]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
DDF: explain why spare_refs are ignored.
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2006-2014 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 taken from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
26 */
27
28 #define HAVE_STDINT_H 1
29 #include "mdadm.h"
30 #include "mdmon.h"
31 #include "sha1.h"
32 #include <values.h>
33
34 /* a non-official T10 name for creation GUIDs */
35 static char T10[] = "Linux-MD";
36
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
40 */
41 #define DECADE (3600*24*(365*10+2))
42 unsigned long crc32(
43 unsigned long crc,
44 const unsigned char *buf,
45 unsigned len);
46
47 #define DDF_NOTFOUND (~0U)
48 #define DDF_CONTAINER (DDF_NOTFOUND-1)
49
50 /* Default for safe_mode_delay. Same value as for IMSM.
51 */
52 static const int DDF_SAFE_MODE_DELAY = 4000;
53
54 /* The DDF metadata handling.
55 * DDF metadata lives at the end of the device.
56 * The last 512 byte block provides an 'anchor' which is used to locate
57 * the rest of the metadata which usually lives immediately behind the anchor.
58 *
59 * Note:
60 * - all multibyte numeric fields are bigendian.
61 * - all strings are space padded.
62 *
63 */
64
65 typedef struct __be16 {
66 __u16 _v16;
67 } be16;
68 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
69 #define be16_and(x, y) ((x)._v16 & (y)._v16)
70 #define be16_or(x, y) ((x)._v16 | (y)._v16)
71 #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16)
72 #define be16_set(x, y) ((x)._v16 |= (y)._v16)
73
74 typedef struct __be32 {
75 __u32 _v32;
76 } be32;
77 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
78
79 typedef struct __be64 {
80 __u64 _v64;
81 } be64;
82 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
83
84 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
85 static inline be16 cpu_to_be16(__u16 x)
86 {
87 be16 be = { ._v16 = __cpu_to_be16(x) };
88 return be;
89 }
90
91 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
92 static inline be32 cpu_to_be32(__u32 x)
93 {
94 be32 be = { ._v32 = __cpu_to_be32(x) };
95 return be;
96 }
97
98 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
99 static inline be64 cpu_to_be64(__u64 x)
100 {
101 be64 be = { ._v64 = __cpu_to_be64(x) };
102 return be;
103 }
104
105 /* Primary Raid Level (PRL) */
106 #define DDF_RAID0 0x00
107 #define DDF_RAID1 0x01
108 #define DDF_RAID3 0x03
109 #define DDF_RAID4 0x04
110 #define DDF_RAID5 0x05
111 #define DDF_RAID1E 0x11
112 #define DDF_JBOD 0x0f
113 #define DDF_CONCAT 0x1f
114 #define DDF_RAID5E 0x15
115 #define DDF_RAID5EE 0x25
116 #define DDF_RAID6 0x06
117
118 /* Raid Level Qualifier (RLQ) */
119 #define DDF_RAID0_SIMPLE 0x00
120 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
121 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
122 #define DDF_RAID3_0 0x00 /* parity in first extent */
123 #define DDF_RAID3_N 0x01 /* parity in last extent */
124 #define DDF_RAID4_0 0x00 /* parity in first extent */
125 #define DDF_RAID4_N 0x01 /* parity in last extent */
126 /* these apply to raid5e and raid5ee as well */
127 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
128 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
129 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
130 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
131
132 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
133 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
134
135 /* Secondary RAID Level (SRL) */
136 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
137 #define DDF_2MIRRORED 0x01
138 #define DDF_2CONCAT 0x02
139 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
140
141 /* Magic numbers */
142 #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11)
143 #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111)
144 #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222)
145 #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333)
146 #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD)
147 #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE)
148 #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555)
149 #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888)
150 #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0)
151 #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C)
152
153 #define DDF_GUID_LEN 24
154 #define DDF_REVISION_0 "01.00.00"
155 #define DDF_REVISION_2 "01.02.00"
156
157 struct ddf_header {
158 be32 magic; /* DDF_HEADER_MAGIC */
159 be32 crc;
160 char guid[DDF_GUID_LEN];
161 char revision[8]; /* 01.02.00 */
162 be32 seq; /* starts at '1' */
163 be32 timestamp;
164 __u8 openflag;
165 __u8 foreignflag;
166 __u8 enforcegroups;
167 __u8 pad0; /* 0xff */
168 __u8 pad1[12]; /* 12 * 0xff */
169 /* 64 bytes so far */
170 __u8 header_ext[32]; /* reserved: fill with 0xff */
171 be64 primary_lba;
172 be64 secondary_lba;
173 __u8 type;
174 __u8 pad2[3]; /* 0xff */
175 be32 workspace_len; /* sectors for vendor space -
176 * at least 32768(sectors) */
177 be64 workspace_lba;
178 be16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
179 be16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
180 be16 max_partitions; /* i.e. max num of configuration
181 record entries per disk */
182 be16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
183 *12/512) */
184 be16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
185 __u8 pad3[54]; /* 0xff */
186 /* 192 bytes so far */
187 be32 controller_section_offset;
188 be32 controller_section_length;
189 be32 phys_section_offset;
190 be32 phys_section_length;
191 be32 virt_section_offset;
192 be32 virt_section_length;
193 be32 config_section_offset;
194 be32 config_section_length;
195 be32 data_section_offset;
196 be32 data_section_length;
197 be32 bbm_section_offset;
198 be32 bbm_section_length;
199 be32 diag_space_offset;
200 be32 diag_space_length;
201 be32 vendor_offset;
202 be32 vendor_length;
203 /* 256 bytes so far */
204 __u8 pad4[256]; /* 0xff */
205 };
206
207 /* type field */
208 #define DDF_HEADER_ANCHOR 0x00
209 #define DDF_HEADER_PRIMARY 0x01
210 #define DDF_HEADER_SECONDARY 0x02
211
212 /* The content of the 'controller section' - global scope */
213 struct ddf_controller_data {
214 be32 magic; /* DDF_CONTROLLER_MAGIC */
215 be32 crc;
216 char guid[DDF_GUID_LEN];
217 struct controller_type {
218 be16 vendor_id;
219 be16 device_id;
220 be16 sub_vendor_id;
221 be16 sub_device_id;
222 } type;
223 char product_id[16];
224 __u8 pad[8]; /* 0xff */
225 __u8 vendor_data[448];
226 };
227
228 /* The content of phys_section - global scope */
229 struct phys_disk {
230 be32 magic; /* DDF_PHYS_RECORDS_MAGIC */
231 be32 crc;
232 be16 used_pdes; /* This is a counter, not a max - the list
233 * of used entries may not be dense */
234 be16 max_pdes;
235 __u8 pad[52];
236 struct phys_disk_entry {
237 char guid[DDF_GUID_LEN];
238 be32 refnum;
239 be16 type;
240 be16 state;
241 be64 config_size; /* DDF structures must be after here */
242 char path[18]; /* Another horrible structure really
243 * but is "used for information
244 * purposes only" */
245 __u8 pad[6];
246 } entries[0];
247 };
248
249 /* phys_disk_entry.type is a bitmap - bigendian remember */
250 #define DDF_Forced_PD_GUID 1
251 #define DDF_Active_in_VD 2
252 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
253 #define DDF_Spare 8 /* overrides Global_spare */
254 #define DDF_Foreign 16
255 #define DDF_Legacy 32 /* no DDF on this device */
256
257 #define DDF_Interface_mask 0xf00
258 #define DDF_Interface_SCSI 0x100
259 #define DDF_Interface_SAS 0x200
260 #define DDF_Interface_SATA 0x300
261 #define DDF_Interface_FC 0x400
262
263 /* phys_disk_entry.state is a bigendian bitmap */
264 #define DDF_Online 1
265 #define DDF_Failed 2 /* overrides 1,4,8 */
266 #define DDF_Rebuilding 4
267 #define DDF_Transition 8
268 #define DDF_SMART 16
269 #define DDF_ReadErrors 32
270 #define DDF_Missing 64
271
272 /* The content of the virt_section global scope */
273 struct virtual_disk {
274 be32 magic; /* DDF_VIRT_RECORDS_MAGIC */
275 be32 crc;
276 be16 populated_vdes;
277 be16 max_vdes;
278 __u8 pad[52];
279 struct virtual_entry {
280 char guid[DDF_GUID_LEN];
281 be16 unit;
282 __u16 pad0; /* 0xffff */
283 be16 guid_crc;
284 be16 type;
285 __u8 state;
286 __u8 init_state;
287 __u8 pad1[14];
288 char name[16];
289 } entries[0];
290 };
291
292 /* virtual_entry.type is a bitmap - bigendian */
293 #define DDF_Shared 1
294 #define DDF_Enforce_Groups 2
295 #define DDF_Unicode 4
296 #define DDF_Owner_Valid 8
297
298 /* virtual_entry.state is a bigendian bitmap */
299 #define DDF_state_mask 0x7
300 #define DDF_state_optimal 0x0
301 #define DDF_state_degraded 0x1
302 #define DDF_state_deleted 0x2
303 #define DDF_state_missing 0x3
304 #define DDF_state_failed 0x4
305 #define DDF_state_part_optimal 0x5
306
307 #define DDF_state_morphing 0x8
308 #define DDF_state_inconsistent 0x10
309
310 /* virtual_entry.init_state is a bigendian bitmap */
311 #define DDF_initstate_mask 0x03
312 #define DDF_init_not 0x00
313 #define DDF_init_quick 0x01 /* initialisation is progress.
314 * i.e. 'state_inconsistent' */
315 #define DDF_init_full 0x02
316
317 #define DDF_access_mask 0xc0
318 #define DDF_access_rw 0x00
319 #define DDF_access_ro 0x80
320 #define DDF_access_blocked 0xc0
321
322 /* The content of the config_section - local scope
323 * It has multiple records each config_record_len sectors
324 * They can be vd_config or spare_assign
325 */
326
327 struct vd_config {
328 be32 magic; /* DDF_VD_CONF_MAGIC */
329 be32 crc;
330 char guid[DDF_GUID_LEN];
331 be32 timestamp;
332 be32 seqnum;
333 __u8 pad0[24];
334 be16 prim_elmnt_count;
335 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
336 __u8 prl;
337 __u8 rlq;
338 __u8 sec_elmnt_count;
339 __u8 sec_elmnt_seq;
340 __u8 srl;
341 be64 blocks; /* blocks per component could be different
342 * on different component devices...(only
343 * for concat I hope) */
344 be64 array_blocks; /* blocks in array */
345 __u8 pad1[8];
346 be32 spare_refs[8]; /* This is used to detect missing spares.
347 * As we don't have an interface for that
348 * the values are ignored.
349 */
350 __u8 cache_pol[8];
351 __u8 bg_rate;
352 __u8 pad2[3];
353 __u8 pad3[52];
354 __u8 pad4[192];
355 __u8 v0[32]; /* reserved- 0xff */
356 __u8 v1[32]; /* reserved- 0xff */
357 __u8 v2[16]; /* reserved- 0xff */
358 __u8 v3[16]; /* reserved- 0xff */
359 __u8 vendor[32];
360 be32 phys_refnum[0]; /* refnum of each disk in sequence */
361 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
362 bvd are always the same size */
363 };
364 #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe])
365
366 /* vd_config.cache_pol[7] is a bitmap */
367 #define DDF_cache_writeback 1 /* else writethrough */
368 #define DDF_cache_wadaptive 2 /* only applies if writeback */
369 #define DDF_cache_readahead 4
370 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
371 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
372 #define DDF_cache_wallowed 32 /* enable write caching */
373 #define DDF_cache_rallowed 64 /* enable read caching */
374
375 struct spare_assign {
376 be32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
377 be32 crc;
378 be32 timestamp;
379 __u8 reserved[7];
380 __u8 type;
381 be16 populated; /* SAEs used */
382 be16 max; /* max SAEs */
383 __u8 pad[8];
384 struct spare_assign_entry {
385 char guid[DDF_GUID_LEN];
386 be16 secondary_element;
387 __u8 pad[6];
388 } spare_ents[0];
389 };
390 /* spare_assign.type is a bitmap */
391 #define DDF_spare_dedicated 0x1 /* else global */
392 #define DDF_spare_revertible 0x2 /* else committable */
393 #define DDF_spare_active 0x4 /* else not active */
394 #define DDF_spare_affinity 0x8 /* enclosure affinity */
395
396 /* The data_section contents - local scope */
397 struct disk_data {
398 be32 magic; /* DDF_PHYS_DATA_MAGIC */
399 be32 crc;
400 char guid[DDF_GUID_LEN];
401 be32 refnum; /* crc of some magic drive data ... */
402 __u8 forced_ref; /* set when above was not result of magic */
403 __u8 forced_guid; /* set if guid was forced rather than magic */
404 __u8 vendor[32];
405 __u8 pad[442];
406 };
407
408 /* bbm_section content */
409 struct bad_block_log {
410 be32 magic;
411 be32 crc;
412 be16 entry_count;
413 be32 spare_count;
414 __u8 pad[10];
415 be64 first_spare;
416 struct mapped_block {
417 be64 defective_start;
418 be32 replacement_start;
419 be16 remap_count;
420 __u8 pad[2];
421 } entries[0];
422 };
423
424 /* Struct for internally holding ddf structures */
425 /* The DDF structure stored on each device is potentially
426 * quite different, as some data is global and some is local.
427 * The global data is:
428 * - ddf header
429 * - controller_data
430 * - Physical disk records
431 * - Virtual disk records
432 * The local data is:
433 * - Configuration records
434 * - Physical Disk data section
435 * ( and Bad block and vendor which I don't care about yet).
436 *
437 * The local data is parsed into separate lists as it is read
438 * and reconstructed for writing. This means that we only need
439 * to make config changes once and they are automatically
440 * propagated to all devices.
441 * The global (config and disk data) records are each in a list
442 * of separate data structures. When writing we find the entry
443 * or entries applicable to the particular device.
