]> 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
platform-intel - cache 'intel_devices' for a few seconds.
[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 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
51 *
52 * Note:
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
55 *
56 */
57
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
65 #define DDF_JBOD 0x0f
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
70
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
84
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
87
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
93
94 /* Magic numbers */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
105
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
109
110 struct ddf_header {
111 __u32 magic; /* DDF_HEADER_MAGIC */
112 __u32 crc;
113 char guid[DDF_GUID_LEN];
114 char revision[8]; /* 01.02.00 */
115 __u32 seq; /* starts at '1' */
116 __u32 timestamp;
117 __u8 openflag;
118 __u8 foreignflag;
119 __u8 enforcegroups;
120 __u8 pad0; /* 0xff */
121 __u8 pad1[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext[32]; /* reserved: fill with 0xff */
124 __u64 primary_lba;
125 __u64 secondary_lba;
126 __u8 type;
127 __u8 pad2[3]; /* 0xff */
128 __u32 workspace_len; /* sectors for vendor space -
129 * at least 32768(sectors) */
130 __u64 workspace_lba;
131 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
136 *12/512) */
137 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset;
141 __u32 controller_section_length;
142 __u32 phys_section_offset;
143 __u32 phys_section_length;
144 __u32 virt_section_offset;
145 __u32 virt_section_length;
146 __u32 config_section_offset;
147 __u32 config_section_length;
148 __u32 data_section_offset;
149 __u32 data_section_length;
150 __u32 bbm_section_offset;
151 __u32 bbm_section_length;
152 __u32 diag_space_offset;
153 __u32 diag_space_length;
154 __u32 vendor_offset;
155 __u32 vendor_length;
156 /* 256 bytes so far */
157 __u8 pad4[256]; /* 0xff */
158 };
159
160 /* type field */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
164
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data {
167 __u32 magic; /* DDF_CONTROLLER_MAGIC */
168 __u32 crc;
169 char guid[DDF_GUID_LEN];
170 struct controller_type {
171 __u16 vendor_id;
172 __u16 device_id;
173 __u16 sub_vendor_id;
174 __u16 sub_device_id;
175 } type;
176 char product_id[16];
177 __u8 pad[8]; /* 0xff */
178 __u8 vendor_data[448];
179 };
180
181 /* The content of phys_section - global scope */
182 struct phys_disk {
183 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
184 __u32 crc;
185 __u16 used_pdes;
186 __u16 max_pdes;
187 __u8 pad[52];
188 struct phys_disk_entry {
189 char guid[DDF_GUID_LEN];
190 __u32 refnum;
191 __u16 type;
192 __u16 state;
193 __u64 config_size; /* DDF structures must be after here */
194 char path[18]; /* another horrible structure really */
195 __u8 pad[6];
196 } entries[0];
197 };
198
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
206
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
212
213 /* phys_disk_entry.state is a bigendian bitmap */
214 #define DDF_Online 1
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
218 #define DDF_SMART 16
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
221
222 /* The content of the virt_section global scope */
223 struct virtual_disk {
224 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
225 __u32 crc;
226 __u16 populated_vdes;
227 __u16 max_vdes;
228 __u8 pad[52];
229 struct virtual_entry {
230 char guid[DDF_GUID_LEN];
231 __u16 unit;
232 __u16 pad0; /* 0xffff */
233 __u16 guid_crc;
234 __u16 type;
235 __u8 state;
236 __u8 init_state;
237 __u8 pad1[14];
238 char name[16];
239 } entries[0];
240 };
241
242 /* virtual_entry.type is a bitmap - bigendian */
243 #define DDF_Shared 1
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
247
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
256
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
259
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
266
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
271
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
275 */
276
277 struct vd_config {
278 __u32 magic; /* DDF_VD_CONF_MAGIC */
279 __u32 crc;
280 char guid[DDF_GUID_LEN];
281 __u32 timestamp;
282 __u32 seqnum;
283 __u8 pad0[24];
284 __u16 prim_elmnt_count;
285 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
286 __u8 prl;
287 __u8 rlq;
288 __u8 sec_elmnt_count;
289 __u8 sec_elmnt_seq;
290 __u8 srl;
291 __u64 blocks; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks; /* blocks in array */
295 __u8 pad1[8];
296 __u32 spare_refs[8];
297 __u8 cache_pol[8];
298 __u8 bg_rate;
299 __u8 pad2[3];
300 __u8 pad3[52];
301 __u8 pad4[192];
302 __u8 v0[32]; /* reserved- 0xff */
303 __u8 v1[32]; /* reserved- 0xff */
304 __u8 v2[16]; /* reserved- 0xff */
305 __u8 v3[16]; /* reserved- 0xff */
306 __u8 vendor[32];
307 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
310 };
311
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
320
321 struct spare_assign {
322 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
323 __u32 crc;
324 __u32 timestamp;
325 __u8 reserved[7];
326 __u8 type;
327 __u16 populated; /* SAEs used */
328 __u16 max; /* max SAEs */
329 __u8 pad[8];
330 struct spare_assign_entry {
331 char guid[DDF_GUID_LEN];
332 __u16 secondary_element;
333 __u8 pad[6];
334 } spare_ents[0];
335 };
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
341
342 /* The data_section contents - local scope */
343 struct disk_data {
344 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
345 __u32 crc;
346 char guid[DDF_GUID_LEN];
347 __u32 refnum; /* crc of some magic drive data ... */
348 __u8 forced_ref; /* set when above was not result of magic */
349 __u8 forced_guid; /* set if guid was forced rather than magic */
350 __u8 vendor[32];
351 __u8 pad[442];
352 };
353
354 /* bbm_section content */
355 struct bad_block_log {
356 __u32 magic;
357 __u32 crc;
358 __u16 entry_count;
359 __u32 spare_count;
360 __u8 pad[10];
361 __u64 first_spare;
362 struct mapped_block {
363 __u64 defective_start;
364 __u32 replacement_start;
365 __u16 remap_count;
366 __u8 pad[2];
367 } entries[0];
368 };
369
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
374 * - ddf header
375 * - controller_data
376 * - Physical disk records
377 * - Virtual disk records
378 * The local data is:
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
382 *
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
391 */
392 struct ddf_super {
393 struct ddf_header anchor, primary, secondary;
394 struct ddf_controller_data controller;
395 struct ddf_header *active;
396 struct phys_disk *phys;
397 struct virtual_disk *virt;
398 int pdsize, vdsize;
399 unsigned int max_part, mppe, conf_rec_len;
400 int currentdev;
401 int updates_pending;
402 struct vcl {
403 union {
404 char space[512];
405 struct {
406 struct vcl *next;
407 __u64 *lba_offset; /* location in 'conf' of
408 * the lba table */
409 unsigned int vcnum; /* index into ->virt */
410 __u64 *block_sizes; /* NULL if all the same */
411 };
412 };
413 struct vd_config conf;
414 } *conflist, *currentconf;
415 struct dl {
416 union {
417 char space[512];
418 struct {
419 struct dl *next;
420 int major, minor;
421 char *devname;
422 int fd;
423 unsigned long long size; /* sectors */
424 int pdnum; /* index in ->phys */
425 struct spare_assign *spare;
426 void *mdupdate; /* hold metadata update */
427
428 /* These fields used by auto-layout */
429 int raiddisk; /* slot to fill in autolayout */
430 __u64 esize;
431 };
432 };
433 struct disk_data disk;
434 struct vcl *vlist[0]; /* max_part in size */
435 } *dlist, *add_list;
436 };
437
438 #ifndef offsetof
439 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
440 #endif
441
442 static unsigned int calc_crc(void *buf, int len)
443 {
444 /* crcs are always at the same place as in the ddf_header */
445 struct ddf_header *ddf = buf;
446 __u32 oldcrc = ddf->crc;
447 __u32 newcrc;
448 ddf->crc = 0xffffffff;
449
450 newcrc = crc32(0, buf, len);
451 ddf->crc = oldcrc;
452 /* The crc is store (like everything) bigendian, so convert
453 * here for simplicity
454 */
455 return __cpu_to_be32(newcrc);
456 }
457
458 static int load_ddf_header(int fd, unsigned long long lba,
459 unsigned long long size,
460 int type,
461 struct ddf_header *hdr, struct ddf_header *anchor)
462 {
463 /* read a ddf header (primary or secondary) from fd/lba
464 * and check that it is consistent with anchor
465 * Need to check:
466 * magic, crc, guid, rev, and LBA's header_type, and
467 * everything after header_type must be the same
468 */
469 if (lba >= size-1)
470 return 0;
471
472 if (lseek64(fd, lba<<9, 0) < 0)
473 return 0;
474
475 if (read(fd, hdr, 512) != 512)
476 return 0;
477
478 if (hdr->magic != DDF_HEADER_MAGIC)
479 return 0;
480 if (calc_crc(hdr, 512) != hdr->crc)
481 return 0;
482 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
483 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
484 anchor->primary_lba != hdr->primary_lba ||
485 anchor->secondary_lba != hdr->secondary_lba ||
486 hdr->type != type ||
487 memcmp(anchor->pad2, hdr->pad2, 512 -
488 offsetof(struct ddf_header, pad2)) != 0)
489 return 0;
490
491 /* Looks good enough to me... */
492 return 1;
493 }
494
495 static void *load_section(int fd, struct ddf_super *super, void *buf,
496 __u32 offset_be, __u32 len_be, int check)
497 {
498 unsigned long long offset = __be32_to_cpu(offset_be);
499 unsigned long long len = __be32_to_cpu(len_be);
500 int dofree = (buf == NULL);
501
502 if (check)
503 if (len != 2 && len != 8 && len != 32
504 && len != 128 && len != 512)
505 return NULL;
506
507 if (len > 1024)
508 return NULL;
509 if (buf) {
510 /* All pre-allocated sections are a single block */
511 if (len != 1)
512 return NULL;
513 } else if (posix_memalign(&buf, 512, len<<9) != 0)
514 buf = NULL;
515
516 if (!buf)
517 return NULL;
518
519 if (super->active->type == 1)
520 offset += __be64_to_cpu(super->active->primary_lba);
521 else
522 offset += __be64_to_cpu(super->active->secondary_lba);
523
524 if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
525 if (dofree)
526 free(buf);
527 return NULL;
528 }
529 if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
530 if (dofree)
531 free(buf);
532 return NULL;
533 }
534 return buf;
535 }
536
537 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
538 {
539 unsigned long long dsize;
540
541 get_dev_size(fd, NULL, &dsize);
542
543 if (lseek64(fd, dsize-512, 0) < 0) {
544 if (devname)
545 pr_err("Cannot seek to anchor block on %s: %s\n",
546 devname, strerror(errno));
547 return 1;
548 }
549 if (read(fd, &super->anchor, 512) != 512) {
550 if (devname)
551 pr_err("Cannot read anchor block on %s: %s\n",
552 devname, strerror(errno));
553 return 1;
554 }
555 if (super->anchor.magic != DDF_HEADER_MAGIC) {
556 if (devname)
557 pr_err("no DDF anchor found on %s\n",
558 devname);
559 return 2;
560 }
561 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
562 if (devname)
563 pr_err("bad CRC on anchor on %s\n",
564 devname);
565 return 2;
566 }
567 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
568 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
569 if (devname)
570 pr_err("can only support super revision"
571 " %.8s and earlier, not %.8s on %s\n",
572 DDF_REVISION_2, super->anchor.revision,devname);
573 return 2;
574 }
575 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
576 dsize >> 9, 1,
577 &super->primary, &super->anchor) == 0) {
578 if (devname)
579 pr_err("Failed to load primary DDF header "
580 "on %s\n", devname);
581 return 2;
582 }
583 super->active = &super->primary;
584 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
585 dsize >> 9, 2,
586 &super->secondary, &super->anchor)) {
587 if ((__be32_to_cpu(super->primary.seq)
588 < __be32_to_cpu(super->secondary.seq) &&
589 !super->secondary.openflag)
590 || (__be32_to_cpu(super->primary.seq)
591 == __be32_to_cpu(super->secondary.seq) &&
592 super->primary.openflag && !super->secondary.openflag)
593 )
594 super->active = &super->secondary;
595 }
596 return 0;
597 }
598
599 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
600 {
601 void *ok;
602 ok = load_section(fd, super, &super->controller,
603 super->active->controller_section_offset,
604 super->active->controller_section_length,
605 0);
606 super->phys = load_section(fd, super, NULL,
607 super->active->phys_section_offset,
608 super->active->phys_section_length,
609 1);
610 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
611
612 super->virt = load_section(fd, super, NULL,
613 super->active->virt_section_offset,
614 super->active->virt_section_length,
615 1);
616 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
617 if (!ok ||
618 !super->phys ||
619 !super->virt) {
620 free(super->phys);
621 free(super->virt);
622 super->phys = NULL;
623 super->virt = NULL;
624 return 2;
625 }
626 super->conflist = NULL;
627 super->dlist = NULL;
628
629 super->max_part = __be16_to_cpu(super->active->max_partitions);
630 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
631 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
632 return 0;
633 }
634
635 static int load_ddf_local(int fd, struct ddf_super *super,
636 char *devname, int keep)
637 {
638 struct dl *dl;
639 struct stat stb;
640 char *conf;
641 unsigned int i;
642 unsigned int confsec;
643 int vnum;
644 unsigned int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
645 unsigned long long dsize;
646
647 /* First the local disk info */
648 if (posix_memalign((void**)&dl, 512,
649 sizeof(*dl) +
650 (super->max_part) * sizeof(dl->vlist[0])) != 0) {
651 pr_err("%s could not allocate disk info buffer\n",
652 __func__);
653 return 1;
654 }
655
656 load_section(fd, super, &dl->disk,
657 super->active->data_section_offset,
658 super->active->data_section_length,
659 0);
660 dl->devname = devname ? xstrdup(devname) : NULL;
661
662 fstat(fd, &stb);
663 dl->major = major(stb.st_rdev);
664 dl->minor = minor(stb.st_rdev);
665 dl->next = super->dlist;
666 dl->fd = keep ? fd : -1;
667
668 dl->size = 0;
669 if (get_dev_size(fd, devname, &dsize))
670 dl->size = dsize >> 9;
671 dl->spare = NULL;
672 for (i = 0 ; i < super->max_part ; i++)
673 dl->vlist[i] = NULL;
674 super->dlist = dl;
675 dl->pdnum = -1;
676 for (i = 0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
677 if (memcmp(super->phys->entries[i].guid,
678 dl->disk.guid, DDF_GUID_LEN) == 0)
679 dl->pdnum = i;
680
