]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
Switch open_subarray to use the new load_container
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neil@brown.name>
23 *
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
26 */
27
28 #define HAVE_STDINT_H 1
29 #include "mdadm.h"
30 #include "mdmon.h"
31 #include "sha1.h"
32 #include <values.h>
33
34 /* a non-official T10 name for creation GUIDs */
35 static char T10[] = "Linux-MD";
36
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
40 */
41 #define DECADE (3600*24*(365*10+2))
42 unsigned long crc32(
43 unsigned long crc,
44 const unsigned char *buf,
45 unsigned len);
46
47 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
51 *
52 * Note:
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
55 *
56 */
57
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
65 #define DDF_JBOD 0x0f
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
70
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
84
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
87
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
93
94 /* Magic numbers */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
105
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
109
110 struct ddf_header {
111 __u32 magic; /* DDF_HEADER_MAGIC */
112 __u32 crc;
113 char guid[DDF_GUID_LEN];
114 char revision[8]; /* 01.02.00 */
115 __u32 seq; /* starts at '1' */
116 __u32 timestamp;
117 __u8 openflag;
118 __u8 foreignflag;
119 __u8 enforcegroups;
120 __u8 pad0; /* 0xff */
121 __u8 pad1[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext[32]; /* reserved: fill with 0xff */
124 __u64 primary_lba;
125 __u64 secondary_lba;
126 __u8 type;
127 __u8 pad2[3]; /* 0xff */
128 __u32 workspace_len; /* sectors for vendor space -
129 * at least 32768(sectors) */
130 __u64 workspace_lba;
131 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
136 *12/512) */
137 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset;
141 __u32 controller_section_length;
142 __u32 phys_section_offset;
143 __u32 phys_section_length;
144 __u32 virt_section_offset;
145 __u32 virt_section_length;
146 __u32 config_section_offset;
147 __u32 config_section_length;
148 __u32 data_section_offset;
149 __u32 data_section_length;
150 __u32 bbm_section_offset;
151 __u32 bbm_section_length;
152 __u32 diag_space_offset;
153 __u32 diag_space_length;
154 __u32 vendor_offset;
155 __u32 vendor_length;
156 /* 256 bytes so far */
157 __u8 pad4[256]; /* 0xff */
158 };
159
160 /* type field */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
164
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data {
167 __u32 magic; /* DDF_CONTROLLER_MAGIC */
168 __u32 crc;
169 char guid[DDF_GUID_LEN];
170 struct controller_type {
171 __u16 vendor_id;
172 __u16 device_id;
173 __u16 sub_vendor_id;
174 __u16 sub_device_id;
175 } type;
176 char product_id[16];
177 __u8 pad[8]; /* 0xff */
178 __u8 vendor_data[448];
179 };
180
181 /* The content of phys_section - global scope */
182 struct phys_disk {
183 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
184 __u32 crc;
185 __u16 used_pdes;
186 __u16 max_pdes;
187 __u8 pad[52];
188 struct phys_disk_entry {
189 char guid[DDF_GUID_LEN];
190 __u32 refnum;
191 __u16 type;
192 __u16 state;
193 __u64 config_size; /* DDF structures must be after here */
194 char path[18]; /* another horrible structure really */
195 __u8 pad[6];
196 } entries[0];
197 };
198
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
206
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
212
213 /* phys_disk_entry.state is a bigendian bitmap */
214 #define DDF_Online 1
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
218 #define DDF_SMART 16
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
221
222 /* The content of the virt_section global scope */
223 struct virtual_disk {
224 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
225 __u32 crc;
226 __u16 populated_vdes;
227 __u16 max_vdes;
228 __u8 pad[52];
229 struct virtual_entry {
230 char guid[DDF_GUID_LEN];
231 __u16 unit;
232 __u16 pad0; /* 0xffff */
233 __u16 guid_crc;
234 __u16 type;
235 __u8 state;
236 __u8 init_state;
237 __u8 pad1[14];
238 char name[16];
239 } entries[0];
240 };
241
242 /* virtual_entry.type is a bitmap - bigendian */
243 #define DDF_Shared 1
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
247
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
256
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
259
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
266
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
271
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
275 */
276
277 struct vd_config {
278 __u32 magic; /* DDF_VD_CONF_MAGIC */
279 __u32 crc;
280 char guid[DDF_GUID_LEN];
281 __u32 timestamp;
282 __u32 seqnum;
283 __u8 pad0[24];
284 __u16 prim_elmnt_count;
285 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
286 __u8 prl;
287 __u8 rlq;
288 __u8 sec_elmnt_count;
289 __u8 sec_elmnt_seq;
290 __u8 srl;
291 __u64 blocks; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks; /* blocks in array */
295 __u8 pad1[8];
296 __u32 spare_refs[8];
297 __u8 cache_pol[8];
298 __u8 bg_rate;
299 __u8 pad2[3];
300 __u8 pad3[52];
301 __u8 pad4[192];
302 __u8 v0[32]; /* reserved- 0xff */
303 __u8 v1[32]; /* reserved- 0xff */
304 __u8 v2[16]; /* reserved- 0xff */
305 __u8 v3[16]; /* reserved- 0xff */
306 __u8 vendor[32];
307 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
310 };
311
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
320
321 struct spare_assign {
322 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
323 __u32 crc;
324 __u32 timestamp;
325 __u8 reserved[7];
326 __u8 type;
327 __u16 populated; /* SAEs used */
328 __u16 max; /* max SAEs */
329 __u8 pad[8];
330 struct spare_assign_entry {
331 char guid[DDF_GUID_LEN];
332 __u16 secondary_element;
333 __u8 pad[6];
334 } spare_ents[0];
335 };
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
341
342 /* The data_section contents - local scope */
343 struct disk_data {
344 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
345 __u32 crc;
346 char guid[DDF_GUID_LEN];
347 __u32 refnum; /* crc of some magic drive data ... */
348 __u8 forced_ref; /* set when above was not result of magic */
349 __u8 forced_guid; /* set if guid was forced rather than magic */
350 __u8 vendor[32];
351 __u8 pad[442];
352 };
353
354 /* bbm_section content */
355 struct bad_block_log {
356 __u32 magic;
357 __u32 crc;
358 __u16 entry_count;
359 __u32 spare_count;
360 __u8 pad[10];
361 __u64 first_spare;
362 struct mapped_block {
363 __u64 defective_start;
364 __u32 replacement_start;
365 __u16 remap_count;
366 __u8 pad[2];
367 } entries[0];
368 };
369
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
374 * - ddf header
375 * - controller_data
376 * - Physical disk records
377 * - Virtual disk records
378 * The local data is:
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
382 *
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
391 */
392 struct ddf_super {
393 struct ddf_header anchor, primary, secondary;
394 struct ddf_controller_data controller;
395 struct ddf_header *active;
396 struct phys_disk *phys;
397 struct virtual_disk *virt;
398 int pdsize, vdsize;
399 unsigned int max_part, mppe, conf_rec_len;
400 int currentdev;
401 int updates_pending;
402 struct vcl {
403 union {
404 char space[512];
405 struct {
406 struct vcl *next;
407 __u64 *lba_offset; /* location in 'conf' of
408 * the lba table */
409 unsigned int vcnum; /* index into ->virt */
410 __u64 *block_sizes; /* NULL if all the same */
411 };
412 };
413 struct vd_config conf;
414 } *conflist, *currentconf;
415 struct dl {
416 union {
417 char space[512];
418 struct {
419 struct dl *next;
420 int major, minor;
421 char *devname;
422 int fd;
423 unsigned long long size; /* sectors */
424 int pdnum; /* index in ->phys */
425 struct spare_assign *spare;
426 void *mdupdate; /* hold metadata update */
427
428 /* These fields used by auto-layout */
429 int raiddisk; /* slot to fill in autolayout */
430 __u64 esize;
431 };
432 };
433 struct disk_data disk;
434 struct vcl *vlist[0]; /* max_part in size */
435 } *dlist, *add_list;
436 };
437
438 #ifndef offsetof
439 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
440 #endif
441
442
443 static unsigned int calc_crc(void *buf, int len)
444 {
445 /* crcs are always at the same place as in the ddf_header */
446 struct ddf_header *ddf = buf;
447 __u32 oldcrc = ddf->crc;
448 __u32 newcrc;
449 ddf->crc = 0xffffffff;
450
451 newcrc = crc32(0, buf, len);
452 ddf->crc = oldcrc;
453 /* The crc is store (like everything) bigendian, so convert
454 * here for simplicity
455 */
456 return __cpu_to_be32(newcrc);
457 }
458
459 static int load_ddf_header(int fd, unsigned long long lba,
460 unsigned long long size,
461 int type,
462 struct ddf_header *hdr, struct ddf_header *anchor)
463 {
464 /* read a ddf header (primary or secondary) from fd/lba
465 * and check that it is consistent with anchor
466 * Need to check:
467 * magic, crc, guid, rev, and LBA's header_type, and
468 * everything after header_type must be the same
469 */
470 if (lba >= size-1)
471 return 0;
472
473 if (lseek64(fd, lba<<9, 0) < 0)
474 return 0;
475
476 if (read(fd, hdr, 512) != 512)
477 return 0;
478
479 if (hdr->magic != DDF_HEADER_MAGIC)
480 return 0;
481 if (calc_crc(hdr, 512) != hdr->crc)
482 return 0;
483 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
484 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
485 anchor->primary_lba != hdr->primary_lba ||
486 anchor->secondary_lba != hdr->secondary_lba ||
487 hdr->type != type ||
488 memcmp(anchor->pad2, hdr->pad2, 512 -
489 offsetof(struct ddf_header, pad2)) != 0)
490 return 0;
491
492 /* Looks good enough to me... */
493 return 1;
494 }
495
496 static void *load_section(int fd, struct ddf_super *super, void *buf,
497 __u32 offset_be, __u32 len_be, int check)
498 {
499 unsigned long long offset = __be32_to_cpu(offset_be);
500 unsigned long long len = __be32_to_cpu(len_be);
501 int dofree = (buf == NULL);
502
503 if (check)
504 if (len != 2 && len != 8 && len != 32
505 && len != 128 && len != 512)
506 return NULL;
507
508 if (len > 1024)
509 return NULL;
510 if (buf) {
511 /* All pre-allocated sections are a single block */
512 if (len != 1)
513 return NULL;
514 } else if (posix_memalign(&buf, 512, len<<9) != 0)
515 buf = NULL;
516
517 if (!buf)
518 return NULL;
519
520 if (super->active->type == 1)
521 offset += __be64_to_cpu(super->active->primary_lba);
522 else
523 offset += __be64_to_cpu(super->active->secondary_lba);
524
525 if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
526 if (dofree)
527 free(buf);
528 return NULL;
529 }
530 if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
531 if (dofree)
532 free(buf);
533 return NULL;
534 }
535 return buf;
536 }
537
538 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
539 {
540 unsigned long long dsize;
541
542 get_dev_size(fd, NULL, &dsize);
543
544 if (lseek64(fd, dsize-512, 0) < 0) {
545 if (devname)
546 fprintf(stderr,
547 Name": Cannot seek to anchor block on %s: %s\n",
548 devname, strerror(errno));
549 return 1;
550 }
551 if (read(fd, &super->anchor, 512) != 512) {
552 if (devname)
553 fprintf(stderr,
554 Name ": Cannot read anchor block on %s: %s\n",
555 devname, strerror(errno));
556 return 1;
557 }
558 if (super->anchor.magic != DDF_HEADER_MAGIC) {
559 if (devname)
560 fprintf(stderr, Name ": no DDF anchor found on %s\n",
561 devname);
562 return 2;
563 }
564 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
565 if (devname)
566 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
567 devname);
568 return 2;
569 }
570 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
571 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
572 if (devname)
573 fprintf(stderr, Name ": can only support super revision"
574 " %.8s and earlier, not %.8s on %s\n",
575 DDF_REVISION_2, super->anchor.revision,devname);
576 return 2;
577 }
578 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
579 dsize >> 9, 1,
580 &super->primary, &super->anchor) == 0) {
581 if (devname)
582 fprintf(stderr,
583 Name ": Failed to load primary DDF header "
584 "on %s\n", devname);
585 return 2;
586 }
587 super->active = &super->primary;
588 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
589 dsize >> 9, 2,
590 &super->secondary, &super->anchor)) {
591 if ((__be32_to_cpu(super->primary.seq)
592 < __be32_to_cpu(super->secondary.seq) &&
593 !super->secondary.openflag)
594 || (__be32_to_cpu(super->primary.seq)
595 == __be32_to_cpu(super->secondary.seq) &&
596 super->primary.openflag && !super->secondary.openflag)
597 )
598 super->active = &super->secondary;
599 }
600 return 0;
601 }
602
603 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
604 {
605 void *ok;
606 ok = load_section(fd, super, &super->controller,
607 super->active->controller_section_offset,
608 super->active->controller_section_length,
609 0);
610 super->phys = load_section(fd, super, NULL,
611 super->active->phys_section_offset,
612 super->active->phys_section_length,
613 1);
614 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
615
616 super->virt = load_section(fd, super, NULL,
617 super->active->virt_section_offset,
618 super->active->virt_section_length,
619 1);
620 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
621 if (!ok ||
622 !super->phys ||
623 !super->virt) {
624 free(super->phys);
625 free(super->virt);
626 super->phys = NULL;
627 super->virt = NULL;
628 return 2;
629 }
630 super->conflist = NULL;
631 super->dlist = NULL;
632
633 super->max_part = __be16_to_cpu(super->active->max_partitions);
634 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
635 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
636 return 0;
637 }
638
639 static int load_ddf_local(int fd, struct ddf_super *super,
640 char *devname, int keep)
641 {
642 struct dl *dl;
643 struct stat stb;
644 char *conf;
645 unsigned int i;
646 unsigned int confsec;
647 int vnum;
648 unsigned int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
649 unsigned long long dsize;
650
651 /* First the local disk info */
652 if (posix_memalign((void**)&dl, 512,
653 sizeof(*dl) +
654 (super->max_part) * sizeof(dl->vlist[0])) != 0) {
655 fprintf(stderr, Name ": %s could not allocate disk info buffer\n",
656 __func__);
657 return 1;
658 }
659
660 load_section(fd, super, &dl->disk,
661 super->active->data_section_offset,
662 super->active->data_section_length,
663 0);
664 dl->devname = devname ? strdup(devname) : NULL;
665
666 fstat(fd, &stb);
667 dl->major = major(stb.st_rdev);
668 dl->minor = minor(stb.st_rdev);
669 dl->next = super->dlist;