444 */
445 struct ddf_super {
446 struct ddf_header anchor, primary, secondary;
447 struct ddf_controller_data controller;
448 struct ddf_header *active;
449 struct phys_disk *phys;
450 struct virtual_disk *virt;
451 char *conf;
452 int pdsize, vdsize;
453 unsigned int max_part, mppe, conf_rec_len;
454 int currentdev;
455 int updates_pending;
456 struct vcl {
457 union {
458 char space[512];
459 struct {
460 struct vcl *next;
461 unsigned int vcnum; /* index into ->virt */
462 /* For an array with a secondary level there are
463 * multiple vd_config structures, all with the same
464 * guid but with different sec_elmnt_seq.
465 * One of these structures is in 'conf' below.
466 * The others are in other_bvds, not in any
467 * particular order.
468 */
469 struct vd_config **other_bvds;
470 __u64 *block_sizes; /* NULL if all the same */
471 };
472 };
473 struct vd_config conf;
474 } *conflist, *currentconf;
475 struct dl {
476 union {
477 char space[512];
478 struct {
479 struct dl *next;
480 int major, minor;
481 char *devname;
482 int fd;
483 unsigned long long size; /* sectors */
484 be64 primary_lba; /* sectors */
485 be64 secondary_lba; /* sectors */
486 be64 workspace_lba; /* sectors */
487 int pdnum; /* index in ->phys */
488 struct spare_assign *spare;
489 void *mdupdate; /* hold metadata update */
490
491 /* These fields used by auto-layout */
492 int raiddisk; /* slot to fill in autolayout */
493 __u64 esize;
494 int displayed;
495 };
496 };
497 struct disk_data disk;
498 struct vcl *vlist[0]; /* max_part in size */
499 } *dlist, *add_list;
500 };
501
502 #ifndef MDASSEMBLE
503 static int load_super_ddf_all(struct supertype *st, int fd,
504 void **sbp, char *devname);
505 static int get_svd_state(const struct ddf_super *, const struct vcl *);
506 static int
507 validate_geometry_ddf_container(struct supertype *st,
508 int level, int layout, int raiddisks,
509 int chunk, unsigned long long size,
510 unsigned long long data_offset,
511 char *dev, unsigned long long *freesize,
512 int verbose);
513
514 static int validate_geometry_ddf_bvd(struct supertype *st,
515 int level, int layout, int raiddisks,
516 int *chunk, unsigned long long size,
517 unsigned long long data_offset,
518 char *dev, unsigned long long *freesize,
519 int verbose);
520 #endif
521
522 static void free_super_ddf(struct supertype *st);
523 static int all_ff(const char *guid);
524 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
525 be32 refnum, unsigned int nmax,
526 const struct vd_config **bvd,
527 unsigned int *idx);
528 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map);
529 static void uuid_from_ddf_guid(const char *guid, int uuid[4]);
530 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
531 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i);
532 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map);
533 static int init_super_ddf_bvd(struct supertype *st,
534 mdu_array_info_t *info,
535 unsigned long long size,
536 char *name, char *homehost,
537 int *uuid, unsigned long long data_offset);
538
539 #ifndef offsetof
540 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
541 #endif
542
543 #if DEBUG
544 static void pr_state(struct ddf_super *ddf, const char *msg)
545 {
546 unsigned int i;
547 dprintf("%s/%s: ", __func__, msg);
548 for (i = 0; i < be16_to_cpu(ddf->active->max_vd_entries); i++) {
549 if (all_ff(ddf->virt->entries[i].guid))
550 continue;
551 dprintf("%u(s=%02x i=%02x) ", i,
552 ddf->virt->entries[i].state,
553 ddf->virt->entries[i].init_state);
554 }
555 dprintf("\n");
556 }
557 #else
558 static void pr_state(const struct ddf_super *ddf, const char *msg) {}
559 #endif
560
561 static void _ddf_set_updates_pending(struct ddf_super *ddf, struct vd_config *vc,
562 const char *func)
563 {
564 if (vc) {
565 vc->timestamp = cpu_to_be32(time(0)-DECADE);
566 vc->seqnum = cpu_to_be32(be32_to_cpu(vc->seqnum) + 1);
567 }
568 if (ddf->updates_pending)
569 return;
570 ddf->updates_pending = 1;
571 ddf->active->seq = cpu_to_be32((be32_to_cpu(ddf->active->seq)+1));
572 pr_state(ddf, func);
573 }
574
575 #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__)
576
577 static be32 calc_crc(void *buf, int len)
578 {
579 /* crcs are always at the same place as in the ddf_header */
580 struct ddf_header *ddf = buf;
581 be32 oldcrc = ddf->crc;
582 __u32 newcrc;
583 ddf->crc = cpu_to_be32(0xffffffff);
584
585 newcrc = crc32(0, buf, len);
586 ddf->crc = oldcrc;
587 /* The crc is stored (like everything) bigendian, so convert
588 * here for simplicity
589 */
590 return cpu_to_be32(newcrc);
591 }
592
593 #define DDF_INVALID_LEVEL 0xff
594 #define DDF_NO_SECONDARY 0xff
595 static int err_bad_md_layout(const mdu_array_info_t *array)
596 {
597 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
598 array->level, array->layout, array->raid_disks);
599 return -1;
600 }
601
602 static int layout_md2ddf(const mdu_array_info_t *array,
603 struct vd_config *conf)
604 {
605 be16 prim_elmnt_count = cpu_to_be16(array->raid_disks);
606 __u8 prl = DDF_INVALID_LEVEL, rlq = 0;
607 __u8 sec_elmnt_count = 1;
608 __u8 srl = DDF_NO_SECONDARY;
609
610 switch (array->level) {
611 case LEVEL_LINEAR:
612 prl = DDF_CONCAT;
613 break;
614 case 0:
615 rlq = DDF_RAID0_SIMPLE;
616 prl = DDF_RAID0;
617 break;
618 case 1:
619 switch (array->raid_disks) {
620 case 2:
621 rlq = DDF_RAID1_SIMPLE;
622 break;
623 case 3:
624 rlq = DDF_RAID1_MULTI;
625 break;
626 default:
627 return err_bad_md_layout(array);
628 }
629 prl = DDF_RAID1;
630 break;
631 case 4:
632 if (array->layout != 0)
633 return err_bad_md_layout(array);
634 rlq = DDF_RAID4_N;
635 prl = DDF_RAID4;
636 break;
637 case 5:
638 switch (array->layout) {
639 case ALGORITHM_LEFT_ASYMMETRIC:
640 rlq = DDF_RAID5_N_RESTART;
641 break;
642 case ALGORITHM_RIGHT_ASYMMETRIC:
643 rlq = DDF_RAID5_0_RESTART;
644 break;
645 case ALGORITHM_LEFT_SYMMETRIC:
646 rlq = DDF_RAID5_N_CONTINUE;
647 break;
648 case ALGORITHM_RIGHT_SYMMETRIC:
649 /* not mentioned in standard */
650 default:
651 return err_bad_md_layout(array);
652 }
653 prl = DDF_RAID5;
654 break;
655 case 6:
656 switch (array->layout) {
657 case ALGORITHM_ROTATING_N_RESTART:
658 rlq = DDF_RAID5_N_RESTART;
659 break;
660 case ALGORITHM_ROTATING_ZERO_RESTART:
661 rlq = DDF_RAID6_0_RESTART;
662 break;
663 case ALGORITHM_ROTATING_N_CONTINUE:
664 rlq = DDF_RAID5_N_CONTINUE;
665 break;
666 default:
667 return err_bad_md_layout(array);
668 }
669 prl = DDF_RAID6;
670 break;
671 case 10:
672 if (array->raid_disks % 2 == 0 && array->layout == 0x102) {
673 rlq = DDF_RAID1_SIMPLE;
674 prim_elmnt_count = cpu_to_be16(2);
675 sec_elmnt_count = array->raid_disks / 2;
676 } else if (array->raid_disks % 3 == 0
677 && array->layout == 0x103) {
678 rlq = DDF_RAID1_MULTI;
679 prim_elmnt_count = cpu_to_be16(3);
680 sec_elmnt_count = array->raid_disks / 3;
681 } else
682 return err_bad_md_layout(array);
683 srl = DDF_2SPANNED;
684 prl = DDF_RAID1;
685 break;
686 default:
687 return err_bad_md_layout(array);
688 }
689 conf->prl = prl;
690 conf->prim_elmnt_count = prim_elmnt_count;
691 conf->rlq = rlq;
692 conf->srl = srl;
693 conf->sec_elmnt_count = sec_elmnt_count;
694 return 0;
695 }
696
697 static int err_bad_ddf_layout(const struct vd_config *conf)
698 {
699 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
700 conf->prl, conf->rlq, be16_to_cpu(conf->prim_elmnt_count));
701 return -1;
702 }
703
704 static int layout_ddf2md(const struct vd_config *conf,
705 mdu_array_info_t *array)
706 {
707 int level = LEVEL_UNSUPPORTED;
708 int layout = 0;
709 int raiddisks = be16_to_cpu(conf->prim_elmnt_count);
710
711 if (conf->sec_elmnt_count > 1) {
712 /* see also check_secondary() */
713 if (conf->prl != DDF_RAID1 ||
714 (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED)) {
715 pr_err("Unsupported secondary RAID level %u/%u\n",
716 conf->prl, conf->srl);
717 return -1;
718 }
719 if (raiddisks == 2 && conf->rlq == DDF_RAID1_SIMPLE)
720 layout = 0x102;
721 else if (raiddisks == 3 && conf->rlq == DDF_RAID1_MULTI)
722 layout = 0x103;
723 else
724 return err_bad_ddf_layout(conf);
725 raiddisks *= conf->sec_elmnt_count;
726 level = 10;
727 goto good;
728 }
729
730 switch (conf->prl) {
731 case DDF_CONCAT:
732 level = LEVEL_LINEAR;
733 break;
734 case DDF_RAID0:
735 if (conf->rlq != DDF_RAID0_SIMPLE)
736 return err_bad_ddf_layout(conf);
737 level = 0;
738 break;
739 case DDF_RAID1:
740 if (!((conf->rlq == DDF_RAID1_SIMPLE && raiddisks == 2) ||
741 (conf->rlq == DDF_RAID1_MULTI && raiddisks == 3)))
742 return err_bad_ddf_layout(conf);
743 level = 1;
744 break;
745 case DDF_RAID4:
746 if (conf->rlq != DDF_RAID4_N)
747 return err_bad_ddf_layout(conf);
748 level = 4;
749 break;
750 case DDF_RAID5:
751 switch (conf->rlq) {
752 case DDF_RAID5_N_RESTART:
753 layout = ALGORITHM_LEFT_ASYMMETRIC;
754 break;
755 case DDF_RAID5_0_RESTART:
756 layout = ALGORITHM_RIGHT_ASYMMETRIC;
757 break;
758 case DDF_RAID5_N_CONTINUE:
759 layout = ALGORITHM_LEFT_SYMMETRIC;
760 break;
761 default:
762 return err_bad_ddf_layout(conf);
763 }
764 level = 5;
765 break;
766 case DDF_RAID6:
767 switch (conf->rlq) {
768 case DDF_RAID5_N_RESTART:
769 layout = ALGORITHM_ROTATING_N_RESTART;
770 break;
771 case DDF_RAID6_0_RESTART:
772 layout = ALGORITHM_ROTATING_ZERO_RESTART;
773 break;
774 case DDF_RAID5_N_CONTINUE:
775 layout = ALGORITHM_ROTATING_N_CONTINUE;
776 break;
777 default:
778 return err_bad_ddf_layout(conf);
779 }
780 level = 6;
781 break;
782 default:
783 return err_bad_ddf_layout(conf);
784 };
785
786 good:
787 array->level = level;
788 array->layout = layout;
789 array->raid_disks = raiddisks;
790 return 0;
791 }
792
793 static int load_ddf_header(int fd, unsigned long long lba,
794 unsigned long long size,
795 int type,
796 struct ddf_header *hdr, struct ddf_header *anchor)
797 {
798 /* read a ddf header (primary or secondary) from fd/lba
799 * and check that it is consistent with anchor
800 * Need to check:
801 * magic, crc, guid, rev, and LBA's header_type, and
802 * everything after header_type must be the same
803 */
804 if (lba >= size-1)
805 return 0;
806
807 if (lseek64(fd, lba<<9, 0) < 0)
808 return 0;
809
810 if (read(fd, hdr, 512) != 512)
811 return 0;
812
813 if (!be32_eq(hdr->magic, DDF_HEADER_MAGIC)) {
814 pr_err("%s: bad header magic\n", __func__);
815 return 0;
816 }
817 if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) {
818 pr_err("%s: bad CRC\n", __func__);
819 return 0;
820 }
821 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
822 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
823 !be64_eq(anchor->primary_lba, hdr->primary_lba) ||
824 !be64_eq(anchor->secondary_lba, hdr->secondary_lba) ||
825 hdr->type != type ||
826 memcmp(anchor->pad2, hdr->pad2, 512 -
827 offsetof(struct ddf_header, pad2)) != 0) {
828 pr_err("%s: header mismatch\n", __func__);
829 return 0;
830 }
831
832 /* Looks good enough to me... */
833 return 1;
834 }
835
836 static void *load_section(int fd, struct ddf_super *super, void *buf,
837 be32 offset_be, be32 len_be, int check)
838 {
839 unsigned long long offset = be32_to_cpu(offset_be);
840 unsigned long long len = be32_to_cpu(len_be);
841 int dofree = (buf == NULL);
842
843 if (check)
844 if (len != 2 && len != 8 && len != 32
845 && len != 128 && len != 512)
846 return NULL;
847
848 if (len > 1024)
849 return NULL;
850 if (!buf && posix_memalign(&buf, 512, len<<9) != 0)
851 buf = NULL;
852
853 if (!buf)
854 return NULL;
855
856 if (super->active->type == 1)
857 offset += be64_to_cpu(super->active->primary_lba);
858 else
859 offset += be64_to_cpu(super->active->secondary_lba);
860
861 if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
862 if (dofree)
863 free(buf);
864 return NULL;
865 }
866 if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
867 if (dofree)
868 free(buf);
869 return NULL;
870 }
871 return buf;
872 }
873
874 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
875 {
876 unsigned long long dsize;
877
878 get_dev_size(fd, NULL, &dsize);
879
880 if (lseek64(fd, dsize-512, 0) < 0) {
881 if (devname)
882 pr_err("Cannot seek to anchor block on %s: %s\n",
883 devname, strerror(errno));
884 return 1;
885 }
886 if (read(fd, &super->anchor, 512) != 512) {
887 if (devname)
888 pr_err("Cannot read anchor block on %s: %s\n",
889 devname, strerror(errno));
890 return 1;
891 }
892 if (!be32_eq(super->anchor.magic, DDF_HEADER_MAGIC)) {
893 if (devname)
894 pr_err("no DDF anchor found on %s\n",
895 devname);
896 return 2;
897 }
898 if (!be32_eq(calc_crc(&super->anchor, 512), super->anchor.crc)) {
899 if (devname)
900 pr_err("bad CRC on anchor on %s\n",
901 devname);
902 return 2;
903 }
904 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
905 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
906 if (devname)
907 pr_err("can only support super revision"
908 " %.8s and earlier, not %.8s on %s\n",
