pd < max_pds ? (int)pd : -1; | |
4022 | if (dl2->spare) { | |
4023 | if (posix_memalign((void **)&dl1->spare, 512, | |
4024 | first->conf_rec_len*512) != 0) { | |
4025 | pr_err("could not allocate spare info buf\n"); | |
4026 | return 3; | |
4027 | } | |
4028 | memcpy(dl1->spare, dl2->spare, first->conf_rec_len*512); | |
4029 | } | |
4030 | for (vd = 0 ; vd < first->max_part ; vd++) { | |
4031 | if (!dl2->vlist[vd]) { | |
4032 | dl1->vlist[vd] = NULL; | |
4033 | continue; | |
4034 | } | |
4035 | for (vl1 = first->conflist; vl1; vl1 = vl1->next) { | |
4036 | if (!memcmp(vl1->conf.guid, | |
4037 | dl2->vlist[vd]->conf.guid, | |
4038 | DDF_GUID_LEN)) | |
4039 | break; | |
4040 | dl1->vlist[vd] = vl1; | |
4041 | } | |
4042 | } | |
4043 | first->dlist = dl1; | |
4044 | dprintf("added disk %d: %08x\n", dl1->pdnum, | |
4045 | be32_to_cpu(dl1->disk.refnum)); | |
4046 | } | |
4047 | ||
4048 | return 0; | |
4049 | } | |
4050 | ||
4051 | /* | |
4052 | * A new array 'a' has been started which claims to be instance 'inst' | |
4053 | * within container 'c'. | |
4054 | * We need to confirm that the array matches the metadata in 'c' so | |
4055 | * that we don't corrupt any metadata. | |
4056 | */ | |
4057 | static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst) | |
4058 | { | |
4059 | struct ddf_super *ddf = c->sb; | |
4060 | int n = atoi(inst); | |
4061 | struct mdinfo *dev; | |
4062 | struct dl *dl; | |
4063 | static const char faulty[] = "faulty"; | |
4064 | ||
4065 | if (all_ff(ddf->virt->entries[n].guid)) { | |
4066 | pr_err("subarray %d doesn't exist\n", n); | |
4067 | return -ENODEV; | |
4068 | } | |
4069 | dprintf("new subarray %d, GUID: %s\n", n, | |
4070 | guid_str(ddf->virt->entries[n].guid)); | |
4071 | for (dev = a->info.devs; dev; dev = dev->next) { | |
4072 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4073 | if (dl->major == dev->disk.major && | |
4074 | dl->minor == dev->disk.minor) | |
4075 | break; | |
4076 | if (!dl || dl->pdnum < 0) { | |
4077 | pr_err("device %d/%d of subarray %d not found in meta data\n", | |
4078 | dev->disk.major, dev->disk.minor, n); | |
4079 | return -1; | |
4080 | } | |
4081 | if ((be16_to_cpu(ddf->phys->entries[dl->pdnum].state) & | |
4082 | (DDF_Online|DDF_Missing|DDF_Failed)) != DDF_Online) { | |
4083 | pr_err("new subarray %d contains broken device %d/%d (%02x)\n", | |
4084 | n, dl->major, dl->minor, | |
4085 | be16_to_cpu(ddf->phys->entries[dl->pdnum].state)); | |
4086 | if (write(dev->state_fd, faulty, sizeof(faulty)-1) != | |
4087 | sizeof(faulty) - 1) | |
4088 | pr_err("Write to state_fd failed\n"); | |
4089 | dev->curr_state = DS_FAULTY; | |
4090 | } | |
4091 | } | |
4092 | a->info.container_member = n; | |
4093 | return 0; | |
4094 | } | |
4095 | ||
4096 | static void handle_missing(struct ddf_super *ddf, struct active_array *a, int inst) | |
4097 | { | |
4098 | /* This member array is being activated. If any devices | |
4099 | * are missing they must now be marked as failed. | |
4100 | */ | |
4101 | struct vd_config *vc; | |
4102 | unsigned int n_bvd; | |
4103 | struct vcl *vcl; | |
4104 | struct dl *dl; | |
4105 | int pd; | |
4106 | int n; | |
4107 | int state; | |
4108 | ||
4109 | for (n = 0; ; n++) { | |
4110 | vc = find_vdcr(ddf, inst, n, &n_bvd, &vcl); | |
4111 | if (!vc) | |
4112 | break; | |
4113 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4114 | if (be32_eq(dl->disk.refnum, vc->phys_refnum[n_bvd])) | |
4115 | break; | |
4116 | if (dl) | |
4117 | /* Found this disk, so not missing */ | |
4118 | continue; | |
4119 | ||
4120 | /* Mark the device as failed/missing. */ | |
4121 | pd = find_phys(ddf, vc->phys_refnum[n_bvd]); | |
4122 | if (pd >= 0 && be16_and(ddf->phys->entries[pd].state, | |
4123 | cpu_to_be16(DDF_Online))) { | |
4124 | be16_clear(ddf->phys->entries[pd].state, | |
4125 | cpu_to_be16(DDF_Online)); | |
4126 | be16_set(ddf->phys->entries[pd].state, | |
4127 | cpu_to_be16(DDF_Failed|DDF_Missing)); | |
4128 | vc->phys_refnum[n_bvd] = cpu_to_be32(0); | |
4129 | ddf_set_updates_pending(ddf, vc); | |
4130 | } | |
4131 | ||
4132 | /* Mark the array as Degraded */ | |
4133 | state = get_svd_state(ddf, vcl); | |
4134 | if (ddf->virt->entries[inst].state != | |
4135 | ((ddf->virt->entries[inst].state & ~DDF_state_mask) | |
4136 | | state)) { | |
4137 | ddf->virt->entries[inst].state = | |
4138 | (ddf->virt->entries[inst].state & ~DDF_state_mask) | |
4139 | | state; | |
4140 | a->check_degraded = 1; | |
4141 | ddf_set_updates_pending(ddf, vc); | |
4142 | } | |
4143 | } | |
4144 | } | |
4145 | ||
4146 | /* | |
4147 | * The array 'a' is to be marked clean in the metadata. | |
4148 | * If '->resync_start' is not ~(unsigned long long)0, then the array is only | |
4149 | * clean up to the point (in sectors). If that cannot be recorded in the | |
4150 | * metadata, then leave it as dirty. | |
4151 | * | |
4152 | * For DDF, we need to clear the DDF_state_inconsistent bit in the | |
4153 | * !global! virtual_disk.virtual_entry structure. | |
4154 | */ | |
4155 | static int ddf_set_array_state(struct active_array *a, int consistent) | |
4156 | { | |
4157 | struct ddf_super *ddf = a->container->sb; | |
4158 | int inst = a->info.container_member; | |
4159 | int old = ddf->virt->entries[inst].state; | |
4160 | if (consistent == 2) { | |
4161 | handle_missing(ddf, a, inst); | |
4162 | consistent = 1; | |
4163 | if (!is_resync_complete(&a->info)) | |
4164 | consistent = 0; | |
4165 | } | |
4166 | if (consistent) | |
4167 | ddf->virt->entries[inst].state &= ~DDF_state_inconsistent; | |
4168 | else | |
