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