681 /* Now the config list. */
682 /* 'conf' is an array of config entries, some of which are
683 * probably invalid. Those which are good need to be copied into
684 * the conflist
685 */
686
687 conf = load_section(fd, super, NULL,
688 super->active->config_section_offset,
689 super->active->config_section_length,
690 0);
691
692 vnum = 0;
693 for (confsec = 0;
694 confsec < __be32_to_cpu(super->active->config_section_length);
695 confsec += super->conf_rec_len) {
696 struct vd_config *vd =
697 (struct vd_config *)((char*)conf + confsec*512);
698 struct vcl *vcl;
699
700 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
701 if (dl->spare)
702 continue;
703 if (posix_memalign((void**)&dl->spare, 512,
704 super->conf_rec_len*512) != 0) {
705 pr_err("%s could not allocate spare info buf\n",
706 __func__);
707 return 1;
708 }
709
710 memcpy(dl->spare, vd, super->conf_rec_len*512);
711 continue;
712 }
713 if (vd->magic != DDF_VD_CONF_MAGIC)
714 continue;
715 for (vcl = super->conflist; vcl; vcl = vcl->next) {
716 if (memcmp(vcl->conf.guid,
717 vd->guid, DDF_GUID_LEN) == 0)
718 break;
719 }
720
721 if (vcl) {
722 dl->vlist[vnum++] = vcl;
723 if (__be32_to_cpu(vd->seqnum) <=
724 __be32_to_cpu(vcl->conf.seqnum))
725 continue;
726 } else {
727 if (posix_memalign((void**)&vcl, 512,
728 (super->conf_rec_len*512 +
729 offsetof(struct vcl, conf))) != 0) {
730 pr_err("%s could not allocate vcl buf\n",
731 __func__);
732 return 1;
733 }
734 vcl->next = super->conflist;
735 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
736 super->conflist = vcl;
737 dl->vlist[vnum++] = vcl;
738 }
739 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
740 vcl->lba_offset = (__u64*)
741 &vcl->conf.phys_refnum[super->mppe];
742
743 for (i=0; i < max_virt_disks ; i++)
744 if (memcmp(super->virt->entries[i].guid,
745 vcl->conf.guid, DDF_GUID_LEN)==0)
746 break;
747 if (i < max_virt_disks)
748 vcl->vcnum = i;
749 }
750 free(conf);
751
752 return 0;
753 }
754
755 #ifndef MDASSEMBLE
756 static int load_super_ddf_all(struct supertype *st, int fd,
757 void **sbp, char *devname);
758 #endif
759
760 static void free_super_ddf(struct supertype *st);
761
762 static int load_super_ddf(struct supertype *st, int fd,
763 char *devname)
764 {
765 unsigned long long dsize;
766 struct ddf_super *super;
767 int rv;
768
769 if (get_dev_size(fd, devname, &dsize) == 0)
770 return 1;
771
772 if (test_partition(fd))
773 /* DDF is not allowed on partitions */
774 return 1;
775
776 /* 32M is a lower bound */
777 if (dsize <= 32*1024*1024) {
778 if (devname)
779 pr_err("%s is too small for ddf: "
780 "size is %llu sectors.\n",
781 devname, dsize>>9);
782 return 1;
783 }
784 if (dsize & 511) {
785 if (devname)
786 pr_err("%s is an odd size for ddf: "
787 "size is %llu bytes.\n",
788 devname, dsize);
789 return 1;
790 }
791
792 free_super_ddf(st);
793
794 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
795 pr_err("malloc of %zu failed.\n",
796 sizeof(*super));
797 return 1;
798 }
799 memset(super, 0, sizeof(*super));
800
801 rv = load_ddf_headers(fd, super, devname);
802 if (rv) {
803 free(super);
804 return rv;
805 }
806
807 /* Have valid headers and have chosen the best. Let's read in the rest*/
808
809 rv = load_ddf_global(fd, super, devname);
810
811 if (rv) {
812 if (devname)
813 pr_err("Failed to load all information "
814 "sections on %s\n", devname);
815 free(super);
816 return rv;
817 }
818
819 rv = load_ddf_local(fd, super, devname, 0);
820
821 if (rv) {
822 if (devname)
823 pr_err("Failed to load all information "
824 "sections on %s\n", devname);
825 free(super);
826 return rv;
827 }
828
829 /* Should possibly check the sections .... */
830
831 st->sb = super;
832 if (st->ss == NULL) {
833 st->ss = &super_ddf;
834 st->minor_version = 0;
835 st->max_devs = 512;
836 }
837 return 0;
838
839 }
840
841 static void free_super_ddf(struct supertype *st)
842 {
843 struct ddf_super *ddf = st->sb;
844 if (ddf == NULL)
845 return;
846 free(ddf->phys);
847 free(ddf->virt);
848 while (ddf->conflist) {
849 struct vcl *v = ddf->conflist;
850 ddf->conflist = v->next;
851 if (v->block_sizes)
852 free(v->block_sizes);
853 free(v);
854 }
855 while (ddf->dlist) {
856 struct dl *d = ddf->dlist;
857 ddf->dlist = d->next;
858 if (d->fd >= 0)
859 close(d->fd);
860 if (d->spare)
861 free(d->spare);
862 free(d);
863 }
864 while (ddf->add_list) {
865 struct dl *d = ddf->add_list;
866 ddf->add_list = d->next;
867 if (d->fd >= 0)
868 close(d->fd);
869 if (d->spare)
870 free(d->spare);
871 free(d);
872 }
873 free(ddf);
874 st->sb = NULL;
875 }
876
877 static struct supertype *match_metadata_desc_ddf(char *arg)
878 {
879 /* 'ddf' only support containers */
880 struct supertype *st;
881 if (strcmp(arg, "ddf") != 0 &&
882 strcmp(arg, "default") != 0
883 )
884 return NULL;
885
886 st = xcalloc(1, sizeof(*st));
887 st->container_dev = NoMdDev;
888 st->ss = &super_ddf;
889 st->max_devs = 512;
890 st->minor_version = 0;
891 st->sb = NULL;
892 return st;
893 }
894
895 #ifndef MDASSEMBLE
896
897 static mapping_t ddf_state[] = {
898 { "Optimal", 0},
899 { "Degraded", 1},
900 { "Deleted", 2},
901 { "Missing", 3},
902 { "Failed", 4},
903 { "Partially Optimal", 5},
904 { "-reserved-", 6},
905 { "-reserved-", 7},
906 { NULL, 0}
907 };
908
909 static mapping_t ddf_init_state[] = {
910 { "Not Initialised", 0},
911 { "QuickInit in Progress", 1},
912 { "Fully Initialised", 2},
913 { "*UNKNOWN*", 3},
914 { NULL, 0}
915 };
916 static mapping_t ddf_access[] = {
917 { "Read/Write", 0},
918 { "Reserved", 1},
919 { "Read Only", 2},
920 { "Blocked (no access)", 3},
921 { NULL ,0}
922 };
923
924 static mapping_t ddf_level[] = {
925 { "RAID0", DDF_RAID0},
926 { "RAID1", DDF_RAID1},
927 { "RAID3", DDF_RAID3},
928 { "RAID4", DDF_RAID4},
929 { "RAID5", DDF_RAID5},
930 { "RAID1E",DDF_RAID1E},
931 { "JBOD", DDF_JBOD},
932 { "CONCAT",DDF_CONCAT},
933 { "RAID5E",DDF_RAID5E},
934 { "RAID5EE",DDF_RAID5EE},
935 { "RAID6", DDF_RAID6},
936 { NULL, 0}
937 };
938 static mapping_t ddf_sec_level[] = {
939 { "Striped", DDF_2STRIPED},
940 { "Mirrored", DDF_2MIRRORED},
941 { "Concat", DDF_2CONCAT},
942 { "Spanned", DDF_2SPANNED},
943 { NULL, 0}
944 };
945 #endif
946
947 struct num_mapping {
948 int num1, num2;
949 };
950 static struct num_mapping ddf_level_num[] = {
951 { DDF_RAID0, 0 },
952 { DDF_RAID1, 1 },
953 { DDF_RAID3, LEVEL_UNSUPPORTED },
954 { DDF_RAID4, 4 },
955 { DDF_RAID5, 5 },
956 { DDF_RAID1E, LEVEL_UNSUPPORTED },
957 { DDF_JBOD, LEVEL_UNSUPPORTED },
958 { DDF_CONCAT, LEVEL_LINEAR },
959 { DDF_RAID5E, LEVEL_UNSUPPORTED },
960 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
961 { DDF_RAID6, 6},
962 { MAXINT, MAXINT }
963 };
964
965 static int map_num1(struct num_mapping *map, int num)
966 {
967 int i;
968 for (i=0 ; map[i].num1 != MAXINT; i++)
969 if (map[i].num1 == num)
970 break;
971 return map[i].num2;
972 }
973
974 static int all_ff(char *guid)
975 {
976 int i;
977 for (i = 0; i < DDF_GUID_LEN; i++)
978 if (guid[i] != (char)0xff)
979 return 0;
980 return 1;
981 }
982
983 #ifndef MDASSEMBLE
984 static void print_guid(char *guid, int tstamp)
985 {
986 /* A GUIDs are part (or all) ASCII and part binary.
987 * They tend to be space padded.
988 * We print the GUID in HEX, then in parentheses add
989 * any initial ASCII sequence, and a possible
990 * time stamp from bytes 16-19
991 */
992 int l = DDF_GUID_LEN;
993 int i;
994
995 for (i=0 ; i<DDF_GUID_LEN ; i++) {
996 if ((i&3)==0 && i != 0) printf(":");
997 printf("%02X", guid[i]&255);
998 }
999
1000 printf("\n (");
1001 while (l && guid[l-1] == ' ')
1002 l--;
1003 for (i=0 ; i<l ; i++) {
1004 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1005 fputc(guid[i], stdout);
1006 else
1007 break;
1008 }
1009 if (tstamp) {
1010 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1011 char tbuf[100];
1012 struct tm *tm;
1013 tm = localtime(&then);
1014 strftime(tbuf, 100, " %D %T",tm);
1015 fputs(tbuf, stdout);
1016 }
1017 printf(")");
1018 }
1019
1020 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1021 {
1022 int crl = sb->conf_rec_len;
1023 struct vcl *vcl;
1024
1025 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1026 unsigned int i;
1027 struct vd_config *vc = &vcl->conf;
1028
1029 if (calc_crc(vc, crl*512) != vc->crc)
1030 continue;
1031 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1032 continue;
1033
1034 /* Ok, we know about this VD, let's give more details */
1035 printf(" Raid Devices[%d] : %d (", n,
1036 __be16_to_cpu(vc->prim_elmnt_count));
1037 for (i = 0; i < __be16_to_cpu(vc->prim_elmnt_count); i++) {
1038 int j;
1039 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1040 for (j=0; j<cnt; j++)
1041 if (vc->phys_refnum[i] == sb->phys->entries[j].refnum)
1042 break;
1043 if (i) printf(" ");
1044 if (j < cnt)
1045 printf("%d", j);
1046 else
1047 printf("--");
1048 }
1049 printf(")\n");
1050 if (vc->chunk_shift != 255)
1051 printf(" Chunk Size[%d] : %d sectors\n", n,
1052 1 << vc->chunk_shift);
1053 printf(" Raid Level[%d] : %s\n", n,
1054 map_num(ddf_level, vc->prl)?:"-unknown-");
1055 if (vc->sec_elmnt_count != 1) {
1056 printf(" Secondary Position[%d] : %d of %d\n", n,
1057 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1058 printf(" Secondary Level[%d] : %s\n", n,
1059 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1060 }
1061 printf(" Device Size[%d] : %llu\n", n,
1062 (unsigned long long)__be64_to_cpu(vc->blocks)/2);
1063 printf(" Array Size[%d] : %llu\n", n,
1064 (unsigned long long)__be64_to_cpu(vc->array_blocks)/2);
1065 }
1066 }
1067
1068 static void examine_vds(struct ddf_super *sb)
1069 {
1070 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1071 int i;
1072 printf(" Virtual Disks : %d\n", cnt);
1073
1074 for (i=0; i<cnt; i++) {
1075 struct virtual_entry *ve = &sb->virt->entries[i];
1076 printf("\n");
1077 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1078 printf("\n");
1079 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1080 printf(" state[%d] : %s, %s%s\n", i,
1081 map_num(ddf_state, ve->state & 7),
1082 (ve->state & 8) ? "Morphing, ": "",
1083 (ve->state & 16)? "Not Consistent" : "Consistent");
1084 printf(" init state[%d] : %s\n", i,
1085 map_num(ddf_init_state, ve->init_state&3));
1086 printf(" access[%d] : %s\n", i,
1087 map_num(ddf_access, (ve->init_state>>6) & 3));
1088 printf(" Name[%d] : %.16s\n", i, ve->name);
1089 examine_vd(i, sb, ve->guid);
1090 }
1091 if (cnt) printf("\n");
1092 }
1093
1094 static void examine_pds(struct ddf_super *sb)
1095 {
1096 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1097 int i;
1098 struct dl *dl;
1099 printf(" Physical Disks : %d\n", cnt);
1100 printf(" Number RefNo Size Device Type/State\n");
1101
1102 for (i=0 ; i<cnt ; i++) {
1103 struct phys_disk_entry *pd = &sb->phys->entries[i];
1104 int type = __be16_to_cpu(pd->type);
1105 int state = __be16_to_cpu(pd->state);
1106
1107 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1108 //printf("\n");
1109 printf(" %3d %08x ", i,
1110 __be32_to_cpu(pd->refnum));
1111 printf("%8lluK ",
1112 (unsigned long long)__be64_to_cpu(pd->config_size)>>1);
1113 for (dl = sb->dlist; dl ; dl = dl->next) {
1114 if (dl->disk.refnum == pd->refnum) {
1115 char *dv = map_dev(dl->major, dl->minor, 0);
1116 if (dv) {
1117 printf("%-15s", dv);
1118 break;
1119 }
1120 }
1121 }
1122 if (!dl)
1123 printf("%15s","");
1124 printf(" %s%s%s%s%s",
1125 (type&2) ? "active":"",
1126 (type&4) ? "Global-Spare":"",
1127 (type&8) ? "spare" : "",
1128 (type&16)? ", foreign" : "",
1129 (type&32)? "pass-through" : "");
1130 if (state & DDF_Failed)
1131 /* This over-rides these three */
1132 state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition);
1133 printf("/%s%s%s%s%s%s%s",
1134 (state&1)? "Online": "Offline",
1135 (state&2)? ", Failed": "",
1136 (state&4)? ", Rebuilding": "",
1137 (state&8)? ", in-transition": "",
1138 (state&16)? ", SMART-errors": "",
1139 (state&32)? ", Unrecovered-Read-Errors": "",
1140 (state&64)? ", Missing" : "");
1141 printf("\n");
1142 }
1143 }
1144
1145 static void examine_super_ddf(struct supertype *st, char *homehost)
1146 {
1147 struct ddf_super *sb = st->sb;
1148
1149 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1150 printf(" Version : %.8s\n", sb->anchor.revision);
1151 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1152 printf("\n");
1153 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1154 printf("\n");
1155 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1156 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1157 ?"yes" : "no");
1158 examine_vds(sb);
1159 examine_pds(sb);
1160 }
1161
1162 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map);
1163
1164 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
1165
1166 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1167 {
1168 /* We just write a generic DDF ARRAY entry
1169 */
1170 struct mdinfo info;
1171 char nbuf[64];
1172 getinfo_super_ddf(st, &info, NULL);
1173 fname_from_uuid(st, &info, nbuf, ':');
1174
1175 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1176 }
1177
1178 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1179 {
1180 /* We just write a generic DDF ARRAY entry
1181 */
1182 struct ddf_super *ddf = st->sb;
1183 struct mdinfo info;
1184 unsigned int i;
1185 char nbuf[64];
1186 getinfo_super_ddf(st, &info, NULL);
1187 fname_from_uuid(st, &info, nbuf, ':');
1188
1189 for (i = 0; i < __be16_to_cpu(ddf->virt->max_vdes); i++) {
1190 struct virtual_entry *ve = &ddf->virt->entries[i];
1191 struct vcl vcl;
1192 char nbuf1[64];
1193 if (all_ff(ve->guid))
1194 continue;
1195 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1196 ddf->currentconf =&vcl;
1197 uuid_from_super_ddf(st, info.uuid);
1198 fname_from_uuid(st, &info, nbuf1, ':');
1199 printf("ARRAY container=%s member=%d UUID=%s\n",
1200 nbuf+5, i, nbuf1+5);
1201 }
1202 }
1203
1204 static void export_examine_super_ddf(struct supertype *st)
1205 {
1206 struct mdinfo info;
1207 char nbuf[64];
1208 getinfo_super_ddf(st, &info, NULL);
1209 fname_from_uuid(st, &info, nbuf, ':');
1210 printf("MD_METADATA=ddf\n");
1211 printf("MD_LEVEL=container\n");
1212 printf("MD_UUID=%s\n", nbuf+5);
1213 }
1214
1215 static void detail_super_ddf(struct supertype *st, char *homehost)
1216 {
1217 /* FIXME later
1218 * Could print DDF GUID
1219 * Need to find which array
1220 * If whole, briefly list all arrays
1221 * If one, give name
1222 */
1223 }
1224
1225 static void brief_detail_super_ddf(struct supertype *st)
1226 {
1227 /* FIXME I really need to know which array we are detailing.