4169 | ddf->virt->entries[inst].state |= DDF_state_inconsistent; | |
4170 | if (old != ddf->virt->entries[inst].state) | |
4171 | ddf_set_updates_pending(ddf, NULL); | |
4172 | ||
4173 | old = ddf->virt->entries[inst].init_state; | |
4174 | ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask; | |
4175 | if (is_resync_complete(&a->info)) | |
4176 | ddf->virt->entries[inst].init_state |= DDF_init_full; | |
4177 | else if (a->info.resync_start == 0) | |
4178 | ddf->virt->entries[inst].init_state |= DDF_init_not; | |
4179 | else | |
4180 | ddf->virt->entries[inst].init_state |= DDF_init_quick; | |
4181 | if (old != ddf->virt->entries[inst].init_state) | |
4182 | ddf_set_updates_pending(ddf, NULL); | |
4183 | ||
4184 | dprintf("ddf mark %d/%s (%d) %s %llu\n", inst, | |
4185 | guid_str(ddf->virt->entries[inst].guid), a->curr_state, | |
4186 | consistent?"clean":"dirty", | |
4187 | a->info.resync_start); | |
4188 | return consistent; | |
4189 | } | |
4190 | ||
4191 | static int get_bvd_state(const struct ddf_super *ddf, | |
4192 | const struct vd_config *vc) | |
4193 | { | |
4194 | unsigned int i, n_bvd, working = 0; | |
4195 | unsigned int n_prim = be16_to_cpu(vc->prim_elmnt_count); | |
4196 | int pd, st, state; | |
4197 | char *avail = xcalloc(1, n_prim); | |
4198 | mdu_array_info_t array; | |
4199 | ||
4200 | layout_ddf2md(vc, &array); | |
4201 | ||
4202 | for (i = 0; i < n_prim; i++) { | |
4203 | if (!find_index_in_bvd(ddf, vc, i, &n_bvd)) | |
4204 | continue; | |
4205 | pd = find_phys(ddf, vc->phys_refnum[n_bvd]); | |
4206 | if (pd < 0) | |
4207 | continue; | |
4208 | st = be16_to_cpu(ddf->phys->entries[pd].state); | |
4209 | if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding)) == | |
4210 | DDF_Online) { | |
4211 | working++; | |
4212 | avail[i] = 1; | |
4213 | } | |
4214 | } | |
4215 | ||
4216 | state = DDF_state_degraded; | |
4217 | if (working == n_prim) | |
4218 | state = DDF_state_optimal; | |
4219 | else | |
4220 | switch (vc->prl) { | |
4221 | case DDF_RAID0: | |
4222 | case DDF_CONCAT: | |
4223 | case DDF_JBOD: | |
4224 | state = DDF_state_failed; | |
4225 | break; | |
4226 | case DDF_RAID1: | |
4227 | if (working == 0) | |
4228 | state = DDF_state_failed; | |
4229 | else if (working >= 2) | |
4230 | state = DDF_state_part_optimal; | |
4231 | break; | |
4232 | case DDF_RAID1E: | |
4233 | if (!enough(10, n_prim, array.layout, 1, avail)) | |
4234 | state = DDF_state_failed; | |
4235 | break; | |
4236 | case DDF_RAID4: | |
4237 | case DDF_RAID5: | |
4238 | if (working < n_prim - 1) | |
4239 | state = DDF_state_failed; | |
4240 | break; | |
4241 | case DDF_RAID6: | |
4242 | if (working < n_prim - 2) | |
4243 | state = DDF_state_failed; | |
4244 | else if (working == n_prim - 1) | |
4245 | state = DDF_state_part_optimal; | |
4246 | break; | |
4247 | } | |
4248 | return state; | |
4249 | } | |
4250 | ||
4251 | static int secondary_state(int state, int other, int seclevel) | |
4252 | { | |
4253 | if (state == DDF_state_optimal && other == DDF_state_optimal) | |
4254 | return DDF_state_optimal; | |
4255 | if (seclevel == DDF_2MIRRORED) { | |
4256 | if (state == DDF_state_optimal || other == DDF_state_optimal) | |
4257 | return DDF_state_part_optimal; | |
4258 | if (state == DDF_state_failed && other == DDF_state_failed) | |
4259 | return DDF_state_failed; | |
4260 | return DDF_state_degraded; | |
4261 | } else { | |
4262 | if (state == DDF_state_failed || other == DDF_state_failed) | |
4263 | return DDF_state_failed; | |
4264 | if (state == DDF_state_degraded || other == DDF_state_degraded) | |
4265 | return DDF_state_degraded; | |
4266 | return DDF_state_part_optimal; | |
4267 | } | |
4268 | } | |
4269 | ||
4270 | static int get_svd_state(const struct ddf_super *ddf, const struct vcl *vcl) | |
4271 | { | |
4272 | int state = get_bvd_state(ddf, &vcl->conf); | |
4273 | unsigned int i; | |
4274 | for (i = 1; i < vcl->conf.sec_elmnt_count; i++) { | |
4275 | state = secondary_state( | |
4276 | state, | |
4277 | get_bvd_state(ddf, vcl->other_bvds[i-1]), | |
4278 | vcl->conf.srl); | |
4279 | } | |
4280 | return state; | |
4281 | } | |
4282 | ||
4283 | /* | |
4284 | * The state of each disk is stored in the global phys_disk structure | |
4285 | * in phys_disk.entries[n].state. | |
4286 | * This makes various combinations awkward. | |
4287 | * - When a device fails in any array, it must be failed in all arrays | |
4288 | * that include a part of this device. | |
4289 | * - When a component is rebuilding, we cannot include it officially in the | |
4290 | * array unless this is the only array that uses the device. | |
4291 | * | |
4292 | * So: when transitioning: | |
4293 | * Online -> failed, just set failed flag. monitor will propagate | |
4294 | * spare -> online, the device might need to be added to the array. | |
4295 | * spare -> failed, just set failed. Don't worry if in array or not. | |
4296 | */ | |
4297 | static void ddf_set_disk(struct active_array *a, int n, int state) | |
4298 | { | |
4299 | struct ddf_super *ddf = a->container->sb; | |
4300 | unsigned int inst = a->info.container_member, n_bvd; | |
4301 | struct vcl *vcl; | |
4302 | struct vd_config *vc = find_vdcr(ddf, inst, (unsigned int)n, | |
4303 | &n_bvd, &vcl); | |
4304 | int pd; | |
4305 | struct mdinfo *mdi; | |
4306 | struct dl *dl; | |
4307 | int update = 0; | |
4308 | ||
4309 | dprintf("%d to %x\n", n, state); | |
4310 | if (vc == NULL) { | |
4311 | dprintf("ddf: cannot find instance %d!!\n", inst); | |
4312 | return; | |
4313 | } | |
4314 | /* Find the matching slot in 'info'. */ | |