670 dl->fd = keep ? fd : -1;
671
672 dl->size = 0;
673 if (get_dev_size(fd, devname, &dsize))
674 dl->size = dsize >> 9;
675 dl->spare = NULL;
676 for (i = 0 ; i < super->max_part ; i++)
677 dl->vlist[i] = NULL;
678 super->dlist = dl;
679 dl->pdnum = -1;
680 for (i = 0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
681 if (memcmp(super->phys->entries[i].guid,
682 dl->disk.guid, DDF_GUID_LEN) == 0)
683 dl->pdnum = i;
684
685 /* Now the config list. */
686 /* 'conf' is an array of config entries, some of which are
687 * probably invalid. Those which are good need to be copied into
688 * the conflist
689 */
690
691 conf = load_section(fd, super, NULL,
692 super->active->config_section_offset,
693 super->active->config_section_length,
694 0);
695
696 vnum = 0;
697 for (confsec = 0;
698 confsec < __be32_to_cpu(super->active->config_section_length);
699 confsec += super->conf_rec_len) {
700 struct vd_config *vd =
701 (struct vd_config *)((char*)conf + confsec*512);
702 struct vcl *vcl;
703
704 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
705 if (dl->spare)
706 continue;
707 if (posix_memalign((void**)&dl->spare, 512,
708 super->conf_rec_len*512) != 0) {
709 fprintf(stderr, Name
710 ": %s could not allocate spare info buf\n",
711 __func__);
712 return 1;
713 }
714
715 memcpy(dl->spare, vd, super->conf_rec_len*512);
716 continue;
717 }
718 if (vd->magic != DDF_VD_CONF_MAGIC)
719 continue;
720 for (vcl = super->conflist; vcl; vcl = vcl->next) {
721 if (memcmp(vcl->conf.guid,
722 vd->guid, DDF_GUID_LEN) == 0)
723 break;
724 }
725
726 if (vcl) {
727 dl->vlist[vnum++] = vcl;
728 if (__be32_to_cpu(vd->seqnum) <=
729 __be32_to_cpu(vcl->conf.seqnum))
730 continue;
731 } else {
732 if (posix_memalign((void**)&vcl, 512,
733 (super->conf_rec_len*512 +
734 offsetof(struct vcl, conf))) != 0) {
735 fprintf(stderr, Name
736 ": %s could not allocate vcl buf\n",
737 __func__);
738 return 1;
739 }
740 vcl->next = super->conflist;
741 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
742 super->conflist = vcl;
743 dl->vlist[vnum++] = vcl;
744 }
745 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
746 vcl->lba_offset = (__u64*)
747 &vcl->conf.phys_refnum[super->mppe];
748
749 for (i=0; i < max_virt_disks ; i++)
750 if (memcmp(super->virt->entries[i].guid,
751 vcl->conf.guid, DDF_GUID_LEN)==0)
752 break;
753 if (i < max_virt_disks)
754 vcl->vcnum = i;
755 }
756 free(conf);
757
758 return 0;
759 }
760
761 #ifndef MDASSEMBLE
762 static int load_super_ddf_all(struct supertype *st, int fd,
763 void **sbp, char *devname);
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) == 0)
778 return 0;
779 #endif
780
781 if (get_dev_size(fd, devname, &dsize) == 0)
782 return 1;
783
784 if (test_partition(fd))
785 /* DDF is not allowed on partitions */
786 return 1;
787
788 /* 32M is a lower bound */
789 if (dsize <= 32*1024*1024) {
790 if (devname)
791 fprintf(stderr,
792 Name ": %s is too small for ddf: "
793 "size is %llu sectors.\n",
794 devname, dsize>>9);
795 return 1;
796 }
797 if (dsize & 511) {
798 if (devname)
799 fprintf(stderr,
800 Name ": %s is an odd size for ddf: "
801 "size is %llu bytes.\n",
802 devname, dsize);
803 return 1;
804 }
805
806 free_super_ddf(st);
807
808 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
809 fprintf(stderr, Name ": malloc of %zu failed.\n",
810 sizeof(*super));
811 return 1;
812 }
813 memset(super, 0, sizeof(*super));
814
815 rv = load_ddf_headers(fd, super, devname);
816 if (rv) {
817 free(super);
818 return rv;
819 }
820
821 /* Have valid headers and have chosen the best. Let's read in the rest*/
822
823 rv = load_ddf_global(fd, super, devname);
824
825 if (rv) {
826 if (devname)
827 fprintf(stderr,
828 Name ": Failed to load all information "
829 "sections on %s\n", devname);
830 free(super);
831 return rv;
832 }
833
834 rv = load_ddf_local(fd, super, devname, 0);
835
836 if (rv) {
837 if (devname)
838 fprintf(stderr,
839 Name ": Failed to load all information "
840 "sections on %s\n", devname);
841 free(super);
842 return rv;
843 }
844
845 /* Should possibly check the sections .... */
846
847 st->sb = super;
848 if (st->ss == NULL) {
849 st->ss = &super_ddf;
850 st->minor_version = 0;
851 st->max_devs = 512;
852 }
853 st->loaded_container = 0;
854 return 0;
855
856 }
857
858 static void free_super_ddf(struct supertype *st)
859 {
860 struct ddf_super *ddf = st->sb;
861 if (ddf == NULL)
862 return;
863 free(ddf->phys);
864 free(ddf->virt);
865 while (ddf->conflist) {
866 struct vcl *v = ddf->conflist;
867 ddf->conflist = v->next;
868 if (v->block_sizes)
869 free(v->block_sizes);
870 free(v);
871 }
872 while (ddf->dlist) {
873 struct dl *d = ddf->dlist;
874 ddf->dlist = d->next;
875 if (d->fd >= 0)
876 close(d->fd);
877 if (d->spare)
878 free(d->spare);
879 free(d);
880 }
881 free(ddf);
882 st->sb = NULL;
883 }
884
885 static struct supertype *match_metadata_desc_ddf(char *arg)
886 {
887 /* 'ddf' only support containers */
888 struct supertype *st;
889 if (strcmp(arg, "ddf") != 0 &&
890 strcmp(arg, "default") != 0
891 )
892 return NULL;
893
894 st = malloc(sizeof(*st));
895 memset(st, 0, sizeof(*st));
896 st->container_dev = NoMdDev;
897 st->ss = &super_ddf;
898 st->max_devs = 512;
899 st->minor_version = 0;
900 st->sb = NULL;
901 return st;
902 }
903
904
905 #ifndef MDASSEMBLE
906
907 static mapping_t ddf_state[] = {
908 { "Optimal", 0},
909 { "Degraded", 1},
910 { "Deleted", 2},
911 { "Missing", 3},
912 { "Failed", 4},
913 { "Partially Optimal", 5},
914 { "-reserved-", 6},
915 { "-reserved-", 7},
916 { NULL, 0}
917 };
918
919 static mapping_t ddf_init_state[] = {
920 { "Not Initialised", 0},
921 { "QuickInit in Progress", 1},
922 { "Fully Initialised", 2},
923 { "*UNKNOWN*", 3},
924 { NULL, 0}
925 };
926 static mapping_t ddf_access[] = {
927 { "Read/Write", 0},
928 { "Reserved", 1},
929 { "Read Only", 2},
930 { "Blocked (no access)", 3},
931 { NULL ,0}
932 };
933
934 static mapping_t ddf_level[] = {
935 { "RAID0", DDF_RAID0},
936 { "RAID1", DDF_RAID1},
937 { "RAID3", DDF_RAID3},
938 { "RAID4", DDF_RAID4},
939 { "RAID5", DDF_RAID5},
940 { "RAID1E",DDF_RAID1E},
941 { "JBOD", DDF_JBOD},
942 { "CONCAT",DDF_CONCAT},
943 { "RAID5E",DDF_RAID5E},
944 { "RAID5EE",DDF_RAID5EE},
945 { "RAID6", DDF_RAID6},
946 { NULL, 0}
947 };
948 static mapping_t ddf_sec_level[] = {
949 { "Striped", DDF_2STRIPED},
950 { "Mirrored", DDF_2MIRRORED},
951 { "Concat", DDF_2CONCAT},
952 { "Spanned", DDF_2SPANNED},
953 { NULL, 0}
954 };
955 #endif
956
957 struct num_mapping {
958 int num1, num2;
959 };
960 static struct num_mapping ddf_level_num[] = {
961 { DDF_RAID0, 0 },
962 { DDF_RAID1, 1 },
963 { DDF_RAID3, LEVEL_UNSUPPORTED },
964 { DDF_RAID4, 4 },
965 { DDF_RAID5, 5 },
966 { DDF_RAID1E, LEVEL_UNSUPPORTED },
967 { DDF_JBOD, LEVEL_UNSUPPORTED },
968 { DDF_CONCAT, LEVEL_LINEAR },
969 { DDF_RAID5E, LEVEL_UNSUPPORTED },
970 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
971 { DDF_RAID6, 6},
972 { MAXINT, MAXINT }
973 };
974
975 static int map_num1(struct num_mapping *map, int num)
976 {
977 int i;
978 for (i=0 ; map[i].num1 != MAXINT; i++)
979 if (map[i].num1 == num)
980 break;
981 return map[i].num2;
982 }
983
984 static int all_ff(char *guid)
985 {
986 int i;
987 for (i = 0; i < DDF_GUID_LEN; i++)
988 if (guid[i] != (char)0xff)
989 return 0;
990 return 1;
991 }
992
993 #ifndef MDASSEMBLE
994 static void print_guid(char *guid, int tstamp)
995 {
996 /* A GUIDs are part (or all) ASCII and part binary.
997 * They tend to be space padded.
998 * We print the GUID in HEX, then in parentheses add
999 * any initial ASCII sequence, and a possible
1000 * time stamp from bytes 16-19
1001 */
1002 int l = DDF_GUID_LEN;
1003 int i;
1004
1005 for (i=0 ; i<DDF_GUID_LEN ; i++) {
1006 if ((i&3)==0 && i != 0) printf(":");
1007 printf("%02X", guid[i]&255);
1008 }
1009
1010 printf("\n (");
1011 while (l && guid[l-1] == ' ')
1012 l--;
1013 for (i=0 ; i<l ; i++) {
1014 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1015 fputc(guid[i], stdout);
1016 else
1017 break;
1018 }
1019 if (tstamp) {
1020 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1021 char tbuf[100];
1022 struct tm *tm;
1023 tm = localtime(&then);
1024 strftime(tbuf, 100, " %D %T",tm);
1025 fputs(tbuf, stdout);
1026 }
1027 printf(")");
1028 }
1029
1030 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1031 {
1032 int crl = sb->conf_rec_len;
1033 struct vcl *vcl;
1034
1035 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1036 unsigned int i;
1037 struct vd_config *vc = &vcl->conf;
1038
1039 if (calc_crc(vc, crl*512) != vc->crc)
1040 continue;
1041 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1042 continue;
1043
1044 /* Ok, we know about this VD, let's give more details */
1045 printf(" Raid Devices[%d] : %d (", n,
1046 __be16_to_cpu(vc->prim_elmnt_count));
1047 for (i = 0; i < __be16_to_cpu(vc->prim_elmnt_count); i++) {
1048 int j;
1049 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1050 for (j=0; j<cnt; j++)
1051 if (vc->phys_refnum[i] == sb->phys->entries[j].refnum)
1052 break;
1053 if (i) printf(" ");
1054 if (j < cnt)
1055 printf("%d", j);
1056 else
1057 printf("--");
1058 }
1059 printf(")\n");
1060 if (vc->chunk_shift != 255)
1061 printf(" Chunk Size[%d] : %d sectors\n", n,
1062 1 << vc->chunk_shift);
1063 printf(" Raid Level[%d] : %s\n", n,
1064 map_num(ddf_level, vc->prl)?:"-unknown-");
1065 if (vc->sec_elmnt_count != 1) {
1066 printf(" Secondary Position[%d] : %d of %d\n", n,
1067 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1068 printf(" Secondary Level[%d] : %s\n", n,
1069 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1070 }
1071 printf(" Device Size[%d] : %llu\n", n,
1072 (unsigned long long)__be64_to_cpu(vc->blocks)/2);
1073 printf(" Array Size[%d] : %llu\n", n,
1074 (unsigned long long)__be64_to_cpu(vc->array_blocks)/2);
1075 }
1076 }
1077
1078 static void examine_vds(struct ddf_super *sb)
1079 {
1080 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1081 int i;
1082 printf(" Virtual Disks : %d\n", cnt);
1083
1084 for (i=0; i<cnt; i++) {
1085 struct virtual_entry *ve = &sb->virt->entries[i];
1086 printf("\n");
1087 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1088 printf("\n");
1089 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1090 printf(" state[%d] : %s, %s%s\n", i,
1091 map_num(ddf_state, ve->state & 7),
1092 (ve->state & 8) ? "Morphing, ": "",
1093 (ve->state & 16)? "Not Consistent" : "Consistent");
1094 printf(" init state[%d] : %s\n", i,
1095 map_num(ddf_init_state, ve->init_state&3));
1096 printf(" access[%d] : %s\n", i,
1097 map_num(ddf_access, (ve->init_state>>6) & 3));
1098 printf(" Name[%d] : %.16s\n", i, ve->name);
1099 examine_vd(i, sb, ve->guid);
1100 }
1101 if (cnt) printf("\n");
1102 }
1103
1104 static void examine_pds(struct ddf_super *sb)
1105 {
1106 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1107 int i;
1108 struct dl *dl;
1109 printf(" Physical Disks : %d\n", cnt);
1110 printf(" Number RefNo Size Device Type/State\n");
1111
1112 for (i=0 ; i<cnt ; i++) {
1113 struct phys_disk_entry *pd = &sb->phys->entries[i];
1114 int type = __be16_to_cpu(pd->type);
1115 int state = __be16_to_cpu(pd->state);
1116
1117 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1118 //printf("\n");
1119 printf(" %3d %08x ", i,
1120 __be32_to_cpu(pd->refnum));
1121 printf("%8lluK ",
1122 (unsigned long long)__be64_to_cpu(pd->config_size)>>1);
1123 for (dl = sb->dlist; dl ; dl = dl->next) {
1124 if (dl->disk.refnum == pd->refnum) {
1125 char *dv = map_dev(dl->major, dl->minor, 0);
1126 if (dv) {
1127 printf("%-15s", dv);
1128 break;
1129 }
1130 }
1131 }
1132 if (!dl)
1133 printf("%15s","");
1134 printf(" %s%s%s%s%s",
1135 (type&2) ? "active":"",
1136 (type&4) ? "Global-Spare":"",
1137 (type&8) ? "spare" : "",
1138 (type&16)? ", foreign" : "",
1139 (type&32)? "pass-through" : "");
1140 printf("/%s%s%s%s%s%s%s",
1141 (state&1)? "Online": "Offline",
1142 (state&2)? ", Failed": "",
1143 (state&4)? ", Rebuilding": "",
1144 (state&8)? ", in-transition": "",
1145 (state&16)? ", SMART-errors": "",
1146 (state&32)? ", Unrecovered-Read-Errors": "",
1147 (state&64)? ", Missing" : "");
1148 printf("\n");
1149 }
1150 }
1151
1152 static void examine_super_ddf(struct supertype *st, char *homehost)
1153 {
1154 struct ddf_super *sb = st->sb;
1155
1156 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1157 printf(" Version : %.8s\n", sb->anchor.revision);
1158 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1159 printf("\n");
1160 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1161 printf("\n");
1162 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1163 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1164 ?"yes" : "no");
1165 examine_vds(sb);
1166 examine_pds(sb);
1167 }
1168
1169 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map);
1170
1171 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
1172
1173 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1174 {
1175 /* We just write a generic DDF ARRAY entry
1176 */
1177 struct mdinfo info;
1178 char nbuf[64];
1179 getinfo_super_ddf(st, &info, NULL);
1180 fname_from_uuid(st, &info, nbuf, ':');
1181
1182 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1183 }
1184
1185 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1186 {
1187 /* We just write a generic DDF ARRAY entry
1188 */
1189 struct ddf_super *ddf = st->sb;
1190 struct mdinfo info;
1191 unsigned int i;
1192 char nbuf[64];
1193 getinfo_super_ddf(st, &info, NULL);
1194 fname_from_uuid(st, &info, nbuf, ':');
1195
1196 for (i = 0; i < __be16_to_cpu(ddf->virt->max_vdes); i++) {
1197 struct virtual_entry *ve = &ddf->virt->entries[i];
1198 struct vcl vcl;
1199 char nbuf1[64];
1200 if (all_ff(ve->guid))
1201 continue;
1202 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1203 ddf->currentconf =&vcl;
1204 uuid_from_super_ddf(st, info.uuid);
1205 fname_from_uuid(st, &info, nbuf1, ':');
1206 printf("ARRAY container=%s member=%d UUID=%s\n",
1207 nbuf+5, i, nbuf1+5);
1208 }
1209 }
1210
1211 static void export_examine_super_ddf(struct supertype *st)
1212 {
1213 struct mdinfo info;
1214 char nbuf[64];
1215 getinfo_super_ddf(st, &info, NULL);
1216 fname_from_uuid(st, &info, nbuf, ':');
1217 printf("MD_METADATA=ddf\n");
1218 printf("MD_LEVEL=container\n");
1219 printf("MD_UUID=%s\n", nbuf+5);
1220 }
1221
1222
1223 static void detail_super_ddf(struct supertype *st, char *homehost)
1224 {
1225 /* FIXME later
1226 * Could print DDF GUID
1227 * Need to find which array
1228 * If whole, briefly list all arrays
1229 * If one, give name
1230 */
1231 }
1232
1233 static void brief_detail_super_ddf(struct supertype *st)
1234 {
1235 /* FIXME I really need to know which array we are detailing.