909 DDF_REVISION_2, super->anchor.revision,devname);
910 return 2;
911 }
912 super->active = NULL;
913 if (load_ddf_header(fd, be64_to_cpu(super->anchor.primary_lba),
914 dsize >> 9, 1,
915 &super->primary, &super->anchor) == 0) {
916 if (devname)
917 pr_err("Failed to load primary DDF header "
918 "on %s\n", devname);
919 } else
920 super->active = &super->primary;
921
922 if (load_ddf_header(fd, be64_to_cpu(super->anchor.secondary_lba),
923 dsize >> 9, 2,
924 &super->secondary, &super->anchor)) {
925 if (super->active == NULL
926 || (be32_to_cpu(super->primary.seq)
927 < be32_to_cpu(super->secondary.seq) &&
928 !super->secondary.openflag)
929 || (be32_to_cpu(super->primary.seq)
930 == be32_to_cpu(super->secondary.seq) &&
931 super->primary.openflag && !super->secondary.openflag)
932 )
933 super->active = &super->secondary;
934 } else if (devname &&
935 be64_to_cpu(super->anchor.secondary_lba) != ~(__u64)0)
936 pr_err("Failed to load secondary DDF header on %s\n",
937 devname);
938 if (super->active == NULL)
939 return 2;
940 return 0;
941 }
942
943 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
944 {
945 void *ok;
946 ok = load_section(fd, super, &super->controller,
947 super->active->controller_section_offset,
948 super->active->controller_section_length,
949 0);
950 super->phys = load_section(fd, super, NULL,
951 super->active->phys_section_offset,
952 super->active->phys_section_length,
953 1);
954 super->pdsize = be32_to_cpu(super->active->phys_section_length) * 512;
955
956 super->virt = load_section(fd, super, NULL,
957 super->active->virt_section_offset,
958 super->active->virt_section_length,
959 1);
960 super->vdsize = be32_to_cpu(super->active->virt_section_length) * 512;
961 if (!ok ||
962 !super->phys ||
963 !super->virt) {
964 free(super->phys);
965 free(super->virt);
966 super->phys = NULL;
967 super->virt = NULL;
968 return 2;
969 }
970 super->conflist = NULL;
971 super->dlist = NULL;
972
973 super->max_part = be16_to_cpu(super->active->max_partitions);
974 super->mppe = be16_to_cpu(super->active->max_primary_element_entries);
975 super->conf_rec_len = be16_to_cpu(super->active->config_record_len);
976 return 0;
977 }
978
979 #define DDF_UNUSED_BVD 0xff
980 static int alloc_other_bvds(const struct ddf_super *ddf, struct vcl *vcl)
981 {
982 unsigned int n_vds = vcl->conf.sec_elmnt_count - 1;
983 unsigned int i, vdsize;
984 void *p;
985 if (n_vds == 0) {
986 vcl->other_bvds = NULL;
987 return 0;
988 }
989 vdsize = ddf->conf_rec_len * 512;
990 if (posix_memalign(&p, 512, n_vds *
991 (vdsize + sizeof(struct vd_config *))) != 0)
992 return -1;
993 vcl->other_bvds = (struct vd_config **) (p + n_vds * vdsize);
994 for (i = 0; i < n_vds; i++) {
995 vcl->other_bvds[i] = p + i * vdsize;
996 memset(vcl->other_bvds[i], 0, vdsize);
997 vcl->other_bvds[i]->sec_elmnt_seq = DDF_UNUSED_BVD;
998 }
999 return 0;
1000 }
1001
1002 static void add_other_bvd(struct vcl *vcl, struct vd_config *vd,
1003 unsigned int len)
1004 {
1005 int i;
1006 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1007 if (vcl->other_bvds[i]->sec_elmnt_seq == vd->sec_elmnt_seq)
1008 break;
1009
1010 if (i < vcl->conf.sec_elmnt_count-1) {
1011 if (be32_to_cpu(vd->seqnum) <=
1012 be32_to_cpu(vcl->other_bvds[i]->seqnum))
1013 return;
1014 } else {
1015 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1016 if (vcl->other_bvds[i]->sec_elmnt_seq == DDF_UNUSED_BVD)
1017 break;
1018 if (i == vcl->conf.sec_elmnt_count-1) {
1019 pr_err("no space for sec level config %u, count is %u\n",
1020 vd->sec_elmnt_seq, vcl->conf.sec_elmnt_count);
1021 return;
1022 }
1023 }
1024 memcpy(vcl->other_bvds[i], vd, len);
1025 }
1026
1027 static int load_ddf_local(int fd, struct ddf_super *super,
1028 char *devname, int keep)
1029 {
1030 struct dl *dl;
1031 struct stat stb;
1032 char *conf;
1033 unsigned int i;
1034 unsigned int confsec;
1035 int vnum;
1036 unsigned int max_virt_disks =
1037 be16_to_cpu(super->active->max_vd_entries);
1038 unsigned long long dsize;
1039
1040 /* First the local disk info */
1041 if (posix_memalign((void**)&dl, 512,
1042 sizeof(*dl) +
1043 (super->max_part) * sizeof(dl->vlist[0])) != 0) {
1044 pr_err("%s could not allocate disk info buffer\n",
1045 __func__);
1046 return 1;
1047 }
1048
1049 load_section(fd, super, &dl->disk,
1050 super->active->data_section_offset,
1051 super->active->data_section_length,
1052 0);
1053 dl->devname = devname ? xstrdup(devname) : NULL;
1054
1055 fstat(fd, &stb);
1056 dl->major = major(stb.st_rdev);
1057 dl->minor = minor(stb.st_rdev);
1058 dl->next = super->dlist;
1059 dl->fd = keep ? fd : -1;
1060
1061 dl->size = 0;
1062 if (get_dev_size(fd, devname, &dsize))
1063 dl->size = dsize >> 9;
1064 /* If the disks have different sizes, the LBAs will differ
1065 * between phys disks.
1066 * At this point here, the values in super->active must be valid
1067 * for this phys disk. */
1068 dl->primary_lba = super->active->primary_lba;
1069 dl->secondary_lba = super->active->secondary_lba;
1070 dl->workspace_lba = super->active->workspace_lba;
1071 dl->spare = NULL;
1072 for (i = 0 ; i < super->max_part ; i++)
1073 dl->vlist[i] = NULL;
1074 super->dlist = dl;
1075 dl->pdnum = -1;
1076 for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++)
1077 if (memcmp(super->phys->entries[i].guid,
1078 dl->disk.guid, DDF_GUID_LEN) == 0)
1079 dl->pdnum = i;
1080
1081 /* Now the config list. */
1082 /* 'conf' is an array of config entries, some of which are
1083 * probably invalid. Those which are good need to be copied into
1084 * the conflist
1085 */
1086
1087 conf = load_section(fd, super, super->conf,
1088 super->active->config_section_offset,
1089 super->active->config_section_length,
1090 0);
1091 super->conf = conf;
1092 vnum = 0;
1093 for (confsec = 0;
1094 confsec < be32_to_cpu(super->active->config_section_length);
1095 confsec += super->conf_rec_len) {
1096 struct vd_config *vd =
1097 (struct vd_config *)((char*)conf + confsec*512);
1098 struct vcl *vcl;
1099
1100 if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) {
1101 if (dl->spare)
1102 continue;
1103 if (posix_memalign((void**)&dl->spare, 512,
1104 super->conf_rec_len*512) != 0) {
1105 pr_err("%s could not allocate spare info buf\n",
1106 __func__);
1107 return 1;
1108 }
1109
1110 memcpy(dl->spare, vd, super->conf_rec_len*512);
1111 continue;
1112 }
1113 if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC))
1114 /* Must be vendor-unique - I cannot handle those */
1115 continue;
1116
1117 for (vcl = super->conflist; vcl; vcl = vcl->next) {
1118 if (memcmp(vcl->conf.guid,
1119 vd->guid, DDF_GUID_LEN) == 0)
1120 break;
1121 }
1122
1123 if (vcl) {
1124 dl->vlist[vnum++] = vcl;
1125 if (vcl->other_bvds != NULL &&
1126 vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) {
1127 add_other_bvd(vcl, vd, super->conf_rec_len*512);
1128 continue;
1129 }
1130 if (be32_to_cpu(vd->seqnum) <=
1131 be32_to_cpu(vcl->conf.seqnum))
1132 continue;
1133 } else {
1134 if (posix_memalign((void**)&vcl, 512,
1135 (super->conf_rec_len*512 +
1136 offsetof(struct vcl, conf))) != 0) {
1137 pr_err("%s could not allocate vcl buf\n",
1138 __func__);
1139 return 1;
1140 }
1141 vcl->next = super->conflist;
1142 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1143 vcl->conf.sec_elmnt_count = vd->sec_elmnt_count;
1144 if (alloc_other_bvds(super, vcl) != 0) {
1145 pr_err("%s could not allocate other bvds\n",
1146 __func__);
1147 free(vcl);
1148 return 1;
1149 };
1150 super->conflist = vcl;
1151 dl->vlist[vnum++] = vcl;
1152 }
1153 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
1154 for (i=0; i < max_virt_disks ; i++)
1155 if (memcmp(super->virt->entries[i].guid,
1156 vcl->conf.guid, DDF_GUID_LEN)==0)
1157 break;
1158 if (i < max_virt_disks)
1159 vcl->vcnum = i;
1160 }
1161
1162 return 0;
1163 }
1164
1165 static int load_super_ddf(struct supertype *st, int fd,
1166 char *devname)
1167 {
1168 unsigned long long dsize;
1169 struct ddf_super *super;
1170 int rv;
1171
1172 if (get_dev_size(fd, devname, &dsize) == 0)
1173 return 1;
1174
1175 if (test_partition(fd))
1176 /* DDF is not allowed on partitions */
1177 return 1;
1178
1179 /* 32M is a lower bound */
1180 if (dsize <= 32*1024*1024) {
1181 if (devname)
1182 pr_err("%s is too small for ddf: "
1183 "size is %llu sectors.\n",
1184 devname, dsize>>9);
1185 return 1;
1186 }
1187 if (dsize & 511) {
1188 if (devname)
1189 pr_err("%s is an odd size for ddf: "
1190 "size is %llu bytes.\n",
1191 devname, dsize);
1192 return 1;
1193 }
1194
1195 free_super_ddf(st);
1196
1197 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
1198 pr_err("malloc of %zu failed.\n",
1199 sizeof(*super));
1200 return 1;
1201 }
1202 memset(super, 0, sizeof(*super));
1203
1204 rv = load_ddf_headers(fd, super, devname);
1205 if (rv) {
1206 free(super);
1207 return rv;
1208 }
1209
1210 /* Have valid headers and have chosen the best. Let's read in the rest*/
1211
1212 rv = load_ddf_global(fd, super, devname);
1213
1214 if (rv) {
1215 if (devname)
1216 pr_err("Failed to load all information "
1217 "sections on %s\n", devname);
1218 free(super);
1219 return rv;
1220 }
1221
1222 rv = load_ddf_local(fd, super, devname, 0);
1223
1224 if (rv) {
1225 if (devname)
1226 pr_err("Failed to load all information "
1227 "sections on %s\n", devname);
1228 free(super);
1229 return rv;
1230 }
1231
1232 /* Should possibly check the sections .... */
1233
1234 st->sb = super;
1235 if (st->ss == NULL) {
1236 st->ss = &super_ddf;
1237 st->minor_version = 0;
1238 st->max_devs = 512;
1239 }
1240 return 0;
1241
1242 }
1243
1244 static void free_super_ddf(struct supertype *st)
1245 {
1246 struct ddf_super *ddf = st->sb;
1247 if (ddf == NULL)
1248 return;
1249 free(ddf->phys);
1250 free(ddf->virt);
1251 free(ddf->conf);
1252 while (ddf->conflist) {
1253 struct vcl *v = ddf->conflist;
1254 ddf->conflist = v->next;
1255 if (v->block_sizes)
1256 free(v->block_sizes);
1257 if (v->other_bvds)
1258 /*
1259 v->other_bvds[0] points to beginning of buffer,
1260 see alloc_other_bvds()
1261 */
1262 free(v->other_bvds[0]);
1263 free(v);
1264 }
1265 while (ddf->dlist) {
1266 struct dl *d = ddf->dlist;
1267 ddf->dlist = d->next;
1268 if (d->fd >= 0)
1269 close(d->fd);
1270 if (d->spare)
1271 free(d->spare);
1272 free(d);
1273 }
1274 while (ddf->add_list) {
1275 struct dl *d = ddf->add_list;
1276 ddf->add_list = d->next;
1277 if (d->fd >= 0)
1278 close(d->fd);
1279 if (d->spare)
1280 free(d->spare);
1281 free(d);
1282 }
1283 free(ddf);
1284 st->sb = NULL;
1285 }
1286
1287 static struct supertype *match_metadata_desc_ddf(char *arg)
1288 {
1289 /* 'ddf' only supports containers */
1290 struct supertype *st;
1291 if (strcmp(arg, "ddf") != 0 &&
1292 strcmp(arg, "default") != 0
1293 )
1294 return NULL;
1295
1296 st = xcalloc(1, sizeof(*st));
1297 st->ss = &super_ddf;
1298 st->max_devs = 512;
1299 st->minor_version = 0;
1300 st->sb = NULL;
1301 return st;
1302 }
1303
1304 #ifndef MDASSEMBLE
1305
1306 static mapping_t ddf_state[] = {
1307 { "Optimal", 0},
1308 { "Degraded", 1},
1309 { "Deleted", 2},
1310 { "Missing", 3},
1311 { "Failed", 4},
1312 { "Partially Optimal", 5},
1313 { "-reserved-", 6},
1314 { "-reserved-", 7},
1315 { NULL, 0}
1316 };
1317
1318 static mapping_t ddf_init_state[] = {
1319 { "Not Initialised", 0},
1320 { "QuickInit in Progress", 1},
1321 { "Fully Initialised", 2},
1322 { "*UNKNOWN*", 3},
1323 { NULL, 0}
1324 };
1325 static mapping_t ddf_access[] = {
1326 { "Read/Write", 0},
1327 { "Reserved", 1},
1328 { "Read Only", 2},
1329 { "Blocked (no access)", 3},
1330 { NULL ,0}
1331 };
1332
1333 static mapping_t ddf_level[] = {
1334 { "RAID0", DDF_RAID0},
1335 { "RAID1", DDF_RAID1},
1336 { "RAID3", DDF_RAID3},
1337 { "RAID4", DDF_RAID4},
1338 { "RAID5", DDF_RAID5},
1339 { "RAID1E",DDF_RAID1E},
1340 { "JBOD", DDF_JBOD},
1341 { "CONCAT",DDF_CONCAT},
1342 { "RAID5E",DDF_RAID5E},
1343 { "RAID5EE",DDF_RAID5EE},
1344 { "RAID6", DDF_RAID6},
1345 { NULL, 0}
1346 };
1347 static mapping_t ddf_sec_level[] = {
1348 { "Striped", DDF_2STRIPED},
1349 { "Mirrored", DDF_2MIRRORED},
1350 { "Concat", DDF_2CONCAT},
1351 { "Spanned", DDF_2SPANNED},
1352 { NULL, 0}
1353 };
1354 #endif
1355
1356 static int all_ff(const char *guid)
1357 {
1358 int i;
1359 for (i = 0; i < DDF_GUID_LEN; i++)
1360 if (guid[i] != (char)0xff)
1361 return 0;
1362 return 1;
1363 }
1364
1365 static const char *guid_str(const char *guid)
1366 {
1367 static char buf[DDF_GUID_LEN*2+1];
1368 int i;
1369 char *p = buf;
1370 for (i = 0; i < DDF_GUID_LEN; i++) {
1371 unsigned char c = guid[i];
1372 if (c >= 32 && c < 127)
1373 p += sprintf(p, "%c", c);
1374 else
1375 p += sprintf(p, "%02x", c);
1376 }
1377 *p = '\0';
1378 return (const char *) buf;
1379 }
1380
1381 #ifndef MDASSEMBLE
1382 static void print_guid(char *guid, int tstamp)
1383 {
1384 /* A GUIDs are part (or all) ASCII and part binary.