4315 | for (mdi = a->info.devs; mdi; mdi = mdi->next) | |
4316 | if (mdi->disk.raid_disk == n) | |
4317 | break; | |
4318 | if (!mdi) { | |
4319 | pr_err("cannot find raid disk %d\n", n); | |
4320 | return; | |
4321 | } | |
4322 | ||
4323 | /* and find the 'dl' entry corresponding to that. */ | |
4324 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4325 | if (mdi->state_fd >= 0 && | |
4326 | mdi->disk.major == dl->major && | |
4327 | mdi->disk.minor == dl->minor) | |
4328 | break; | |
4329 | if (!dl) { | |
4330 | pr_err("cannot find raid disk %d (%d/%d)\n", | |
4331 | n, mdi->disk.major, mdi->disk.minor); | |
4332 | return; | |
4333 | } | |
4334 | ||
4335 | pd = find_phys(ddf, vc->phys_refnum[n_bvd]); | |
4336 | if (pd < 0 || pd != dl->pdnum) { | |
4337 | /* disk doesn't currently exist or has changed. | |
4338 | * If it is now in_sync, insert it. */ | |
4339 | dprintf("phys disk not found for %d: %d/%d ref %08x\n", | |
4340 | dl->pdnum, dl->major, dl->minor, | |
4341 | be32_to_cpu(dl->disk.refnum)); | |
4342 | dprintf("array %u disk %u ref %08x pd %d\n", | |
4343 | inst, n_bvd, | |
4344 | be32_to_cpu(vc->phys_refnum[n_bvd]), pd); | |
4345 | if ((state & DS_INSYNC) && ! (state & DS_FAULTY) && | |
4346 | dl->pdnum >= 0) { | |
4347 | pd = dl->pdnum; | |
4348 | vc->phys_refnum[n_bvd] = dl->disk.refnum; | |
4349 | LBA_OFFSET(ddf, vc)[n_bvd] = | |
4350 | cpu_to_be64(mdi->data_offset); | |
4351 | be16_clear(ddf->phys->entries[pd].type, | |
4352 | cpu_to_be16(DDF_Global_Spare)); | |
4353 | be16_set(ddf->phys->entries[pd].type, | |
4354 | cpu_to_be16(DDF_Active_in_VD)); | |
4355 | update = 1; | |
4356 | } | |
4357 | } else { | |
4358 | be16 old = ddf->phys->entries[pd].state; | |
4359 | if (state & DS_FAULTY) | |
4360 | be16_set(ddf->phys->entries[pd].state, | |
4361 | cpu_to_be16(DDF_Failed)); | |
4362 | if (state & DS_INSYNC) { | |
4363 | be16_set(ddf->phys->entries[pd].state, | |
4364 | cpu_to_be16(DDF_Online)); | |
4365 | be16_clear(ddf->phys->entries[pd].state, | |
4366 | cpu_to_be16(DDF_Rebuilding)); | |
4367 | } | |
4368 | if (!be16_eq(old, ddf->phys->entries[pd].state)) | |
4369 | update = 1; | |
4370 | } | |
4371 | ||
4372 | dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n, | |
4373 | be32_to_cpu(dl->disk.refnum), state, | |
4374 | be16_to_cpu(ddf->phys->entries[pd].state)); | |
4375 | ||
4376 | /* Now we need to check the state of the array and update | |
4377 | * virtual_disk.entries[n].state. | |
4378 | * It needs to be one of "optimal", "degraded", "failed". | |
4379 | * I don't understand 'deleted' or 'missing'. | |
4380 | */ | |
4381 | state = get_svd_state(ddf, vcl); | |
4382 | ||
4383 | if (ddf->virt->entries[inst].state != | |
4384 | ((ddf->virt->entries[inst].state & ~DDF_state_mask) | |
4385 | | state)) { | |
4386 | ddf->virt->entries[inst].state = | |
4387 | (ddf->virt->entries[inst].state & ~DDF_state_mask) | |
4388 | | state; | |
4389 | update = 1; | |
4390 | } | |
4391 | if (update) | |
4392 | ddf_set_updates_pending(ddf, vc); | |
4393 | } | |
4394 | ||
4395 | static void ddf_sync_metadata(struct supertype *st) | |
4396 | { | |
4397 | /* | |
4398 | * Write all data to all devices. | |
4399 | * Later, we might be able to track whether only local changes | |
4400 | * have been made, or whether any global data has been changed, | |
4401 | * but ddf is sufficiently weird that it probably always | |
4402 | * changes global data .... | |
4403 | */ | |
4404 | struct ddf_super *ddf = st->sb; | |
4405 | if (!ddf->updates_pending) | |
4406 | return; | |
4407 | ddf->updates_pending = 0; | |
4408 | __write_init_super_ddf(st); | |
4409 | dprintf("ddf: sync_metadata\n"); | |
4410 | } | |
4411 | ||
4412 | static int del_from_conflist(struct vcl **list, const char *guid) | |
4413 | { | |
4414 | struct vcl **p; | |
4415 | int found = 0; | |
4416 | for (p = list; p && *p; p = &((*p)->next)) | |
4417 | if (!memcmp((*p)->conf.guid, guid, DDF_GUID_LEN)) { | |
4418 | found = 1; | |
4419 | *p = (*p)->next; | |
4420 | } | |
4421 | return found; | |
4422 | } | |
4423 | ||
4424 | static int _kill_subarray_ddf(struct ddf_super *ddf, const char *guid) | |
4425 | { | |
4426 | struct dl *dl; | |
4427 | unsigned int vdnum, i; | |
4428 | vdnum = find_vde_by_guid(ddf, guid); | |
4429 | if (vdnum == DDF_NOTFOUND) { | |
4430 | pr_err("could not find VD %s\n", guid_str(guid)); | |
4431 | return -1; | |
4432 | } | |
4433 | if (del_from_conflist(&ddf->conflist, guid) == 0) { | |
4434 | pr_err("could not find conf %s\n", guid_str(guid)); | |
4435 | return -1; | |
4436 | } | |
4437 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4438 | for (i = 0; i < ddf->max_part; i++) | |
4439 | if (dl->vlist[i] != NULL && | |
4440 | !memcmp(dl->vlist[i]->conf.guid, guid, | |
4441 | DDF_GUID_LEN)) | |
4442 | dl->vlist[i] = NULL; | |
4443 | memset(ddf->virt->entries[vdnum].guid, 0xff, DDF_GUID_LEN); | |
4444 | dprintf("deleted %s\n", guid_str(guid)); | |
4445 | return 0; | |
4446 | } | |
4447 | ||
4448 | static int kill_subarray_ddf(struct supertype *st) | |
4449 | { | |
4450 | struct ddf_super *ddf = st->sb; | |
4451 | /* | |
4452 | * currentconf is set in container_content_ddf, | |
4453 | * called with subarray arg | |
4454 | */ | |
4455 | struct vcl *victim = ddf->currentconf; | |
4456 | struct vd_config *conf; | |
4457 | unsigned int vdnum; | |
4458 | ||
4459 | ddf->currentconf = NULL; | |
4460 | if (!victim) { | |
4461 | pr_err("nothing to kill\n"); | |
4462 | return -1; | |
4463 | } | |
4464 | conf = &victim->conf; | |
4465 | vdnum = find_vde_by_guid(ddf, conf->guid); | |
4466 | if (vdnum == DDF_NOTFOUND) { | |