1228 * Can that be stored in ddf_super??
1229 */
1230 // struct ddf_super *ddf = st->sb;
1231 struct mdinfo info;
1232 char nbuf[64];
1233 getinfo_super_ddf(st, &info, NULL);
1234 fname_from_uuid(st, &info, nbuf,':');
1235 printf(" UUID=%s", nbuf + 5);
1236 }
1237 #endif
1238
1239 static int match_home_ddf(struct supertype *st, char *homehost)
1240 {
1241 /* It matches 'this' host if the controller is a
1242 * Linux-MD controller with vendor_data matching
1243 * the hostname
1244 */
1245 struct ddf_super *ddf = st->sb;
1246 unsigned int len;
1247
1248 if (!homehost)
1249 return 0;
1250 len = strlen(homehost);
1251
1252 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1253 len < sizeof(ddf->controller.vendor_data) &&
1254 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1255 ddf->controller.vendor_data[len] == 0);
1256 }
1257
1258 #ifndef MDASSEMBLE
1259 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst)
1260 {
1261 struct vcl *v;
1262
1263 for (v = ddf->conflist; v; v = v->next)
1264 if (inst == v->vcnum)
1265 return &v->conf;
1266 return NULL;
1267 }
1268 #endif
1269
1270 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1271 {
1272 /* Find the entry in phys_disk which has the given refnum
1273 * and return it's index
1274 */
1275 unsigned int i;
1276 for (i = 0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1277 if (ddf->phys->entries[i].refnum == phys_refnum)
1278 return i;
1279 return -1;
1280 }
1281
1282 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1283 {
1284 /* The uuid returned here is used for:
1285 * uuid to put into bitmap file (Create, Grow)
1286 * uuid for backup header when saving critical section (Grow)
1287 * comparing uuids when re-adding a device into an array
1288 * In these cases the uuid required is that of the data-array,
1289 * not the device-set.
1290 * uuid to recognise same set when adding a missing device back
1291 * to an array. This is a uuid for the device-set.
1292 *
1293 * For each of these we can make do with a truncated
1294 * or hashed uuid rather than the original, as long as
1295 * everyone agrees.
1296 * In the case of SVD we assume the BVD is of interest,
1297 * though that might be the case if a bitmap were made for
1298 * a mirrored SVD - worry about that later.
1299 * So we need to find the VD configuration record for the
1300 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1301 * The first 16 bytes of the sha1 of these is used.
1302 */
1303 struct ddf_super *ddf = st->sb;
1304 struct vcl *vcl = ddf->currentconf;
1305 char *guid;
1306 char buf[20];
1307 struct sha1_ctx ctx;
1308
1309 if (vcl)
1310 guid = vcl->conf.guid;
1311 else
1312 guid = ddf->anchor.guid;
1313
1314 sha1_init_ctx(&ctx);
1315 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1316 sha1_finish_ctx(&ctx, buf);
1317 memcpy(uuid, buf, 4*4);
1318 }
1319
1320 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map);
1321
1322 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1323 {
1324 struct ddf_super *ddf = st->sb;
1325 int map_disks = info->array.raid_disks;
1326 __u32 *cptr;
1327
1328 if (ddf->currentconf) {
1329 getinfo_super_ddf_bvd(st, info, map);
1330 return;
1331 }
1332 memset(info, 0, sizeof(*info));
1333
1334 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1335 info->array.level = LEVEL_CONTAINER;
1336 info->array.layout = 0;
1337 info->array.md_minor = -1;
1338 cptr = (__u32 *)(ddf->anchor.guid + 16);
1339 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
1340
1341 info->array.utime = 0;
1342 info->array.chunk_size = 0;
1343 info->container_enough = 1;
1344
1345 info->disk.major = 0;
1346 info->disk.minor = 0;
1347 if (ddf->dlist) {
1348 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1349 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1350
1351 info->data_offset = __be64_to_cpu(ddf->phys->
1352 entries[info->disk.raid_disk].
1353 config_size);
1354 info->component_size = ddf->dlist->size - info->data_offset;
1355 } else {
1356 info->disk.number = -1;
1357 info->disk.raid_disk = -1;
1358 // info->disk.raid_disk = find refnum in the table and use index;
1359 }
1360 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
1361
1362 info->recovery_start = MaxSector;
1363 info->reshape_active = 0;
1364 info->recovery_blocked = 0;
1365 info->name[0] = 0;
1366
1367 info->array.major_version = -1;
1368 info->array.minor_version = -2;
1369 strcpy(info->text_version, "ddf");
1370 info->safe_mode_delay = 0;
1371
1372 uuid_from_super_ddf(st, info->uuid);
1373
1374 if (map) {
1375 int i;
1376 for (i = 0 ; i < map_disks; i++) {
1377 if (i < info->array.raid_disks &&
1378 (__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Online) &&
1379 !(__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Failed))
1380 map[i] = 1;
1381 else
1382 map[i] = 0;
1383 }
1384 }
1385 }
1386
1387 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1388
1389 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
1390 {
1391 struct ddf_super *ddf = st->sb;
1392 struct vcl *vc = ddf->currentconf;
1393 int cd = ddf->currentdev;
1394 int j;
1395 struct dl *dl;
1396 int map_disks = info->array.raid_disks;
1397 __u32 *cptr;
1398
1399 memset(info, 0, sizeof(*info));
1400 /* FIXME this returns BVD info - what if we want SVD ?? */
1401
1402 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1403 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1404 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1405 info->array.raid_disks);
1406 info->array.md_minor = -1;
1407 cptr = (__u32 *)(vc->conf.guid + 16);
1408 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
1409 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1410 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1411 info->custom_array_size = 0;
1412
1413 if (cd >= 0 && (unsigned)cd < ddf->mppe) {
1414 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1415 if (vc->block_sizes)
1416 info->component_size = vc->block_sizes[cd];
1417 else
1418 info->component_size = __be64_to_cpu(vc->conf.blocks);
1419 }
1420
1421 for (dl = ddf->dlist; dl ; dl = dl->next)
1422 if (dl->raiddisk == ddf->currentdev)
1423 break;
1424
1425 info->disk.major = 0;
1426 info->disk.minor = 0;
1427 info->disk.state = 0;
1428 if (dl) {
1429 info->disk.major = dl->major;
1430 info->disk.minor = dl->minor;
1431 info->disk.raid_disk = dl->raiddisk;
1432 info->disk.number = dl->pdnum;
1433 info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
1434 }
1435
1436 info->container_member = ddf->currentconf->vcnum;
1437
1438 info->recovery_start = MaxSector;
1439 info->resync_start = 0;
1440 info->reshape_active = 0;
1441 info->recovery_blocked = 0;
1442 if (!(ddf->virt->entries[info->container_member].state
1443 & DDF_state_inconsistent) &&
1444 (ddf->virt->entries[info->container_member].init_state
1445 & DDF_initstate_mask)
1446 == DDF_init_full)
1447 info->resync_start = MaxSector;
1448
1449 uuid_from_super_ddf(st, info->uuid);
1450
1451 info->array.major_version = -1;
1452 info->array.minor_version = -2;
1453 sprintf(info->text_version, "/%s/%d",
1454 devnum2devname(st->container_dev),
1455 info->container_member);
1456 info->safe_mode_delay = 200;
1457
1458 memcpy(info->name, ddf->virt->entries[info->container_member].name, 16);
1459 info->name[16]=0;
1460 for(j=0; j<16; j++)
1461 if (info->name[j] == ' ')
1462 info->name[j] = 0;
1463
1464 if (map)
1465 for (j = 0; j < map_disks; j++) {
1466 map[j] = 0;
1467 if (j < info->array.raid_disks) {
1468 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
1469 if (i >= 0 &&
1470 (__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Online) &&
1471 !(__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Failed))
1472 map[i] = 1;
1473 }
1474 }
1475 }
1476
1477 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1478 char *update,
1479 char *devname, int verbose,
1480 int uuid_set, char *homehost)
1481 {
1482 /* For 'assemble' and 'force' we need to return non-zero if any
1483 * change was made. For others, the return value is ignored.
1484 * Update options are:
1485 * force-one : This device looks a bit old but needs to be included,
1486 * update age info appropriately.
1487 * assemble: clear any 'faulty' flag to allow this device to
1488 * be assembled.
1489 * force-array: Array is degraded but being forced, mark it clean
1490 * if that will be needed to assemble it.
1491 *
1492 * newdev: not used ????
1493 * grow: Array has gained a new device - this is currently for
1494 * linear only
1495 * resync: mark as dirty so a resync will happen.
1496 * uuid: Change the uuid of the array to match what is given
1497 * homehost: update the recorded homehost
1498 * name: update the name - preserving the homehost
1499 * _reshape_progress: record new reshape_progress position.
1500 *
1501 * Following are not relevant for this version:
1502 * sparc2.2 : update from old dodgey metadata
1503 * super-minor: change the preferred_minor number
1504 * summaries: update redundant counters.
1505 */
1506 int rv = 0;
1507 // struct ddf_super *ddf = st->sb;
1508 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1509 // struct virtual_entry *ve = find_ve(ddf);
1510
1511 /* we don't need to handle "force-*" or "assemble" as
1512 * there is no need to 'trick' the kernel. We the metadata is
1513 * first updated to activate the array, all the implied modifications
1514 * will just happen.
1515 */
1516
1517 if (strcmp(update, "grow") == 0) {
1518 /* FIXME */
1519 } else if (strcmp(update, "resync") == 0) {
1520 // info->resync_checkpoint = 0;
1521 } else if (strcmp(update, "homehost") == 0) {
1522 /* homehost is stored in controller->vendor_data,
1523 * or it is when we are the vendor
1524 */
1525 // if (info->vendor_is_local)
1526 // strcpy(ddf->controller.vendor_data, homehost);
1527 rv = -1;
1528 } else if (strcmp(update, "name") == 0) {
1529 /* name is stored in virtual_entry->name */
1530 // memset(ve->name, ' ', 16);
1531 // strncpy(ve->name, info->name, 16);
1532 rv = -1;
1533 } else if (strcmp(update, "_reshape_progress") == 0) {
1534 /* We don't support reshape yet */
1535 } else if (strcmp(update, "assemble") == 0 ) {
1536 /* Do nothing, just succeed */
1537 rv = 0;
1538 } else
1539 rv = -1;
1540
1541 // update_all_csum(ddf);
1542
1543 return rv;
1544 }
1545
1546 static void make_header_guid(char *guid)
1547 {
1548 __u32 stamp;
1549 /* Create a DDF Header of Virtual Disk GUID */
1550
1551 /* 24 bytes of fiction required.
1552 * first 8 are a 'vendor-id' - "Linux-MD"
1553 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1554 * Remaining 8 random number plus timestamp
1555 */
1556 memcpy(guid, T10, sizeof(T10));
1557 stamp = __cpu_to_be32(0xdeadbeef);
1558 memcpy(guid+8, &stamp, 4);
1559 stamp = __cpu_to_be32(0);
1560 memcpy(guid+12, &stamp, 4);
1561 stamp = __cpu_to_be32(time(0) - DECADE);
1562 memcpy(guid+16, &stamp, 4);
1563 stamp = random32();
1564 memcpy(guid+20, &stamp, 4);
1565 }
1566
1567 static int init_super_ddf_bvd(struct supertype *st,
1568 mdu_array_info_t *info,
1569 unsigned long long size,
1570 char *name, char *homehost,
1571 int *uuid, unsigned long long data_offset);
1572
1573 static int init_super_ddf(struct supertype *st,
1574 mdu_array_info_t *info,
1575 unsigned long long size, char *name, char *homehost,
1576 int *uuid, unsigned long long data_offset)
1577 {
1578 /* This is primarily called by Create when creating a new array.
1579 * We will then get add_to_super called for each component, and then
1580 * write_init_super called to write it out to each device.
1581 * For DDF, Create can create on fresh devices or on a pre-existing
1582 * array.
1583 * To create on a pre-existing array a different method will be called.
1584 * This one is just for fresh drives.
1585 *
1586 * We need to create the entire 'ddf' structure which includes:
1587 * DDF headers - these are easy.
1588 * Controller data - a Sector describing this controller .. not that
1589 * this is a controller exactly.
1590 * Physical Disk Record - one entry per device, so
1591 * leave plenty of space.
1592 * Virtual Disk Records - again, just leave plenty of space.
1593 * This just lists VDs, doesn't give details
1594 * Config records - describes the VDs that use this disk
1595 * DiskData - describes 'this' device.
1596 * BadBlockManagement - empty
1597 * Diag Space - empty
1598 * Vendor Logs - Could we put bitmaps here?
1599 *
1600 */
1601 struct ddf_super *ddf;
1602 char hostname[17];
1603 int hostlen;
1604 int max_phys_disks, max_virt_disks;
1605 unsigned long long sector;
1606 int clen;
1607 int i;
1608 int pdsize, vdsize;
1609 struct phys_disk *pd;
1610 struct virtual_disk *vd;
1611
1612 if (data_offset != INVALID_SECTORS) {
1613 fprintf(stderr, Name ": data-offset not supported by DDF\n");
1614 return 0;
1615 }
1616
1617 if (st->sb)
1618 return init_super_ddf_bvd(st, info, size, name, homehost, uuid,
1619 data_offset);
1620
1621 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
1622 pr_err("%s could not allocate superblock\n", __func__);
1623 return 0;
1624 }
1625 memset(ddf, 0, sizeof(*ddf));
1626 ddf->dlist = NULL; /* no physical disks yet */
1627 ddf->conflist = NULL; /* No virtual disks yet */
1628 st->sb = ddf;
1629
1630 if (info == NULL) {
1631 /* zeroing superblock */
1632 return 0;
1633 }
1634
1635 /* At least 32MB *must* be reserved for the ddf. So let's just
1636 * start 32MB from the end, and put the primary header there.