1236 * Can that be stored in ddf_super??
1237 */
1238 // struct ddf_super *ddf = st->sb;
1239 struct mdinfo info;
1240 char nbuf[64];
1241 getinfo_super_ddf(st, &info, NULL);
1242 fname_from_uuid(st, &info, nbuf,':');
1243 printf(" UUID=%s", nbuf + 5);
1244 }
1245 #endif
1246
1247 static int match_home_ddf(struct supertype *st, char *homehost)
1248 {
1249 /* It matches 'this' host if the controller is a
1250 * Linux-MD controller with vendor_data matching
1251 * the hostname
1252 */
1253 struct ddf_super *ddf = st->sb;
1254 unsigned int len;
1255
1256 if (!homehost)
1257 return 0;
1258 len = strlen(homehost);
1259
1260 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1261 len < sizeof(ddf->controller.vendor_data) &&
1262 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1263 ddf->controller.vendor_data[len] == 0);
1264 }
1265
1266 #ifndef MDASSEMBLE
1267 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst)
1268 {
1269 struct vcl *v;
1270
1271 for (v = ddf->conflist; v; v = v->next)
1272 if (inst == v->vcnum)
1273 return &v->conf;
1274 return NULL;
1275 }
1276 #endif
1277
1278 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1279 {
1280 /* Find the entry in phys_disk which has the given refnum
1281 * and return it's index
1282 */
1283 unsigned int i;
1284 for (i = 0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1285 if (ddf->phys->entries[i].refnum == phys_refnum)
1286 return i;
1287 return -1;
1288 }
1289
1290 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1291 {
1292 /* The uuid returned here is used for:
1293 * uuid to put into bitmap file (Create, Grow)
1294 * uuid for backup header when saving critical section (Grow)
1295 * comparing uuids when re-adding a device into an array
1296 * In these cases the uuid required is that of the data-array,
1297 * not the device-set.
1298 * uuid to recognise same set when adding a missing device back
1299 * to an array. This is a uuid for the device-set.
1300 *
1301 * For each of these we can make do with a truncated
1302 * or hashed uuid rather than the original, as long as
1303 * everyone agrees.
1304 * In the case of SVD we assume the BVD is of interest,
1305 * though that might be the case if a bitmap were made for
1306 * a mirrored SVD - worry about that later.
1307 * So we need to find the VD configuration record for the
1308 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1309 * The first 16 bytes of the sha1 of these is used.
1310 */
1311 struct ddf_super *ddf = st->sb;
1312 struct vcl *vcl = ddf->currentconf;
1313 char *guid;
1314 char buf[20];
1315 struct sha1_ctx ctx;
1316
1317 if (vcl)
1318 guid = vcl->conf.guid;
1319 else
1320 guid = ddf->anchor.guid;
1321
1322 sha1_init_ctx(&ctx);
1323 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1324 sha1_finish_ctx(&ctx, buf);
1325 memcpy(uuid, buf, 4*4);
1326 }
1327
1328 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map);
1329
1330 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1331 {
1332 struct ddf_super *ddf = st->sb;
1333 int map_disks = info->array.raid_disks;
1334
1335 if (ddf->currentconf) {
1336 getinfo_super_ddf_bvd(st, info, map);
1337 return;
1338 }
1339
1340 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1341 info->array.level = LEVEL_CONTAINER;
1342 info->array.layout = 0;
1343 info->array.md_minor = -1;
1344 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1345 (ddf->anchor.guid+16));
1346 info->array.utime = 0;
1347 info->array.chunk_size = 0;
1348 info->container_enough = 1;
1349
1350
1351 info->disk.major = 0;
1352 info->disk.minor = 0;
1353 if (ddf->dlist) {
1354 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1355 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1356
1357 info->data_offset = __be64_to_cpu(ddf->phys->
1358 entries[info->disk.raid_disk].
1359 config_size);
1360 info->component_size = ddf->dlist->size - info->data_offset;
1361 } else {
1362 info->disk.number = -1;
1363 info->disk.raid_disk = -1;
1364 // info->disk.raid_disk = find refnum in the table and use index;
1365 }
1366 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
1367
1368
1369 info->recovery_start = MaxSector;
1370 info->reshape_active = 0;
1371 info->name[0] = 0;
1372
1373 info->array.major_version = -1;
1374 info->array.minor_version = -2;
1375 strcpy(info->text_version, "ddf");
1376 info->safe_mode_delay = 0;
1377
1378 uuid_from_super_ddf(st, info->uuid);
1379
1380 if (map) {
1381 int i;
1382 for (i = 0 ; i < map_disks; i++) {
1383 if (i < info->array.raid_disks &&
1384 (__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Online) &&
1385 !(__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Failed))
1386 map[i] = 1;
1387 else
1388 map[i] = 0;
1389 }
1390 }
1391 }
1392
1393 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1394
1395 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
1396 {
1397 struct ddf_super *ddf = st->sb;
1398 struct vcl *vc = ddf->currentconf;
1399 int cd = ddf->currentdev;
1400 int j;
1401 struct dl *dl;
1402 int map_disks = info->array.raid_disks;
1403
1404 /* FIXME this returns BVD info - what if we want SVD ?? */
1405
1406 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1407 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1408 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1409 info->array.raid_disks);
1410 info->array.md_minor = -1;
1411 info->array.ctime = DECADE +
1412 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1413 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1414 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1415 info->custom_array_size = 0;
1416
1417 if (cd >= 0 && (unsigned)cd < ddf->mppe) {
1418 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1419 if (vc->block_sizes)
1420 info->component_size = vc->block_sizes[cd];
1421 else
1422 info->component_size = __be64_to_cpu(vc->conf.blocks);
1423 }
1424
1425 for (dl = ddf->dlist; dl ; dl = dl->next)
1426 if (dl->raiddisk == info->disk.raid_disk)
1427 break;
1428 info->disk.major = 0;
1429 info->disk.minor = 0;
1430 if (dl) {
1431 info->disk.major = dl->major;
1432 info->disk.minor = dl->minor;
1433 }
1434 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1435 // info->disk.raid_disk = find refnum in the table and use index;
1436 // info->disk.state = ???;
1437
1438 info->container_member = ddf->currentconf->vcnum;
1439
1440 info->recovery_start = MaxSector;
1441 info->resync_start = 0;
1442 info->reshape_active = 0;
1443 if (!(ddf->virt->entries[info->container_member].state
1444 & DDF_state_inconsistent) &&
1445 (ddf->virt->entries[info->container_member].init_state
1446 & DDF_initstate_mask)
1447 == DDF_init_full)
1448 info->resync_start = MaxSector;
1449
1450 uuid_from_super_ddf(st, info->uuid);
1451
1452 info->array.major_version = -1;
1453 info->array.minor_version = -2;
1454 sprintf(info->text_version, "/%s/%d",
1455 devnum2devname(st->container_dev),
1456 info->container_member);
1457 info->safe_mode_delay = 200;
1458
1459 memcpy(info->name, ddf->virt->entries[info->container_member].name, 16);
1460 info->name[16]=0;
1461 for(j=0; j<16; j++)
1462 if (info->name[j] == ' ')
1463 info->name[j] = 0;
1464
1465 if (map)
1466 for (j = 0; j < map_disks; j++) {
1467 map[j] = 0;
1468 if (j < info->array.raid_disks) {
1469 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
1470 if (i >= 0 &&
1471 (__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Online) &&
1472 !(__be16_to_cpu(ddf->phys->entries[i].state) & DDF_Failed))
1473 map[i] = 1;
1474 }
1475 }
1476 }
1477
1478
1479 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1480 char *update,
1481 char *devname, int verbose,
1482 int uuid_set, char *homehost)
1483 {
1484 /* For 'assemble' and 'force' we need to return non-zero if any
1485 * change was made. For others, the return value is ignored.
1486 * Update options are:
1487 * force-one : This device looks a bit old but needs to be included,
1488 * update age info appropriately.
1489 * assemble: clear any 'faulty' flag to allow this device to
1490 * be assembled.
1491 * force-array: Array is degraded but being forced, mark it clean
1492 * if that will be needed to assemble it.
1493 *
1494 * newdev: not used ????
1495 * grow: Array has gained a new device - this is currently for
1496 * linear only
1497 * resync: mark as dirty so a resync will happen.
1498 * uuid: Change the uuid of the array to match what is given
1499 * homehost: update the recorded homehost
1500 * name: update the name - preserving the homehost
1501 * _reshape_progress: record new reshape_progress position.
1502 *
1503 * Following are not relevant for this version:
1504 * sparc2.2 : update from old dodgey metadata
1505 * super-minor: change the preferred_minor number
1506 * summaries: update redundant counters.
1507 */
1508 int rv = 0;
1509 // struct ddf_super *ddf = st->sb;
1510 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1511 // struct virtual_entry *ve = find_ve(ddf);
1512
1513 /* we don't need to handle "force-*" or "assemble" as
1514 * there is no need to 'trick' the kernel. We the metadata is
1515 * first updated to activate the array, all the implied modifications
1516 * will just happen.
1517 */
1518
1519 if (strcmp(update, "grow") == 0) {
1520 /* FIXME */
1521 } else if (strcmp(update, "resync") == 0) {
1522 // info->resync_checkpoint = 0;
1523 } else if (strcmp(update, "homehost") == 0) {
1524 /* homehost is stored in controller->vendor_data,
1525 * or it is when we are the vendor
1526 */
1527 // if (info->vendor_is_local)
1528 // strcpy(ddf->controller.vendor_data, homehost);
1529 rv = -1;
1530 } if (strcmp(update, "name") == 0) {
1531 /* name is stored in virtual_entry->name */
1532 // memset(ve->name, ' ', 16);
1533 // strncpy(ve->name, info->name, 16);
1534 rv = -1;
1535 } if (strcmp(update, "_reshape_progress") == 0) {
1536 /* We don't support reshape yet */
1537 } else
1538 rv = -1;
1539
1540 // update_all_csum(ddf);
1541
1542 return rv;
1543 }
1544
1545 static void make_header_guid(char *guid)
1546 {
1547 __u32 stamp;
1548 /* Create a DDF Header of Virtual Disk GUID */
1549
1550 /* 24 bytes of fiction required.