4467 | pr_err("could not find VD %s\n", guid_str(conf->guid)); | |
4468 | return -1; | |
4469 | } | |
4470 | if (st->update_tail) { | |
4471 | struct virtual_disk *vd; | |
4472 | int len = sizeof(struct virtual_disk) | |
4473 | + sizeof(struct virtual_entry); | |
4474 | vd = xmalloc(len); | |
4475 | if (vd == NULL) { | |
4476 | pr_err("failed to allocate %d bytes\n", len); | |
4477 | return -1; | |
4478 | } | |
4479 | memset(vd, 0 , len); | |
4480 | vd->magic = DDF_VIRT_RECORDS_MAGIC; | |
4481 | vd->populated_vdes = cpu_to_be16(0); | |
4482 | memcpy(vd->entries[0].guid, conf->guid, DDF_GUID_LEN); | |
4483 | /* we use DDF_state_deleted as marker */ | |
4484 | vd->entries[0].state = DDF_state_deleted; | |
4485 | append_metadata_update(st, vd, len); | |
4486 | } else { | |
4487 | _kill_subarray_ddf(ddf, conf->guid); | |
4488 | ddf_set_updates_pending(ddf, NULL); | |
4489 | ddf_sync_metadata(st); | |
4490 | } | |
4491 | return 0; | |
4492 | } | |
4493 | ||
4494 | static void copy_matching_bvd(struct ddf_super *ddf, | |
4495 | struct vd_config *conf, | |
4496 | const struct metadata_update *update) | |
4497 | { | |
4498 | unsigned int mppe = | |
4499 | be16_to_cpu(ddf->anchor.max_primary_element_entries); | |
4500 | unsigned int len = ddf->conf_rec_len * 512; | |
4501 | char *p; | |
4502 | struct vd_config *vc; | |
4503 | for (p = update->buf; p < update->buf + update->len; p += len) { | |
4504 | vc = (struct vd_config *) p; | |
4505 | if (vc->sec_elmnt_seq == conf->sec_elmnt_seq) { | |
4506 | memcpy(conf->phys_refnum, vc->phys_refnum, | |
4507 | mppe * (sizeof(__u32) + sizeof(__u64))); | |
4508 | return; | |
4509 | } | |
4510 | } | |
4511 | pr_err("no match for BVD %d of %s in update\n", | |
4512 | conf->sec_elmnt_seq, guid_str(conf->guid)); | |
4513 | } | |
4514 | ||
4515 | static void ddf_process_phys_update(struct supertype *st, | |
4516 | struct metadata_update *update) | |
4517 | { | |
4518 | struct ddf_super *ddf = st->sb; | |
4519 | struct phys_disk *pd; | |
4520 | unsigned int ent; | |
4521 | ||
4522 | pd = (struct phys_disk*)update->buf; | |
4523 | ent = be16_to_cpu(pd->used_pdes); | |
4524 | if (ent >= be16_to_cpu(ddf->phys->max_pdes)) | |
4525 | return; | |
4526 | if (be16_and(pd->entries[0].state, cpu_to_be16(DDF_Missing))) { | |
4527 | struct dl **dlp; | |
4528 | /* removing this disk. */ | |
4529 | be16_set(ddf->phys->entries[ent].state, | |
4530 | cpu_to_be16(DDF_Missing)); | |
4531 | for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) { | |
4532 | struct dl *dl = *dlp; | |
4533 | if (dl->pdnum == (signed)ent) { | |
4534 | close(dl->fd); | |
4535 | dl->fd = -1; | |
4536 | *dlp = dl->next; | |
4537 | update->space = dl->devname; | |
4538 | *(void**)dl = update->space_list; | |
4539 | update->space_list = (void**)dl; | |
4540 | break; | |
4541 | } | |
4542 | } | |
4543 | ddf_set_updates_pending(ddf, NULL); | |
4544 | return; | |
4545 | } | |
4546 | if (!all_ff(ddf->phys->entries[ent].guid)) | |
4547 | return; | |
4548 | ddf->phys->entries[ent] = pd->entries[0]; | |
4549 | ddf->phys->used_pdes = cpu_to_be16 | |
4550 | (1 + be16_to_cpu(ddf->phys->used_pdes)); | |
4551 | ddf_set_updates_pending(ddf, NULL); | |
4552 | if (ddf->add_list) { | |
4553 | struct active_array *a; | |
4554 | struct dl *al = ddf->add_list; | |
4555 | ddf->add_list = al->next; | |
4556 | ||
4557 | al->next = ddf->dlist; | |
4558 | ddf->dlist = al; | |
4559 | ||
4560 | /* As a device has been added, we should check | |
4561 | * for any degraded devices that might make | |
4562 | * use of this spare */ | |
4563 | for (a = st->arrays ; a; a=a->next) | |
4564 | a->check_degraded = 1; | |
4565 | } | |
4566 | } | |
4567 | ||
4568 | static void ddf_process_virt_update(struct supertype *st, | |
4569 | struct metadata_update *update) | |
4570 | { | |
4571 | struct ddf_super *ddf = st->sb; | |
4572 | struct virtual_disk *vd; | |
4573 | unsigned int ent; | |
4574 | ||
4575 | vd = (struct virtual_disk*)update->buf; | |
4576 | ||
4577 | if (vd->entries[0].state == DDF_state_deleted) { | |
4578 | if (_kill_subarray_ddf(ddf, vd->entries[0].guid)) | |
4579 | return; | |
4580 | } else { | |
4581 | ent = find_vde_by_guid(ddf, vd->entries[0].guid); | |
4582 | if (ent != DDF_NOTFOUND) { | |
4583 | dprintf("VD %s exists already in slot %d\n", | |
4584 | guid_str(vd->entries[0].guid), | |
4585 | ent); | |
4586 | return; | |
4587 | } | |
4588 | ent = find_unused_vde(ddf); | |
4589 | if (ent == DDF_NOTFOUND) | |
4590 | return; | |
4591 | ddf->virt->entries[ent] = vd->entries[0]; | |
4592 | ddf->virt->populated_vdes = | |
4593 | cpu_to_be16( | |
4594 | 1 + be16_to_cpu( | |
4595 | ddf->virt->populated_vdes)); | |
4596 | dprintf("added VD %s in slot %d(s=%02x i=%02x)\n", | |
4597 | guid_str(vd->entries[0].guid), ent, | |
4598 | ddf->virt->entries[ent].state, | |
4599 | ddf->virt->entries[ent].init_state); | |
4600 | } | |
4601 | ddf_set_updates_pending(ddf, NULL); | |
4602 | } | |
4603 | ||
4604 | static void ddf_remove_failed(struct ddf_super *ddf) | |
4605 | { | |
4606 | /* Now remove any 'Failed' devices that are not part | |
4607 | * of any VD. They will have the Transition flag set. | |
4608 | * Once done, we need to update all dl->pdnum numbers. | |
4609 | */ | |
4610 | unsigned int pdnum; | |
4611 | unsigned int pd2 = 0; | |
4612 | struct dl *dl; | |
4613 | ||
4614 | for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes); | |
4615 | pdnum++) { | |
4616 | if (be32_to_cpu(ddf->phys->entries[pdnum].refnum) == | |
4617 | 0xFFFFFFFF) | |
4618 | continue; | |
4619 | if (be16_and(ddf->phys->entries[pdnum].state, | |