1637 * Don't do secondary for now.
1638 * We don't know exactly where that will be yet as it could be
1639 * different on each device. To just set up the lengths.
1640 *
1641 */
1642
1643 ddf->anchor.magic = DDF_HEADER_MAGIC;
1644 make_header_guid(ddf->anchor.guid);
1645
1646 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1647 ddf->anchor.seq = __cpu_to_be32(1);
1648 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1649 ddf->anchor.openflag = 0xFF;
1650 ddf->anchor.foreignflag = 0;
1651 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1652 ddf->anchor.pad0 = 0xff;
1653 memset(ddf->anchor.pad1, 0xff, 12);
1654 memset(ddf->anchor.header_ext, 0xff, 32);
1655 ddf->anchor.primary_lba = ~(__u64)0;
1656 ddf->anchor.secondary_lba = ~(__u64)0;
1657 ddf->anchor.type = DDF_HEADER_ANCHOR;
1658 memset(ddf->anchor.pad2, 0xff, 3);
1659 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1660 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1661 of 32M reserved.. */
1662 max_phys_disks = 1023; /* Should be enough */
1663 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1664 max_virt_disks = 255;
1665 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1666 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1667 ddf->max_part = 64;
1668 ddf->mppe = 256;
1669 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1670 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1671 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1672 memset(ddf->anchor.pad3, 0xff, 54);
1673 /* controller sections is one sector long immediately
1674 * after the ddf header */
1675 sector = 1;
1676 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1677 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1678 sector += 1;
1679
1680 /* phys is 8 sectors after that */
1681 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1682 sizeof(struct phys_disk_entry)*max_phys_disks,
1683 512);
1684 switch(pdsize/512) {
1685 case 2: case 8: case 32: case 128: case 512: break;
1686 default: abort();
1687 }
1688 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1689 ddf->anchor.phys_section_length =
1690 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1691 sector += pdsize/512;
1692
1693 /* virt is another 32 sectors */
1694 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1695 sizeof(struct virtual_entry) * max_virt_disks,
1696 512);
1697 switch(vdsize/512) {
1698 case 2: case 8: case 32: case 128: case 512: break;
1699 default: abort();
1700 }
1701 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1702 ddf->anchor.virt_section_length =
1703 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1704 sector += vdsize/512;
1705
1706 clen = ddf->conf_rec_len * (ddf->max_part+1);
1707 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1708 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1709 sector += clen;
1710
1711 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1712 ddf->anchor.data_section_length = __cpu_to_be32(1);
1713 sector += 1;
1714
1715 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1716 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1717 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1718 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1719 ddf->anchor.vendor_length = __cpu_to_be32(0);
1720 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1721
1722 memset(ddf->anchor.pad4, 0xff, 256);
1723
1724 memcpy(&ddf->primary, &ddf->anchor, 512);
1725 memcpy(&ddf->secondary, &ddf->anchor, 512);
1726
1727 ddf->primary.openflag = 1; /* I guess.. */
1728 ddf->primary.type = DDF_HEADER_PRIMARY;
1729
1730 ddf->secondary.openflag = 1; /* I guess.. */
1731 ddf->secondary.type = DDF_HEADER_SECONDARY;
1732
1733 ddf->active = &ddf->primary;
1734
1735 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1736
1737 /* 24 more bytes of fiction required.
1738 * first 8 are a 'vendor-id' - "Linux-MD"
1739 * Remaining 16 are serial number.... maybe a hostname would do?
1740 */
1741 memcpy(ddf->controller.guid, T10, sizeof(T10));
1742 gethostname(hostname, sizeof(hostname));
1743 hostname[sizeof(hostname) - 1] = 0;
1744 hostlen = strlen(hostname);
1745 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1746 for (i = strlen(T10) ; i+hostlen < 24; i++)
1747 ddf->controller.guid[i] = ' ';
1748
1749 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1750 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1751 ddf->controller.type.sub_vendor_id = 0;
1752 ddf->controller.type.sub_device_id = 0;
1753 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1754 memset(ddf->controller.pad, 0xff, 8);
1755 memset(ddf->controller.vendor_data, 0xff, 448);
1756 if (homehost && strlen(homehost) < 440)
1757 strcpy((char*)ddf->controller.vendor_data, homehost);
1758
1759 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
1760 pr_err("%s could not allocate pd\n", __func__);
1761 return 0;
1762 }
1763 ddf->phys = pd;
1764 ddf->pdsize = pdsize;
1765
1766 memset(pd, 0xff, pdsize);
1767 memset(pd, 0, sizeof(*pd));
1768 pd->magic = DDF_PHYS_RECORDS_MAGIC;
1769 pd->used_pdes = __cpu_to_be16(0);
1770 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1771 memset(pd->pad, 0xff, 52);
1772
1773 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
1774 pr_err("%s could not allocate vd\n", __func__);
1775 return 0;
1776 }
1777 ddf->virt = vd;
1778 ddf->vdsize = vdsize;
1779 memset(vd, 0, vdsize);
1780 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1781 vd->populated_vdes = __cpu_to_be16(0);
1782 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1783 memset(vd->pad, 0xff, 52);
1784
1785 for (i=0; i<max_virt_disks; i++)
1786 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1787
1788 st->sb = ddf;
1789 ddf->updates_pending = 1;
1790 return 1;
1791 }
1792
1793 static int chunk_to_shift(int chunksize)
1794 {
1795 return ffs(chunksize/512)-1;
1796 }
1797
1798 static int level_to_prl(int level)
1799 {
1800 switch (level) {
1801 case LEVEL_LINEAR: return DDF_CONCAT;
1802 case 0: return DDF_RAID0;
1803 case 1: return DDF_RAID1;
1804 case 4: return DDF_RAID4;
1805 case 5: return DDF_RAID5;
1806 case 6: return DDF_RAID6;
1807 default: return -1;
1808 }
1809 }
1810
1811 static int layout_to_rlq(int level, int layout, int raiddisks)
1812 {
1813 switch(level) {
1814 case 0:
1815 return DDF_RAID0_SIMPLE;
1816 case 1:
1817 switch(raiddisks) {
1818 case 2: return DDF_RAID1_SIMPLE;
1819 case 3: return DDF_RAID1_MULTI;
1820 default: return -1;
1821 }
1822 case 4:
1823 switch(layout) {
1824 case 0: return DDF_RAID4_N;
1825 }
1826 break;
1827 case 5:
1828 switch(layout) {
1829 case ALGORITHM_LEFT_ASYMMETRIC:
1830 return DDF_RAID5_N_RESTART;
1831 case ALGORITHM_RIGHT_ASYMMETRIC:
1832 return DDF_RAID5_0_RESTART;
1833 case ALGORITHM_LEFT_SYMMETRIC:
1834 return DDF_RAID5_N_CONTINUE;
1835 case ALGORITHM_RIGHT_SYMMETRIC:
1836 return -1; /* not mentioned in standard */
1837 }
1838 case 6:
1839 switch(layout) {
1840 case ALGORITHM_ROTATING_N_RESTART:
1841 return DDF_RAID5_N_RESTART;
1842 case ALGORITHM_ROTATING_ZERO_RESTART:
1843 return DDF_RAID6_0_RESTART;
1844 case ALGORITHM_ROTATING_N_CONTINUE:
1845 return DDF_RAID5_N_CONTINUE;
1846 }
1847 }
1848 return -1;
1849 }
1850
1851 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1852 {
1853 switch(prl) {
1854 case DDF_RAID0:
1855 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1856 case DDF_RAID1:
1857 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1858 on raiddisks*/
1859 case DDF_RAID4:
1860 switch(rlq) {
1861 case DDF_RAID4_N:
1862 return 0;
1863 default:
1864 /* not supported */
1865 return -1; /* FIXME this isn't checked */
1866 }
1867 case DDF_RAID5:
1868 switch(rlq) {
1869 case DDF_RAID5_N_RESTART:
1870 return ALGORITHM_LEFT_ASYMMETRIC;
1871 case DDF_RAID5_0_RESTART:
1872 return ALGORITHM_RIGHT_ASYMMETRIC;
1873 case DDF_RAID5_N_CONTINUE:
1874 return ALGORITHM_LEFT_SYMMETRIC;
1875 default:
1876 return -1;
1877 }
1878 case DDF_RAID6:
1879 switch(rlq) {
1880 case DDF_RAID5_N_RESTART:
1881 return ALGORITHM_ROTATING_N_RESTART;
1882 case DDF_RAID6_0_RESTART:
1883 return ALGORITHM_ROTATING_ZERO_RESTART;
1884 case DDF_RAID5_N_CONTINUE:
1885 return ALGORITHM_ROTATING_N_CONTINUE;
1886 default:
1887 return -1;
1888 }
1889 }
1890 return -1;
1891 }
1892
1893 #ifndef MDASSEMBLE
1894 struct extent {
1895 unsigned long long start, size;
1896 };
1897 static int cmp_extent(const void *av, const void *bv)
1898 {
1899 const struct extent *a = av;
1900 const struct extent *b = bv;
1901 if (a->start < b->start)
1902 return -1;
1903 if (a->start > b->start)
1904 return 1;
1905 return 0;
1906 }
1907
1908 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1909 {
1910 /* find a list of used extents on the give physical device
1911 * (dnum) of the given ddf.
1912 * Return a malloced array of 'struct extent'
1913
1914 * FIXME ignore DDF_Legacy devices?
1915
1916 */
1917 struct extent *rv;
1918 int n = 0;
1919 unsigned int i, j;
1920
1921 rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2));
1922
1923 for (i = 0; i < ddf->max_part; i++) {
1924 struct vcl *v = dl->vlist[i];
1925 if (v == NULL)
1926 continue;
1927 for (j = 0; j < v->conf.prim_elmnt_count; j++)
1928 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1929 /* This device plays role 'j' in 'v'. */
1930 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1931 rv[n].size = __be64_to_cpu(v->conf.blocks);
1932 n++;
1933 break;
1934 }
1935 }
1936 qsort(rv, n, sizeof(*rv), cmp_extent);
1937
1938 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1939 rv[n].size = 0;
1940 return rv;
1941 }
1942 #endif
1943
1944 static int init_super_ddf_bvd(struct supertype *st,
1945 mdu_array_info_t *info,
1946 unsigned long long size,
1947 char *name, char *homehost,
1948 int *uuid, unsigned long long data_offset)
1949 {
1950 /* We are creating a BVD inside a pre-existing container.
1951 * so st->sb is already set.
1952 * We need to create a new vd_config and a new virtual_entry
1953 */
1954 struct ddf_super *ddf = st->sb;
1955 unsigned int venum;
1956 struct virtual_entry *ve;
1957 struct vcl *vcl;
1958 struct vd_config *vc;
1959
1960 if (__be16_to_cpu(ddf->virt->populated_vdes)
1961 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1962 pr_err("This ddf already has the "
1963 "maximum of %d virtual devices\n",
1964 __be16_to_cpu(ddf->virt->max_vdes));
1965 return 0;
1966 }
1967
1968 if (name)
1969 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1970 if (!all_ff(ddf->virt->entries[venum].guid)) {
1971 char *n = ddf->virt->entries[venum].name;
1972
1973 if (strncmp(name, n, 16) == 0) {
1974 pr_err("This ddf already"
1975 " has an array called %s\n",
1976 name);
1977 return 0;
1978 }
1979 }
1980
1981 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1982 if (all_ff(ddf->virt->entries[venum].guid))
1983 break;
1984 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1985 pr_err("Cannot find spare slot for "
1986 "virtual disk - DDF is corrupt\n");
1987 return 0;
1988 }
1989 ve = &ddf->virt->entries[venum];
1990
1991 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1992 * timestamp, random number
1993 */
1994 make_header_guid(ve->guid);
1995 ve->unit = __cpu_to_be16(info->md_minor);
1996 ve->pad0 = 0xFFFF;
1997 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1998 ve->type = 0;
1999 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
2000 if (info->state & 1) /* clean */
2001 ve->init_state = DDF_init_full;
2002 else
2003 ve->init_state = DDF_init_not;
2004
2005 memset(ve->pad1, 0xff, 14);
2006 memset(ve->name, ' ', 16);
2007 if (name)
2008 strncpy(ve->name, name, 16);
2009 ddf->virt->populated_vdes =
2010 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
2011
2012 /* Now create a new vd_config */
2013 if (posix_memalign((void**)&vcl, 512,
2014 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
2015 pr_err("%s could not allocate vd_config\n", __func__);
2016 return 0;
2017 }
2018 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
2019 vcl->vcnum = venum;
2020 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2021
2022 vc = &vcl->conf;
2023
2024 vc->magic = DDF_VD_CONF_MAGIC;
2025 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2026 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
2027 vc->seqnum = __cpu_to_be32(1);
2028 memset(vc->pad0, 0xff, 24);
2029 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
2030 vc->chunk_shift = chunk_to_shift(info->chunk_size);
2031 vc->prl = level_to_prl(info->level);
2032 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
2033 vc->sec_elmnt_count = 1;
2034 vc->sec_elmnt_seq = 0;
2035 vc->srl = 0;
2036 vc->blocks = __cpu_to_be64(info->size * 2);
2037 vc->array_blocks = __cpu_to_be64(
2038 calc_array_size(info->level, info->raid_disks, info->layout,
2039 info->chunk_size, info->size*2));
2040 memset(vc->pad1, 0xff, 8);
2041 vc->spare_refs[0] = 0xffffffff;
2042 vc->spare_refs[1] = 0xffffffff;
2043 vc->spare_refs[2] = 0xffffffff;
2044 vc->spare_refs[3] = 0xffffffff;
2045 vc->spare_refs[4] = 0xffffffff;
2046 vc->spare_refs[5] = 0xffffffff;
2047 vc->spare_refs[6] = 0xffffffff;
2048 vc->spare_refs[7] = 0xffffffff;
2049 memset(vc->cache_pol, 0, 8);
2050 vc->bg_rate = 0x80;
2051 memset(vc->pad2, 0xff, 3);
2052 memset(vc->pad3, 0xff, 52);
2053 memset(vc->pad4, 0xff, 192);
2054 memset(vc->v0, 0xff, 32);
2055 memset(vc->v1, 0xff, 32);
2056 memset(vc->v2, 0xff, 16);
2057 memset(vc->v3, 0xff, 16);
2058 memset(vc->vendor, 0xff, 32);
2059
2060 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2061 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2062
2063 vcl->next = ddf->conflist;
2064 ddf->conflist = vcl;
2065 ddf->currentconf = vcl;
2066 ddf->updates_pending = 1;
2067 return 1;
2068 }
2069
2070 #ifndef MDASSEMBLE
2071 static void add_to_super_ddf_bvd(struct supertype *st,
2072 mdu_disk_info_t *dk, int fd, char *devname)
2073 {
2074 /* fd and devname identify a device with-in the ddf container (st).
2075 * dk identifies a location in the new BVD.
2076 * We need to find suitable free space in that device and update
2077 * the phys_refnum and lba_offset for the newly created vd_config.