1551 * first 8 are a 'vendor-id' - "Linux-MD"
1552 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1553 * Remaining 8 random number plus timestamp
1554 */
1555 memcpy(guid, T10, sizeof(T10));
1556 stamp = __cpu_to_be32(0xdeadbeef);
1557 memcpy(guid+8, &stamp, 4);
1558 stamp = __cpu_to_be32(0);
1559 memcpy(guid+12, &stamp, 4);
1560 stamp = __cpu_to_be32(time(0) - DECADE);
1561 memcpy(guid+16, &stamp, 4);
1562 stamp = random32();
1563 memcpy(guid+20, &stamp, 4);
1564 }
1565
1566 static int init_super_ddf_bvd(struct supertype *st,
1567 mdu_array_info_t *info,
1568 unsigned long long size,
1569 char *name, char *homehost,
1570 int *uuid);
1571
1572 static int init_super_ddf(struct supertype *st,
1573 mdu_array_info_t *info,
1574 unsigned long long size, char *name, char *homehost,
1575 int *uuid)
1576 {
1577 /* This is primarily called by Create when creating a new array.
1578 * We will then get add_to_super called for each component, and then
1579 * write_init_super called to write it out to each device.
1580 * For DDF, Create can create on fresh devices or on a pre-existing
1581 * array.
1582 * To create on a pre-existing array a different method will be called.
1583 * This one is just for fresh drives.
1584 *
1585 * We need to create the entire 'ddf' structure which includes:
1586 * DDF headers - these are easy.
1587 * Controller data - a Sector describing this controller .. not that
1588 * this is a controller exactly.
1589 * Physical Disk Record - one entry per device, so
1590 * leave plenty of space.
1591 * Virtual Disk Records - again, just leave plenty of space.
1592 * This just lists VDs, doesn't give details
1593 * Config records - describes the VDs that use this disk
1594 * DiskData - describes 'this' device.
1595 * BadBlockManagement - empty
1596 * Diag Space - empty
1597 * Vendor Logs - Could we put bitmaps here?
1598 *
1599 */
1600 struct ddf_super *ddf;
1601 char hostname[17];
1602 int hostlen;
1603 int max_phys_disks, max_virt_disks;
1604 unsigned long long sector;
1605 int clen;
1606 int i;
1607 int pdsize, vdsize;
1608 struct phys_disk *pd;
1609 struct virtual_disk *vd;
1610
1611 if (st->sb)
1612 return init_super_ddf_bvd(st, info, size, name, homehost, uuid);
1613
1614 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
1615 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
1616 return 0;
1617 }
1618 memset(ddf, 0, sizeof(*ddf));
1619 ddf->dlist = NULL; /* no physical disks yet */
1620 ddf->conflist = NULL; /* No virtual disks yet */
1621 st->sb = ddf;
1622
1623 if (info == NULL) {
1624 /* zeroing superblock */
1625 return 0;
1626 }
1627
1628 /* At least 32MB *must* be reserved for the ddf. So let's just
1629 * start 32MB from the end, and put the primary header there.
1630 * Don't do secondary for now.
1631 * We don't know exactly where that will be yet as it could be
1632 * different on each device. To just set up the lengths.
1633 *
1634 */
1635
1636 ddf->anchor.magic = DDF_HEADER_MAGIC;
1637 make_header_guid(ddf->anchor.guid);
1638
1639 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1640 ddf->anchor.seq = __cpu_to_be32(1);
1641 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1642 ddf->anchor.openflag = 0xFF;
1643 ddf->anchor.foreignflag = 0;
1644 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1645 ddf->anchor.pad0 = 0xff;
1646 memset(ddf->anchor.pad1, 0xff, 12);
1647 memset(ddf->anchor.header_ext, 0xff, 32);
1648 ddf->anchor.primary_lba = ~(__u64)0;
1649 ddf->anchor.secondary_lba = ~(__u64)0;
1650 ddf->anchor.type = DDF_HEADER_ANCHOR;
1651 memset(ddf->anchor.pad2, 0xff, 3);
1652 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1653 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1654 of 32M reserved.. */
1655 max_phys_disks = 1023; /* Should be enough */
1656 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1657 max_virt_disks = 255;
1658 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1659 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1660 ddf->max_part = 64;
1661 ddf->mppe = 256;
1662 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1663 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1664 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1665 memset(ddf->anchor.pad3, 0xff, 54);
1666 /* controller sections is one sector long immediately
1667 * after the ddf header */
1668 sector = 1;
1669 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1670 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1671 sector += 1;
1672
1673 /* phys is 8 sectors after that */
1674 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1675 sizeof(struct phys_disk_entry)*max_phys_disks,
1676 512);
1677 switch(pdsize/512) {
1678 case 2: case 8: case 32: case 128: case 512: break;
1679 default: abort();
1680 }
1681 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1682 ddf->anchor.phys_section_length =
1683 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1684 sector += pdsize/512;
1685
1686 /* virt is another 32 sectors */
1687 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1688 sizeof(struct virtual_entry) * max_virt_disks,
1689 512);
1690 switch(vdsize/512) {
1691 case 2: case 8: case 32: case 128: case 512: break;
1692 default: abort();
1693 }
1694 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1695 ddf->anchor.virt_section_length =
1696 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1697 sector += vdsize/512;
1698
1699 clen = ddf->conf_rec_len * (ddf->max_part+1);
1700 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1701 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1702 sector += clen;
1703
1704 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1705 ddf->anchor.data_section_length = __cpu_to_be32(1);
1706 sector += 1;
1707
1708 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1709 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1710 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1711 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1712 ddf->anchor.vendor_length = __cpu_to_be32(0);
1713 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1714
1715 memset(ddf->anchor.pad4, 0xff, 256);
1716
1717 memcpy(&ddf->primary, &ddf->anchor, 512);
1718 memcpy(&ddf->secondary, &ddf->anchor, 512);
1719
1720 ddf->primary.openflag = 1; /* I guess.. */
1721 ddf->primary.type = DDF_HEADER_PRIMARY;
1722
1723 ddf->secondary.openflag = 1; /* I guess.. */
1724 ddf->secondary.type = DDF_HEADER_SECONDARY;
1725
1726 ddf->active = &ddf->primary;
1727
1728 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1729
1730 /* 24 more bytes of fiction required.
1731 * first 8 are a 'vendor-id' - "Linux-MD"
1732 * Remaining 16 are serial number.... maybe a hostname would do?
1733 */
1734 memcpy(ddf->controller.guid, T10, sizeof(T10));
1735 gethostname(hostname, sizeof(hostname));
1736 hostname[sizeof(hostname) - 1] = 0;
1737 hostlen = strlen(hostname);
1738 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1739 for (i = strlen(T10) ; i+hostlen < 24; i++)
1740 ddf->controller.guid[i] = ' ';
1741
1742 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1743 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1744 ddf->controller.type.sub_vendor_id = 0;
1745 ddf->controller.type.sub_device_id = 0;
1746 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1747 memset(ddf->controller.pad, 0xff, 8);
1748 memset(ddf->controller.vendor_data, 0xff, 448);
1749 if (homehost && strlen(homehost) < 440)
1750 strcpy((char*)ddf->controller.vendor_data, homehost);
1751
1752 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
1753 fprintf(stderr, Name ": %s could not allocate pd\n", __func__);
1754 return 0;
1755 }
1756 ddf->phys = pd;
1757 ddf->pdsize = pdsize;
1758
1759 memset(pd, 0xff, pdsize);
1760 memset(pd, 0, sizeof(*pd));
1761 pd->magic = DDF_PHYS_RECORDS_MAGIC;
1762 pd->used_pdes = __cpu_to_be16(0);
1763 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1764 memset(pd->pad, 0xff, 52);
1765
1766 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
1767 fprintf(stderr, Name ": %s could not allocate vd\n", __func__);
1768 return 0;
1769 }
1770 ddf->virt = vd;
1771 ddf->vdsize = vdsize;
1772 memset(vd, 0, vdsize);
1773 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1774 vd->populated_vdes = __cpu_to_be16(0);
1775 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1776 memset(vd->pad, 0xff, 52);
1777
1778 for (i=0; i<max_virt_disks; i++)
1779 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1780
1781 st->sb = ddf;
1782 ddf->updates_pending = 1;
1783 return 1;
1784 }
1785
1786 static int chunk_to_shift(int chunksize)
1787 {
1788 return ffs(chunksize/512)-1;
1789 }
1790
1791 static int level_to_prl(int level)
1792 {
1793 switch (level) {
1794 case LEVEL_LINEAR: return DDF_CONCAT;
1795 case 0: return DDF_RAID0;
1796 case 1: return DDF_RAID1;
1797 case 4: return DDF_RAID4;
1798 case 5: return DDF_RAID5;
1799 case 6: return DDF_RAID6;
1800 default: return -1;
1801 }
1802 }
1803 static int layout_to_rlq(int level, int layout, int raiddisks)
1804 {
1805 switch(level) {
1806 case 0:
1807 return DDF_RAID0_SIMPLE;
1808 case 1:
1809 switch(raiddisks) {
1810 case 2: return DDF_RAID1_SIMPLE;
1811 case 3: return DDF_RAID1_MULTI;
1812 default: return -1;
1813 }
1814 case 4:
1815 switch(layout) {
1816 case 0: return DDF_RAID4_N;
1817 }
1818 break;
1819 case 5:
1820 switch(layout) {
1821 case ALGORITHM_LEFT_ASYMMETRIC:
1822 return DDF_RAID5_N_RESTART;
1823 case ALGORITHM_RIGHT_ASYMMETRIC:
1824 return DDF_RAID5_0_RESTART;
1825 case ALGORITHM_LEFT_SYMMETRIC:
1826 return DDF_RAID5_N_CONTINUE;
1827 case ALGORITHM_RIGHT_SYMMETRIC:
1828 return -1; /* not mentioned in standard */
1829 }
1830 case 6:
1831 switch(layout) {
1832 case ALGORITHM_ROTATING_N_RESTART:
1833 return DDF_RAID5_N_RESTART;
1834 case ALGORITHM_ROTATING_ZERO_RESTART:
1835 return DDF_RAID6_0_RESTART;
1836 case ALGORITHM_ROTATING_N_CONTINUE:
1837 return DDF_RAID5_N_CONTINUE;
1838 }
1839 }
1840 return -1;
1841 }
1842
1843 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1844 {
1845 switch(prl) {
1846 case DDF_RAID0:
1847 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1848 case DDF_RAID1:
1849 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1850 on raiddisks*/
1851 case DDF_RAID4:
1852 switch(rlq) {
1853 case DDF_RAID4_N:
1854 return 0;
1855 default:
1856 /* not supported */
1857 return -1; /* FIXME this isn't checked */
1858 }
1859 case DDF_RAID5:
1860 switch(rlq) {
1861 case DDF_RAID5_N_RESTART:
1862 return ALGORITHM_LEFT_ASYMMETRIC;
1863 case DDF_RAID5_0_RESTART:
1864 return ALGORITHM_RIGHT_ASYMMETRIC;
1865 case DDF_RAID5_N_CONTINUE:
1866 return ALGORITHM_LEFT_SYMMETRIC;
1867 default:
1868 return -1;
1869 }
1870 case DDF_RAID6:
1871 switch(rlq) {
1872 case DDF_RAID5_N_RESTART:
1873 return ALGORITHM_ROTATING_N_RESTART;
1874 case DDF_RAID6_0_RESTART:
1875 return ALGORITHM_ROTATING_ZERO_RESTART;
1876 case DDF_RAID5_N_CONTINUE:
1877 return ALGORITHM_ROTATING_N_CONTINUE;
1878 default:
1879 return -1;
1880 }
1881 }
1882 return -1;
1883 }
1884
1885 #ifndef MDASSEMBLE
1886 struct extent {
1887 unsigned long long start, size;
1888 };
1889 static int cmp_extent(const void *av, const void *bv)
1890 {
1891 const struct extent *a = av;
1892 const struct extent *b = bv;
1893 if (a->start < b->start)
1894 return -1;
1895 if (a->start > b->start)
1896 return 1;
1897 return 0;
1898 }
1899
1900 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1901 {
1902 /* find a list of used extents on the give physical device
1903 * (dnum) of the given ddf.
1904 * Return a malloced array of 'struct extent'
1905
1906 FIXME ignore DDF_Legacy devices?
1907
1908 */
1909 struct extent *rv;
1910 int n = 0;
1911 unsigned int i, j;
1912
1913 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1914 if (!rv)
1915 return NULL;
1916
1917 for (i = 0; i < ddf->max_part; i++) {
1918 struct vcl *v = dl->vlist[i];
1919 if (v == NULL)
1920 continue;
1921 for (j = 0; j < v->conf.prim_elmnt_count; j++)
1922 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1923 /* This device plays role 'j' in 'v'. */
1924 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1925 rv[n].size = __be64_to_cpu(v->conf.blocks);
1926 n++;
1927 break;
1928 }
1929 }
1930 qsort(rv, n, sizeof(*rv), cmp_extent);
1931
1932 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1933 rv[n].size = 0;
1934 return rv;
1935 }
1936 #endif
1937
1938 static int init_super_ddf_bvd(struct supertype *st,
1939 mdu_array_info_t *info,
1940 unsigned long long size,
1941 char *name, char *homehost,
1942 int *uuid)
1943 {
1944 /* We are creating a BVD inside a pre-existing container.
1945 * so st->sb is already set.