4620 | cpu_to_be16(DDF_Failed)) && | |
4621 | be16_and(ddf->phys->entries[pdnum].state, | |
4622 | cpu_to_be16(DDF_Transition))) { | |
4623 | /* skip this one unless in dlist*/ | |
4624 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4625 | if (dl->pdnum == (int)pdnum) | |
4626 | break; | |
4627 | if (!dl) | |
4628 | continue; | |
4629 | } | |
4630 | if (pdnum == pd2) | |
4631 | pd2++; | |
4632 | else { | |
4633 | ddf->phys->entries[pd2] = | |
4634 | ddf->phys->entries[pdnum]; | |
4635 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4636 | if (dl->pdnum == (int)pdnum) | |
4637 | dl->pdnum = pd2; | |
4638 | pd2++; | |
4639 | } | |
4640 | } | |
4641 | ddf->phys->used_pdes = cpu_to_be16(pd2); | |
4642 | while (pd2 < pdnum) { | |
4643 | memset(ddf->phys->entries[pd2].guid, 0xff, | |
4644 | DDF_GUID_LEN); | |
4645 | pd2++; | |
4646 | } | |
4647 | } | |
4648 | ||
4649 | static void ddf_update_vlist(struct ddf_super *ddf, struct dl *dl) | |
4650 | { | |
4651 | struct vcl *vcl; | |
4652 | unsigned int vn = 0; | |
4653 | int in_degraded = 0; | |
4654 | ||
4655 | if (dl->pdnum < 0) | |
4656 | return; | |
4657 | for (vcl = ddf->conflist; vcl ; vcl = vcl->next) { | |
4658 | unsigned int dn, ibvd; | |
4659 | const struct vd_config *conf; | |
4660 | int vstate; | |
4661 | dn = get_pd_index_from_refnum(vcl, | |
4662 | dl->disk.refnum, | |
4663 | ddf->mppe, | |
4664 | &conf, &ibvd); | |
4665 | if (dn == DDF_NOTFOUND) | |
4666 | continue; | |
4667 | dprintf("dev %d/%08x has %s (sec=%u) at %d\n", | |
4668 | dl->pdnum, | |
4669 | be32_to_cpu(dl->disk.refnum), | |
4670 | guid_str(conf->guid), | |
4671 | conf->sec_elmnt_seq, vn); | |
4672 | /* Clear the Transition flag */ | |
4673 | if (be16_and | |
4674 | (ddf->phys->entries[dl->pdnum].state, | |
4675 | cpu_to_be16(DDF_Failed))) | |
4676 | be16_clear(ddf->phys | |
4677 | ->entries[dl->pdnum].state, | |
4678 | cpu_to_be16(DDF_Transition)); | |
4679 | dl->vlist[vn++] = vcl; | |
4680 | vstate = ddf->virt->entries[vcl->vcnum].state | |
4681 | & DDF_state_mask; | |
4682 | if (vstate == DDF_state_degraded || | |
4683 | vstate == DDF_state_part_optimal) | |
4684 | in_degraded = 1; | |
4685 | } | |
4686 | while (vn < ddf->max_part) | |
4687 | dl->vlist[vn++] = NULL; | |
4688 | if (dl->vlist[0]) { | |
4689 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4690 | cpu_to_be16(DDF_Global_Spare)); | |
4691 | if (!be16_and(ddf->phys | |
4692 | ->entries[dl->pdnum].type, | |
4693 | cpu_to_be16(DDF_Active_in_VD))) { | |
4694 | be16_set(ddf->phys | |
4695 | ->entries[dl->pdnum].type, | |
4696 | cpu_to_be16(DDF_Active_in_VD)); | |
4697 | if (in_degraded) | |
4698 | be16_set(ddf->phys | |
4699 | ->entries[dl->pdnum] | |
4700 | .state, | |
4701 | cpu_to_be16 | |
4702 | (DDF_Rebuilding)); | |
4703 | } | |
4704 | } | |
4705 | if (dl->spare) { | |
4706 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4707 | cpu_to_be16(DDF_Global_Spare)); | |
4708 | be16_set(ddf->phys->entries[dl->pdnum].type, | |
4709 | cpu_to_be16(DDF_Spare)); | |
4710 | } | |
4711 | if (!dl->vlist[0] && !dl->spare) { | |
4712 | be16_set(ddf->phys->entries[dl->pdnum].type, | |
4713 | cpu_to_be16(DDF_Global_Spare)); | |
4714 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4715 | cpu_to_be16(DDF_Spare)); | |
4716 | be16_clear(ddf->phys->entries[dl->pdnum].type, | |
4717 | cpu_to_be16(DDF_Active_in_VD)); | |
4718 | } | |
4719 | } | |
4720 | ||
4721 | static void ddf_process_conf_update(struct supertype *st, | |
4722 | struct metadata_update *update) | |
4723 | { | |
4724 | struct ddf_super *ddf = st->sb; | |
4725 | struct vd_config *vc; | |
4726 | struct vcl *vcl; | |
4727 | struct dl *dl; | |
4728 | unsigned int ent; | |
4729 | unsigned int pdnum, len; | |
4730 | ||
4731 | vc = (struct vd_config*)update->buf; | |
4732 | len = ddf->conf_rec_len * 512; | |
4733 | if ((unsigned int)update->len != len * vc->sec_elmnt_count) { | |
4734 | pr_err("%s: insufficient data (%d) for %u BVDs\n", | |
4735 | guid_str(vc->guid), update->len, | |
4736 | vc->sec_elmnt_count); | |
4737 | return; | |
4738 | } | |
4739 | for (vcl = ddf->conflist; vcl ; vcl = vcl->next) | |
4740 | if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0) | |
4741 | break; | |
4742 | dprintf("conf update for %s (%s)\n", | |
4743 | guid_str(vc->guid), (vcl ? "old" : "new")); | |
4744 | if (vcl) { | |
4745 | /* An update, just copy the phys_refnum and lba_offset | |
4746 | * fields | |
4747 | */ | |
4748 | unsigned int i; | |
4749 | unsigned int k; | |
4750 | copy_matching_bvd(ddf, &vcl->conf, update); | |
4751 | for (k = 0; k < be16_to_cpu(vc->prim_elmnt_count); k++) | |
4752 | dprintf("BVD %u has %08x at %llu\n", 0, | |
4753 | be32_to_cpu(vcl->conf.phys_refnum[k]), | |
4754 | be64_to_cpu(LBA_OFFSET(ddf, | |
4755 | &vcl->conf)[k])); | |
4756 | for (i = 1; i < vc->sec_elmnt_count; i++) { | |
4757 | copy_matching_bvd(ddf, vcl->other_bvds[i-1], | |
4758 | update); | |
4759 | for (k = 0; k < be16_to_cpu( | |
4760 | vc->prim_elmnt_count); k++) | |
4761 | dprintf("BVD %u has %08x at %llu\n", i, | |
4762 | be32_to_cpu | |
4763 | (vcl->other_bvds[i-1]-> | |
4764 | phys_refnum[k]), | |
4765 | be64_to_cpu | |
4766 | (LBA_OFFSET | |
4767 | (ddf, | |
4768 | vcl->other_bvds[i-1])[k])); | |
4769 | } | |
4770 | } else { | |
4771 | /* A new VD_CONF */ | |
4772 | unsigned int i; | |
4773 | if (!update->space) | |
4774 | return; | |
4775 | vcl = update->space; | |
4776 | update->space = NULL; | |
4777 | vcl->next = ddf->conflist; | |
4778 | memcpy(&vcl->conf, vc, len); | |
4779 | ent = find_vde_by_guid(ddf, vc->guid); | |