2078 * We might also want to update the type in the phys_disk
2079 * section.
2080 *
2081 * Alternately: fd == -1 and we have already chosen which device to
2082 * use and recorded in dlist->raid_disk;
2083 */
2084 struct dl *dl;
2085 struct ddf_super *ddf = st->sb;
2086 struct vd_config *vc;
2087 __u64 *lba_offset;
2088 unsigned int working;
2089 unsigned int i;
2090 unsigned long long blocks, pos, esize;
2091 struct extent *ex;
2092
2093 if (fd == -1) {
2094 for (dl = ddf->dlist; dl ; dl = dl->next)
2095 if (dl->raiddisk == dk->raid_disk)
2096 break;
2097 } else {
2098 for (dl = ddf->dlist; dl ; dl = dl->next)
2099 if (dl->major == dk->major &&
2100 dl->minor == dk->minor)
2101 break;
2102 }
2103 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
2104 return;
2105
2106 vc = &ddf->currentconf->conf;
2107 lba_offset = ddf->currentconf->lba_offset;
2108
2109 ex = get_extents(ddf, dl);
2110 if (!ex)
2111 return;
2112
2113 i = 0; pos = 0;
2114 blocks = __be64_to_cpu(vc->blocks);
2115 if (ddf->currentconf->block_sizes)
2116 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2117
2118 do {
2119 esize = ex[i].start - pos;
2120 if (esize >= blocks)
2121 break;
2122 pos = ex[i].start + ex[i].size;
2123 i++;
2124 } while (ex[i-1].size);
2125
2126 free(ex);
2127 if (esize < blocks)
2128 return;
2129
2130 ddf->currentdev = dk->raid_disk;
2131 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
2132 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
2133
2134 for (i = 0; i < ddf->max_part ; i++)
2135 if (dl->vlist[i] == NULL)
2136 break;
2137 if (i == ddf->max_part)
2138 return;
2139 dl->vlist[i] = ddf->currentconf;
2140
2141 if (fd >= 0)
2142 dl->fd = fd;
2143 if (devname)
2144 dl->devname = devname;
2145
2146 /* Check how many working raid_disks, and if we can mark
2147 * array as optimal yet
2148 */
2149 working = 0;
2150
2151 for (i = 0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
2152 if (vc->phys_refnum[i] != 0xffffffff)
2153 working++;
2154
2155 /* Find which virtual_entry */
2156 i = ddf->currentconf->vcnum;
2157 if (working == __be16_to_cpu(vc->prim_elmnt_count))
2158 ddf->virt->entries[i].state =
2159 (ddf->virt->entries[i].state & ~DDF_state_mask)
2160 | DDF_state_optimal;
2161
2162 if (vc->prl == DDF_RAID6 &&
2163 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
2164 ddf->virt->entries[i].state =
2165 (ddf->virt->entries[i].state & ~DDF_state_mask)
2166 | DDF_state_part_optimal;
2167
2168 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
2169 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
2170 ddf->updates_pending = 1;
2171 }
2172
2173 /* add a device to a container, either while creating it or while
2174 * expanding a pre-existing container
2175 */
2176 static int add_to_super_ddf(struct supertype *st,
2177 mdu_disk_info_t *dk, int fd, char *devname,
2178 unsigned long long data_offset)
2179 {
2180 struct ddf_super *ddf = st->sb;
2181 struct dl *dd;
2182 time_t now;
2183 struct tm *tm;
2184 unsigned long long size;
2185 struct phys_disk_entry *pde;
2186 unsigned int n, i;
2187 struct stat stb;
2188 __u32 *tptr;
2189
2190 if (ddf->currentconf) {
2191 add_to_super_ddf_bvd(st, dk, fd, devname);
2192 return 0;
2193 }
2194
2195 /* This is device numbered dk->number. We need to create
2196 * a phys_disk entry and a more detailed disk_data entry.
2197 */
2198 fstat(fd, &stb);
2199 if (posix_memalign((void**)&dd, 512,
2200 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2201 pr_err("%s could allocate buffer for new disk, aborting\n",
2202 __func__);
2203 return 1;
2204 }
2205 dd->major = major(stb.st_rdev);
2206 dd->minor = minor(stb.st_rdev);
2207 dd->devname = devname;
2208 dd->fd = fd;
2209 dd->spare = NULL;
2210
2211 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2212 now = time(0);
2213 tm = localtime(&now);
2214 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2215 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2216 tptr = (__u32 *)(dd->disk.guid + 16);
2217 *tptr++ = random32();
2218 *tptr = random32();
2219
2220 do {
2221 /* Cannot be bothered finding a CRC of some irrelevant details*/
2222 dd->disk.refnum = random32();
2223 for (i = __be16_to_cpu(ddf->active->max_pd_entries);
2224 i > 0; i--)
2225 if (ddf->phys->entries[i-1].refnum == dd->disk.refnum)
2226 break;
2227 } while (i > 0);
2228
2229 dd->disk.forced_ref = 1;
2230 dd->disk.forced_guid = 1;
2231 memset(dd->disk.vendor, ' ', 32);
2232 memcpy(dd->disk.vendor, "Linux", 5);
2233 memset(dd->disk.pad, 0xff, 442);
2234 for (i = 0; i < ddf->max_part ; i++)
2235 dd->vlist[i] = NULL;
2236
2237 n = __be16_to_cpu(ddf->phys->used_pdes);
2238 pde = &ddf->phys->entries[n];
2239 dd->pdnum = n;
2240
2241 if (st->update_tail) {
2242 int len = (sizeof(struct phys_disk) +
2243 sizeof(struct phys_disk_entry));
2244 struct phys_disk *pd;
2245
2246 pd = xmalloc(len);
2247 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2248 pd->used_pdes = __cpu_to_be16(n);
2249 pde = &pd->entries[0];
2250 dd->mdupdate = pd;
2251 } else {
2252 n++;
2253 ddf->phys->used_pdes = __cpu_to_be16(n);
2254 }
2255
2256 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2257 pde->refnum = dd->disk.refnum;
2258 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2259 pde->state = __cpu_to_be16(DDF_Online);
2260 get_dev_size(fd, NULL, &size);
2261 /* We are required to reserve 32Meg, and record the size in sectors */
2262 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2263 sprintf(pde->path, "%17.17s","Information: nil") ;
2264 memset(pde->pad, 0xff, 6);
2265
2266 dd->size = size >> 9;
2267 if (st->update_tail) {
2268 dd->next = ddf->add_list;
2269 ddf->add_list = dd;
2270 } else {
2271 dd->next = ddf->dlist;
2272 ddf->dlist = dd;
2273 ddf->updates_pending = 1;
2274 }
2275
2276 return 0;
2277 }
2278
2279 static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk)
2280 {
2281 struct ddf_super *ddf = st->sb;
2282 struct dl *dl;
2283
2284 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2285 * disappeared from the container.
2286 * We need to arrange that it disappears from the metadata and
2287 * internal data structures too.
2288 * Most of the work is done by ddf_process_update which edits
2289 * the metadata and closes the file handle and attaches the memory
2290 * where free_updates will free it.
2291 */
2292 for (dl = ddf->dlist; dl ; dl = dl->next)
2293 if (dl->major == dk->major &&
2294 dl->minor == dk->minor)
2295 break;
2296 if (!dl)
2297 return -1;
2298
2299 if (st->update_tail) {
2300 int len = (sizeof(struct phys_disk) +
2301 sizeof(struct phys_disk_entry));
2302 struct phys_disk *pd;
2303
2304 pd = xmalloc(len);
2305 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2306 pd->used_pdes = __cpu_to_be16(dl->pdnum);
2307 pd->entries[0].state = __cpu_to_be16(DDF_Missing);
2308 append_metadata_update(st, pd, len);
2309 }
2310 return 0;
2311 }
2312
2313 /*
2314 * This is the write_init_super method for a ddf container. It is
2315 * called when creating a container or adding another device to a
2316 * container.
2317 */
2318 #define NULL_CONF_SZ 4096
2319
2320 static int __write_init_super_ddf(struct supertype *st)
2321 {
2322
2323 struct ddf_super *ddf = st->sb;
2324 int i;
2325 struct dl *d;
2326 int n_config;
2327 int conf_size;
2328 int attempts = 0;
2329 int successes = 0;
2330 unsigned long long size, sector;
2331 char *null_aligned;
2332
2333 if (posix_memalign((void**)&null_aligned, 4096, NULL_CONF_SZ) != 0) {
2334 return -ENOMEM;
2335 }
2336 memset(null_aligned, 0xff, NULL_CONF_SZ);
2337
2338 /* try to write updated metadata,
2339 * if we catch a failure move on to the next disk
2340 */
2341 for (d = ddf->dlist; d; d=d->next) {
2342 int fd = d->fd;
2343
2344 if (fd < 0)
2345 continue;
2346
2347 attempts++;
2348 /* We need to fill in the primary, (secondary) and workspace
2349 * lba's in the headers, set their checksums,
2350 * Also checksum phys, virt....
2351 *
2352 * Then write everything out, finally the anchor is written.
2353 */
2354 get_dev_size(fd, NULL, &size);
2355 size /= 512;
2356 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2357 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2358 ddf->anchor.seq = __cpu_to_be32(1);
2359 memcpy(&ddf->primary, &ddf->anchor, 512);
2360 memcpy(&ddf->secondary, &ddf->anchor, 512);
2361
2362 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2363 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2364 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2365
2366 ddf->primary.openflag = 0;
2367 ddf->primary.type = DDF_HEADER_PRIMARY;
2368
2369 ddf->secondary.openflag = 0;
2370 ddf->secondary.type = DDF_HEADER_SECONDARY;
2371
2372 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2373 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2374
2375 sector = size - 16*1024*2;
2376 lseek64(fd, sector<<9, 0);
2377 if (write(fd, &ddf->primary, 512) < 0)
2378 continue;
2379
2380 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2381 if (write(fd, &ddf->controller, 512) < 0)
2382 continue;
2383
2384 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2385
2386 if (write(fd, ddf->phys, ddf->pdsize) < 0)
2387 continue;
2388
2389 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2390 if (write(fd, ddf->virt, ddf->vdsize) < 0)
2391 continue;
2392
2393 /* Now write lots of config records. */
2394 n_config = ddf->max_part;
2395 conf_size = ddf->conf_rec_len * 512;
2396 for (i = 0 ; i <= n_config ; i++) {
2397 struct vcl *c = d->vlist[i];
2398 if (i == n_config)
2399 c = (struct vcl*)d->spare;
2400
2401 if (c) {
2402 c->conf.crc = calc_crc(&c->conf, conf_size);
2403 if (write(fd, &c->conf, conf_size) < 0)
2404 break;
2405 } else {
2406 unsigned int togo = conf_size;
2407 while (togo > NULL_CONF_SZ) {
2408 if (write(fd, null_aligned, NULL_CONF_SZ) < 0)
2409 break;
2410 togo -= NULL_CONF_SZ;
2411 }
2412 if (write(fd, null_aligned, togo) < 0)
2413 break;
2414 }
2415 }
2416 if (i <= n_config)
2417 continue;
2418 d->disk.crc = calc_crc(&d->disk, 512);
2419 if (write(fd, &d->disk, 512) < 0)
2420 continue;
2421
2422 /* Maybe do the same for secondary */
2423
2424 lseek64(fd, (size-1)*512, SEEK_SET);
2425 if (write(fd, &ddf->anchor, 512) < 0)
2426 continue;
2427 successes++;
2428 }
2429 free(null_aligned);
2430
2431 return attempts != successes;
2432 }
2433
2434 static int write_init_super_ddf(struct supertype *st)
2435 {
2436 struct ddf_super *ddf = st->sb;
2437 struct vcl *currentconf = ddf->currentconf;
2438
2439 /* we are done with currentconf reset it to point st at the container */
2440 ddf->currentconf = NULL;
2441
2442 if (st->update_tail) {
2443 /* queue the virtual_disk and vd_config as metadata updates */
2444 struct virtual_disk *vd;
2445 struct vd_config *vc;
2446 int len;
2447
2448 if (!currentconf) {
2449 int len = (sizeof(struct phys_disk) +
2450 sizeof(struct phys_disk_entry));
2451
2452 /* adding a disk to the container. */
2453 if (!ddf->add_list)
2454 return 0;
2455
2456 append_metadata_update(st, ddf->add_list->mdupdate, len);
2457 ddf->add_list->mdupdate = NULL;
2458 return 0;
2459 }
2460
2461 /* Newly created VD */
2462
2463 /* First the virtual disk. We have a slightly fake header */
2464 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2465 vd = xmalloc(len);
2466 *vd = *ddf->virt;
2467 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
2468 vd->populated_vdes = __cpu_to_be16(currentconf->vcnum);
2469 append_metadata_update(st, vd, len);
2470
2471 /* Then the vd_config */
2472 len = ddf->conf_rec_len * 512;
2473 vc = xmalloc(len);
2474 memcpy(vc, &currentconf->conf, len);
2475 append_metadata_update(st, vc, len);
2476
2477 /* FIXME I need to close the fds! */
2478 return 0;
2479 } else {
2480 struct dl *d;
2481 for (d = ddf->dlist; d; d=d->next)
2482 while (Kill(d->devname, NULL, 0, -1, 1) == 0);
2483 return __write_init_super_ddf(st);
2484 }
2485 }
2486
2487 #endif
2488
2489 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize,
2490 unsigned long long data_offset)
2491 {
2492 /* We must reserve the last 32Meg */
2493 if (devsize <= 32*1024*2)
2494 return 0;
2495 return devsize - 32*1024*2;
2496 }
2497
2498 #ifndef MDASSEMBLE
2499
2500 static int reserve_space(struct supertype *st, int raiddisks,
2501 unsigned long long size, int chunk,
2502 unsigned long long *freesize)
2503 {
2504 /* Find 'raiddisks' spare extents at least 'size' big (but
2505 * only caring about multiples of 'chunk') and remember
2506 * them.
2507 * If the cannot be found, fail.
2508 */
2509 struct dl *dl;
2510 struct ddf_super *ddf = st->sb;
2511 int cnt = 0;
2512
2513 for (dl = ddf->dlist; dl ; dl=dl->next) {
2514 dl->raiddisk = -1;
2515 dl->esize = 0;
2516 }
2517 /* Now find largest extent on each device */
2518 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2519 struct extent *e = get_extents(ddf, dl);
2520 unsigned long long pos = 0;
2521 int i = 0;
2522 int found = 0;
2523 unsigned long long minsize = size;
2524
2525 if (size == 0)
2526 minsize = chunk;
2527
2528 if (!e)
2529 continue;
2530 do {
2531 unsigned long long esize;
2532 esize = e[i].start - pos;
2533 if (esize >= minsize) {
2534 found = 1;
2535 minsize = esize;
2536 }
2537 pos = e[i].start + e[i].size;
2538 i++;
2539 } while (e[i-1].size);
2540 if (found) {
2541 cnt++;
2542 dl->esize = minsize;
2543 }
2544 free(e);
2545 }
2546 if (cnt < raiddisks) {
2547 pr_err("not enough devices with space to create array.\n");
2548 return 0; /* No enough free spaces large enough */
2549 }
2550 if (size == 0) {
2551 /* choose the largest size of which there are at least 'raiddisk' */
2552 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2553 struct dl *dl2;
2554 if (dl->esize <= size)
2555 continue;
2556 /* This is bigger than 'size', see if there are enough */
2557 cnt = 0;
2558 for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next)
2559 if (dl2->esize >= dl->esize)
2560 cnt++;
2561 if (cnt >= raiddisks)
2562 size = dl->esize;
2563 }
2564 if (chunk) {
2565 size = size / chunk;
2566 size *= chunk;
2567 }
2568 *freesize = size;
2569 if (size < 32) {
2570 pr_err("not enough spare devices to create array.\n");
2571 return 0;
2572 }
2573 }
2574 /* We have a 'size' of which there are enough spaces.