1946 * We need to create a new vd_config and a new virtual_entry
1947 */
1948 struct ddf_super *ddf = st->sb;
1949 unsigned int venum;
1950 struct virtual_entry *ve;
1951 struct vcl *vcl;
1952 struct vd_config *vc;
1953
1954 if (__be16_to_cpu(ddf->virt->populated_vdes)
1955 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1956 fprintf(stderr, Name": This ddf already has the "
1957 "maximum of %d virtual devices\n",
1958 __be16_to_cpu(ddf->virt->max_vdes));
1959 return 0;
1960 }
1961
1962 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1963 if (all_ff(ddf->virt->entries[venum].guid))
1964 break;
1965 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1966 fprintf(stderr, Name ": Cannot find spare slot for "
1967 "virtual disk - DDF is corrupt\n");
1968 return 0;
1969 }
1970 ve = &ddf->virt->entries[venum];
1971
1972 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1973 * timestamp, random number
1974 */
1975 make_header_guid(ve->guid);
1976 ve->unit = __cpu_to_be16(info->md_minor);
1977 ve->pad0 = 0xFFFF;
1978 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1979 ve->type = 0;
1980 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1981 if (info->state & 1) /* clean */
1982 ve->init_state = DDF_init_full;
1983 else
1984 ve->init_state = DDF_init_not;
1985
1986 memset(ve->pad1, 0xff, 14);
1987 memset(ve->name, ' ', 16);
1988 if (name)
1989 strncpy(ve->name, name, 16);
1990 ddf->virt->populated_vdes =
1991 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1992
1993 /* Now create a new vd_config */
1994 if (posix_memalign((void**)&vcl, 512,
1995 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
1996 fprintf(stderr, Name ": %s could not allocate vd_config\n", __func__);
1997 return 0;
1998 }
1999 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
2000 vcl->vcnum = venum;
2001 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2002
2003 vc = &vcl->conf;
2004
2005 vc->magic = DDF_VD_CONF_MAGIC;
2006 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2007 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
2008 vc->seqnum = __cpu_to_be32(1);
2009 memset(vc->pad0, 0xff, 24);
2010 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
2011 vc->chunk_shift = chunk_to_shift(info->chunk_size);
2012 vc->prl = level_to_prl(info->level);
2013 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
2014 vc->sec_elmnt_count = 1;
2015 vc->sec_elmnt_seq = 0;
2016 vc->srl = 0;
2017 vc->blocks = __cpu_to_be64(info->size * 2);
2018 vc->array_blocks = __cpu_to_be64(
2019 calc_array_size(info->level, info->raid_disks, info->layout,
2020 info->chunk_size, info->size*2));
2021 memset(vc->pad1, 0xff, 8);
2022 vc->spare_refs[0] = 0xffffffff;
2023 vc->spare_refs[1] = 0xffffffff;
2024 vc->spare_refs[2] = 0xffffffff;
2025 vc->spare_refs[3] = 0xffffffff;
2026 vc->spare_refs[4] = 0xffffffff;
2027 vc->spare_refs[5] = 0xffffffff;
2028 vc->spare_refs[6] = 0xffffffff;
2029 vc->spare_refs[7] = 0xffffffff;
2030 memset(vc->cache_pol, 0, 8);
2031 vc->bg_rate = 0x80;
2032 memset(vc->pad2, 0xff, 3);
2033 memset(vc->pad3, 0xff, 52);
2034 memset(vc->pad4, 0xff, 192);
2035 memset(vc->v0, 0xff, 32);
2036 memset(vc->v1, 0xff, 32);
2037 memset(vc->v2, 0xff, 16);
2038 memset(vc->v3, 0xff, 16);
2039 memset(vc->vendor, 0xff, 32);
2040
2041 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2042 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2043
2044 vcl->next = ddf->conflist;
2045 ddf->conflist = vcl;
2046 ddf->currentconf = vcl;
2047 ddf->updates_pending = 1;
2048 return 1;
2049 }
2050
2051 #ifndef MDASSEMBLE
2052 static void add_to_super_ddf_bvd(struct supertype *st,
2053 mdu_disk_info_t *dk, int fd, char *devname)
2054 {
2055 /* fd and devname identify a device with-in the ddf container (st).
2056 * dk identifies a location in the new BVD.
2057 * We need to find suitable free space in that device and update
2058 * the phys_refnum and lba_offset for the newly created vd_config.
2059 * We might also want to update the type in the phys_disk
2060 * section.
2061 *
2062 * Alternately: fd == -1 and we have already chosen which device to
2063 * use and recorded in dlist->raid_disk;
2064 */
2065 struct dl *dl;
2066 struct ddf_super *ddf = st->sb;
2067 struct vd_config *vc;
2068 __u64 *lba_offset;
2069 unsigned int working;
2070 unsigned int i;
2071 unsigned long long blocks, pos, esize;
2072 struct extent *ex;
2073
2074 if (fd == -1) {
2075 for (dl = ddf->dlist; dl ; dl = dl->next)
2076 if (dl->raiddisk == dk->raid_disk)
2077 break;
2078 } else {
2079 for (dl = ddf->dlist; dl ; dl = dl->next)
2080 if (dl->major == dk->major &&
2081 dl->minor == dk->minor)
2082 break;
2083 }
2084 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
2085 return;
2086
2087 vc = &ddf->currentconf->conf;
2088 lba_offset = ddf->currentconf->lba_offset;
2089
2090 ex = get_extents(ddf, dl);
2091 if (!ex)
2092 return;
2093
2094 i = 0; pos = 0;
2095 blocks = __be64_to_cpu(vc->blocks);
2096 if (ddf->currentconf->block_sizes)
2097 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2098
2099 do {
2100 esize = ex[i].start - pos;
2101 if (esize >= blocks)
2102 break;
2103 pos = ex[i].start + ex[i].size;
2104 i++;
2105 } while (ex[i-1].size);
2106
2107 free(ex);
2108 if (esize < blocks)
2109 return;
2110
2111 ddf->currentdev = dk->raid_disk;
2112 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
2113 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
2114
2115 for (i = 0; i < ddf->max_part ; i++)
2116 if (dl->vlist[i] == NULL)
2117 break;
2118 if (i == ddf->max_part)
2119 return;
2120 dl->vlist[i] = ddf->currentconf;
2121
2122 if (fd >= 0)
2123 dl->fd = fd;
2124 if (devname)
2125 dl->devname = devname;
2126
2127 /* Check how many working raid_disks, and if we can mark
2128 * array as optimal yet
2129 */
2130 working = 0;
2131
2132 for (i = 0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
2133 if (vc->phys_refnum[i] != 0xffffffff)
2134 working++;
2135
2136 /* Find which virtual_entry */
2137 i = ddf->currentconf->vcnum;
2138 if (working == __be16_to_cpu(vc->prim_elmnt_count))
2139 ddf->virt->entries[i].state =
2140 (ddf->virt->entries[i].state & ~DDF_state_mask)
2141 | DDF_state_optimal;
2142
2143 if (vc->prl == DDF_RAID6 &&
2144 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
2145 ddf->virt->entries[i].state =
2146 (ddf->virt->entries[i].state & ~DDF_state_mask)
2147 | DDF_state_part_optimal;
2148
2149 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
2150 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
2151 ddf->updates_pending = 1;
2152 }
2153
2154 /* add a device to a container, either while creating it or while
2155 * expanding a pre-existing container
2156 */
2157 static int add_to_super_ddf(struct supertype *st,
2158 mdu_disk_info_t *dk, int fd, char *devname)
2159 {
2160 struct ddf_super *ddf = st->sb;
2161 struct dl *dd;
2162 time_t now;
2163 struct tm *tm;
2164 unsigned long long size;
2165 struct phys_disk_entry *pde;
2166 unsigned int n, i;
2167 struct stat stb;
2168
2169 if (ddf->currentconf) {
2170 add_to_super_ddf_bvd(st, dk, fd, devname);
2171 return 0;
2172 }
2173
2174 /* This is device numbered dk->number. We need to create
2175 * a phys_disk entry and a more detailed disk_data entry.
2176 */
2177 fstat(fd, &stb);
2178 if (posix_memalign((void**)&dd, 512,
2179 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2180 fprintf(stderr, Name
2181 ": %s could allocate buffer for new disk, aborting\n",
2182 __func__);
2183 return 1;
2184 }
2185 dd->major = major(stb.st_rdev);
2186 dd->minor = minor(stb.st_rdev);
2187 dd->devname = devname;
2188 dd->fd = fd;
2189 dd->spare = NULL;
2190
2191 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2192 now = time(0);
2193 tm = localtime(&now);
2194 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2195 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2196 *(__u32*)(dd->disk.guid + 16) = random32();
2197 *(__u32*)(dd->disk.guid + 20) = random32();
2198
2199 do {
2200 /* Cannot be bothered finding a CRC of some irrelevant details*/
2201 dd->disk.refnum = random32();
2202 for (i = __be16_to_cpu(ddf->active->max_pd_entries);
2203 i > 0; i--)
2204 if (ddf->phys->entries[i-1].refnum == dd->disk.refnum)
2205 break;
2206 } while (i > 0);
2207
2208 dd->disk.forced_ref = 1;
2209 dd->disk.forced_guid = 1;
2210 memset(dd->disk.vendor, ' ', 32);
2211 memcpy(dd->disk.vendor, "Linux", 5);
2212 memset(dd->disk.pad, 0xff, 442);
2213 for (i = 0; i < ddf->max_part ; i++)
2214 dd->vlist[i] = NULL;
2215
2216 n = __be16_to_cpu(ddf->phys->used_pdes);
2217 pde = &ddf->phys->entries[n];
2218 dd->pdnum = n;
2219
2220 if (st->update_tail) {
2221 int len = (sizeof(struct phys_disk) +
2222 sizeof(struct phys_disk_entry));
2223 struct phys_disk *pd;
2224
2225 pd = malloc(len);
2226 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2227 pd->used_pdes = __cpu_to_be16(n);
2228 pde = &pd->entries[0];
2229 dd->mdupdate = pd;
2230 } else {
2231 n++;
2232 ddf->phys->used_pdes = __cpu_to_be16(n);
2233 }
2234
2235 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2236 pde->refnum = dd->disk.refnum;
2237 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2238 pde->state = __cpu_to_be16(DDF_Online);
2239 get_dev_size(fd, NULL, &size);
2240 /* We are required to reserve 32Meg, and record the size in sectors */
2241 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2242 sprintf(pde->path, "%17.17s","Information: nil") ;
2243 memset(pde->pad, 0xff, 6);
2244
2245 dd->size = size >> 9;
2246 if (st->update_tail) {
2247 dd->next = ddf->add_list;
2248 ddf->add_list = dd;
2249 } else {
2250 dd->next = ddf->dlist;
2251 ddf->dlist = dd;
2252 ddf->updates_pending = 1;
2253 }
2254
2255 return 0;
2256 }
2257
2258 /*
2259 * This is the write_init_super method for a ddf container. It is
2260 * called when creating a container or adding another device to a
2261 * container.
2262 */
2263
2264 static unsigned char null_conf[4096+512];
2265
2266 static int __write_init_super_ddf(struct supertype *st, int do_close)
2267 {
2268
2269 struct ddf_super *ddf = st->sb;
2270 int i;
2271 struct dl *d;
2272 int n_config;
2273 int conf_size;
2274 int attempts = 0;
2275 int successes = 0;
2276 unsigned long long size, sector;
2277
2278 /* try to write updated metadata,
2279 * if we catch a failure move on to the next disk
2280 */
2281 for (d = ddf->dlist; d; d=d->next) {
2282 int fd = d->fd;
2283
2284 if (fd < 0)
2285 continue;
2286
2287 attempts++;
2288 /* We need to fill in the primary, (secondary) and workspace
2289 * lba's in the headers, set their checksums,
2290 * Also checksum phys, virt....
2291 *
2292 * Then write everything out, finally the anchor is written.
2293 */
2294 get_dev_size(fd, NULL, &size);
2295 size /= 512;
2296 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2297 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2298 ddf->anchor.seq = __cpu_to_be32(1);
2299 memcpy(&ddf->primary, &ddf->anchor, 512);
2300 memcpy(&ddf->secondary, &ddf->anchor, 512);
2301
2302 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2303 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2304 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2305
2306 ddf->primary.openflag = 0;
2307 ddf->primary.type = DDF_HEADER_PRIMARY;
2308
2309 ddf->secondary.openflag = 0;
2310 ddf->secondary.type = DDF_HEADER_SECONDARY;
2311
2312 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2313 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2314
2315 sector = size - 16*1024*2;
2316 lseek64(fd, sector<<9, 0);
2317 if (write(fd, &ddf->primary, 512) < 0)
2318 continue;
2319
2320 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2321 if (write(fd, &ddf->controller, 512) < 0)
2322 continue;
2323
2324 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2325
2326 if (write(fd, ddf->phys, ddf->pdsize) < 0)
2327 continue;
2328
2329 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2330 if (write(fd, ddf->virt, ddf->vdsize) < 0)
2331 continue;
2332
2333 /* Now write lots of config records. */
2334 n_config = ddf->max_part;
2335 conf_size = ddf->conf_rec_len * 512;
2336 for (i = 0 ; i <= n_config ; i++) {
2337 struct vcl *c = d->vlist[i];
2338 if (i == n_config)
2339 c = (struct vcl*)d->spare;
2340
2341 if (c) {
2342 c->conf.crc = calc_crc(&c->conf, conf_size);
2343 if (write(fd, &c->conf, conf_size) < 0)
2344 break;
2345 } else {
2346 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2347 if (null_conf[0] != 0xff)
2348 memset(null_conf, 0xff, sizeof(null_conf));
2349 unsigned int togo = conf_size;
2350 while (togo > sizeof(null_conf)-512) {
2351 if (write(fd, null_aligned, sizeof(null_conf)-512) < 0)
2352 break;
2353 togo -= sizeof(null_conf)-512;
2354 }
2355 if (write(fd, null_aligned, togo) < 0)
2356 break;
2357 }
2358 }
2359 if (i <= n_config)
2360 continue;
2361 d->disk.crc = calc_crc(&d->disk, 512);
2362 if (write(fd, &d->disk, 512) < 0)
2363 continue;
2364
2365 /* Maybe do the same for secondary */
2366
2367 lseek64(fd, (size-1)*512, SEEK_SET);
2368 if (write(fd, &ddf->anchor, 512) < 0)
2369 continue;
2370 successes++;
2371 }
2372
2373 if (do_close)
2374 for (d = ddf->dlist; d; d=d->next) {
2375 close(d->fd);
2376 d->fd = -1;
2377 }
2378
2379 return attempts != successes;
2380 }
2381
2382 static int write_init_super_ddf(struct supertype *st)
2383 {
2384 struct ddf_super *ddf = st->sb;
2385 struct vcl *currentconf = ddf->currentconf;
2386
2387 /* we are done with currentconf reset it to point st at the container */
2388 ddf->currentconf = NULL;
2389
2390 if (st->update_tail) {
2391 /* queue the virtual_disk and vd_config as metadata updates */
2392 struct virtual_disk *vd;
2393 struct vd_config *vc;
2394 int len;
2395
2396 if (!currentconf) {
2397 int len = (sizeof(struct phys_disk) +
2398 sizeof(struct phys_disk_entry));
2399
2400 /* adding a disk to the container. */
2401 if (!ddf->add_list)
2402 return 0;
2403
2404 append_metadata_update(st, ddf->add_list->mdupdate, len);
2405 ddf->add_list->mdupdate = NULL;
2406 return 0;
2407 }
2408
2409 /* Newly created VD */
2410
2411 /* First the virtual disk. We have a slightly fake header */
2412 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2413 vd = malloc(len);
2414 *vd = *ddf->virt;
2415 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
2416 vd->populated_vdes = __cpu_to_be16(currentconf->vcnum);
2417 append_metadata_update(st, vd, len);
2418
2419 /* Then the vd_config */
2420 len = ddf->conf_rec_len * 512;
2421 vc = malloc(len);
2422 memcpy(vc, &currentconf->conf, len);
2423 append_metadata_update(st, vc, len);
2424
2425 /* FIXME I need to close the fds! */
2426 return 0;
2427 } else {
2428 struct dl *d;
2429 for (d = ddf->dlist; d; d=d->next)
2430 while (Kill(d->devname, NULL, 0, 1, 1) == 0);
2431 return __write_init_super_ddf(st, 1);
2432 }
2433 }
2434
2435 #endif
2436
2437 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2438 {
2439 /* We must reserve the last 32Meg */
2440 if (devsize <= 32*1024*2)
2441 return 0;
2442 return devsize - 32*1024*2;
2443 }
2444
2445 #ifndef MDASSEMBLE
2446
2447 static int reserve_space(struct supertype *st, int raiddisks,
2448 unsigned long long size, int chunk,
2449 unsigned long long *freesize)
2450 {
2451 /* Find 'raiddisks' spare extents at least 'size' big (but
2452 * only caring about multiples of 'chunk') and remember
2453 * them.