4780 | if (ent == DDF_NOTFOUND) | |
4781 | return; | |
4782 | vcl->vcnum = ent; | |
4783 | ddf->conflist = vcl; | |
4784 | for (i = 1; i < vc->sec_elmnt_count; i++) | |
4785 | memcpy(vcl->other_bvds[i-1], | |
4786 | update->buf + len * i, len); | |
4787 | } | |
4788 | /* Set DDF_Transition on all Failed devices - to help | |
4789 | * us detect those that are no longer in use | |
4790 | */ | |
4791 | for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes); | |
4792 | pdnum++) | |
4793 | if (be16_and(ddf->phys->entries[pdnum].state, | |
4794 | cpu_to_be16(DDF_Failed))) | |
4795 | be16_set(ddf->phys->entries[pdnum].state, | |
4796 | cpu_to_be16(DDF_Transition)); | |
4797 | ||
4798 | /* Now make sure vlist is correct for each dl. */ | |
4799 | for (dl = ddf->dlist; dl; dl = dl->next) | |
4800 | ddf_update_vlist(ddf, dl); | |
4801 | ddf_remove_failed(ddf); | |
4802 | ||
4803 | ddf_set_updates_pending(ddf, vc); | |
4804 | } | |
4805 | ||
4806 | static void ddf_process_update(struct supertype *st, | |
4807 | struct metadata_update *update) | |
4808 | { | |
4809 | /* Apply this update to the metadata. | |
4810 | * The first 4 bytes are a DDF_*_MAGIC which guides | |
4811 | * our actions. | |
4812 | * Possible update are: | |
4813 | * DDF_PHYS_RECORDS_MAGIC | |
4814 | * Add a new physical device or remove an old one. | |
4815 | * Changes to this record only happen implicitly. | |
4816 | * used_pdes is the device number. | |
4817 | * DDF_VIRT_RECORDS_MAGIC | |
4818 | * Add a new VD. Possibly also change the 'access' bits. | |
4819 | * populated_vdes is the entry number. | |
4820 | * DDF_VD_CONF_MAGIC | |
4821 | * New or updated VD. the VIRT_RECORD must already | |
4822 | * exist. For an update, phys_refnum and lba_offset | |
4823 | * (at least) are updated, and the VD_CONF must | |
4824 | * be written to precisely those devices listed with | |
4825 | * a phys_refnum. | |
4826 | * DDF_SPARE_ASSIGN_MAGIC | |
4827 | * replacement Spare Assignment Record... but for which device? | |
4828 | * | |
4829 | * So, e.g.: | |
4830 | * - to create a new array, we send a VIRT_RECORD and | |
4831 | * a VD_CONF. Then assemble and start the array. | |
4832 | * - to activate a spare we send a VD_CONF to add the phys_refnum | |
4833 | * and offset. This will also mark the spare as active with | |
4834 | * a spare-assignment record. | |
4835 | */ | |
4836 | be32 *magic = (be32 *)update->buf; | |
4837 | ||
4838 | dprintf("Process update %x\n", be32_to_cpu(*magic)); | |
4839 | ||
4840 | if (be32_eq(*magic, DDF_PHYS_RECORDS_MAGIC)) { | |
4841 | if (update->len == (sizeof(struct phys_disk) + | |
4842 | sizeof(struct phys_disk_entry))) | |
4843 | ddf_process_phys_update(st, update); | |
4844 | } else if (be32_eq(*magic, DDF_VIRT_RECORDS_MAGIC)) { | |
4845 | if (update->len == (sizeof(struct virtual_disk) + | |
4846 | sizeof(struct virtual_entry))) | |
4847 | ddf_process_virt_update(st, update); | |
4848 | } else if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) { | |
4849 | ddf_process_conf_update(st, update); | |
4850 | } | |
4851 | /* case DDF_SPARE_ASSIGN_MAGIC */ | |
4852 | } | |
4853 | ||
4854 | static int ddf_prepare_update(struct supertype *st, | |
4855 | struct metadata_update *update) | |
4856 | { | |
4857 | /* This update arrived at managemon. | |
4858 | * We are about to pass it to monitor. | |
4859 | * If a malloc is needed, do it here. | |
4860 | */ | |
4861 | struct ddf_super *ddf = st->sb; | |
4862 | be32 *magic; | |
4863 | if (update->len < 4) | |
4864 | return 0; | |
4865 | magic = (be32 *)update->buf; | |
4866 | if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) { | |
4867 | struct vcl *vcl; | |
4868 | struct vd_config *conf; | |
4869 | if (update->len < (int)sizeof(*conf)) | |
4870 | return 0; | |
4871 | conf = (struct vd_config *) update->buf; | |
4872 | if (posix_memalign(&update->space, 512, | |
4873 | offsetof(struct vcl, conf) | |
4874 | + ddf->conf_rec_len * 512) != 0) { | |
4875 | update->space = NULL; | |
4876 | return 0; | |
4877 | } | |
4878 | vcl = update->space; | |
4879 | vcl->conf.sec_elmnt_count = conf->sec_elmnt_count; | |
4880 | if (alloc_other_bvds(ddf, vcl) != 0) { | |
4881 | free(update->space); | |
4882 | update->space = NULL; | |
4883 | return 0; | |
4884 | } | |
4885 | } | |
4886 | return 1; | |
4887 | } | |
4888 | ||
4889 | /* | |
4890 | * Check degraded state of a RAID10. | |
4891 | * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error | |
4892 | */ | |
4893 | static int raid10_degraded(struct mdinfo *info) | |
4894 | { | |
4895 | int n_prim, n_bvds; | |
4896 | int i; | |
4897 | struct mdinfo *d; | |
4898 | char *found; | |
4899 | int ret = -1; | |
4900 | ||
4901 | n_prim = info->array.layout & ~0x100; | |
4902 | n_bvds = info->array.raid_disks / n_prim; | |
4903 | found = xmalloc(n_bvds); | |
4904 | if (found == NULL) | |
4905 | return ret; | |
4906 | memset(found, 0, n_bvds); | |
4907 | for (d = info->devs; d; d = d->next) { | |
4908 | i = d->disk.raid_disk / n_prim; | |
4909 | if (i >= n_bvds) { | |
4910 | pr_err("BUG: invalid raid disk\n"); | |
4911 | goto out; | |
4912 | } | |
4913 | if (d->state_fd > 0) | |
4914 | found[i]++; | |
4915 | } | |
4916 | ret = 2; | |
4917 | for (i = 0; i < n_bvds; i++) | |
4918 | if (!found[i]) { | |
4919 | dprintf("BVD %d/%d failed\n", i, n_bvds); | |
4920 | ret = 0; | |
4921 | goto out; | |
4922 | } else if (found[i] < n_prim) { | |
4923 | dprintf("BVD %d/%d degraded\n", i, n_bvds); | |
4924 | ret = 1; | |
4925 | } | |
4926 | out: | |
4927 | free(found); | |
4928 | return ret; | |
4929 | } | |
4930 | ||
4931 | /* | |