2575 * We simply do a first-fit */
2576 cnt = 0;
2577 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
2578 if (dl->esize < size)
2579 continue;
2580
2581 dl->raiddisk = cnt;
2582 cnt++;
2583 }
2584 return 1;
2585 }
2586
2587 static int
2588 validate_geometry_ddf_container(struct supertype *st,
2589 int level, int layout, int raiddisks,
2590 int chunk, unsigned long long size,
2591 unsigned long long data_offset,
2592 char *dev, unsigned long long *freesize,
2593 int verbose);
2594
2595 static int validate_geometry_ddf_bvd(struct supertype *st,
2596 int level, int layout, int raiddisks,
2597 int *chunk, unsigned long long size,
2598 unsigned long long data_offset,
2599 char *dev, unsigned long long *freesize,
2600 int verbose);
2601
2602 static int validate_geometry_ddf(struct supertype *st,
2603 int level, int layout, int raiddisks,
2604 int *chunk, unsigned long long size,
2605 unsigned long long data_offset,
2606 char *dev, unsigned long long *freesize,
2607 int verbose)
2608 {
2609 int fd;
2610 struct mdinfo *sra;
2611 int cfd;
2612
2613 /* ddf potentially supports lots of things, but it depends on
2614 * what devices are offered (and maybe kernel version?)
2615 * If given unused devices, we will make a container.
2616 * If given devices in a container, we will make a BVD.
2617 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2618 */
2619
2620 if (chunk && *chunk == UnSet)
2621 *chunk = DEFAULT_CHUNK;
2622
2623 if (level == -1000000) level = LEVEL_CONTAINER;
2624 if (level == LEVEL_CONTAINER) {
2625 /* Must be a fresh device to add to a container */
2626 return validate_geometry_ddf_container(st, level, layout,
2627 raiddisks, chunk?*chunk:0,
2628 size, data_offset, dev,
2629 freesize,
2630 verbose);
2631 }
2632
2633 if (!dev) {
2634 /* Initial sanity check. Exclude illegal levels. */
2635 int i;
2636 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2637 if (ddf_level_num[i].num2 == level)
2638 break;
2639 if (ddf_level_num[i].num1 == MAXINT) {
2640 if (verbose)
2641 pr_err("DDF does not support level %d arrays\n",
2642 level);
2643 return 0;
2644 }
2645 /* Should check layout? etc */
2646
2647 if (st->sb && freesize) {
2648 /* --create was given a container to create in.
2649 * So we need to check that there are enough
2650 * free spaces and return the amount of space.
2651 * We may as well remember which drives were
2652 * chosen so that add_to_super/getinfo_super
2653 * can return them.
2654 */
2655 return reserve_space(st, raiddisks, size, chunk?*chunk:0, freesize);
2656 }
2657 return 1;
2658 }
2659
2660 if (st->sb) {
2661 /* A container has already been opened, so we are
2662 * creating in there. Maybe a BVD, maybe an SVD.
2663 * Should make a distinction one day.
2664 */
2665 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2666 chunk, size, data_offset, dev,
2667 freesize,
2668 verbose);
2669 }
2670 /* This is the first device for the array.
2671 * If it is a container, we read it in and do automagic allocations,
2672 * no other devices should be given.
2673 * Otherwise it must be a member device of a container, and we
2674 * do manual allocation.
2675 * Later we should check for a BVD and make an SVD.
2676 */
2677 fd = open(dev, O_RDONLY|O_EXCL, 0);
2678 if (fd >= 0) {
2679 sra = sysfs_read(fd, 0, GET_VERSION);
2680 close(fd);
2681 if (sra && sra->array.major_version == -1 &&
2682 strcmp(sra->text_version, "ddf") == 0) {
2683
2684 /* load super */
2685 /* find space for 'n' devices. */
2686 /* remember the devices */
2687 /* Somehow return the fact that we have enough */
2688 }
2689
2690 if (verbose)
2691 pr_err("ddf: Cannot create this array "
2692 "on device %s - a container is required.\n",
2693 dev);
2694 return 0;
2695 }
2696 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2697 if (verbose)
2698 pr_err("ddf: Cannot open %s: %s\n",
2699 dev, strerror(errno));
2700 return 0;
2701 }
2702 /* Well, it is in use by someone, maybe a 'ddf' container. */
2703 cfd = open_container(fd);
2704 if (cfd < 0) {
2705 close(fd);
2706 if (verbose)
2707 pr_err("ddf: Cannot use %s: %s\n",
2708 dev, strerror(EBUSY));
2709 return 0;
2710 }
2711 sra = sysfs_read(cfd, 0, GET_VERSION);
2712 close(fd);
2713 if (sra && sra->array.major_version == -1 &&
2714 strcmp(sra->text_version, "ddf") == 0) {
2715 /* This is a member of a ddf container. Load the container
2716 * and try to create a bvd
2717 */
2718 struct ddf_super *ddf;
2719 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) {
2720 st->sb = ddf;
2721 st->container_dev = fd2devnum(cfd);
2722 close(cfd);
2723 return validate_geometry_ddf_bvd(st, level, layout,
2724 raiddisks, chunk, size,
2725 data_offset,
2726 dev, freesize,
2727 verbose);
2728 }
2729 close(cfd);
2730 } else /* device may belong to a different container */
2731 return 0;
2732
2733 return 1;
2734 }
2735
2736 static int
2737 validate_geometry_ddf_container(struct supertype *st,
2738 int level, int layout, int raiddisks,
2739 int chunk, unsigned long long size,
2740 unsigned long long data_offset,
2741 char *dev, unsigned long long *freesize,
2742 int verbose)
2743 {
2744 int fd;
2745 unsigned long long ldsize;
2746
2747 if (level != LEVEL_CONTAINER)
2748 return 0;
2749 if (!dev)
2750 return 1;
2751
2752 fd = open(dev, O_RDONLY|O_EXCL, 0);
2753 if (fd < 0) {
2754 if (verbose)
2755 pr_err("ddf: Cannot open %s: %s\n",
2756 dev, strerror(errno));
2757 return 0;
2758 }
2759 if (!get_dev_size(fd, dev, &ldsize)) {
2760 close(fd);
2761 return 0;
2762 }
2763 close(fd);
2764
2765 *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS);
2766 if (*freesize == 0)
2767 return 0;
2768
2769 return 1;
2770 }
2771
2772 static int validate_geometry_ddf_bvd(struct supertype *st,
2773 int level, int layout, int raiddisks,
2774 int *chunk, unsigned long long size,
2775 unsigned long long data_offset,
2776 char *dev, unsigned long long *freesize,
2777 int verbose)
2778 {
2779 struct stat stb;
2780 struct ddf_super *ddf = st->sb;
2781 struct dl *dl;
2782 unsigned long long pos = 0;
2783 unsigned long long maxsize;
2784 struct extent *e;
2785 int i;
2786 /* ddf/bvd supports lots of things, but not containers */
2787 if (level == LEVEL_CONTAINER) {
2788 if (verbose)
2789 pr_err("DDF cannot create a container within an container\n");
2790 return 0;
2791 }
2792 /* We must have the container info already read in. */
2793 if (!ddf)
2794 return 0;
2795
2796 if (!dev) {
2797 /* General test: make sure there is space for
2798 * 'raiddisks' device extents of size 'size'.
2799 */
2800 unsigned long long minsize = size;
2801 int dcnt = 0;
2802 if (minsize == 0)
2803 minsize = 8;
2804 for (dl = ddf->dlist; dl ; dl = dl->next)
2805 {
2806 int found = 0;
2807 pos = 0;
2808
2809 i = 0;
2810 e = get_extents(ddf, dl);
2811 if (!e) continue;
2812 do {
2813 unsigned long long esize;
2814 esize = e[i].start - pos;
2815 if (esize >= minsize)
2816 found = 1;
2817 pos = e[i].start + e[i].size;
2818 i++;
2819 } while (e[i-1].size);
2820 if (found)
2821 dcnt++;
2822 free(e);
2823 }
2824 if (dcnt < raiddisks) {
2825 if (verbose)
2826 pr_err("ddf: Not enough devices with "
2827 "space for this array (%d < %d)\n",
2828 dcnt, raiddisks);
2829 return 0;
2830 }
2831 return 1;
2832 }
2833 /* This device must be a member of the set */
2834 if (stat(dev, &stb) < 0)
2835 return 0;
2836 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2837 return 0;
2838 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2839 if (dl->major == (int)major(stb.st_rdev) &&
2840 dl->minor == (int)minor(stb.st_rdev))
2841 break;
2842 }
2843 if (!dl) {
2844 if (verbose)
2845 pr_err("ddf: %s is not in the "
2846 "same DDF set\n",
2847 dev);
2848 return 0;
2849 }
2850 e = get_extents(ddf, dl);
2851 maxsize = 0;
2852 i = 0;
2853 if (e) do {
2854 unsigned long long esize;
2855 esize = e[i].start - pos;
2856 if (esize >= maxsize)
2857 maxsize = esize;
2858 pos = e[i].start + e[i].size;
2859 i++;
2860 } while (e[i-1].size);
2861 *freesize = maxsize;
2862 // FIXME here I am
2863
2864 return 1;
2865 }
2866
2867 static int load_super_ddf_all(struct supertype *st, int fd,
2868 void **sbp, char *devname)
2869 {
2870 struct mdinfo *sra;
2871 struct ddf_super *super;
2872 struct mdinfo *sd, *best = NULL;
2873 int bestseq = 0;
2874 int seq;
2875 char nm[20];
2876 int dfd;
2877
2878 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2879 if (!sra)
2880 return 1;
2881 if (sra->array.major_version != -1 ||
2882 sra->array.minor_version != -2 ||
2883 strcmp(sra->text_version, "ddf") != 0)
2884 return 1;
2885
2886 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2887 return 1;
2888 memset(super, 0, sizeof(*super));
2889
2890 /* first, try each device, and choose the best ddf */
2891 for (sd = sra->devs ; sd ; sd = sd->next) {
2892 int rv;
2893 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2894 dfd = dev_open(nm, O_RDONLY);
2895 if (dfd < 0)
2896 return 2;
2897 rv = load_ddf_headers(dfd, super, NULL);
2898 close(dfd);
2899 if (rv == 0) {
2900 seq = __be32_to_cpu(super->active->seq);
2901 if (super->active->openflag)
2902 seq--;
2903 if (!best || seq > bestseq) {
2904 bestseq = seq;
2905 best = sd;
2906 }
2907 }
2908 }
2909 if (!best)
2910 return 1;
2911 /* OK, load this ddf */
2912 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2913 dfd = dev_open(nm, O_RDONLY);
2914 if (dfd < 0)
2915 return 1;
2916 load_ddf_headers(dfd, super, NULL);
2917 load_ddf_global(dfd, super, NULL);
2918 close(dfd);
2919 /* Now we need the device-local bits */
2920 for (sd = sra->devs ; sd ; sd = sd->next) {
2921 int rv;
2922
2923 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2924 dfd = dev_open(nm, O_RDWR);
2925 if (dfd < 0)
2926 return 2;
2927 rv = load_ddf_headers(dfd, super, NULL);
2928 if (rv == 0)
2929 rv = load_ddf_local(dfd, super, NULL, 1);
2930 if (rv)
2931 return 1;
2932 }
2933
2934 *sbp = super;
2935 if (st->ss == NULL) {
2936 st->ss = &super_ddf;
2937 st->minor_version = 0;
2938 st->max_devs = 512;
2939 }
2940 st->container_dev = fd2devnum(fd);
2941 return 0;
2942 }
2943
2944 static int load_container_ddf(struct supertype *st, int fd,
2945 char *devname)
2946 {
2947 return load_super_ddf_all(st, fd, &st->sb, devname);
2948 }
2949
2950 #endif /* MDASSEMBLE */
2951
2952 static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray)
2953 {
2954 /* Given a container loaded by load_super_ddf_all,
2955 * extract information about all the arrays into
2956 * an mdinfo tree.
2957 *
2958 * For each vcl in conflist: create an mdinfo, fill it in,
2959 * then look for matching devices (phys_refnum) in dlist
2960 * and create appropriate device mdinfo.