2454 * If the cannot be found, fail.
2455 */
2456 struct dl *dl;
2457 struct ddf_super *ddf = st->sb;
2458 int cnt = 0;
2459
2460 for (dl = ddf->dlist; dl ; dl=dl->next) {
2461 dl->raiddisk = -1;
2462 dl->esize = 0;
2463 }
2464 /* Now find largest extent on each device */
2465 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2466 struct extent *e = get_extents(ddf, dl);
2467 unsigned long long pos = 0;
2468 int i = 0;
2469 int found = 0;
2470 unsigned long long minsize = size;
2471
2472 if (size == 0)
2473 minsize = chunk;
2474
2475 if (!e)
2476 continue;
2477 do {
2478 unsigned long long esize;
2479 esize = e[i].start - pos;
2480 if (esize >= minsize) {
2481 found = 1;
2482 minsize = esize;
2483 }
2484 pos = e[i].start + e[i].size;
2485 i++;
2486 } while (e[i-1].size);
2487 if (found) {
2488 cnt++;
2489 dl->esize = minsize;
2490 }
2491 free(e);
2492 }
2493 if (cnt < raiddisks) {
2494 fprintf(stderr, Name ": not enough devices with space to create array.\n");
2495 return 0; /* No enough free spaces large enough */
2496 }
2497 if (size == 0) {
2498 /* choose the largest size of which there are at least 'raiddisk' */
2499 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2500 struct dl *dl2;
2501 if (dl->esize <= size)
2502 continue;
2503 /* This is bigger than 'size', see if there are enough */
2504 cnt = 0;
2505 for (dl2 = dl; dl2 ; dl2=dl2->next)
2506 if (dl2->esize >= dl->esize)
2507 cnt++;
2508 if (cnt >= raiddisks)
2509 size = dl->esize;
2510 }
2511 if (chunk) {
2512 size = size / chunk;
2513 size *= chunk;
2514 }
2515 *freesize = size;
2516 if (size < 32) {
2517 fprintf(stderr, Name ": not enough spare devices to create array.\n");
2518 return 0;
2519 }
2520 }
2521 /* We have a 'size' of which there are enough spaces.
2522 * We simply do a first-fit */
2523 cnt = 0;
2524 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
2525 if (dl->esize < size)
2526 continue;
2527
2528 dl->raiddisk = cnt;
2529 cnt++;
2530 }
2531 return 1;
2532 }
2533
2534
2535
2536 static int
2537 validate_geometry_ddf_container(struct supertype *st,
2538 int level, int layout, int raiddisks,
2539 int chunk, unsigned long long size,
2540 char *dev, unsigned long long *freesize,
2541 int verbose);
2542
2543 static int validate_geometry_ddf_bvd(struct supertype *st,
2544 int level, int layout, int raiddisks,
2545 int chunk, unsigned long long size,
2546 char *dev, unsigned long long *freesize,
2547 int verbose);
2548
2549 static int validate_geometry_ddf(struct supertype *st,
2550 int level, int layout, int raiddisks,
2551 int chunk, unsigned long long size,
2552 char *dev, unsigned long long *freesize,
2553 int verbose)
2554 {
2555 int fd;
2556 struct mdinfo *sra;
2557 int cfd;
2558
2559 /* ddf potentially supports lots of things, but it depends on
2560 * what devices are offered (and maybe kernel version?)
2561 * If given unused devices, we will make a container.
2562 * If given devices in a container, we will make a BVD.
2563 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2564 */
2565
2566 if (level == LEVEL_CONTAINER) {
2567 /* Must be a fresh device to add to a container */
2568 return validate_geometry_ddf_container(st, level, layout,
2569 raiddisks, chunk,
2570 size, dev, freesize,
2571 verbose);
2572 }
2573
2574 if (!dev) {
2575 /* Initial sanity check. Exclude illegal levels. */
2576 int i;
2577 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2578 if (ddf_level_num[i].num2 == level)
2579 break;
2580 if (ddf_level_num[i].num1 == MAXINT) {
2581 if (verbose)
2582 fprintf(stderr, Name ": DDF does not support level %d arrays\n",
2583 level);
2584 return 0;
2585 }
2586 /* Should check layout? etc */
2587
2588 if (st->sb && freesize) {
2589 /* --create was given a container to create in.
2590 * So we need to check that there are enough
2591 * free spaces and return the amount of space.
2592 * We may as well remember which drives were
2593 * chosen so that add_to_super/getinfo_super
2594 * can return them.
2595 */
2596 return reserve_space(st, raiddisks, size, chunk, freesize);
2597 }
2598 return 1;
2599 }
2600
2601 if (st->sb) {
2602 /* A container has already been opened, so we are
2603 * creating in there. Maybe a BVD, maybe an SVD.
2604 * Should make a distinction one day.
2605 */
2606 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2607 chunk, size, dev, freesize,
2608 verbose);
2609 }
2610 /* This is the first device for the array.
2611 * If it is a container, we read it in and do automagic allocations,
2612 * no other devices should be given.
2613 * Otherwise it must be a member device of a container, and we
2614 * do manual allocation.
2615 * Later we should check for a BVD and make an SVD.
2616 */
2617 fd = open(dev, O_RDONLY|O_EXCL, 0);
2618 if (fd >= 0) {
2619 sra = sysfs_read(fd, 0, GET_VERSION);
2620 close(fd);
2621 if (sra && sra->array.major_version == -1 &&
2622 strcmp(sra->text_version, "ddf") == 0) {
2623
2624 /* load super */
2625 /* find space for 'n' devices. */
2626 /* remember the devices */
2627 /* Somehow return the fact that we have enough */
2628 }
2629
2630 if (verbose)
2631 fprintf(stderr,
2632 Name ": ddf: Cannot create this array "
2633 "on device %s - a container is required.\n",
2634 dev);
2635 return 0;
2636 }
2637 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2638 if (verbose)
2639 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2640 dev, strerror(errno));
2641 return 0;
2642 }
2643 /* Well, it is in use by someone, maybe a 'ddf' container. */
2644 cfd = open_container(fd);
2645 if (cfd < 0) {
2646 close(fd);
2647 if (verbose)
2648 fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2649 dev, strerror(EBUSY));
2650 return 0;
2651 }
2652 sra = sysfs_read(cfd, 0, GET_VERSION);
2653 close(fd);
2654 if (sra && sra->array.major_version == -1 &&
2655 strcmp(sra->text_version, "ddf") == 0) {
2656 /* This is a member of a ddf container. Load the container
2657 * and try to create a bvd
2658 */
2659 struct ddf_super *ddf;
2660 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) {
2661 st->sb = ddf;
2662 st->container_dev = fd2devnum(cfd);
2663 close(cfd);
2664 return validate_geometry_ddf_bvd(st, level, layout,
2665 raiddisks, chunk, size,
2666 dev, freesize,
2667 verbose);
2668 }
2669 close(cfd);
2670 } else /* device may belong to a different container */
2671 return 0;
2672
2673 return 1;
2674 }
2675
2676 static int
2677 validate_geometry_ddf_container(struct supertype *st,
2678 int level, int layout, int raiddisks,
2679 int chunk, unsigned long long size,
2680 char *dev, unsigned long long *freesize,
2681 int verbose)
2682 {
2683 int fd;
2684 unsigned long long ldsize;
2685
2686 if (level != LEVEL_CONTAINER)
2687 return 0;
2688 if (!dev)
2689 return 1;
2690
2691 fd = open(dev, O_RDONLY|O_EXCL, 0);
2692 if (fd < 0) {
2693 if (verbose)
2694 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2695 dev, strerror(errno));
2696 return 0;
2697 }
2698 if (!get_dev_size(fd, dev, &ldsize)) {
2699 close(fd);
2700 return 0;
2701 }
2702 close(fd);
2703
2704 *freesize = avail_size_ddf(st, ldsize >> 9);
2705 if (*freesize == 0)
2706 return 0;
2707
2708 return 1;
2709 }
2710
2711 static int validate_geometry_ddf_bvd(struct supertype *st,
2712 int level, int layout, int raiddisks,
2713 int chunk, unsigned long long size,
2714 char *dev, unsigned long long *freesize,
2715 int verbose)
2716 {
2717 struct stat stb;
2718 struct ddf_super *ddf = st->sb;
2719 struct dl *dl;
2720 unsigned long long pos = 0;
2721 unsigned long long maxsize;
2722 struct extent *e;
2723 int i;
2724 /* ddf/bvd supports lots of things, but not containers */
2725 if (level == LEVEL_CONTAINER) {
2726 if (verbose)
2727 fprintf(stderr, Name ": DDF cannot create a container within an container\n");
2728 return 0;
2729 }
2730 /* We must have the container info already read in. */
2731 if (!ddf)
2732 return 0;
2733
2734 if (!dev) {
2735 /* General test: make sure there is space for
2736 * 'raiddisks' device extents of size 'size'.
2737 */
2738 unsigned long long minsize = size;
2739 int dcnt = 0;
2740 if (minsize == 0)
2741 minsize = 8;
2742 for (dl = ddf->dlist; dl ; dl = dl->next)
2743 {
2744 int found = 0;
2745 pos = 0;
2746
2747 i = 0;
2748 e = get_extents(ddf, dl);
2749 if (!e) continue;
2750 do {
2751 unsigned long long esize;
2752 esize = e[i].start - pos;
2753 if (esize >= minsize)
2754 found = 1;
2755 pos = e[i].start + e[i].size;
2756 i++;
2757 } while (e[i-1].size);
2758 if (found)
2759 dcnt++;
2760 free(e);
2761 }
2762 if (dcnt < raiddisks) {
2763 if (verbose)
2764 fprintf(stderr,
2765 Name ": ddf: Not enough devices with "
2766 "space for this array (%d < %d)\n",
2767 dcnt, raiddisks);
2768 return 0;
2769 }
2770 return 1;
2771 }
2772 /* This device must be a member of the set */
2773 if (stat(dev, &stb) < 0)
2774 return 0;
2775 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2776 return 0;
2777 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2778 if (dl->major == (int)major(stb.st_rdev) &&
2779 dl->minor == (int)minor(stb.st_rdev))
2780 break;
2781 }
2782 if (!dl) {
2783 if (verbose)
2784 fprintf(stderr, Name ": ddf: %s is not in the "
2785 "same DDF set\n",
2786 dev);
2787 return 0;
2788 }
2789 e = get_extents(ddf, dl);
2790 maxsize = 0;
2791 i = 0;
2792 if (e) do {
2793 unsigned long long esize;
2794 esize = e[i].start - pos;
2795 if (esize >= maxsize)
2796 maxsize = esize;
2797 pos = e[i].start + e[i].size;
2798 i++;
2799 } while (e[i-1].size);
2800 *freesize = maxsize;
2801 // FIXME here I am
2802
2803 return 1;
2804 }
2805
2806 static int load_super_ddf_all(struct supertype *st, int fd,
2807 void **sbp, char *devname)
2808 {
2809 struct mdinfo *sra;
2810 struct ddf_super *super;
2811 struct mdinfo *sd, *best = NULL;
2812 int bestseq = 0;
2813 int seq;
2814 char nm[20];
2815 int dfd;
2816
2817 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2818 if (!sra)
2819 return 1;
2820 if (sra->array.major_version != -1 ||
2821 sra->array.minor_version != -2 ||
2822 strcmp(sra->text_version, "ddf") != 0)
2823 return 1;
2824
2825 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2826 return 1;
2827 memset(super, 0, sizeof(*super));
2828
2829 /* first, try each device, and choose the best ddf */
2830 for (sd = sra->devs ; sd ; sd = sd->next) {
2831 int rv;
2832 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2833 dfd = dev_open(nm, O_RDONLY);
2834 if (dfd < 0)
2835 return 2;
2836 rv = load_ddf_headers(dfd, super, NULL);
2837 close(dfd);
2838 if (rv == 0) {
2839 seq = __be32_to_cpu(super->active->seq);
2840 if (super->active->openflag)
2841 seq--;
2842 if (!best || seq > bestseq) {
2843 bestseq = seq;
2844 best = sd;
2845 }
2846 }
2847 }
2848 if (!best)
2849 return 1;
2850 /* OK, load this ddf */
2851 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2852 dfd = dev_open(nm, O_RDONLY);
2853 if (dfd < 0)
2854 return 1;
2855 load_ddf_headers(dfd, super, NULL);
2856 load_ddf_global(dfd, super, NULL);
2857 close(dfd);
2858 /* Now we need the device-local bits */
2859 for (sd = sra->devs ; sd ; sd = sd->next) {
2860 int rv;
2861
2862 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2863 dfd = dev_open(nm, O_RDWR);
2864 if (dfd < 0)
2865 return 2;
2866 rv = load_ddf_headers(dfd, super, NULL);
2867 if (rv == 0)
2868 rv = load_ddf_local(dfd, super, NULL, 1);
2869 if (rv)
2870 return 1;
2871 }
2872
2873 *sbp = super;
2874 if (st->ss == NULL) {
2875 st->ss = &super_ddf;
2876 st->minor_version = 0;
2877 st->max_devs = 512;
2878 st->container_dev = fd2devnum(fd);
2879 }
2880 st->loaded_container = 1;
2881 return 0;
2882 }
2883
2884 static int load_container_ddf(struct supertype *st, int fd,
2885 char *devname)
2886 {
2887 return load_super_ddf_all(st, fd, &st->sb, devname);
2888 }
2889
2890 #endif /* MDASSEMBLE */
2891
2892 static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray)
2893 {
2894 /* Given a container loaded by load_super_ddf_all,
2895 * extract information about all the arrays into
2896 * an mdinfo tree.