4932 | * Check if the array 'a' is degraded but not failed. | |
4933 | * If it is, find as many spares as are available and needed and | |
4934 | * arrange for their inclusion. | |
4935 | * We only choose devices which are not already in the array, | |
4936 | * and prefer those with a spare-assignment to this array. | |
4937 | * Otherwise we choose global spares - assuming always that | |
4938 | * there is enough room. | |
4939 | * For each spare that we assign, we return an 'mdinfo' which | |
4940 | * describes the position for the device in the array. | |
4941 | * We also add to 'updates' a DDF_VD_CONF_MAGIC update with | |
4942 | * the new phys_refnum and lba_offset values. | |
4943 | * | |
4944 | * Only worry about BVDs at the moment. | |
4945 | */ | |
4946 | static struct mdinfo *ddf_activate_spare(struct active_array *a, | |
4947 | struct metadata_update **updates) | |
4948 | { | |
4949 | int working = 0; | |
4950 | struct mdinfo *d; | |
4951 | struct ddf_super *ddf = a->container->sb; | |
4952 | int global_ok = 0; | |
4953 | struct mdinfo *rv = NULL; | |
4954 | struct mdinfo *di; | |
4955 | struct metadata_update *mu; | |
4956 | struct dl *dl; | |
4957 | int i; | |
4958 | unsigned int j; | |
4959 | struct vcl *vcl; | |
4960 | struct vd_config *vc; | |
4961 | unsigned int n_bvd; | |
4962 | ||
4963 | for (d = a->info.devs ; d ; d = d->next) { | |
4964 | if ((d->curr_state & DS_FAULTY) && | |
4965 | d->state_fd >= 0) | |
4966 | /* wait for Removal to happen */ | |
4967 | return NULL; | |
4968 | if (d->state_fd >= 0) | |
4969 | working ++; | |
4970 | } | |
4971 | ||
4972 | dprintf("working=%d (%d) level=%d\n", working, | |
4973 | a->info.array.raid_disks, | |
4974 | a->info.array.level); | |
4975 | if (working == a->info.array.raid_disks) | |
4976 | return NULL; /* array not degraded */ | |
4977 | switch (a->info.array.level) { | |
4978 | case 1: | |
4979 | if (working == 0) | |
4980 | return NULL; /* failed */ | |
4981 | break; | |
4982 | case 4: | |
4983 | case 5: | |
4984 | if (working < a->info.array.raid_disks - 1) | |
4985 | return NULL; /* failed */ | |
4986 | break; | |
4987 | case 6: | |
4988 | if (working < a->info.array.raid_disks - 2) | |
4989 | return NULL; /* failed */ | |
4990 | break; | |
4991 | case 10: | |
4992 | if (raid10_degraded(&a->info) < 1) | |
4993 | return NULL; | |
4994 | break; | |
4995 | default: /* concat or stripe */ | |
4996 | return NULL; /* failed */ | |
4997 | } | |
4998 | ||
4999 | /* For each slot, if it is not working, find a spare */ | |
5000 | dl = ddf->dlist; | |
5001 | for (i = 0; i < a->info.array.raid_disks; i++) { | |
5002 | for (d = a->info.devs ; d ; d = d->next) | |
5003 | if (d->disk.raid_disk == i) | |
5004 | break; | |
5005 | dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0); | |
5006 | if (d && (d->state_fd >= 0)) | |
5007 | continue; | |
5008 | ||
5009 | /* OK, this device needs recovery. Find a spare */ | |
5010 | again: | |
5011 | for ( ; dl ; dl = dl->next) { | |
5012 | unsigned long long esize; | |
5013 | unsigned long long pos; | |
5014 | struct mdinfo *d2; | |
5015 | int is_global = 0; | |
5016 | int is_dedicated = 0; | |
5017 | be16 state; | |
5018 | ||
5019 | if (dl->pdnum < 0) | |
5020 | continue; | |
5021 | state = ddf->phys->entries[dl->pdnum].state; | |
5022 | if (be16_and(state, | |
5023 | cpu_to_be16(DDF_Failed|DDF_Missing)) || | |
5024 | !be16_and(state, | |
5025 | cpu_to_be16(DDF_Online))) | |
5026 | continue; | |
5027 | ||
5028 | /* If in this array, skip */ | |
5029 | for (d2 = a->info.devs ; d2 ; d2 = d2->next) | |
5030 | if (d2->state_fd >= 0 && | |
5031 | d2->disk.major == dl->major && | |
5032 | d2->disk.minor == dl->minor) { | |
5033 | dprintf("%x:%x (%08x) already in array\n", | |
5034 | dl->major, dl->minor, | |
5035 | be32_to_cpu(dl->disk.refnum)); | |
5036 | break; | |
5037 | } | |
5038 | if (d2) | |
5039 | continue; | |
5040 | if (be16_and(ddf->phys->entries[dl->pdnum].type, | |
5041 | cpu_to_be16(DDF_Spare))) { | |
5042 | /* Check spare assign record */ | |
5043 | if (dl->spare) { | |
5044 | if (dl->spare->type & DDF_spare_dedicated) { | |
5045 | /* check spare_ents for guid */ | |
5046 | unsigned int j; | |
5047 | for (j = 0 ; | |
5048 | j < be16_to_cpu | |
5049 | (dl->spare | |
5050 | ->populated); | |
5051 | j++) { | |
5052 | if (memcmp(dl->spare->spare_ents[j].guid, | |
5053 | ddf->virt->entries[a->info.container_member].guid, | |
5054 | DDF_GUID_LEN) == 0) | |
5055 | is_dedicated = 1; | |
5056 | } | |
5057 | } else | |
5058 | is_global = 1; | |
5059 | } | |
5060 | } else if (be16_and(ddf->phys->entries[dl->pdnum].type, | |
5061 | cpu_to_be16(DDF_Global_Spare))) { | |
5062 | is_global = 1; | |
5063 | } else if (!be16_and(ddf->phys | |
5064 | ->entries[dl->pdnum].state, | |
5065 | cpu_to_be16(DDF_Failed))) { | |
5066 | /* we can possibly use some of this */ | |
5067 | is_global = 1; | |
5068 | } | |
5069 | if ( ! (is_dedicated || | |
5070 | (is_global && global_ok))) { | |
5071 | dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor, | |
5072 | is_dedicated, is_global); | |
5073 | continue; | |
5074 | } | |
5075 | ||
5076 | /* We are allowed to use this device - is there space? | |
5077 | * We need a->info.component_size sectors */ | |
5078 | esize = a->info.component_size; | |
5079 | pos = find_space(ddf, dl, INVALID_SECTORS, &esize); | |
5080 | ||
5081 | if (esize < a->info.component_size) { | |
5082 | dprintf("%x:%x has no room: %llu %llu\n", | |
5083 | dl->major, dl->minor, | |
5084 | esize, a->info.component_size); | |
5085 | /* No room */ | |
5086 | continue; | |
