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