1385 * They tend to be space padded.
1386 * We print the GUID in HEX, then in parentheses add
1387 * any initial ASCII sequence, and a possible
1388 * time stamp from bytes 16-19
1389 */
1390 int l = DDF_GUID_LEN;
1391 int i;
1392
1393 for (i=0 ; i<DDF_GUID_LEN ; i++) {
1394 if ((i&3)==0 && i != 0) printf(":");
1395 printf("%02X", guid[i]&255);
1396 }
1397
1398 printf("\n (");
1399 while (l && guid[l-1] == ' ')
1400 l--;
1401 for (i=0 ; i<l ; i++) {
1402 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1403 fputc(guid[i], stdout);
1404 else
1405 break;
1406 }
1407 if (tstamp) {
1408 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1409 char tbuf[100];
1410 struct tm *tm;
1411 tm = localtime(&then);
1412 strftime(tbuf, 100, " %D %T",tm);
1413 fputs(tbuf, stdout);
1414 }
1415 printf(")");
1416 }
1417
1418 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1419 {
1420 int crl = sb->conf_rec_len;
1421 struct vcl *vcl;
1422
1423 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1424 unsigned int i;
1425 struct vd_config *vc = &vcl->conf;
1426
1427 if (!be32_eq(calc_crc(vc, crl*512), vc->crc))
1428 continue;
1429 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1430 continue;
1431
1432 /* Ok, we know about this VD, let's give more details */
1433 printf(" Raid Devices[%d] : %d (", n,
1434 be16_to_cpu(vc->prim_elmnt_count));
1435 for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) {
1436 int j;
1437 int cnt = be16_to_cpu(sb->phys->max_pdes);
1438 for (j=0; j<cnt; j++)
1439 if (be32_eq(vc->phys_refnum[i],
1440 sb->phys->entries[j].refnum))
1441 break;
1442 if (i) printf(" ");
1443 if (j < cnt)
1444 printf("%d", j);
1445 else
1446 printf("--");
1447 }
1448 printf(")\n");
1449 if (vc->chunk_shift != 255)
1450 printf(" Chunk Size[%d] : %d sectors\n", n,
1451 1 << vc->chunk_shift);
1452 printf(" Raid Level[%d] : %s\n", n,
1453 map_num(ddf_level, vc->prl)?:"-unknown-");
1454 if (vc->sec_elmnt_count != 1) {
1455 printf(" Secondary Position[%d] : %d of %d\n", n,
1456 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1457 printf(" Secondary Level[%d] : %s\n", n,
1458 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1459 }
1460 printf(" Device Size[%d] : %llu\n", n,
1461 be64_to_cpu(vc->blocks)/2);
1462 printf(" Array Size[%d] : %llu\n", n,
1463 be64_to_cpu(vc->array_blocks)/2);
1464 }
1465 }
1466
1467 static void examine_vds(struct ddf_super *sb)
1468 {
1469 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1470 unsigned int i;
1471 printf(" Virtual Disks : %d\n", cnt);
1472
1473 for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) {
1474 struct virtual_entry *ve = &sb->virt->entries[i];
1475 if (all_ff(ve->guid))
1476 continue;
1477 printf("\n");
1478 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1479 printf("\n");
1480 printf(" unit[%d] : %d\n", i, be16_to_cpu(ve->unit));
1481 printf(" state[%d] : %s, %s%s\n", i,
1482 map_num(ddf_state, ve->state & 7),
1483 (ve->state & DDF_state_morphing) ? "Morphing, ": "",
1484 (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent");
1485 printf(" init state[%d] : %s\n", i,
1486 map_num(ddf_init_state, ve->init_state&DDF_initstate_mask));
1487 printf(" access[%d] : %s\n", i,
1488 map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6));
1489 printf(" Name[%d] : %.16s\n", i, ve->name);
1490 examine_vd(i, sb, ve->guid);
1491 }
1492 if (cnt) printf("\n");
1493 }
1494
1495 static void examine_pds(struct ddf_super *sb)
1496 {
1497 int cnt = be16_to_cpu(sb->phys->max_pdes);
1498 int i;
1499 struct dl *dl;
1500 int unlisted = 0;
1501 printf(" Physical Disks : %d\n", cnt);
1502 printf(" Number RefNo Size Device Type/State\n");
1503
1504 for (dl = sb->dlist; dl; dl = dl->next)
1505 dl->displayed = 0;
1506
1507 for (i=0 ; i<cnt ; i++) {
1508 struct phys_disk_entry *pd = &sb->phys->entries[i];
1509 int type = be16_to_cpu(pd->type);
1510 int state = be16_to_cpu(pd->state);
1511
1512 if (be32_to_cpu(pd->refnum) == 0xffffffff)
1513 /* Not in use */
1514 continue;
1515 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1516 //printf("\n");
1517 printf(" %3d %08x ", i,
1518 be32_to_cpu(pd->refnum));
1519 printf("%8lluK ",
1520 be64_to_cpu(pd->config_size)>>1);
1521 for (dl = sb->dlist; dl ; dl = dl->next) {
1522 if (be32_eq(dl->disk.refnum, pd->refnum)) {
1523 char *dv = map_dev(dl->major, dl->minor, 0);
1524 if (dv) {
1525 printf("%-15s", dv);
1526 break;
1527 }
1528 }
1529 }
1530 if (!dl)
1531 printf("%15s","");
1532 else
1533 dl->displayed = 1;
1534 printf(" %s%s%s%s%s",
1535 (type&2) ? "active":"",
1536 (type&4) ? "Global-Spare":"",
1537 (type&8) ? "spare" : "",
1538 (type&16)? ", foreign" : "",
1539 (type&32)? "pass-through" : "");
1540 if (state & DDF_Failed)
1541 /* This over-rides these three */
1542 state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition);
1543 printf("/%s%s%s%s%s%s%s",
1544 (state&1)? "Online": "Offline",
1545 (state&2)? ", Failed": "",
1546 (state&4)? ", Rebuilding": "",
1547 (state&8)? ", in-transition": "",
1548 (state&16)? ", SMART-errors": "",
1549 (state&32)? ", Unrecovered-Read-Errors": "",
1550 (state&64)? ", Missing" : "");
1551 printf("\n");
1552 }
1553 for (dl = sb->dlist; dl; dl = dl->next) {
1554 char *dv;
1555 if (dl->displayed)
1556 continue;
1557 if (!unlisted)
1558 printf(" Physical disks not in metadata!:\n");
1559 unlisted = 1;
1560 dv = map_dev(dl->major, dl->minor, 0);
1561 printf(" %08x %s\n", be32_to_cpu(dl->disk.refnum),
1562 dv ? dv : "-unknown-");
1563 }
1564 if (unlisted)
1565 printf("\n");
1566 }
1567
1568 static void examine_super_ddf(struct supertype *st, char *homehost)
1569 {
1570 struct ddf_super *sb = st->sb;
1571
1572 printf(" Magic : %08x\n", be32_to_cpu(sb->anchor.magic));
1573 printf(" Version : %.8s\n", sb->anchor.revision);
1574 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1575 printf("\n");
1576 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1577 printf("\n");
1578 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1579 printf(" Redundant hdr : %s\n", (be32_eq(sb->secondary.magic,
1580 DDF_HEADER_MAGIC)
1581 ?"yes" : "no"));
1582 examine_vds(sb);
1583 examine_pds(sb);
1584 }
1585
1586 static unsigned int get_vd_num_of_subarray(struct supertype *st)
1587 {
1588 /*
1589 * Figure out the VD number for this supertype.
1590 * Returns DDF_CONTAINER for the container itself,
1591 * and DDF_NOTFOUND on error.
1592 */
1593 struct ddf_super *ddf = st->sb;
1594 struct mdinfo *sra;
1595 char *sub, *end;
1596 unsigned int vcnum;
1597
1598 if (*st->container_devnm == '\0')
1599 return DDF_CONTAINER;
1600
1601 sra = sysfs_read(-1, st->devnm, GET_VERSION);
1602 if (!sra || sra->array.major_version != -1 ||
1603 sra->array.minor_version != -2 ||
1604 !is_subarray(sra->text_version))
1605 return DDF_NOTFOUND;
1606
1607 sub = strchr(sra->text_version + 1, '/');
1608 if (sub != NULL)
1609 vcnum = strtoul(sub + 1, &end, 10);
1610 if (sub == NULL || *sub == '\0' || *end != '\0' ||
1611 vcnum >= be16_to_cpu(ddf->active->max_vd_entries))
1612 return DDF_NOTFOUND;
1613
1614 return vcnum;
1615 }
1616
1617 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1618 {
1619 /* We just write a generic DDF ARRAY entry
1620 */
1621 struct mdinfo info;
1622 char nbuf[64];
1623 getinfo_super_ddf(st, &info, NULL);
1624 fname_from_uuid(st, &info, nbuf, ':');
1625
1626 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1627 }
1628
1629 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1630 {
1631 /* We write a DDF ARRAY member entry for each vd, identifying container
1632 * by uuid and member by unit number and uuid.
1633 */
1634 struct ddf_super *ddf = st->sb;
1635 struct mdinfo info;
1636 unsigned int i;
1637 char nbuf[64];
1638 getinfo_super_ddf(st, &info, NULL);
1639 fname_from_uuid(st, &info, nbuf, ':');
1640
1641 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
1642 struct virtual_entry *ve = &ddf->virt->entries[i];
1643 struct vcl vcl;
1644 char nbuf1[64];
1645 char namebuf[17];
1646 if (all_ff(ve->guid))
1647 continue;
1648 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1649 ddf->currentconf =&vcl;
1650 vcl.vcnum = i;
1651 uuid_from_super_ddf(st, info.uuid);
1652 fname_from_uuid(st, &info, nbuf1, ':');
1653 _ddf_array_name(namebuf, ddf, i);
1654 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1655 namebuf[0] == '\0' ? "" : " /dev/md/", namebuf,
1656 nbuf+5, i, nbuf1+5);
1657 }
1658 }
1659
1660 static void export_examine_super_ddf(struct supertype *st)
1661 {
1662 struct mdinfo info;
1663 char nbuf[64];
1664 getinfo_super_ddf(st, &info, NULL);
1665 fname_from_uuid(st, &info, nbuf, ':');
1666 printf("MD_METADATA=ddf\n");
1667 printf("MD_LEVEL=container\n");
1668 printf("MD_UUID=%s\n", nbuf+5);
1669 printf("MD_DEVICES=%u\n",
1670 be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes));
1671 }
1672
1673 static int copy_metadata_ddf(struct supertype *st, int from, int to)
1674 {
1675 void *buf;
1676 unsigned long long dsize, offset;
1677 int bytes;
1678 struct ddf_header *ddf;
1679 int written = 0;
1680
1681 /* The meta consists of an anchor, a primary, and a secondary.
1682 * This all lives at the end of the device.