5087 | } | |
5088 | ||
5089 | /* Cool, we have a device with some space at pos */ | |
5090 | di = xcalloc(1, sizeof(*di)); | |
5091 | di->disk.number = i; | |
5092 | di->disk.raid_disk = i; | |
5093 | di->disk.major = dl->major; | |
5094 | di->disk.minor = dl->minor; | |
5095 | di->disk.state = 0; | |
5096 | di->recovery_start = 0; | |
5097 | di->data_offset = pos; | |
5098 | di->component_size = a->info.component_size; | |
5099 | di->next = rv; | |
5100 | rv = di; | |
5101 | dprintf("%x:%x (%08x) to be %d at %llu\n", | |
5102 | dl->major, dl->minor, | |
5103 | be32_to_cpu(dl->disk.refnum), i, pos); | |
5104 | ||
5105 | break; | |
5106 | } | |
5107 | if (!dl && ! global_ok) { | |
5108 | /* not enough dedicated spares, try global */ | |
5109 | global_ok = 1; | |
5110 | dl = ddf->dlist; | |
5111 | goto again; | |
5112 | } | |
5113 | } | |
5114 | ||
5115 | if (!rv) | |
5116 | /* No spares found */ | |
5117 | return rv; | |
5118 | /* Now 'rv' has a list of devices to return. | |
5119 | * Create a metadata_update record to update the | |
5120 | * phys_refnum and lba_offset values | |
5121 | */ | |
5122 | vc = find_vdcr(ddf, a->info.container_member, rv->disk.raid_disk, | |
5123 | &n_bvd, &vcl); | |
5124 | if (vc == NULL) | |
5125 | return NULL; | |
5126 | ||
5127 | mu = xmalloc(sizeof(*mu)); | |
5128 | if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) { | |
5129 | free(mu); | |
5130 | mu = NULL; | |
5131 | } | |
5132 | ||
5133 | mu->len = ddf->conf_rec_len * 512 * vcl->conf.sec_elmnt_count; | |
5134 | mu->buf = xmalloc(mu->len); | |
5135 | mu->space = NULL; | |
5136 | mu->space_list = NULL; | |
5137 | mu->next = *updates; | |
5138 | memcpy(mu->buf, &vcl->conf, ddf->conf_rec_len * 512); | |
5139 | for (j = 1; j < vcl->conf.sec_elmnt_count; j++) | |
5140 | memcpy(mu->buf + j * ddf->conf_rec_len * 512, | |
5141 | vcl->other_bvds[j-1], ddf->conf_rec_len * 512); | |
5142 | ||
5143 | vc = (struct vd_config*)mu->buf; | |
5144 | for (di = rv ; di ; di = di->next) { | |
5145 | unsigned int i_sec, i_prim; | |
5146 | i_sec = di->disk.raid_disk | |
5147 | / be16_to_cpu(vcl->conf.prim_elmnt_count); | |
5148 | i_prim = di->disk.raid_disk | |
5149 | % be16_to_cpu(vcl->conf.prim_elmnt_count); | |
5150 | vc = (struct vd_config *)(mu->buf | |
5151 | + i_sec * ddf->conf_rec_len * 512); | |
5152 | for (dl = ddf->dlist; dl; dl = dl->next) | |
5153 | if (dl->major == di->disk.major && | |
5154 | dl->minor == di->disk.minor) | |
5155 | break; | |
5156 | if (!dl || dl->pdnum < 0) { | |
5157 | pr_err("BUG: can't find disk %d (%d/%d)\n", | |
5158 | di->disk.raid_disk, | |
5159 | di->disk.major, di->disk.minor); | |
5160 | return NULL; | |
5161 | } | |
5162 | vc->phys_refnum[i_prim] = ddf->phys->entries[dl->pdnum].refnum; | |
5163 | LBA_OFFSET(ddf, vc)[i_prim] = cpu_to_be64(di->data_offset); | |
5164 | dprintf("BVD %u gets %u: %08x at %llu\n", i_sec, i_prim, | |
5165 | be32_to_cpu(vc->phys_refnum[i_prim]), | |
5166 | be64_to_cpu(LBA_OFFSET(ddf, vc)[i_prim])); | |
5167 | } | |
5168 | *updates = mu; | |
5169 | return rv; | |
5170 | } | |
5171 | ||
5172 | static int ddf_level_to_layout(int level) | |
5173 | { | |
5174 | switch(level) { | |
5175 | case 0: | |
5176 | case 1: | |
5177 | return 0; | |
5178 | case 5: | |
5179 | return ALGORITHM_LEFT_SYMMETRIC; | |
5180 | case 6: | |
5181 | return ALGORITHM_ROTATING_N_CONTINUE; | |
5182 | case 10: | |
5183 | return 0x102; | |
5184 | default: | |
5185 | return UnSet; | |
5186 | } | |
5187 | } | |
5188 | ||
5189 | static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk) | |
5190 | { | |
5191 | if (level && *level == UnSet) | |
5192 | *level = LEVEL_CONTAINER; | |
5193 | ||
5194 | if (level && layout && *layout == UnSet) | |
5195 | *layout = ddf_level_to_layout(*level); | |
5196 | } | |
5197 | ||
5198 | struct superswitch super_ddf = { | |
5199 | .examine_super = examine_super_ddf, | |
5200 | .brief_examine_super = brief_examine_super_ddf, | |
5201 | .brief_examine_subarrays = brief_examine_subarrays_ddf, | |
5202 | .export_examine_super = export_examine_super_ddf, | |
5203 | .detail_super = detail_super_ddf, | |
5204 | .brief_detail_super = brief_detail_super_ddf, | |
5205 | .validate_geometry = validate_geometry_ddf, | |
5206 | .write_init_super = write_init_super_ddf, | |
5207 | .add_to_super = add_to_super_ddf, | |
5208 | .remove_from_super = remove_from_super_ddf, | |
5209 | .load_container = load_container_ddf, | |
5210 | .copy_metadata = copy_metadata_ddf, | |
5211 | .kill_subarray = kill_subarray_ddf, | |
5212 | .match_home = match_home_ddf, | |
5213 | .uuid_from_super= uuid_from_super_ddf, | |
5214 | .getinfo_super = getinfo_super_ddf, | |
5215 | .update_super = update_super_ddf, | |
5216 | ||
5217 | .avail_size = avail_size_ddf, | |
5218 | ||
5219 | .compare_super = compare_super_ddf, | |
5220 | ||
5221 | .load_super = load_super_ddf, | |
5222 | .init_super = init_super_ddf, | |
5223 | .store_super = store_super_ddf, | |
5224 | .free_super = free_super_ddf, | |
5225 | .match_metadata_desc = match_metadata_desc_ddf, | |
5226 | .container_content = container_content_ddf, | |
5227 | .default_geometry = default_geometry_ddf, | |
5228 | ||
5229 | .external = 1, | |
5230 | ||
5231 | /* for mdmon */ | |
5232 | .open_new = ddf_open_new, | |
5233 | .set_array_state= ddf_set_array_state, | |
5234 | .set_disk = ddf_set_disk, | |
5235 | .sync_metadata = ddf_sync_metadata, | |
5236 | .process_update = ddf_process_update, | |
5237 | .prepare_update = ddf_prepare_update, | |
5238 | .activate_spare = ddf_activate_spare, | |
5239 | .name = "ddf", | |
5240 | }; |