2897 *
2898 * For each vcl in conflist: create an mdinfo, fill it in,
2899 * then look for matching devices (phys_refnum) in dlist
2900 * and create appropriate device mdinfo.
2901 */
2902 struct ddf_super *ddf = st->sb;
2903 struct mdinfo *rest = NULL;
2904 struct vcl *vc;
2905
2906 for (vc = ddf->conflist ; vc ; vc=vc->next)
2907 {
2908 unsigned int i;
2909 unsigned int j;
2910 struct mdinfo *this;
2911 char *ep;
2912
2913 if (subarray &&
2914 (strtoul(subarray, &ep, 10) != vc->vcnum ||
2915 *ep != '\0'))
2916 continue;
2917
2918 this = malloc(sizeof(*this));
2919 memset(this, 0, sizeof(*this));
2920 this->next = rest;
2921 rest = this;
2922
2923 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2924 this->array.raid_disks =
2925 __be16_to_cpu(vc->conf.prim_elmnt_count);
2926 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2927 this->array.raid_disks);
2928 this->array.md_minor = -1;
2929 this->array.major_version = -1;
2930 this->array.minor_version = -2;
2931 this->array.ctime = DECADE +
2932 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2933 this->array.utime = DECADE +
2934 __be32_to_cpu(vc->conf.timestamp);
2935 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2936
2937 i = vc->vcnum;
2938 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2939 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2940 DDF_init_full) {
2941 this->array.state = 0;
2942 this->resync_start = 0;
2943 } else {
2944 this->array.state = 1;
2945 this->resync_start = MaxSector;
2946 }
2947 memcpy(this->name, ddf->virt->entries[i].name, 16);
2948 this->name[16]=0;
2949 for(j=0; j<16; j++)
2950 if (this->name[j] == ' ')
2951 this->name[j] = 0;
2952
2953 memset(this->uuid, 0, sizeof(this->uuid));
2954 this->component_size = __be64_to_cpu(vc->conf.blocks);
2955 this->array.size = this->component_size / 2;
2956 this->container_member = i;
2957
2958 ddf->currentconf = vc;
2959 uuid_from_super_ddf(st, this->uuid);
2960 ddf->currentconf = NULL;
2961
2962 sprintf(this->text_version, "/%s/%d",
2963 devnum2devname(st->container_dev),
2964 this->container_member);
2965
2966 for (i = 0 ; i < ddf->mppe ; i++) {
2967 struct mdinfo *dev;
2968 struct dl *d;
2969
2970 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2971 continue;
2972
2973 this->array.working_disks++;
2974
2975 for (d = ddf->dlist; d ; d=d->next)
2976 if (d->disk.refnum == vc->conf.phys_refnum[i])
2977 break;
2978 if (d == NULL)
2979 /* Haven't found that one yet, maybe there are others */
2980 continue;
2981
2982 dev = malloc(sizeof(*dev));
2983 memset(dev, 0, sizeof(*dev));
2984 dev->next = this->devs;
2985 this->devs = dev;
2986
2987 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2988 dev->disk.major = d->major;
2989 dev->disk.minor = d->minor;
2990 dev->disk.raid_disk = i;
2991 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2992 dev->recovery_start = MaxSector;
2993
2994 dev->events = __be32_to_cpu(ddf->primary.seq);
2995 dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2996 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2997 if (d->devname)
2998 strcpy(dev->name, d->devname);
2999 }
3000 }
3001 return rest;
3002 }
3003
3004 static int store_super_ddf(struct supertype *st, int fd)
3005 {
3006 struct ddf_super *ddf = st->sb;
3007 unsigned long long dsize;
3008 void *buf;
3009 int rc;
3010
3011 if (!ddf)
3012 return 1;
3013
3014 /* ->dlist and ->conflist will be set for updates, currently not
3015 * supported
3016 */
3017 if (ddf->dlist || ddf->conflist)
3018 return 1;
3019
3020 if (!get_dev_size(fd, NULL, &dsize))
3021 return 1;
3022
3023 if (posix_memalign(&buf, 512, 512) != 0)
3024 return 1;
3025 memset(buf, 0, 512);
3026
3027 lseek64(fd, dsize-512, 0);
3028 rc = write(fd, buf, 512);
3029 free(buf);
3030 if (rc < 0)
3031 return 1;
3032 return 0;
3033 }
3034
3035 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3036 {
3037 /*
3038 * return:
3039 * 0 same, or first was empty, and second was copied
3040 * 1 second had wrong number
3041 * 2 wrong uuid
3042 * 3 wrong other info
3043 */
3044 struct ddf_super *first = st->sb;
3045 struct ddf_super *second = tst->sb;
3046
3047 if (!first) {
3048 st->sb = tst->sb;
3049 tst->sb = NULL;
3050 return 0;
3051 }
3052
3053 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3054 return 2;
3055
3056 /* FIXME should I look at anything else? */
3057 return 0;
3058 }
3059
3060 #ifndef MDASSEMBLE
3061 /*
3062 * A new array 'a' has been started which claims to be instance 'inst'
3063 * within container 'c'.
3064 * We need to confirm that the array matches the metadata in 'c' so
3065 * that we don't corrupt any metadata.
3066 */
3067 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
3068 {
3069 dprintf("ddf: open_new %s\n", inst);
3070 a->info.container_member = atoi(inst);
3071 return 0;
3072 }
3073
3074 /*
3075 * The array 'a' is to be marked clean in the metadata.
3076 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3077 * clean up to the point (in sectors). If that cannot be recorded in the
3078 * metadata, then leave it as dirty.
3079 *
3080 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3081 * !global! virtual_disk.virtual_entry structure.
3082 */
3083 static int ddf_set_array_state(struct active_array *a, int consistent)
3084 {
3085 struct ddf_super *ddf = a->container->sb;
3086 int inst = a->info.container_member;
3087 int old = ddf->virt->entries[inst].state;
3088 if (consistent == 2) {
3089 /* Should check if a recovery should be started FIXME */
3090 consistent = 1;
3091 if (!is_resync_complete(&a->info))
3092 consistent = 0;
3093 }
3094 if (consistent)
3095 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
3096 else
3097 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
3098 if (old != ddf->virt->entries[inst].state)
3099 ddf->updates_pending = 1;
3100
3101 old = ddf->virt->entries[inst].init_state;
3102 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
3103 if (is_resync_complete(&a->info))
3104 ddf->virt->entries[inst].init_state |= DDF_init_full;
3105 else if (a->info.resync_start == 0)
3106 ddf->virt->entries[inst].init_state |= DDF_init_not;
3107 else
3108 ddf->virt->entries[inst].init_state |= DDF_init_quick;
3109 if (old != ddf->virt->entries[inst].init_state)
3110 ddf->updates_pending = 1;
3111
3112 dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
3113 a->info.resync_start);
3114 return consistent;
3115 }
3116
3117 /*
3118 * The state of each disk is stored in the global phys_disk structure
3119 * in phys_disk.entries[n].state.
3120 * This makes various combinations awkward.
3121 * - When a device fails in any array, it must be failed in all arrays
3122 * that include a part of this device.
3123 * - When a component is rebuilding, we cannot include it officially in the
3124 * array unless this is the only array that uses the device.
3125 *
3126 * So: when transitioning:
3127 * Online -> failed, just set failed flag. monitor will propagate
3128 * spare -> online, the device might need to be added to the array.
3129 * spare -> failed, just set failed. Don't worry if in array or not.
3130 */
3131 static void ddf_set_disk(struct active_array *a, int n, int state)
3132 {
3133 struct ddf_super *ddf = a->container->sb;
3134 unsigned int inst = a->info.container_member;
3135 struct vd_config *vc = find_vdcr(ddf, inst);
3136 int pd = find_phys(ddf, vc->phys_refnum[n]);
3137 int i, st, working;
3138
3139 if (vc == NULL) {
3140 dprintf("ddf: cannot find instance %d!!\n", inst);
3141 return;
3142 }
3143 if (pd < 0) {
3144 /* disk doesn't currently exist. If it is now in_sync,
3145 * insert it. */
3146 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
3147 /* Find dev 'n' in a->info->devs, determine the
3148 * ddf refnum, and set vc->phys_refnum and update
3149 * phys->entries[]
3150 */
3151 /* FIXME */
3152 }
3153 } else {
3154 int old = ddf->phys->entries[pd].state;
3155 if (state & DS_FAULTY)
3156 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
3157 if (state & DS_INSYNC) {
3158 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
3159 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
3160 }
3161 if (old != ddf->phys->entries[pd].state)
3162 ddf->updates_pending = 1;
3163 }
3164
3165 dprintf("ddf: set_disk %d to %x\n", n, state);
3166
3167 /* Now we need to check the state of the array and update
3168 * virtual_disk.entries[n].state.
3169 * It needs to be one of "optimal", "degraded", "failed".
3170 * I don't understand 'deleted' or 'missing'.
3171 */
3172 working = 0;
3173 for (i=0; i < a->info.array.raid_disks; i++) {
3174 pd = find_phys(ddf, vc->phys_refnum[i]);
3175 if (pd < 0)
3176 continue;
3177 st = __be16_to_cpu(ddf->phys->entries[pd].state);
3178 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3179 == DDF_Online)
3180 working++;
3181 }
3182 state = DDF_state_degraded;
3183 if (working == a->info.array.raid_disks)
3184 state = DDF_state_optimal;
3185 else switch(vc->prl) {
3186 case DDF_RAID0:
3187 case DDF_CONCAT:
3188 case DDF_JBOD:
3189 state = DDF_state_failed;
3190 break;
3191 case DDF_RAID1:
3192 if (working == 0)
3193 state = DDF_state_failed;
3194 break;
3195 case DDF_RAID4:
3196 case DDF_RAID5:
3197 if (working < a->info.array.raid_disks-1)
3198 state = DDF_state_failed;
3199 break;
3200 case DDF_RAID6:
3201 if (working < a->info.array.raid_disks-2)
3202 state = DDF_state_failed;
3203 else if (working == a->info.array.raid_disks-1)
3204 state = DDF_state_part_optimal;
3205 break;
3206 }
3207
3208 if (ddf->virt->entries[inst].state !=
3209 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
3210 | state)) {
3211
3212 ddf->virt->entries[inst].state =
3213 (ddf->virt->entries[inst].state & ~DDF_state_mask)
3214 | state;
3215 ddf->updates_pending = 1;
3216 }
3217
3218 }
3219
3220 static void ddf_sync_metadata(struct supertype *st)
3221 {
3222
3223 /*
3224 * Write all data to all devices.
3225 * Later, we might be able to track whether only local changes
3226 * have been made, or whether any global data has been changed,
3227 * but ddf is sufficiently weird that it probably always
3228 * changes global data ....
3229 */
3230 struct ddf_super *ddf = st->sb;
3231 if (!ddf->updates_pending)
3232 return;
3233 ddf->updates_pending = 0;
3234 __write_init_super_ddf(st, 0);
3235 dprintf("ddf: sync_metadata\n");
3236 }
3237
3238 static void ddf_process_update(struct supertype *st,
3239 struct metadata_update *update)
3240 {
3241 /* Apply this update to the metadata.
3242 * The first 4 bytes are a DDF_*_MAGIC which guides
3243 * our actions.
3244 * Possible update are:
3245 * DDF_PHYS_RECORDS_MAGIC
3246 * Add a new physical device. Changes to this record
3247 * only happen implicitly.
3248 * used_pdes is the device number.
3249 * DDF_VIRT_RECORDS_MAGIC
3250 * Add a new VD. Possibly also change the 'access' bits.
3251 * populated_vdes is the entry number.
3252 * DDF_VD_CONF_MAGIC
3253 * New or updated VD. the VIRT_RECORD must already
3254 * exist. For an update, phys_refnum and lba_offset
3255 * (at least) are updated, and the VD_CONF must
3256 * be written to precisely those devices listed with
3257 * a phys_refnum.
3258 * DDF_SPARE_ASSIGN_MAGIC
3259 * replacement Spare Assignment Record... but for which device?
3260 *
3261 * So, e.g.:
3262 * - to create a new array, we send a VIRT_RECORD and
3263 * a VD_CONF. Then assemble and start the array.
3264 * - to activate a spare we send a VD_CONF to add the phys_refnum
3265 * and offset. This will also mark the spare as active with
3266 * a spare-assignment record.