2961 */
2962 struct ddf_super *ddf = st->sb;
2963 struct mdinfo *rest = NULL;
2964 struct vcl *vc;
2965
2966 for (vc = ddf->conflist ; vc ; vc=vc->next)
2967 {
2968 unsigned int i;
2969 unsigned int j;
2970 struct mdinfo *this;
2971 char *ep;
2972 __u32 *cptr;
2973
2974 if (subarray &&
2975 (strtoul(subarray, &ep, 10) != vc->vcnum ||
2976 *ep != '\0'))
2977 continue;
2978
2979 this = xcalloc(1, sizeof(*this));
2980 this->next = rest;
2981 rest = this;
2982
2983 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2984 this->array.raid_disks =
2985 __be16_to_cpu(vc->conf.prim_elmnt_count);
2986 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2987 this->array.raid_disks);
2988 this->array.md_minor = -1;
2989 this->array.major_version = -1;
2990 this->array.minor_version = -2;
2991 cptr = (__u32 *)(vc->conf.guid + 16);
2992 this->array.ctime = DECADE + __be32_to_cpu(*cptr);
2993 this->array.utime = DECADE +
2994 __be32_to_cpu(vc->conf.timestamp);
2995 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2996
2997 i = vc->vcnum;
2998 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2999 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
3000 DDF_init_full) {
3001 this->array.state = 0;
3002 this->resync_start = 0;
3003 } else {
3004 this->array.state = 1;
3005 this->resync_start = MaxSector;
3006 }
3007 memcpy(this->name, ddf->virt->entries[i].name, 16);
3008 this->name[16]=0;
3009 for(j=0; j<16; j++)
3010 if (this->name[j] == ' ')
3011 this->name[j] = 0;
3012
3013 memset(this->uuid, 0, sizeof(this->uuid));
3014 this->component_size = __be64_to_cpu(vc->conf.blocks);
3015 this->array.size = this->component_size / 2;
3016 this->container_member = i;
3017
3018 ddf->currentconf = vc;
3019 uuid_from_super_ddf(st, this->uuid);
3020 ddf->currentconf = NULL;
3021
3022 sprintf(this->text_version, "/%s/%d",
3023 devnum2devname(st->container_dev),
3024 this->container_member);
3025
3026 for (i = 0 ; i < ddf->mppe ; i++) {
3027 struct mdinfo *dev;
3028 struct dl *d;
3029 int stt;
3030 int pd;
3031
3032 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
3033 continue;
3034
3035 for (pd = __be16_to_cpu(ddf->phys->used_pdes);
3036 pd--;)
3037 if (ddf->phys->entries[pd].refnum
3038 == vc->conf.phys_refnum[i])
3039 break;
3040 if (pd < 0)
3041 continue;
3042
3043 stt = __be16_to_cpu(ddf->phys->entries[pd].state);
3044 if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3045 != DDF_Online)
3046 continue;
3047
3048 this->array.working_disks++;
3049
3050 for (d = ddf->dlist; d ; d=d->next)
3051 if (d->disk.refnum == vc->conf.phys_refnum[i])
3052 break;
3053 if (d == NULL)
3054 /* Haven't found that one yet, maybe there are others */
3055 continue;
3056
3057 dev = xcalloc(1, sizeof(*dev));
3058 dev->next = this->devs;
3059 this->devs = dev;
3060
3061 dev->disk.number = __be32_to_cpu(d->disk.refnum);
3062 dev->disk.major = d->major;
3063 dev->disk.minor = d->minor;
3064 dev->disk.raid_disk = i;
3065 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
3066 dev->recovery_start = MaxSector;
3067
3068 dev->events = __be32_to_cpu(ddf->primary.seq);
3069 dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
3070 dev->component_size = __be64_to_cpu(vc->conf.blocks);
3071 if (d->devname)
3072 strcpy(dev->name, d->devname);
3073 }
3074 }
3075 return rest;
3076 }
3077
3078 static int store_super_ddf(struct supertype *st, int fd)
3079 {
3080 struct ddf_super *ddf = st->sb;
3081 unsigned long long dsize;
3082 void *buf;
3083 int rc;
3084
3085 if (!ddf)
3086 return 1;
3087
3088 /* ->dlist and ->conflist will be set for updates, currently not
3089 * supported
3090 */
3091 if (ddf->dlist || ddf->conflist)
3092 return 1;
3093
3094 if (!get_dev_size(fd, NULL, &dsize))
3095 return 1;
3096
3097 if (posix_memalign(&buf, 512, 512) != 0)
3098 return 1;
3099 memset(buf, 0, 512);
3100
3101 lseek64(fd, dsize-512, 0);
3102 rc = write(fd, buf, 512);
3103 free(buf);
3104 if (rc < 0)
3105 return 1;
3106 return 0;
3107 }
3108
3109 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3110 {
3111 /*
3112 * return:
3113 * 0 same, or first was empty, and second was copied
3114 * 1 second had wrong number
3115 * 2 wrong uuid
3116 * 3 wrong other info
3117 */
3118 struct ddf_super *first = st->sb;
3119 struct ddf_super *second = tst->sb;
3120
3121 if (!first) {
3122 st->sb = tst->sb;
3123 tst->sb = NULL;
3124 return 0;
3125 }
3126
3127 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3128 return 2;
3129
3130 /* FIXME should I look at anything else? */
3131 return 0;
3132 }
3133
3134 #ifndef MDASSEMBLE
3135 /*
3136 * A new array 'a' has been started which claims to be instance 'inst'
3137 * within container 'c'.
3138 * We need to confirm that the array matches the metadata in 'c' so
3139 * that we don't corrupt any metadata.
3140 */
3141 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
3142 {
3143 dprintf("ddf: open_new %s\n", inst);
3144 a->info.container_member = atoi(inst);
3145 return 0;
3146 }
3147
3148 /*
3149 * The array 'a' is to be marked clean in the metadata.
3150 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3151 * clean up to the point (in sectors). If that cannot be recorded in the
3152 * metadata, then leave it as dirty.
3153 *
3154 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3155 * !global! virtual_disk.virtual_entry structure.
3156 */
3157 static int ddf_set_array_state(struct active_array *a, int consistent)
3158 {
3159 struct ddf_super *ddf = a->container->sb;
3160 int inst = a->info.container_member;
3161 int old = ddf->virt->entries[inst].state;
3162 if (consistent == 2) {
3163 /* Should check if a recovery should be started FIXME */
3164 consistent = 1;
3165 if (!is_resync_complete(&a->info))
3166 consistent = 0;
3167 }
3168 if (consistent)
3169 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
3170 else
3171 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
3172 if (old != ddf->virt->entries[inst].state)
3173 ddf->updates_pending = 1;
3174
3175 old = ddf->virt->entries[inst].init_state;
3176 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
3177 if (is_resync_complete(&a->info))
3178 ddf->virt->entries[inst].init_state |= DDF_init_full;
3179 else if (a->info.resync_start == 0)
3180 ddf->virt->entries[inst].init_state |= DDF_init_not;
3181 else
3182 ddf->virt->entries[inst].init_state |= DDF_init_quick;
3183 if (old != ddf->virt->entries[inst].init_state)
3184 ddf->updates_pending = 1;
3185
3186 dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
3187 a->info.resync_start);
3188 return consistent;
3189 }
3190
3191 #define container_of(ptr, type, member) ({ \
3192 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
3193 (type *)( (char *)__mptr - offsetof(type,member) );})
3194 /*
3195 * The state of each disk is stored in the global phys_disk structure
3196 * in phys_disk.entries[n].state.
3197 * This makes various combinations awkward.
3198 * - When a device fails in any array, it must be failed in all arrays
3199 * that include a part of this device.
3200 * - When a component is rebuilding, we cannot include it officially in the
3201 * array unless this is the only array that uses the device.
3202 *
3203 * So: when transitioning:
3204 * Online -> failed, just set failed flag. monitor will propagate
3205 * spare -> online, the device might need to be added to the array.
3206 * spare -> failed, just set failed. Don't worry if in array or not.
3207 */
3208 static void ddf_set_disk(struct active_array *a, int n, int state)
3209 {
3210 struct ddf_super *ddf = a->container->sb;
3211 unsigned int inst = a->info.container_member;
3212 struct vd_config *vc = find_vdcr(ddf, inst);
3213 int pd = find_phys(ddf, vc->phys_refnum[n]);
3214 int i, st, working;
3215 struct mdinfo *mdi;
3216 struct dl *dl;
3217
3218 if (vc == NULL) {
3219 dprintf("ddf: cannot find instance %d!!\n", inst);
3220 return;
3221 }
3222 /* Find the matching slot in 'info'. */
3223 for (mdi = a->info.devs; mdi; mdi = mdi->next)
3224 if (mdi->disk.raid_disk == n)
3225 break;
3226 if (!mdi)
3227 return;
3228
3229 /* and find the 'dl' entry corresponding to that. */
3230 for (dl = ddf->dlist; dl; dl = dl->next)
3231 if (mdi->state_fd >= 0 &&
3232 mdi->disk.major == dl->major &&
3233 mdi->disk.minor == dl->minor)
3234 break;
3235 if (!dl)
3236 return;
3237
3238 if (pd < 0 || pd != dl->pdnum) {
3239 /* disk doesn't currently exist or has changed.
3240 * If it is now in_sync, insert it. */
3241 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
3242 struct vcl *vcl;
3243 pd = dl->pdnum;
3244 vc->phys_refnum[n] = dl->disk.refnum;
3245 vcl = container_of(vc, struct vcl, conf);
3246 vcl->lba_offset[n] = mdi->data_offset;
3247 ddf->phys->entries[pd].type &=
3248 ~__cpu_to_be16(DDF_Global_Spare);
3249 ddf->phys->entries[pd].type |=
3250 __cpu_to_be16(DDF_Active_in_VD);
3251 ddf->updates_pending = 1;
3252 }
3253 } else {
3254 int old = ddf->phys->entries[pd].state;
3255 if (state & DS_FAULTY)
3256 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
3257 if (state & DS_INSYNC) {
3258 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
3259 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
3260 }
3261 if (old != ddf->phys->entries[pd].state)
3262 ddf->updates_pending = 1;
3263 }
3264
3265 dprintf("ddf: set_disk %d to %x\n", n, state);
3266
3267 /* Now we need to check the state of the array and update
3268 * virtual_disk.entries[n].state.
3269 * It needs to be one of "optimal", "degraded", "failed".
3270 * I don't understand 'deleted' or 'missing'.
3271 */
3272 working = 0;
3273 for (i=0; i < a->info.array.raid_disks; i++) {
3274 pd = find_phys(ddf, vc->phys_refnum[i]);
3275 if (pd < 0)
3276 continue;
3277 st = __be16_to_cpu(ddf->phys->entries[pd].state);
3278 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3279 == DDF_Online)
3280 working++;
3281 }
3282 state = DDF_state_degraded;
3283 if (working == a->info.array.raid_disks)
3284 state = DDF_state_optimal;
3285 else switch(vc->prl) {
3286 case DDF_RAID0:
3287 case DDF_CONCAT:
3288 case DDF_JBOD:
3289 state = DDF_state_failed;
3290 break;
3291 case DDF_RAID1:
3292 if (working == 0)
3293 state = DDF_state_failed;
3294 else if (working == 2 && state == DDF_state_degraded)
3295 state = DDF_state_part_optimal;
3296 break;
3297 case DDF_RAID4:
3298 case DDF_RAID5:
3299 if (working < a->info.array.raid_disks-1)
3300 state = DDF_state_failed;
3301 break;
3302 case DDF_RAID6:
3303 if (working < a->info.array.raid_disks-2)
3304 state = DDF_state_failed;
3305 else if (working == a->info.array.raid_disks-1)
3306 state = DDF_state_part_optimal;
3307 break;
3308 }
3309
3310 if (ddf->virt->entries[inst].state !=
3311 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
3312 | state)) {
3313
3314 ddf->virt->entries[inst].state =
3315 (ddf->virt->entries[inst].state & ~DDF_state_mask)
3316 | state;
3317 ddf->updates_pending = 1;
3318 }
3319
3320 }
3321
3322 static void ddf_sync_metadata(struct supertype *st)
3323 {
3324
3325 /*
3326 * Write all data to all devices.
3327 * Later, we might be able to track whether only local changes
3328 * have been made, or whether any global data has been changed,
3329 * but ddf is sufficiently weird that it probably always
3330 * changes global data ....
3331 */
3332 struct ddf_super *ddf = st->sb;
3333 if (!ddf->updates_pending)
3334 return;
3335 ddf->updates_pending = 0;
3336 __write_init_super_ddf(st);
3337 dprintf("ddf: sync_metadata\n");
3338 }
3339
3340 static void ddf_process_update(struct supertype *st,
3341 struct metadata_update *update)
3342 {
3343 /* Apply this update to the metadata.
3344 * The first 4 bytes are a DDF_*_MAGIC which guides
3345 * our actions.
3346 * Possible update are:
3347 * DDF_PHYS_RECORDS_MAGIC
3348 * Add a new physical device or remove an old one.
3349 * Changes to this record only happen implicitly.
3350 * used_pdes is the device number.
3351 * DDF_VIRT_RECORDS_MAGIC
3352 * Add a new VD. Possibly also change the 'access' bits.
3353 * populated_vdes is the entry number.
3354 * DDF_VD_CONF_MAGIC
3355 * New or updated VD. the VIRT_RECORD must already
3356 * exist. For an update, phys_refnum and lba_offset
3357 * (at least) are updated, and the VD_CONF must
3358 * be written to precisely those devices listed with
3359 * a phys_refnum.
3360 * DDF_SPARE_ASSIGN_MAGIC
3361 * replacement Spare Assignment Record... but for which device?
3362 *
3363 * So, e.g.:
3364 * - to create a new array, we send a VIRT_RECORD and
3365 * a VD_CONF. Then assemble and start the array.
3366 * - to activate a spare we send a VD_CONF to add the phys_refnum
3367 * and offset. This will also mark the spare as active with
3368 * a spare-assignment record.