1683 * So it is easiest to find the earliest of primary and
1684 * secondary, and copy everything from there.
1685 *
1686 * Anchor is 512 from end. It contains primary_lba and secondary_lba
1687 * we choose one of those
1688 */
1689
1690 if (posix_memalign(&buf, 4096, 4096) != 0)
1691 return 1;
1692
1693 if (!get_dev_size(from, NULL, &dsize))
1694 goto err;
1695
1696 if (lseek64(from, dsize-512, 0) < 0)
1697 goto err;
1698 if (read(from, buf, 512) != 512)
1699 goto err;
1700 ddf = buf;
1701 if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) ||
1702 !be32_eq(calc_crc(ddf, 512), ddf->crc) ||
1703 (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 &&
1704 memcmp(ddf->revision, DDF_REVISION_2, 8) != 0))
1705 goto err;
1706
1707 offset = dsize - 512;
1708 if ((be64_to_cpu(ddf->primary_lba) << 9) < offset)
1709 offset = be64_to_cpu(ddf->primary_lba) << 9;
1710 if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset)
1711 offset = be64_to_cpu(ddf->secondary_lba) << 9;
1712
1713 bytes = dsize - offset;
1714
1715 if (lseek64(from, offset, 0) < 0 ||
1716 lseek64(to, offset, 0) < 0)
1717 goto err;
1718 while (written < bytes) {
1719 int n = bytes - written;
1720 if (n > 4096)
1721 n = 4096;
1722 if (read(from, buf, n) != n)
1723 goto err;
1724 if (write(to, buf, n) != n)
1725 goto err;
1726 written += n;
1727 }
1728 free(buf);
1729 return 0;
1730 err:
1731 free(buf);
1732 return 1;
1733 }
1734
1735 static void detail_super_ddf(struct supertype *st, char *homehost)
1736 {
1737 /* FIXME later
1738 * Could print DDF GUID
1739 * Need to find which array
1740 * If whole, briefly list all arrays
1741 * If one, give name
1742 */
1743 }
1744
1745 static const char *vendors_with_variable_volume_UUID[] = {
1746 "LSI ",
1747 };
1748
1749 static int volume_id_is_reliable(const struct ddf_super *ddf)
1750 {
1751 int n = ARRAY_SIZE(vendors_with_variable_volume_UUID);
1752 int i;
1753 for (i = 0; i < n; i++)
1754 if (!memcmp(ddf->controller.guid,
1755 vendors_with_variable_volume_UUID[i], 8))
1756 return 0;
1757 return 1;
1758 }
1759
1760 static void uuid_of_ddf_subarray(const struct ddf_super *ddf,
1761 unsigned int vcnum, int uuid[4])
1762 {
1763 char buf[DDF_GUID_LEN+18], sha[20], *p;
1764 struct sha1_ctx ctx;
1765 if (volume_id_is_reliable(ddf)) {
1766 uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid);
1767 return;
1768 }
1769 /*
1770 * Some fake RAID BIOSes (in particular, LSI ones) change the
1771 * VD GUID at every boot. These GUIDs are not suitable for
1772 * identifying an array. Luckily the header GUID appears to
1773 * remain constant.
1774 * We construct a pseudo-UUID from the header GUID and those
1775 * properties of the subarray that we expect to remain constant.
1776 */
1777 memset(buf, 0, sizeof(buf));
1778 p = buf;
1779 memcpy(p, ddf->anchor.guid, DDF_GUID_LEN);
1780 p += DDF_GUID_LEN;
1781 memcpy(p, ddf->virt->entries[vcnum].name, 16);
1782 p += 16;
1783 *((__u16 *) p) = vcnum;
1784 sha1_init_ctx(&ctx);
1785 sha1_process_bytes(buf, sizeof(buf), &ctx);
1786 sha1_finish_ctx(&ctx, sha);
1787 memcpy(uuid, sha, 4*4);
1788 }
1789
1790 static void brief_detail_super_ddf(struct supertype *st)
1791 {
1792 struct mdinfo info;
1793 char nbuf[64];
1794 struct ddf_super *ddf = st->sb;
1795 unsigned int vcnum = get_vd_num_of_subarray(st);
1796 if (vcnum == DDF_CONTAINER)
1797 uuid_from_super_ddf(st, info.uuid);
1798 else if (vcnum == DDF_NOTFOUND)
1799 return;
1800 else
1801 uuid_of_ddf_subarray(ddf, vcnum, info.uuid);
1802 fname_from_uuid(st, &info, nbuf,':');
1803 printf(" UUID=%s", nbuf + 5);
1804 }
1805 #endif
1806
1807 static int match_home_ddf(struct supertype *st, char *homehost)
1808 {
1809 /* It matches 'this' host if the controller is a
1810 * Linux-MD controller with vendor_data matching
1811 * the hostname. It would be nice if we could
1812 * test against controller found in /sys or somewhere...
1813 */
1814 struct ddf_super *ddf = st->sb;
1815 unsigned int len;
1816
1817 if (!homehost)
1818 return 0;
1819 len = strlen(homehost);
1820
1821 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1822 len < sizeof(ddf->controller.vendor_data) &&
1823 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1824 ddf->controller.vendor_data[len] == 0);
1825 }
1826
1827 #ifndef MDASSEMBLE
1828 static int find_index_in_bvd(const struct ddf_super *ddf,
1829 const struct vd_config *conf, unsigned int n,
1830 unsigned int *n_bvd)
1831 {
1832 /*
1833 * Find the index of the n-th valid physical disk in this BVD.
1834 * Unused entries can be sprinkled in with the used entries,
1835 * but don't count.
1836 */
1837 unsigned int i, j;
1838 for (i = 0, j = 0;
1839 i < ddf->mppe && j < be16_to_cpu(conf->prim_elmnt_count);
1840 i++) {
1841 if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) {
1842 if (n == j) {
1843 *n_bvd = i;
1844 return 1;
1845 }
1846 j++;
1847 }
1848 }
1849 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1850 __func__, n, be16_to_cpu(conf->prim_elmnt_count));
1851 return 0;
1852 }
1853
1854 /* Given a member array instance number, and a raid disk within that instance,
1855 * find the vd_config structure. The offset of the given disk in the phys_refnum
1856 * table is returned in n_bvd.
1857 * For two-level members with a secondary raid level the vd_config for
1858 * the appropriate BVD is returned.
1859 * The return value is always &vlc->conf, where vlc is returned in last pointer.
1860 */
1861 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst,
1862 unsigned int n,
1863 unsigned int *n_bvd, struct vcl **vcl)
1864 {
1865 struct vcl *v;
1866
1867 for (v = ddf->conflist; v; v = v->next) {
1868 unsigned int nsec, ibvd = 0;
1869 struct vd_config *conf;
1870 if (inst != v->vcnum)
1871 continue;
1872 conf = &v->conf;
1873 if (conf->sec_elmnt_count == 1) {
1874 if (find_index_in_bvd(ddf, conf, n, n_bvd)) {
1875 *vcl = v;
1876 return conf;
1877 } else
1878 goto bad;
1879 }
1880 if (v->other_bvds == NULL) {
1881 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1882 __func__, conf->sec_elmnt_count);
1883 goto bad;
1884 }
1885 nsec = n / be16_to_cpu(conf->prim_elmnt_count);
1886 if (conf->sec_elmnt_seq != nsec) {
1887 for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) {
1888 if (v->other_bvds[ibvd-1]->sec_elmnt_seq
1889 == nsec)
1890 break;
1891 }
1892 if (ibvd == conf->sec_elmnt_count)
1893 goto bad;
1894 conf = v->other_bvds[ibvd-1];
1895 }
1896 if (!find_index_in_bvd(ddf, conf,
1897 n - nsec*conf->sec_elmnt_count, n_bvd))
1898 goto bad;
1899 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1900 , __func__, n, *n_bvd, ibvd, inst);
1901 *vcl = v;
1902 return conf;
1903 }
1904 bad:
1905 pr_err("%s: Could't find disk %d in array %u\n", __func__, n, inst);
1906 return NULL;
1907 }
1908 #endif
1909
1910 static int find_phys(const struct ddf_super *ddf, be32 phys_refnum)
1911 {
1912 /* Find the entry in phys_disk which has the given refnum
1913 * and return it's index
1914 */
1915 unsigned int i;
1916 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++)
1917 if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum))
1918 return i;
1919 return -1;
1920 }
1921
1922 static void uuid_from_ddf_guid(const char *guid, int uuid[4])
1923 {
1924 char buf[20];
1925 struct sha1_ctx ctx;
1926 sha1_init_ctx(&ctx);
1927 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1928 sha1_finish_ctx(&ctx, buf);
1929 memcpy(uuid, buf, 4*4);
1930 }
1931
1932 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1933 {
1934 /* The uuid returned here is used for:
1935 * uuid to put into bitmap file (Create, Grow)
1936 * uuid for backup header when saving critical section (Grow)
1937 * comparing uuids when re-adding a device into an array
1938 * In these cases the uuid required is that of the data-array,
1939 * not the device-set.
1940 * uuid to recognise same set when adding a missing device back
1941 * to an array. This is a uuid for the device-set.
1942 *
1943 * For each of these we can make do with a truncated
1944 * or hashed uuid rather than the original, as long as
1945 * everyone agrees.
1946 * In the case of SVD we assume the BVD is of interest,
1947 * though that might be the case if a bitmap were made for
1948 * a mirrored SVD - worry about that later.
1949 * So we need to find the VD configuration record for the
1950 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1951 * The first 16 bytes of the sha1 of these is used.
1952 */
1953 struct ddf_super *ddf = st->sb;
1954 struct vcl *vcl = ddf->currentconf;
1955
1956 if (vcl)
1957 uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid);
1958 else
1959 uuid_from_ddf_guid(ddf->anchor.guid, uuid);
1960 }
1961
1962 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1963 {
1964 struct ddf_super *ddf = st->sb;
1965 int map_disks = info->array.raid_disks;
1966 __u32 *cptr;
1967
1968 if (ddf->currentconf) {
1969 getinfo_super_ddf_bvd(st, info, map);
1970 return;
1971 }
1972 memset(info, 0, sizeof(*info));
1973
1974 info->array.raid_disks = be16_to_cpu(ddf->phys->used_pdes);
1975 info->array.level = LEVEL_CONTAINER;
1976 info->array.layout = 0;
1977 info->array.md_minor = -1;
1978 cptr = (__u32 *)(ddf->anchor.guid + 16);
1979 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
1980
1981 info->array.chunk_size = 0;
1982 info->container_enough = 1;
1983
1984 info->disk.major = 0;
1985 info->disk.minor = 0;
1986 if (ddf->dlist) {
1987 struct phys_disk_entry *pde = NULL;
1988 info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum);
1989 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1990
1991 info->data_offset = be64_to_cpu(ddf->phys->
1992 entries[info->disk.raid_disk].
1993 config_size);
1994 info->component_size = ddf->dlist->size - info->data_offset;
1995 if (info->disk.raid_disk >= 0)
1996 pde = ddf->phys->entries + info->disk.raid_disk;
1997 if (pde &&
1998 !(be16_to_cpu(pde->state) & DDF_Failed))
1999 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2000 else
2001 info->disk.state = 1 << MD_DISK_FAULTY;
2002
2003 } else {
2004 info->disk.number = -1;
2005 info->disk.raid_disk = -1;
2006 // info->disk.raid_disk = find refnum in the table and use index;
2007 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2008 }
2009 info->events = be32_to_cpu(ddf->active->seq);
2010 info->array.utime = DECADE + be32_to_cpu(ddf->active->timestamp);
2011
2012 info->recovery_start = MaxSector;
2013 info->reshape_active = 0;
2014 info->recovery_blocked = 0;
2015 info->name[0] = 0;
2016
2017 info->array.major_version = -1;
2018 info->array.minor_version = -2;
2019 strcpy(info->text_version, "ddf");
2020 info->safe_mode_delay = 0;
2021
2022 uuid_from_super_ddf(st, info->uuid);
2023
2024 if (map) {
2025 int i;
2026 for (i = 0 ; i < map_disks; i++) {
2027 if (i < info->array.raid_disks &&
2028 !(be16_to_cpu(ddf->phys->entries[i].state)
2029 & DDF_Failed))
2030 map[i] = 1;
2031 else
2032 map[i] = 0;
2033 }
2034 }
2035 }
2036
2037 /* size of name must be at least 17 bytes! */
2038 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i)
2039 {
2040 int j;
2041 memcpy(name, ddf->virt->entries[i].name, 16);
2042 name[16] = 0;
2043 for(j = 0; j < 16; j++)
2044 if (name[j] == ' ')
2045 name[j] = 0;
2046 }
2047
2048 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
2049 {
2050 struct ddf_super *ddf = st->sb;
2051 struct vcl *vc = ddf->currentconf;
2052 int cd = ddf->currentdev;
2053 int n_prim;
2054 int j;
2055 struct dl *dl;
2056 int map_disks = info->array.raid_disks;
2057 __u32 *cptr;
2058 struct vd_config *conf;
2059
2060 memset(info, 0, sizeof(*info));
2061 if (layout_ddf2md(&vc->conf, &info->array) == -1)
2062 return;
2063 info->array.md_minor = -1;
2064 cptr = (__u32 *)(vc->conf.guid + 16);
2065 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
2066 info->array.utime = DECADE + be32_to_cpu(vc->conf.timestamp);
2067 info->array.chunk_size = 512 << vc->conf.chunk_shift;
2068 info->custom_array_size = be64_to_cpu(vc->conf.array_blocks);
2069
2070 conf = &vc->conf;
2071 n_prim = be16_to_cpu(conf->prim_elmnt_count);
2072 if (conf->sec_elmnt_count > 1 && cd >= n_prim) {
2073 int ibvd = cd / n_prim - 1;
2074 cd %= n_prim;
2075 conf = vc->other_bvds[ibvd];
2076 }
2077
2078 if (cd >= 0 && (unsigned)cd < ddf->mppe) {
2079 info->data_offset =
2080 be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]);
2081 if (vc->block_sizes)
2082 info->component_size = vc->block_sizes[cd];
2083 else
2084 info->component_size = be64_to_cpu(conf->blocks);
2085
2086 for (dl = ddf->dlist; dl ; dl = dl->next)
2087 if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd]))
2088 break;
2089 }
2090
2091 info->disk.major = 0;
2092 info->disk.minor = 0;
2093 info->disk.state = 0;
2094 if (dl && dl->pdnum >= 0) {
2095 info->disk.major = dl->major;
2096 info->disk.minor = dl->minor;
2097 info->disk.raid_disk = cd + conf->sec_elmnt_seq
2098 * be16_to_cpu(conf->prim_elmnt_count);
2099 info->disk.number = dl->pdnum;
2100 info->disk.state = 0;
2101 if (info->disk.number >= 0 &&
2102 (be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Online) &&
2103 !(be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Failed))
2104 info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2105 info->events = be32_to_cpu(ddf->active->seq);
2106 }
2107
2108 info->container_member = ddf->currentconf->vcnum;
2109
2110 info->recovery_start = MaxSector;
2111 info->resync_start = 0;
2112 info->reshape_active = 0;
2113 info->recovery_blocked = 0;
2114 if (!(ddf->virt->entries[info->container_member].state
2115 & DDF_state_inconsistent) &&
2116 (ddf->virt->entries[info->container_member].init_state
2117 & DDF_initstate_mask)
2118 == DDF_init_full)
2119 info->resync_start = MaxSector;
2120
2121 uuid_from_super_ddf(st, info->uuid);
2122
2123 info->array.major_version = -1;
2124 info->array.minor_version = -2;
2125 sprintf(info->text_version, "/%s/%d",
2126 st->container_devnm,
2127 info->container_member);
2128 info->safe_mode_delay = DDF_SAFE_MODE_DELAY;
2129
2130 _ddf_array_name(info->name, ddf, info->container_member);
2131
2132 if (map)
2133 for (j = 0; j < map_disks; j++) {
2134 map[j] = 0;
2135 if (j < info->array.raid_disks) {
2136 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
2137 if (i >= 0 &&
2138 (be16_to_cpu(ddf->phys->entries[i].state)
2139 & DDF_Online) &&
2140 !(be16_to_cpu(ddf->phys->entries[i].state)
2141 & DDF_Failed))
2142 map[i] = 1;
2143 }
2144 }
2145 }
2146
2147 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
2148 char *update,
2149 char *devname, int verbose,
2150 int uuid_set, char *homehost)
2151 {
2152 /* For 'assemble' and 'force' we need to return non-zero if any
2153 * change was made. For others, the return value is ignored.