3267 */
3268 struct ddf_super *ddf = st->sb;
3269 __u32 *magic = (__u32*)update->buf;
3270 struct phys_disk *pd;
3271 struct virtual_disk *vd;
3272 struct vd_config *vc;
3273 struct vcl *vcl;
3274 struct dl *dl;
3275 unsigned int mppe;
3276 unsigned int ent;
3277
3278 dprintf("Process update %x\n", *magic);
3279
3280 switch (*magic) {
3281 case DDF_PHYS_RECORDS_MAGIC:
3282
3283 if (update->len != (sizeof(struct phys_disk) +
3284 sizeof(struct phys_disk_entry)))
3285 return;
3286 pd = (struct phys_disk*)update->buf;
3287
3288 ent = __be16_to_cpu(pd->used_pdes);
3289 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
3290 return;
3291 if (!all_ff(ddf->phys->entries[ent].guid))
3292 return;
3293 ddf->phys->entries[ent] = pd->entries[0];
3294 ddf->phys->used_pdes = __cpu_to_be16(1 +
3295 __be16_to_cpu(ddf->phys->used_pdes));
3296 ddf->updates_pending = 1;
3297 if (ddf->add_list) {
3298 struct active_array *a;
3299 struct dl *al = ddf->add_list;
3300 ddf->add_list = al->next;
3301
3302 al->next = ddf->dlist;
3303 ddf->dlist = al;
3304
3305 /* As a device has been added, we should check
3306 * for any degraded devices that might make
3307 * use of this spare */
3308 for (a = st->arrays ; a; a=a->next)
3309 a->check_degraded = 1;
3310 }
3311 break;
3312
3313 case DDF_VIRT_RECORDS_MAGIC:
3314
3315 if (update->len != (sizeof(struct virtual_disk) +
3316 sizeof(struct virtual_entry)))
3317 return;
3318 vd = (struct virtual_disk*)update->buf;
3319
3320 ent = __be16_to_cpu(vd->populated_vdes);
3321 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
3322 return;
3323 if (!all_ff(ddf->virt->entries[ent].guid))
3324 return;
3325 ddf->virt->entries[ent] = vd->entries[0];
3326 ddf->virt->populated_vdes = __cpu_to_be16(1 +
3327 __be16_to_cpu(ddf->virt->populated_vdes));
3328 ddf->updates_pending = 1;
3329 break;
3330
3331 case DDF_VD_CONF_MAGIC:
3332 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
3333
3334 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
3335 if ((unsigned)update->len != ddf->conf_rec_len * 512)
3336 return;
3337 vc = (struct vd_config*)update->buf;
3338 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3339 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
3340 break;
3341 dprintf("vcl = %p\n", vcl);
3342 if (vcl) {
3343 /* An update, just copy the phys_refnum and lba_offset
3344 * fields
3345 */
3346 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
3347 mppe * (sizeof(__u32) + sizeof(__u64)));
3348 } else {
3349 /* A new VD_CONF */
3350 if (!update->space)
3351 return;
3352 vcl = update->space;
3353 update->space = NULL;
3354 vcl->next = ddf->conflist;
3355 memcpy(&vcl->conf, vc, update->len);
3356 vcl->lba_offset = (__u64*)
3357 &vcl->conf.phys_refnum[mppe];
3358 ddf->conflist = vcl;
3359 }
3360 /* Now make sure vlist is correct for each dl. */
3361 for (dl = ddf->dlist; dl; dl = dl->next) {
3362 unsigned int dn;
3363 unsigned int vn = 0;
3364 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3365 for (dn=0; dn < ddf->mppe ; dn++)
3366 if (vcl->conf.phys_refnum[dn] ==
3367 dl->disk.refnum) {
3368 dprintf("dev %d has %p at %d\n",
3369 dl->pdnum, vcl, vn);
3370 dl->vlist[vn++] = vcl;
3371 break;
3372 }
3373 while (vn < ddf->max_part)
3374 dl->vlist[vn++] = NULL;
3375 if (dl->vlist[0]) {
3376 ddf->phys->entries[dl->pdnum].type &=
3377 ~__cpu_to_be16(DDF_Global_Spare);
3378 ddf->phys->entries[dl->pdnum].type |=
3379 __cpu_to_be16(DDF_Active_in_VD);
3380 }
3381 if (dl->spare) {
3382 ddf->phys->entries[dl->pdnum].type &=
3383 ~__cpu_to_be16(DDF_Global_Spare);
3384 ddf->phys->entries[dl->pdnum].type |=
3385 __cpu_to_be16(DDF_Spare);
3386 }
3387 if (!dl->vlist[0] && !dl->spare) {
3388 ddf->phys->entries[dl->pdnum].type |=
3389 __cpu_to_be16(DDF_Global_Spare);
3390 ddf->phys->entries[dl->pdnum].type &=
3391 ~__cpu_to_be16(DDF_Spare |
3392 DDF_Active_in_VD);
3393 }
3394 }
3395 ddf->updates_pending = 1;
3396 break;
3397 case DDF_SPARE_ASSIGN_MAGIC:
3398 default: break;
3399 }
3400 }
3401
3402 static void ddf_prepare_update(struct supertype *st,
3403 struct metadata_update *update)
3404 {
3405 /* This update arrived at managemon.
3406 * We are about to pass it to monitor.
3407 * If a malloc is needed, do it here.
3408 */
3409 struct ddf_super *ddf = st->sb;
3410 __u32 *magic = (__u32*)update->buf;
3411 if (*magic == DDF_VD_CONF_MAGIC)
3412 if (posix_memalign(&update->space, 512,
3413 offsetof(struct vcl, conf)
3414 + ddf->conf_rec_len * 512) != 0)
3415 update->space = NULL;
3416 }
3417
3418 /*
3419 * Check if the array 'a' is degraded but not failed.
3420 * If it is, find as many spares as are available and needed and
3421 * arrange for their inclusion.
3422 * We only choose devices which are not already in the array,
3423 * and prefer those with a spare-assignment to this array.
3424 * otherwise we choose global spares - assuming always that
3425 * there is enough room.
3426 * For each spare that we assign, we return an 'mdinfo' which
3427 * describes the position for the device in the array.
3428 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3429 * the new phys_refnum and lba_offset values.
3430 *
3431 * Only worry about BVDs at the moment.
3432 */
3433 static struct mdinfo *ddf_activate_spare(struct active_array *a,
3434 struct metadata_update **updates)
3435 {
3436 int working = 0;
3437 struct mdinfo *d;
3438 struct ddf_super *ddf = a->container->sb;
3439 int global_ok = 0;
3440 struct mdinfo *rv = NULL;
3441 struct mdinfo *di;
3442 struct metadata_update *mu;
3443 struct dl *dl;
3444 int i;
3445 struct vd_config *vc;
3446 __u64 *lba;
3447
3448 for (d = a->info.devs ; d ; d = d->next) {
3449 if ((d->curr_state & DS_FAULTY) &&
3450 d->state_fd >= 0)
3451 /* wait for Removal to happen */
3452 return NULL;
3453 if (d->state_fd >= 0)
3454 working ++;
3455 }
3456
3457 dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3458 a->info.array.level);
3459 if (working == a->info.array.raid_disks)
3460 return NULL; /* array not degraded */
3461 switch (a->info.array.level) {
3462 case 1:
3463 if (working == 0)
3464 return NULL; /* failed */
3465 break;
3466 case 4:
3467 case 5:
3468 if (working < a->info.array.raid_disks - 1)
3469 return NULL; /* failed */
3470 break;
3471 case 6:
3472 if (working < a->info.array.raid_disks - 2)
3473 return NULL; /* failed */
3474 break;
3475 default: /* concat or stripe */
3476 return NULL; /* failed */
3477 }
3478
3479 /* For each slot, if it is not working, find a spare */
3480 dl = ddf->dlist;
3481 for (i = 0; i < a->info.array.raid_disks; i++) {
3482 for (d = a->info.devs ; d ; d = d->next)
3483 if (d->disk.raid_disk == i)
3484 break;
3485 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3486 if (d && (d->state_fd >= 0))
3487 continue;
3488
3489 /* OK, this device needs recovery. Find a spare */
3490 again:
3491 for ( ; dl ; dl = dl->next) {
3492 unsigned long long esize;
3493 unsigned long long pos;
3494 struct mdinfo *d2;
3495 int is_global = 0;
3496 int is_dedicated = 0;
3497 struct extent *ex;
3498 unsigned int j;
3499 /* If in this array, skip */
3500 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3501 if (d2->disk.major == dl->major &&
3502 d2->disk.minor == dl->minor) {
3503 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3504 break;
3505 }
3506 if (d2)
3507 continue;
3508 if (ddf->phys->entries[dl->pdnum].type &
3509 __cpu_to_be16(DDF_Spare)) {
3510 /* Check spare assign record */
3511 if (dl->spare) {
3512 if (dl->spare->type & DDF_spare_dedicated) {
3513 /* check spare_ents for guid */
3514 for (j = 0 ;
3515 j < __be16_to_cpu(dl->spare->populated);
3516 j++) {
3517 if (memcmp(dl->spare->spare_ents[j].guid,
3518 ddf->virt->entries[a->info.container_member].guid,
3519 DDF_GUID_LEN) == 0)
3520 is_dedicated = 1;
3521 }
3522 } else
3523 is_global = 1;
3524 }
3525 } else if (ddf->phys->entries[dl->pdnum].type &
3526 __cpu_to_be16(DDF_Global_Spare)) {
3527 is_global = 1;
3528 }
3529 if ( ! (is_dedicated ||
3530 (is_global && global_ok))) {
3531 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3532 is_dedicated, is_global);
3533 continue;
3534 }
3535
3536 /* We are allowed to use this device - is there space?
3537 * We need a->info.component_size sectors */
3538 ex = get_extents(ddf, dl);
3539 if (!ex) {
3540 dprintf("cannot get extents\n");
3541 continue;
3542 }
3543 j = 0; pos = 0;
3544 esize = 0;
3545
3546 do {
3547 esize = ex[j].start - pos;
3548 if (esize >= a->info.component_size)
3549 break;
3550 pos = ex[i].start + ex[i].size;
3551 i++;
3552 } while (ex[i-1].size);
3553
3554 free(ex);
3555 if (esize < a->info.component_size) {
3556 dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3557 esize, a->info.component_size);
3558 /* No room */
3559 continue;
3560 }
3561
3562 /* Cool, we have a device with some space at pos */
3563 di = malloc(sizeof(*di));
3564 if (!di)
3565 continue;
3566 memset(di, 0, sizeof(*di));
3567 di->disk.number = i;
3568 di->disk.raid_disk = i;
3569 di->disk.major = dl->major;
3570 di->disk.minor = dl->minor;
3571 di->disk.state = 0;
3572 di->recovery_start = 0;
3573 di->data_offset = pos;
3574 di->component_size = a->info.component_size;
3575 di->container_member = dl->pdnum;
3576 di->next = rv;
3577 rv = di;
3578 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3579 i, pos);
3580
3581 break;
3582 }
3583 if (!dl && ! global_ok) {
3584 /* not enough dedicated spares, try global */
3585 global_ok = 1;
3586 dl = ddf->dlist;
3587 goto again;
3588 }
3589 }
3590
3591 if (!rv)
3592 /* No spares found */
3593 return rv;
3594 /* Now 'rv' has a list of devices to return.
3595 * Create a metadata_update record to update the
3596 * phys_refnum and lba_offset values
3597 */
3598 mu = malloc(sizeof(*mu));
3599 if (mu && posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
3600 free(mu);
3601 mu = NULL;
3602 }
3603 if (!mu) {
3604 while (rv) {
3605 struct mdinfo *n = rv->next;
3606
3607 free(rv);
3608 rv = n;
3609 }
3610 return NULL;
3611 }
3612
3613 mu->buf = malloc(ddf->conf_rec_len * 512);
3614 mu->len = ddf->conf_rec_len;
3615 mu->next = *updates;
3616 vc = find_vdcr(ddf, a->info.container_member);
3617 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3618
3619 vc = (struct vd_config*)mu->buf;
3620 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3621 for (di = rv ; di ; di = di->next) {
3622 vc->phys_refnum[di->disk.raid_disk] =
3623 ddf->phys->entries[dl->pdnum].refnum;
3624 lba[di->disk.raid_disk] = di->data_offset;
3625 }
3626 *updates = mu;
3627 return rv;
3628 }
3629 #endif /* MDASSEMBLE */
3630
3631 static int ddf_level_to_layout(int level)
3632 {
3633 switch(level) {
3634 case 0:
3635 case 1:
3636 return 0;
3637 case 5:
3638 return ALGORITHM_LEFT_SYMMETRIC;
3639 case 6:
3640 return ALGORITHM_ROTATING_N_CONTINUE;
3641 case 10:
3642 return 0x102;
3643 default:
3644 return UnSet;
3645 }
3646 }
3647
3648 struct superswitch super_ddf = {
3649 #ifndef MDASSEMBLE
3650 .examine_super = examine_super_ddf,
3651 .brief_examine_super = brief_examine_super_ddf,
3652 .brief_examine_subarrays = brief_examine_subarrays_ddf,
3653 .export_examine_super = export_examine_super_ddf,
3654 .detail_super = detail_super_ddf,
3655 .brief_detail_super = brief_detail_super_ddf,
3656 .validate_geometry = validate_geometry_ddf,
3657 .write_init_super = write_init_super_ddf,
3658 .add_to_super = add_to_super_ddf,
3659 .load_container = load_container_ddf,
3660 #endif
3661 .match_home = match_home_ddf,
3662 .uuid_from_super= uuid_from_super_ddf,
3663 .getinfo_super = getinfo_super_ddf,
3664 .update_super = update_super_ddf,
3665
3666 .avail_size = avail_size_ddf,
3667
3668 .compare_super = compare_super_ddf,
3669
3670 .load_super = load_super_ddf,
3671 .init_super = init_super_ddf,
3672 .store_super = store_super_ddf,
3673 .free_super = free_super_ddf,
3674 .match_metadata_desc = match_metadata_desc_ddf,
3675 .container_content = container_content_ddf,
3676 .default_layout = ddf_level_to_layout,
3677
3678 .external = 1,
3679
3680 #ifndef MDASSEMBLE
3681 /* for mdmon */
3682 .open_new = ddf_open_new,
3683 .set_array_state= ddf_set_array_state,
3684 .set_disk = ddf_set_disk,
3685 .sync_metadata = ddf_sync_metadata,
3686 .process_update = ddf_process_update,
3687 .prepare_update = ddf_prepare_update,
3688 .activate_spare = ddf_activate_spare,
3689 #endif
3690 .name = "ddf",
3691 };