3369 */
3370 struct ddf_super *ddf = st->sb;
3371 __u32 *magic = (__u32*)update->buf;
3372 struct phys_disk *pd;
3373 struct virtual_disk *vd;
3374 struct vd_config *vc;
3375 struct vcl *vcl;
3376 struct dl *dl;
3377 unsigned int mppe;
3378 unsigned int ent;
3379 unsigned int pdnum, pd2;
3380
3381 dprintf("Process update %x\n", *magic);
3382
3383 switch (*magic) {
3384 case DDF_PHYS_RECORDS_MAGIC:
3385
3386 if (update->len != (sizeof(struct phys_disk) +
3387 sizeof(struct phys_disk_entry)))
3388 return;
3389 pd = (struct phys_disk*)update->buf;
3390
3391 ent = __be16_to_cpu(pd->used_pdes);
3392 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
3393 return;
3394 if (pd->entries[0].state & __cpu_to_be16(DDF_Missing)) {
3395 struct dl **dlp;
3396 /* removing this disk. */
3397 ddf->phys->entries[ent].state |= __cpu_to_be16(DDF_Missing);
3398 for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) {
3399 struct dl *dl = *dlp;
3400 if (dl->pdnum == (signed)ent) {
3401 close(dl->fd);
3402 dl->fd = -1;
3403 /* FIXME this doesn't free
3404 * dl->devname */
3405 update->space = dl;
3406 *dlp = dl->next;
3407 break;
3408 }
3409 }
3410 ddf->updates_pending = 1;
3411 return;
3412 }
3413 if (!all_ff(ddf->phys->entries[ent].guid))
3414 return;
3415 ddf->phys->entries[ent] = pd->entries[0];
3416 ddf->phys->used_pdes = __cpu_to_be16(1 +
3417 __be16_to_cpu(ddf->phys->used_pdes));
3418 ddf->updates_pending = 1;
3419 if (ddf->add_list) {
3420 struct active_array *a;
3421 struct dl *al = ddf->add_list;
3422 ddf->add_list = al->next;
3423
3424 al->next = ddf->dlist;
3425 ddf->dlist = al;
3426
3427 /* As a device has been added, we should check
3428 * for any degraded devices that might make
3429 * use of this spare */
3430 for (a = st->arrays ; a; a=a->next)
3431 a->check_degraded = 1;
3432 }
3433 break;
3434
3435 case DDF_VIRT_RECORDS_MAGIC:
3436
3437 if (update->len != (sizeof(struct virtual_disk) +
3438 sizeof(struct virtual_entry)))
3439 return;
3440 vd = (struct virtual_disk*)update->buf;
3441
3442 ent = __be16_to_cpu(vd->populated_vdes);
3443 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
3444 return;
3445 if (!all_ff(ddf->virt->entries[ent].guid))
3446 return;
3447 ddf->virt->entries[ent] = vd->entries[0];
3448 ddf->virt->populated_vdes = __cpu_to_be16(1 +
3449 __be16_to_cpu(ddf->virt->populated_vdes));
3450 ddf->updates_pending = 1;
3451 break;
3452
3453 case DDF_VD_CONF_MAGIC:
3454 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
3455
3456 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
3457 if ((unsigned)update->len != ddf->conf_rec_len * 512)
3458 return;
3459 vc = (struct vd_config*)update->buf;
3460 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3461 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
3462 break;
3463 dprintf("vcl = %p\n", vcl);
3464 if (vcl) {
3465 /* An update, just copy the phys_refnum and lba_offset
3466 * fields
3467 */
3468 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
3469 mppe * (sizeof(__u32) + sizeof(__u64)));
3470 } else {
3471 /* A new VD_CONF */
3472 if (!update->space)
3473 return;
3474 vcl = update->space;
3475 update->space = NULL;
3476 vcl->next = ddf->conflist;
3477 memcpy(&vcl->conf, vc, update->len);
3478 vcl->lba_offset = (__u64*)
3479 &vcl->conf.phys_refnum[mppe];
3480 for (ent = 0;
3481 ent < __be16_to_cpu(ddf->virt->populated_vdes);
3482 ent++)
3483 if (memcmp(vc->guid, ddf->virt->entries[ent].guid,
3484 DDF_GUID_LEN) == 0) {
3485 vcl->vcnum = ent;
3486 break;
3487 }
3488 ddf->conflist = vcl;
3489 }
3490 /* Set DDF_Transition on all Failed devices - to help
3491 * us detect those that are no longer in use
3492 */
3493 for (pdnum = 0; pdnum < __be16_to_cpu(ddf->phys->used_pdes); pdnum++)
3494 if (ddf->phys->entries[pdnum].state
3495 & __be16_to_cpu(DDF_Failed))
3496 ddf->phys->entries[pdnum].state
3497 |= __be16_to_cpu(DDF_Transition);
3498 /* Now make sure vlist is correct for each dl. */
3499 for (dl = ddf->dlist; dl; dl = dl->next) {
3500 unsigned int dn;
3501 unsigned int vn = 0;
3502 int in_degraded = 0;
3503 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3504 for (dn=0; dn < ddf->mppe ; dn++)
3505 if (vcl->conf.phys_refnum[dn] ==
3506 dl->disk.refnum) {
3507 int vstate;
3508 dprintf("dev %d has %p at %d\n",
3509 dl->pdnum, vcl, vn);
3510 /* Clear the Transition flag */
3511 if (ddf->phys->entries[dl->pdnum].state
3512 & __be16_to_cpu(DDF_Failed))
3513 ddf->phys->entries[dl->pdnum].state &=
3514 ~__be16_to_cpu(DDF_Transition);
3515
3516 dl->vlist[vn++] = vcl;
3517 vstate = ddf->virt->entries[vcl->vcnum].state
3518 & DDF_state_mask;
3519 if (vstate == DDF_state_degraded ||
3520 vstate == DDF_state_part_optimal)
3521 in_degraded = 1;
3522 break;
3523 }
3524 while (vn < ddf->max_part)
3525 dl->vlist[vn++] = NULL;
3526 if (dl->vlist[0]) {
3527 ddf->phys->entries[dl->pdnum].type &=
3528 ~__cpu_to_be16(DDF_Global_Spare);
3529 if (!(ddf->phys->entries[dl->pdnum].type &
3530 __cpu_to_be16(DDF_Active_in_VD))) {
3531 ddf->phys->entries[dl->pdnum].type |=
3532 __cpu_to_be16(DDF_Active_in_VD);
3533 if (in_degraded)
3534 ddf->phys->entries[dl->pdnum].state |=
3535 __cpu_to_be16(DDF_Rebuilding);
3536 }
3537 }
3538 if (dl->spare) {
3539 ddf->phys->entries[dl->pdnum].type &=
3540 ~__cpu_to_be16(DDF_Global_Spare);
3541 ddf->phys->entries[dl->pdnum].type |=
3542 __cpu_to_be16(DDF_Spare);
3543 }
3544 if (!dl->vlist[0] && !dl->spare) {
3545 ddf->phys->entries[dl->pdnum].type |=
3546 __cpu_to_be16(DDF_Global_Spare);
3547 ddf->phys->entries[dl->pdnum].type &=
3548 ~__cpu_to_be16(DDF_Spare |
3549 DDF_Active_in_VD);
3550 }
3551 }
3552
3553 /* Now remove any 'Failed' devices that are not part
3554 * of any VD. They will have the Transition flag set.
3555 * Once done, we need to update all dl->pdnum numbers.
3556 */
3557 pd2 = 0;
3558 for (pdnum = 0; pdnum < __be16_to_cpu(ddf->phys->used_pdes); pdnum++)
3559 if ((ddf->phys->entries[pdnum].state
3560 & __be16_to_cpu(DDF_Failed))
3561 && (ddf->phys->entries[pdnum].state
3562 & __be16_to_cpu(DDF_Transition)))
3563 /* skip this one */;
3564 else if (pdnum == pd2)
3565 pd2++;
3566 else {
3567 ddf->phys->entries[pd2] = ddf->phys->entries[pdnum];
3568 for (dl = ddf->dlist; dl; dl = dl->next)
3569 if (dl->pdnum == (int)pdnum)
3570 dl->pdnum = pd2;
3571 pd2++;
3572 }
3573 ddf->phys->used_pdes = __cpu_to_be16(pd2);
3574 while (pd2 < pdnum) {
3575 memset(ddf->phys->entries[pd2].guid, 0xff, DDF_GUID_LEN);
3576 pd2++;
3577 }
3578
3579 ddf->updates_pending = 1;
3580 break;
3581 case DDF_SPARE_ASSIGN_MAGIC:
3582 default: break;
3583 }
3584 }
3585
3586 static void ddf_prepare_update(struct supertype *st,
3587 struct metadata_update *update)
3588 {
3589 /* This update arrived at managemon.
3590 * We are about to pass it to monitor.
3591 * If a malloc is needed, do it here.
3592 */
3593 struct ddf_super *ddf = st->sb;
3594 __u32 *magic = (__u32*)update->buf;
3595 if (*magic == DDF_VD_CONF_MAGIC)
3596 if (posix_memalign(&update->space, 512,
3597 offsetof(struct vcl, conf)
3598 + ddf->conf_rec_len * 512) != 0)
3599 update->space = NULL;
3600 }
3601
3602 /*
3603 * Check if the array 'a' is degraded but not failed.
3604 * If it is, find as many spares as are available and needed and
3605 * arrange for their inclusion.
3606 * We only choose devices which are not already in the array,
3607 * and prefer those with a spare-assignment to this array.
3608 * otherwise we choose global spares - assuming always that
3609 * there is enough room.
3610 * For each spare that we assign, we return an 'mdinfo' which
3611 * describes the position for the device in the array.
3612 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3613 * the new phys_refnum and lba_offset values.
3614 *
3615 * Only worry about BVDs at the moment.
3616 */
3617 static struct mdinfo *ddf_activate_spare(struct active_array *a,
3618 struct metadata_update **updates)
3619 {
3620 int working = 0;
3621 struct mdinfo *d;
3622 struct ddf_super *ddf = a->container->sb;
3623 int global_ok = 0;
3624 struct mdinfo *rv = NULL;
3625 struct mdinfo *di;
3626 struct metadata_update *mu;
3627 struct dl *dl;
3628 int i;
3629 struct vd_config *vc;
3630 __u64 *lba;
3631
3632 for (d = a->info.devs ; d ; d = d->next) {
3633 if ((d->curr_state & DS_FAULTY) &&
3634 d->state_fd >= 0)
3635 /* wait for Removal to happen */
3636 return NULL;
3637 if (d->state_fd >= 0)
3638 working ++;
3639 }
3640
3641 dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3642 a->info.array.level);
3643 if (working == a->info.array.raid_disks)
3644 return NULL; /* array not degraded */
3645 switch (a->info.array.level) {
3646 case 1:
3647 if (working == 0)
3648 return NULL; /* failed */
3649 break;
3650 case 4:
3651 case 5:
3652 if (working < a->info.array.raid_disks - 1)
3653 return NULL; /* failed */
3654 break;
3655 case 6:
3656 if (working < a->info.array.raid_disks - 2)
3657 return NULL; /* failed */
3658 break;
3659 default: /* concat or stripe */
3660 return NULL; /* failed */
3661 }
3662
3663 /* For each slot, if it is not working, find a spare */
3664 dl = ddf->dlist;
3665 for (i = 0; i < a->info.array.raid_disks; i++) {
3666 for (d = a->info.devs ; d ; d = d->next)
3667 if (d->disk.raid_disk == i)
3668 break;
3669 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3670 if (d && (d->state_fd >= 0))
3671 continue;
3672
3673 /* OK, this device needs recovery. Find a spare */
3674 again:
3675 for ( ; dl ; dl = dl->next) {
3676 unsigned long long esize;
3677 unsigned long long pos;
3678 struct mdinfo *d2;
3679 int is_global = 0;
3680 int is_dedicated = 0;
3681 struct extent *ex;
3682 unsigned int j;
3683 /* If in this array, skip */
3684 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3685 if (d2->state_fd >= 0 &&
3686 d2->disk.major == dl->major &&
3687 d2->disk.minor == dl->minor) {
3688 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3689 break;
3690 }
3691 if (d2)
3692 continue;
3693 if (ddf->phys->entries[dl->pdnum].type &
3694 __cpu_to_be16(DDF_Spare)) {
3695 /* Check spare assign record */
3696 if (dl->spare) {
3697 if (dl->spare->type & DDF_spare_dedicated) {
3698 /* check spare_ents for guid */
3699 for (j = 0 ;
3700 j < __be16_to_cpu(dl->spare->populated);
3701 j++) {
3702 if (memcmp(dl->spare->spare_ents[j].guid,
3703 ddf->virt->entries[a->info.container_member].guid,
3704 DDF_GUID_LEN) == 0)
3705 is_dedicated = 1;
3706 }
3707 } else
3708 is_global = 1;
3709 }
3710 } else if (ddf->phys->entries[dl->pdnum].type &
3711 __cpu_to_be16(DDF_Global_Spare)) {
3712 is_global = 1;
3713 } else if (!(ddf->phys->entries[dl->pdnum].state &
3714 __cpu_to_be16(DDF_Failed))) {
3715 /* we can possibly use some of this */
3716 is_global = 1;
3717 }
3718 if ( ! (is_dedicated ||
3719 (is_global && global_ok))) {
3720 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3721 is_dedicated, is_global);
3722 continue;
3723 }
3724
3725 /* We are allowed to use this device - is there space?
3726 * We need a->info.component_size sectors */
3727 ex = get_extents(ddf, dl);
3728 if (!ex) {
3729 dprintf("cannot get extents\n");
3730 continue;
3731 }
3732 j = 0; pos = 0;
3733 esize = 0;
3734
3735 do {
3736 esize = ex[j].start - pos;
3737 if (esize >= a->info.component_size)
3738 break;
3739 pos = ex[j].start + ex[j].size;
3740 j++;
3741 } while (ex[j-1].size);
3742
3743 free(ex);
3744 if (esize < a->info.component_size) {
3745 dprintf("%x:%x has no room: %llu %llu\n",
3746 dl->major, dl->minor,
3747 esize, a->info.component_size);
3748 /* No room */
3749 continue;
3750 }
3751
3752 /* Cool, we have a device with some space at pos */
3753 di = xcalloc(1, sizeof(*di));
3754 di->disk.number = i;
3755 di->disk.raid_disk = i;
3756 di->disk.major = dl->major;
3757 di->disk.minor = dl->minor;
3758 di->disk.state = 0;
3759 di->recovery_start = 0;
3760 di->data_offset = pos;
3761 di->component_size = a->info.component_size;
3762 di->container_member = dl->pdnum;
3763 di->next = rv;
3764 rv = di;
3765 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3766 i, pos);
3767
3768 break;
3769 }
3770 if (!dl && ! global_ok) {
3771 /* not enough dedicated spares, try global */
3772 global_ok = 1;
3773 dl = ddf->dlist;
3774 goto again;
3775 }
3776 }
3777
3778 if (!rv)
3779 /* No spares found */
3780 return rv;
3781 /* Now 'rv' has a list of devices to return.
3782 * Create a metadata_update record to update the
3783 * phys_refnum and lba_offset values
3784 */
3785 mu = xmalloc(sizeof(*mu));
3786 if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
3787 free(mu);
3788 mu = NULL;
3789 }
3790 mu->buf = xmalloc(ddf->conf_rec_len * 512);
3791 mu->len = ddf->conf_rec_len * 512;
3792 mu->space = NULL;
3793 mu->space_list = NULL;
3794 mu->next = *updates;
3795 vc = find_vdcr(ddf, a->info.container_member);
3796 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3797
3798 vc = (struct vd_config*)mu->buf;
3799 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3800 for (di = rv ; di ; di = di->next) {
3801 vc->phys_refnum[di->disk.raid_disk] =
3802 ddf->phys->entries[dl->pdnum].refnum;
3803 lba[di->disk.raid_disk] = di->data_offset;
3804 }
3805 *updates = mu;
3806 return rv;
3807 }
3808 #endif /* MDASSEMBLE */
3809
3810 static int ddf_level_to_layout(int level)
3811 {
3812 switch(level) {
3813 case 0:
3814 case 1:
3815 return 0;
3816 case 5:
3817 return ALGORITHM_LEFT_SYMMETRIC;
3818 case 6:
3819 return ALGORITHM_ROTATING_N_CONTINUE;
3820 case 10:
3821 return 0x102;
3822 default:
3823 return UnSet;
3824 }
3825 }
3826
3827 static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk)
3828 {
3829 if (level && *level == UnSet)
3830 *level = LEVEL_CONTAINER;
3831
3832 if (level && layout && *layout == UnSet)
3833 *layout = ddf_level_to_layout(*level);
3834 }
3835
3836 struct superswitch super_ddf = {
3837 #ifndef MDASSEMBLE
3838 .examine_super = examine_super_ddf,
3839 .brief_examine_super = brief_examine_super_ddf,
3840 .brief_examine_subarrays = brief_examine_subarrays_ddf,
3841 .export_examine_super = export_examine_super_ddf,
3842 .detail_super = detail_super_ddf,
3843 .brief_detail_super = brief_detail_super_ddf,
3844 .validate_geometry = validate_geometry_ddf,
3845 .write_init_super = write_init_super_ddf,
3846 .add_to_super = add_to_super_ddf,
3847 .remove_from_super = remove_from_super_ddf,
3848 .load_container = load_container_ddf,
3849 #endif
3850 .match_home = match_home_ddf,
3851 .uuid_from_super= uuid_from_super_ddf,
3852 .getinfo_super = getinfo_super_ddf,
3853 .update_super = update_super_ddf,
3854
3855 .avail_size = avail_size_ddf,
3856
3857 .compare_super = compare_super_ddf,
3858
3859 .load_super = load_super_ddf,
3860 .init_super = init_super_ddf,
3861 .store_super = store_super_ddf,
3862 .free_super = free_super_ddf,
3863 .match_metadata_desc = match_metadata_desc_ddf,
3864 .container_content = container_content_ddf,
3865 .default_geometry = default_geometry_ddf,
3866
3867 .external = 1,
3868
3869 #ifndef MDASSEMBLE
3870 /* for mdmon */
3871 .open_new = ddf_open_new,
3872 .set_array_state= ddf_set_array_state,
3873 .set_disk = ddf_set_disk,
3874 .sync_metadata = ddf_sync_metadata,
3875 .process_update = ddf_process_update,
3876 .prepare_update = ddf_prepare_update,
3877 .activate_spare = ddf_activate_spare,
3878 #endif
3879 .name = "ddf",
3880 };
Unnamed repository; edit this file 'description' to name the repository.