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