2154 * Update options are:
2155 * force-one : This device looks a bit old but needs to be included,
2156 * update age info appropriately.
2157 * assemble: clear any 'faulty' flag to allow this device to
2158 * be assembled.
2159 * force-array: Array is degraded but being forced, mark it clean
2160 * if that will be needed to assemble it.
2161 *
2162 * newdev: not used ????
2163 * grow: Array has gained a new device - this is currently for
2164 * linear only
2165 * resync: mark as dirty so a resync will happen.
2166 * uuid: Change the uuid of the array to match what is given
2167 * homehost: update the recorded homehost
2168 * name: update the name - preserving the homehost
2169 * _reshape_progress: record new reshape_progress position.
2170 *
2171 * Following are not relevant for this version:
2172 * sparc2.2 : update from old dodgey metadata
2173 * super-minor: change the preferred_minor number
2174 * summaries: update redundant counters.
2175 */
2176 int rv = 0;
2177 // struct ddf_super *ddf = st->sb;
2178 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2179 // struct virtual_entry *ve = find_ve(ddf);
2180
2181 /* we don't need to handle "force-*" or "assemble" as
2182 * there is no need to 'trick' the kernel. When the metadata is
2183 * first updated to activate the array, all the implied modifications
2184 * will just happen.
2185 */
2186
2187 if (strcmp(update, "grow") == 0) {
2188 /* FIXME */
2189 } else if (strcmp(update, "resync") == 0) {
2190 // info->resync_checkpoint = 0;
2191 } else if (strcmp(update, "homehost") == 0) {
2192 /* homehost is stored in controller->vendor_data,
2193 * or it is when we are the vendor
2194 */
2195 // if (info->vendor_is_local)
2196 // strcpy(ddf->controller.vendor_data, homehost);
2197 rv = -1;
2198 } else if (strcmp(update, "name") == 0) {
2199 /* name is stored in virtual_entry->name */
2200 // memset(ve->name, ' ', 16);
2201 // strncpy(ve->name, info->name, 16);
2202 rv = -1;
2203 } else if (strcmp(update, "_reshape_progress") == 0) {
2204 /* We don't support reshape yet */
2205 } else if (strcmp(update, "assemble") == 0 ) {
2206 /* Do nothing, just succeed */
2207 rv = 0;
2208 } else
2209 rv = -1;
2210
2211 // update_all_csum(ddf);
2212
2213 return rv;
2214 }
2215
2216 static void make_header_guid(char *guid)
2217 {
2218 be32 stamp;
2219 /* Create a DDF Header of Virtual Disk GUID */
2220
2221 /* 24 bytes of fiction required.
2222 * first 8 are a 'vendor-id' - "Linux-MD"
2223 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2224 * Remaining 8 random number plus timestamp
2225 */
2226 memcpy(guid, T10, sizeof(T10));
2227 stamp = cpu_to_be32(0xdeadbeef);
2228 memcpy(guid+8, &stamp, 4);
2229 stamp = cpu_to_be32(0);
2230 memcpy(guid+12, &stamp, 4);
2231 stamp = cpu_to_be32(time(0) - DECADE);
2232 memcpy(guid+16, &stamp, 4);
2233 stamp._v32 = random32();
2234 memcpy(guid+20, &stamp, 4);
2235 }
2236
2237 static unsigned int find_unused_vde(const struct ddf_super *ddf)
2238 {
2239 unsigned int i;
2240 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2241 if (all_ff(ddf->virt->entries[i].guid))
2242 return i;
2243 }
2244 return DDF_NOTFOUND;
2245 }
2246
2247 static unsigned int find_vde_by_name(const struct ddf_super *ddf,
2248 const char *name)
2249 {
2250 unsigned int i;
2251 if (name == NULL)
2252 return DDF_NOTFOUND;
2253 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2254 if (all_ff(ddf->virt->entries[i].guid))
2255 continue;
2256 if (!strncmp(name, ddf->virt->entries[i].name,
2257 sizeof(ddf->virt->entries[i].name)))
2258 return i;
2259 }
2260 return DDF_NOTFOUND;
2261 }
2262
2263 #ifndef MDASSEMBLE
2264 static unsigned int find_vde_by_guid(const struct ddf_super *ddf,
2265 const char *guid)
2266 {
2267 unsigned int i;
2268 if (guid == NULL || all_ff(guid))
2269 return DDF_NOTFOUND;
2270 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++)
2271 if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN))
2272 return i;
2273 return DDF_NOTFOUND;
2274 }
2275 #endif
2276
2277 static int init_super_ddf(struct supertype *st,
2278 mdu_array_info_t *info,
2279 unsigned long long size, char *name, char *homehost,
2280 int *uuid, unsigned long long data_offset)
2281 {
2282 /* This is primarily called by Create when creating a new array.
2283 * We will then get add_to_super called for each component, and then
2284 * write_init_super called to write it out to each device.
2285 * For DDF, Create can create on fresh devices or on a pre-existing
2286 * array.
2287 * To create on a pre-existing array a different method will be called.
2288 * This one is just for fresh drives.
2289 *
2290 * We need to create the entire 'ddf' structure which includes:
2291 * DDF headers - these are easy.
2292 * Controller data - a Sector describing this controller .. not that
2293 * this is a controller exactly.
2294 * Physical Disk Record - one entry per device, so
2295 * leave plenty of space.
2296 * Virtual Disk Records - again, just leave plenty of space.
2297 * This just lists VDs, doesn't give details.
2298 * Config records - describe the VDs that use this disk
2299 * DiskData - describes 'this' device.
2300 * BadBlockManagement - empty
2301 * Diag Space - empty
2302 * Vendor Logs - Could we put bitmaps here?
2303 *
2304 */
2305 struct ddf_super *ddf;
2306 char hostname[17];
2307 int hostlen;
2308 int max_phys_disks, max_virt_disks;
2309 unsigned long long sector;
2310 int clen;
2311 int i;
2312 int pdsize, vdsize;
2313 struct phys_disk *pd;
2314 struct virtual_disk *vd;
2315
2316 if (data_offset != INVALID_SECTORS) {
2317 pr_err("data-offset not supported by DDF\n");
2318 return 0;
2319 }
2320
2321 if (st->sb)
2322 return init_super_ddf_bvd(st, info, size, name, homehost, uuid,
2323 data_offset);
2324
2325 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
2326 pr_err("%s could not allocate superblock\n", __func__);
2327 return 0;
2328 }
2329 memset(ddf, 0, sizeof(*ddf));
2330 st->sb = ddf;
2331
2332 if (info == NULL) {
2333 /* zeroing superblock */
2334 return 0;
2335 }
2336
2337 /* At least 32MB *must* be reserved for the ddf. So let's just
2338 * start 32MB from the end, and put the primary header there.
2339 * Don't do secondary for now.
2340 * We don't know exactly where that will be yet as it could be
2341 * different on each device. So just set up the lengths.
2342 */
2343
2344 ddf->anchor.magic = DDF_HEADER_MAGIC;
2345 make_header_guid(ddf->anchor.guid);
2346
2347 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
2348 ddf->anchor.seq = cpu_to_be32(1);
2349 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
2350 ddf->anchor.openflag = 0xFF;
2351 ddf->anchor.foreignflag = 0;
2352 ddf->anchor.enforcegroups = 0; /* Is this best?? */
2353 ddf->anchor.pad0 = 0xff;
2354 memset(ddf->anchor.pad1, 0xff, 12);
2355 memset(ddf->anchor.header_ext, 0xff, 32);
2356 ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0);
2357 ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0);
2358 ddf->anchor.type = DDF_HEADER_ANCHOR;
2359 memset(ddf->anchor.pad2, 0xff, 3);
2360 ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */
2361 /* Put this at bottom of 32M reserved.. */
2362 ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0);
2363 max_phys_disks = 1023; /* Should be enough, 4095 is also allowed */
2364 ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks);
2365 max_virt_disks = 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2366 ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks);
2367 ddf->max_part = 64;
2368 ddf->anchor.max_partitions = cpu_to_be16(ddf->max_part);
2369 ddf->mppe = 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2370 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
2371 ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len);
2372 ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe);
2373 memset(ddf->anchor.pad3, 0xff, 54);
2374 /* Controller section is one sector long immediately
2375 * after the ddf header */
2376 sector = 1;
2377 ddf->anchor.controller_section_offset = cpu_to_be32(sector);
2378 ddf->anchor.controller_section_length = cpu_to_be32(1);
2379 sector += 1;
2380
2381 /* phys is 8 sectors after that */
2382 pdsize = ROUND_UP(sizeof(struct phys_disk) +
2383 sizeof(struct phys_disk_entry)*max_phys_disks,
2384 512);
2385 switch(pdsize/512) {
2386 case 2: case 8: case 32: case 128: case 512: break;
2387 default: abort();
2388 }
2389 ddf->anchor.phys_section_offset = cpu_to_be32(sector);
2390 ddf->anchor.phys_section_length =
2391 cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
2392 sector += pdsize/512;
2393
2394 /* virt is another 32 sectors */
2395 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
2396 sizeof(struct virtual_entry) * max_virt_disks,
2397 512);
2398 switch(vdsize/512) {
2399 case 2: case 8: case 32: case 128: case 512: break;
2400 default: abort();
2401 }
2402 ddf->anchor.virt_section_offset = cpu_to_be32(sector);
2403 ddf->anchor.virt_section_length =
2404 cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
2405 sector += vdsize/512;
2406
2407 clen = ddf->conf_rec_len * (ddf->max_part+1);
2408 ddf->anchor.config_section_offset = cpu_to_be32(sector);
2409 ddf->anchor.config_section_length = cpu_to_be32(clen);
2410 sector += clen;
2411
2412 ddf->anchor.data_section_offset = cpu_to_be32(sector);
2413 ddf->anchor.data_section_length = cpu_to_be32(1);
2414 sector += 1;
2415
2416 ddf->anchor.bbm_section_length = cpu_to_be32(0);
2417 ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF);
2418 ddf->anchor.diag_space_length = cpu_to_be32(0);
2419 ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF);
2420 ddf->anchor.vendor_length = cpu_to_be32(0);
2421 ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF);
2422
2423 memset(ddf->anchor.pad4, 0xff, 256);
2424
2425 memcpy(&ddf->primary, &ddf->anchor, 512);
2426 memcpy(&ddf->secondary, &ddf->anchor, 512);
2427
2428 ddf->primary.openflag = 1; /* I guess.. */
2429 ddf->primary.type = DDF_HEADER_PRIMARY;
2430
2431 ddf->secondary.openflag = 1; /* I guess.. */
2432 ddf->secondary.type = DDF_HEADER_SECONDARY;
2433
2434 ddf->active = &ddf->primary;
2435
2436 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
2437
2438 /* 24 more bytes of fiction required.
2439 * first 8 are a 'vendor-id' - "Linux-MD"
2440 * Remaining 16 are serial number.... maybe a hostname would do?
2441 */
2442 memcpy(ddf->controller.guid, T10, sizeof(T10));
2443 gethostname(hostname, sizeof(hostname));
2444 hostname[sizeof(hostname) - 1] = 0;
2445 hostlen = strlen(hostname);
2446 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
2447 for (i = strlen(T10) ; i+hostlen < 24; i++)
2448 ddf->controller.guid[i] = ' ';
2449
2450 ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD);
2451 ddf->controller.type.device_id = cpu_to_be16(0xBEEF);
2452 ddf->controller.type.sub_vendor_id = cpu_to_be16(0);
2453 ddf->controller.type.sub_device_id = cpu_to_be16(0);
2454 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
2455 memset(ddf->controller.pad, 0xff, 8);
2456 memset(ddf->controller.vendor_data, 0xff, 448);
2457 if (homehost && strlen(homehost) < 440)
2458 strcpy((char*)ddf->controller.vendor_data, homehost);
2459
2460 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
2461 pr_err("%s could not allocate pd\n", __func__);
2462 return 0;
2463 }
2464 ddf->phys = pd;
2465 ddf->pdsize = pdsize;
2466
2467 memset(pd, 0xff, pdsize);
2468 memset(pd, 0, sizeof(*pd));
2469 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2470 pd->used_pdes = cpu_to_be16(0);
2471 pd->max_pdes = cpu_to_be16(max_phys_disks);
2472 memset(pd->pad, 0xff, 52);
2473 for (i = 0; i < max_phys_disks; i++)
2474 memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN);
2475
2476 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
2477 pr_err("%s could not allocate vd\n", __func__);
2478 return 0;
2479 }
2480 ddf->virt = vd;
2481 ddf->vdsize = vdsize;
2482 memset(vd, 0, vdsize);
2483 vd->magic = DDF_VIRT_RECORDS_MAGIC;
2484 vd->populated_vdes = cpu_to_be16(0);
2485 vd->max_vdes = cpu_to_be16(max_virt_disks);
2486 memset(vd->pad, 0xff, 52);
2487
2488 for (i=0; i<max_virt_disks; i++)
2489 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
2490
2491 st->sb = ddf;
2492 ddf_set_updates_pending(ddf, NULL);
2493 return 1;
2494 }
2495
2496 static int chunk_to_shift(int chunksize)
2497 {
2498 return ffs(chunksize/512)-1;
2499 }
2500
2501 #ifndef MDASSEMBLE
2502 struct extent {
2503 unsigned long long start, size;
2504 };
2505 static int cmp_extent(const void *av, const void *bv)
2506 {
2507 const struct extent *a = av;
2508 const struct extent *b = bv;
2509 if (a->start < b->start)
2510 return -1;
2511 if (a->start > b->start)
2512 return 1;
2513 return 0;
2514 }
2515
2516 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
2517 {
2518 /* Find a list of used extents on the give physical device
2519 * (dnum) of the given ddf.
2520 * Return a malloced array of 'struct extent'
2521 */
2522 struct extent *rv;
2523 int n = 0;
2524 unsigned int i;
2525 __u16 state;
2526
2527 if (dl->pdnum < 0)
2528 return NULL;
2529 state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state);
2530
2531 if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online)
2532 return NULL;
2533
2534 rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2));
2535
2536 for (i = 0; i < ddf->max_part; i++) {
2537 const struct vd_config *bvd;
2538 unsigned int ibvd;
2539 struct vcl *v = dl->vlist[i];
2540 if (v == NULL ||
2541 get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe,
2542 &bvd, &ibvd) == DDF_NOTFOUND)
2543 continue;
2544 rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]);
2545 rv[n].size = be64_to_cpu(bvd->blocks);
2546 n++;
2547 }
2548 qsort(rv, n, sizeof(*rv), cmp_extent);
2549
2550 rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
2551 rv[n].size = 0;
2552 return rv;
2553 }
2554 #endif
2555
2556 static int init_super_ddf_bvd(struct supertype *st,
2557 mdu_array_info_t *info,
2558 unsigned long long size,
2559 char *name, char *homehost,
2560 int *uuid, unsigned long long data_offset)
2561 {
2562 /* We are creating a BVD inside a pre-existing container.
2563 * so st->sb is already set.
2564 * We need to create a new vd_config and a new virtual_entry
2565 */
2566 struct ddf_super *ddf = st->sb;
2567 unsigned int venum, i;
2568 struct virtual_entry *ve;
2569 struct vcl *vcl;
2570 struct vd_config *vc;
2571
2572 if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) {
2573 pr_err("This ddf already has an array called %s\n", name);
2574 return 0;
2575 }
2576 venum = find_unused_vde(ddf);
2577 if (venum == DDF_NOTFOUND) {
2578 pr_err("Cannot find spare slot for virtual disk\n");
2579 return 0;
2580 }
2581 ve = &ddf->virt->entries[venum];
2582
2583 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2584 * timestamp, random number
2585 */
2586 make_header_guid(ve->guid);
2587 ve->unit = cpu_to_be16(info->md_minor);
2588 ve->pad0 = 0xFFFF;
2589 ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid,
2590 DDF_GUID_LEN);
2591 ve->type = cpu_to_be16(0);
2592 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
2593 if (info->state & 1) /* clean */
2594 ve->init_state = DDF_init_full;
2595 else
2596 ve->init_state = DDF_init_not;
2597
2598 memset(ve->pad1, 0xff, 14);
2599 memset(ve->name, ' ', 16);
2600 if (name)
2601 strncpy(ve->name, name, 16);
2602 ddf->virt->populated_vdes =
2603 cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1);
2604
2605 /* Now create a new vd_config */
2606 if (posix_memalign((void**)&vcl, 512,
2607 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
2608 pr_err("%s could not allocate vd_config\n", __func__);
2609 return 0;
2610 }
2611 vcl->vcnum = venum;
2612 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2613 vc = &vcl->conf;
2614
2615 vc->magic = DDF_VD_CONF_MAGIC;
2616 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2617 vc->timestamp = cpu_to_be32(time(0)-DECADE);
2618 vc->seqnum = cpu_to_be32(1);
2619 memset(vc->pad0, 0xff, 24);
2620 vc->chunk_shift = chunk_to_shift(info->chunk_size);
2621 if (layout_md2ddf(info, vc) == -1 ||
2622 be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) {
2623 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2624 __func__, info->level, info->layout, info->raid_disks);
2625 free(vcl);
2626 return 0;
2627 }
2628 vc->sec_elmnt_seq = 0;
2629 if (alloc_other_bvds(ddf, vcl) != 0) {
2630 pr_err("%s could not allocate other bvds\n",
2631 __func__);
2632 free(vcl);
2633 return 0;
2634 }
2635 vc->blocks = cpu_to_be64(info->size * 2);
2636 vc->array_blocks = cpu_to_be64(
2637 calc_array_size(info->level, info->raid_disks, info->layout,
2638 info->chunk_size, info->size*2));
2639 memset(vc->pad1, 0xff, 8);
2640 vc->spare_refs[0] = cpu_to_be32(0xffffffff);
2641 vc->spare_refs[1] = cpu_to_be32(0xffffffff);
2642 vc->spare_refs[2] = cpu_to_be32(0xffffffff);
2643 vc->spare_refs[3] = cpu_to_be32(0xffffffff);
2644 vc->spare_refs[4] = cpu_to_be32(0xffffffff);
2645 vc->spare_refs[5] = cpu_to_be32(0xffffffff);
2646 vc->spare_refs[6] = cpu_to_be32(0xffffffff);
2647 vc->spare_refs[7] = cpu_to_be32(0xffffffff);
2648 memset(vc->cache_pol, 0, 8);
2649 vc->bg_rate = 0x80;
2650 memset(vc->pad2, 0xff, 3);
2651 memset(vc->pad3, 0xff, 52);
2652 memset(vc->pad4, 0xff, 192);
2653 memset(vc->v0, 0xff, 32);
2654 memset(vc->v1, 0xff, 32);
2655 memset(vc->v2, 0xff, 16);
2656 memset(vc->v3, 0xff, 16);
2657 memset(vc->vendor, 0xff, 32);
2658
2659 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2660 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2661
2662 for (i = 1; i < vc->sec_elmnt_count; i++) {
2663 memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512);
2664 vcl->other_bvds[i-1]->sec_elmnt_seq = i;
2665 }
2666
2667 vcl->next = ddf->conflist;
2668 ddf->conflist = vcl;
2669 ddf->currentconf = vcl;
2670 ddf_set_updates_pending(ddf, NULL);
2671 return 1;
2672 }
2673
2674 #ifndef MDASSEMBLE
2675 static void add_to_super_ddf_bvd(struct supertype *st,
2676 mdu_disk_info_t *dk, int fd, char *devname)
2677 {
2678 /* fd and devname identify a device within the ddf container (st).
2679 * dk identifies a location in the new BVD.
2680 * We need to find suitable free space in that device and update
2681 * the phys_refnum and lba_offset for the newly created vd_config.
2682 * We might also want to update the type in the phys_disk
2683 * section.
2684 *
2685 * Alternately: fd == -1 and we have already chosen which device to
2686 * use and recorded in dlist->raid_disk;
2687 */
2688 struct dl *dl;
2689 struct ddf_super *ddf = st->sb;
2690 struct vd_config *vc;
2691 unsigned int i;
2692 unsigned long long blocks, pos, esize;
2693 struct extent *ex;
2694 unsigned int raid_disk = dk->raid_disk;
2695
2696 if (fd == -1) {
2697 for (dl = ddf->dlist; dl ; dl = dl->next)
2698 if (dl->raiddisk == dk->raid_disk)
2699 break;
2700 } else {
2701 for (dl = ddf->dlist; dl ; dl = dl->next)
2702 if (dl->major == dk->major &&
2703 dl->minor == dk->minor)
2704 break;
2705 }
2706 if (!dl || dl->pdnum < 0 || ! (dk->state & (1<<MD_DISK_SYNC)))
2707 return;
2708
2709 vc = &ddf->currentconf->conf;
2710 if (vc->sec_elmnt_count > 1) {
2711 unsigned int n = be16_to_cpu(vc->prim_elmnt_count);
2712 if (raid_disk >= n)
2713 vc = ddf->currentconf->other_bvds[raid_disk / n - 1];
2714 raid_disk %= n;
2715 }
2716
2717 ex = get_extents(ddf, dl);
2718 if (!ex)
2719 return;
2720
2721 i = 0; pos = 0;
2722 blocks = be64_to_cpu(vc->blocks);
2723 if (ddf->currentconf->block_sizes)
2724 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2725
2726 /* First-fit */
2727 do {
2728 esize = ex[i].start - pos;
2729 if (esize >= blocks)
2730 break;
2731 pos = ex[i].start + ex[i].size;
2732 i++;
2733 } while (ex[i-1].size);
2734
2735 free(ex);
2736 if (esize < blocks)
2737 return;
2738
2739 ddf->currentdev = dk->raid_disk;
2740 vc->phys_refnum[raid_disk] = dl->disk.refnum;
2741 LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos);
2742
2743 for (i = 0; i < ddf->max_part ; i++)
2744 if (dl->vlist[i] == NULL)
2745 break;
2746 if (i == ddf->max_part)
2747 return;
2748 dl->vlist[i] = ddf->currentconf;
2749
2750 if (fd >= 0)
2751 dl->fd = fd;
2752 if (devname)
2753 dl->devname = devname;
2754
2755 /* Check if we can mark array as optimal yet */
2756 i = ddf->currentconf->vcnum;
2757 ddf->virt->entries[i].state =
2758 (ddf->virt->entries[i].state & ~DDF_state_mask)
2759 | get_svd_state(ddf, ddf->currentconf);
2760 be16_clear(ddf->phys->entries[dl->pdnum].type,
2761 cpu_to_be16(DDF_Global_Spare));
2762 be16_set(ddf->phys->entries[dl->pdnum].type,
2763 cpu_to_be16(DDF_Active_in_VD));
2764 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2765 __func__, dl->pdnum, be32_to_cpu(dl->disk.refnum),
2766 ddf->currentconf->vcnum, guid_str(vc->guid),
2767 dk->raid_disk);
2768 ddf_set_updates_pending(ddf, vc);
2769 }
2770
2771 static unsigned int find_unused_pde(const struct ddf_super *ddf)
2772 {
2773 unsigned int i;
2774 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) {
2775 if (all_ff(ddf->phys->entries[i].guid))
2776 return i;
2777 }
2778 return DDF_NOTFOUND;
2779 }
2780
2781 static void _set_config_size(struct phys_disk_entry *pde, const struct dl *dl)
2782 {
2783 __u64 cfs, t;
2784 cfs = min(dl->size - 32*1024*2ULL, be64_to_cpu(dl->primary_lba));
2785 t = be64_to_cpu(dl->secondary_lba);
2786 if (t != ~(__u64)0)
2787 cfs = min(cfs, t);
2788 /*
2789 * Some vendor DDF structures interpret workspace_lba
2790 * very differently than we do: Make a sanity check on the value.
2791 */
2792 t = be64_to_cpu(dl->workspace_lba);
2793 if (t < cfs) {
2794 __u64 wsp = cfs - t;
2795 if (wsp > 1024*1024*2ULL && wsp > dl->size / 16) {
2796 pr_err("%s: %x:%x: workspace size 0x%llx too big, ignoring\n",
2797 __func__, dl->major, dl->minor, wsp);
2798 } else
2799 cfs = t;
2800 }
2801 pde->config_size = cpu_to_be64(cfs);
2802 dprintf("%s: %x:%x config_size %llx, DDF structure is %llx blocks\n",
2803 __func__, dl->major, dl->minor, cfs, dl->size-cfs);
2804 }
2805
2806 /* Add a device to a container, either while creating it or while
2807 * expanding a pre-existing container
2808 */
2809 static int add_to_super_ddf(struct supertype *st,
2810 mdu_disk_info_t *dk, int fd, char *devname,
2811 unsigned long long data_offset)
2812 {
2813 struct ddf_super *ddf = st->sb;
2814 struct dl *dd;
2815 time_t now;
2816 struct tm *tm;
2817 unsigned long long size;
2818 struct phys_disk_entry *pde;
2819 unsigned int n, i;
2820 struct stat stb;
2821 __u32 *tptr;
2822
2823 if (ddf->currentconf) {
2824 add_to_super_ddf_bvd(st, dk, fd, devname);
2825 return 0;
2826 }
2827
2828 /* This is device numbered dk->number. We need to create
2829 * a phys_disk entry and a more detailed disk_data entry.
2830 */
2831 fstat(fd, &stb);
2832 n = find_unused_pde(ddf);
2833 if (n == DDF_NOTFOUND) {
2834 pr_err("%s: No free slot in array, cannot add disk\n",
2835 __func__);
2836 return 1;
2837 }
2838 pde = &ddf->phys->entries[n];
2839 get_dev_size(fd, NULL, &size);
2840 if (size <= 32*1024*1024) {
2841 pr_err("%s: device size must be at least 32MB\n",
2842 __func__);
2843 return 1;
2844 }
2845 size >>= 9;
2846
2847 if (posix_memalign((void**)&dd, 512,
2848 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2849 pr_err("%s could allocate buffer for new disk, aborting\n",
2850 __func__);
2851 return 1;
2852 }
2853 dd->major = major(stb.st_rdev);
2854 dd->minor = minor(stb.st_rdev);
2855 dd->devname = devname;
2856 dd->fd = fd;
2857 dd->spare = NULL;
2858
2859 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2860 now = time(0);
2861 tm = localtime(&now);
2862 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2863 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2864 tptr = (__u32 *)(dd->disk.guid + 16);
2865 *tptr++ = random32();
2866 *tptr = random32();
2867
2868 do {
2869 /* Cannot be bothered finding a CRC of some irrelevant details*/
2870 dd->disk.refnum._v32 = random32();
2871 for (i = be16_to_cpu(ddf->active->max_pd_entries);
2872 i > 0; i--)
2873 if (be32_eq(ddf->phys->entries[i-1].refnum,
2874 dd->disk.refnum))
2875 break;
2876 } while (i > 0);
2877
2878 dd->disk.forced_ref = 1;
2879 dd->disk.forced_guid = 1;
2880 memset(dd->disk.vendor, ' ', 32);
2881 memcpy(dd->disk.vendor, "Linux", 5);
2882 memset(dd->disk.pad, 0xff, 442);
2883 for (i = 0; i < ddf->max_part ; i++)
2884 dd->vlist[i] = NULL;
2885
2886 dd->pdnum = n;
2887
2888 if (st->update_tail) {
2889 int len = (sizeof(struct phys_disk) +
2890 sizeof(struct phys_disk_entry));
2891 struct phys_disk *pd;
2892
2893 pd = xmalloc(len);
2894 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2895 pd->used_pdes = cpu_to_be16(n);
2896 pde = &pd->entries[0];
2897 dd->mdupdate = pd;
2898 } else
2899 ddf->phys->used_pdes = cpu_to_be16(
2900 1 + be16_to_cpu(ddf->phys->used_pdes));
2901
2902 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2903 pde->refnum = dd->disk.refnum;
2904 pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2905 pde->state = cpu_to_be16(DDF_Online);
2906 dd->size = size;
2907 /*
2908 * If there is already a device in dlist, try to reserve the same
2909 * amount of workspace. Otherwise, use 32MB.
2910 * We checked disk size above already.
2911 */
2912 #define __calc_lba(new, old, lba, mb) do { \
2913 unsigned long long dif; \
2914 if ((old) != NULL) \
2915 dif = (old)->size - be64_to_cpu((old)->lba); \
2916 else \
2917 dif = (new)->size; \
2918 if ((new)->size > dif) \
2919 (new)->lba = cpu_to_be64((new)->size - dif); \
2920 else \
2921 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2922 } while (0)
2923 __calc_lba(dd, ddf->dlist, workspace_lba, 32);
2924 __calc_lba(dd, ddf->dlist, primary_lba, 16);
2925 if (ddf->dlist == NULL ||
2926 be64_to_cpu(ddf->dlist->secondary_lba) != ~(__u64)0)
2927 __calc_lba(dd, ddf->dlist, secondary_lba, 32);
2928 _set_config_size(pde, dd);
2929
2930 sprintf(pde->path, "%17.17s","Information: nil") ;
2931 memset(pde->pad, 0xff, 6);
2932
2933 if (st->update_tail) {
2934 dd->next = ddf->add_list;
2935 ddf->add_list = dd;
2936 } else {
2937 dd->next = ddf->dlist;
2938 ddf->dlist = dd;
2939 ddf_set_updates_pending(ddf, NULL);
2940 }
2941
2942 return 0;
2943 }
2944
2945 static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk)
2946 {
2947 struct ddf_super *ddf = st->sb;
2948 struct dl *dl;
2949
2950 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2951 * disappeared from the container.
2952 * We need to arrange that it disappears from the metadata and
2953 * internal data structures too.
2954 * Most of the work is done by ddf_process_update which edits
2955 * the metadata and closes the file handle and attaches the memory
2956 * where free_updates will free it.
2957 */
2958 for (dl = ddf->dlist; dl ; dl = dl->next)
2959 if (dl->major == dk->major &&
2960 dl->minor == dk->minor)
2961 break;
2962 if (!dl || dl->pdnum < 0)
2963 return -1;
2964
2965 if (st->update_tail) {
2966 int len = (sizeof(struct phys_disk) +
2967 sizeof(struct phys_disk_entry));
2968 struct phys_disk *pd;
2969
2970 pd = xmalloc(len);
2971 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2972 pd->used_pdes = cpu_to_be16(dl->pdnum);
2973 pd->entries[0].state = cpu_to_be16(DDF_Missing);
2974 append_metadata_update(st, pd, len);
2975 }
2976 return 0;
2977 }
2978 #endif
2979
2980 /*
2981 * This is the write_init_super method for a ddf container. It is
2982 * called when creating a container or adding another device to a
2983 * container.
2984 */
2985
2986 static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type)
2987 {
2988 unsigned long long sector;
2989 struct ddf_header *header;
2990 int fd, i, n_config, conf_size, buf_size;
2991 int ret = 0;
2992 char *conf;
2993
2994 fd = d->fd;
2995
2996 switch (type) {
2997 case DDF_HEADER_PRIMARY:
2998 header = &ddf->primary;
2999 sector = be64_to_cpu(header->primary_lba);
3000 break;
3001 case DDF_HEADER_SECONDARY:
3002 header = &ddf->secondary;
3003 sector = be64_to_cpu(header->secondary_lba);
3004 break;
3005 default:
3006 return 0;
3007 }
3008 if (sector == ~(__u64)0)
3009 return 0;
3010
3011 header->type = type;
3012 header->openflag = 1;
3013 header->crc = calc_crc(header, 512);
3014
3015 lseek64(fd, sector<<9, 0);
3016 if (write(fd, header, 512) < 0)
3017 goto out;
3018
3019 ddf->controller.crc = calc_crc(&ddf->controller, 512);
3020 if (write(fd, &ddf->controller, 512) < 0)
3021 goto out;
3022
3023 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
3024 if (write(fd, ddf->phys, ddf->pdsize) < 0)
3025 goto out;
3026 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
3027 if (write(fd, ddf->virt, ddf->vdsize) < 0)
3028 goto out;
3029
3030 /* Now write lots of config records. */
3031 n_config = ddf->max_part;
3032 conf_size = ddf->conf_rec_len * 512;
3033 conf = ddf->conf;
3034 buf_size = conf_size * (n_config + 1);
3035 if (!conf) {
3036 if (posix_memalign((void**)&conf, 512, buf_size) != 0)
3037 goto out;
3038 ddf->conf = conf;
3039 }
3040 for (i = 0 ; i <= n_config ; i++) {
3041 struct vcl *c;
3042 struct vd_config *vdc = NULL;
3043 if (i == n_config) {
3044 c = (struct vcl *)d->spare;
3045 if (c)
3046 vdc = &c->conf;
3047 } else {
3048 unsigned int dummy;
3049 c = d->vlist[i];
3050 if (c)
3051 get_pd_index_from_refnum(
3052 c, d->disk.refnum,
3053 ddf->mppe,
3054 (const struct vd_config **)&vdc,
3055 &dummy);
3056 }
3057 if (vdc) {
3058 dprintf("writing conf record %i on disk %08x for %s/%u\n",
3059 i, be32_to_cpu(d->disk.refnum),
3060 guid_str(vdc->guid),
3061 vdc->sec_elmnt_seq);
3062 vdc->crc = calc_crc(vdc, conf_size);
3063 memcpy(conf + i*conf_size, vdc, conf_size);
3064 } else
3065 memset(conf + i*conf_size, 0xff, conf_size);
3066 }
3067 if (write(fd, conf, buf_size) != buf_size)
3068 goto out;
3069
3070 d->disk.crc = calc_crc(&d->disk, 512);
3071 if (write(fd, &d->disk, 512) < 0)
3072 goto out;
3073
3074 ret = 1;
3075 out:
3076 header->openflag = 0;
3077 header->crc = calc_crc(header, 512);
3078
3079 lseek64(fd, sector<<9, 0);
3080 if (write(fd, header, 512) < 0)
3081 ret = 0;
3082
3083 return ret;
3084 }
3085
3086 static int _write_super_to_disk(struct ddf_super *ddf, struct dl *d)
3087 {
3088 unsigned long long size;
3089 int fd = d->fd;
3090 if (fd < 0)
3091 return 0;
3092
3093 /* We need to fill in the primary, (secondary) and workspace
3094 * lba's in the headers, set their checksums,
3095 * Also checksum phys, virt....
3096 *
3097 * Then write everything out, finally the anchor is written.
3098 */
3099 get_dev_size(fd, NULL, &size);
3100 size /= 512;
3101 memcpy(&ddf->anchor, ddf->active, 512);
3102 if (be64_to_cpu(d->workspace_lba) != 0ULL)
3103 ddf->anchor.workspace_lba = d->workspace_lba;
3104 else
3105 ddf->anchor.workspace_lba =
3106 cpu_to_be64(size - 32*1024*2);
3107 if (be64_to_cpu(d->primary_lba) != 0ULL)
3108 ddf->anchor.primary_lba = d->primary_lba;
3109 else
3110 ddf->anchor.primary_lba =
3111 cpu_to_be64(size - 16*1024*2);
3112 if (be64_to_cpu(d->secondary_lba) != 0ULL)
3113 ddf->anchor.secondary_lba = d->secondary_lba;
3114 else
3115 ddf->anchor.secondary_lba =
3116 cpu_to_be64(size - 32*1024*2);
3117 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
3118 memcpy(&ddf->primary, &ddf->anchor, 512);
3119 memcpy(&ddf->secondary, &ddf->anchor, 512);
3120
3121 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
3122 ddf->anchor.seq = cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3123 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
3124
3125 if (!__write_ddf_structure(d, ddf, DDF_HEADER_PRIMARY))
3126 return 0;
3127
3128 if (!__write_ddf_structure(d, ddf, DDF_HEADER_SECONDARY))
3129 return 0;
3130
3131 lseek64(fd, (size-1)*512, SEEK_SET);
3132 if (write(fd, &ddf->anchor, 512) < 0)
3133 return 0;
3134
3135 return 1;
3136 }
3137
3138 #ifndef MDASSEMBLE
3139 static int __write_init_super_ddf(struct supertype *st)
3140 {
3141 struct ddf_super *ddf = st->sb;
3142 struct dl *d;
3143 int attempts = 0;
3144 int successes = 0;
3145
3146 pr_state(ddf, __func__);
3147
3148 /* try to write updated metadata,
3149 * if we catch a failure move on to the next disk
3150 */
3151 for (d = ddf->dlist; d; d=d->next) {
3152 attempts++;
3153 successes += _write_super_to_disk(ddf, d);
3154 }
3155
3156 return attempts != successes;
3157 }
3158
3159 static int write_init_super_ddf(struct supertype *st)
3160 {
3161 struct ddf_super *ddf = st->sb;
3162 struct vcl *currentconf = ddf->currentconf;
3163
3164 /* We are done with currentconf - reset it so st refers to the container */
3165 ddf->currentconf = NULL;
3166
3167 if (st->update_tail) {
3168 /* queue the virtual_disk and vd_config as metadata updates */
3169 struct virtual_disk *vd;
3170 struct vd_config *vc;
3171 int len, tlen;
3172 unsigned int i;
3173
3174 if (!currentconf) {
3175 /* Must be adding a physical disk to the container */
3176 int len = (sizeof(struct phys_disk) +
3177 sizeof(struct phys_disk_entry));
3178
3179 /* adding a disk to the container. */
3180 if (!ddf->add_list)
3181 return 0;
3182
3183 append_metadata_update(st, ddf->add_list->mdupdate, len);
3184 ddf->add_list->mdupdate = NULL;
3185 return 0;
3186 }
3187
3188 /* Newly created VD */
3189
3190 /* First the virtual disk. We have a slightly fake header */
3191 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
3192 vd = xmalloc(len);
3193 *vd = *ddf->virt;
3194 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
3195 vd->populated_vdes = cpu_to_be16(currentconf->vcnum);
3196 append_metadata_update(st, vd, len);
3197
3198 /* Then the vd_config */
3199 len = ddf->conf_rec_len * 512;
3200 tlen = len * currentconf->conf.sec_elmnt_count;
3201 vc = xmalloc(tlen);
3202 memcpy(vc, &currentconf->conf, len);
3203 for (i = 1; i < currentconf->conf.sec_elmnt_count; i++)
3204 memcpy((char *)vc + i*len, currentconf->other_bvds[i-1],
3205 len);
3206 append_metadata_update(st, vc, tlen);
3207
3208 /* FIXME I need to close the fds! */
3209 return 0;
3210 } else {
3211 struct dl *d;
3212 if (!currentconf)
3213 for (d = ddf->dlist; d; d=d->next)
3214 while (Kill(d->devname, NULL, 0, -1, 1) == 0);
3215 return __write_init_super_ddf(st);
3216 }
3217 }
3218
3219 #endif
3220
3221 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize,
3222 unsigned long long data_offset)
3223 {
3224 /* We must reserve the last 32Meg */
3225 if (devsize <= 32*1024*2)
3226 return 0;
3227 return devsize - 32*1024*2;
3228 }
3229
3230 #ifndef MDASSEMBLE
3231
3232 static int reserve_space(struct supertype *st, int raiddisks,
3233 unsigned long long size, int chunk,
3234 unsigned long long *freesize)
3235 {
3236 /* Find 'raiddisks' spare extents at least 'size' big (but
3237 * only caring about multiples of 'chunk') and remember
3238 * them. If size==0, find the largest size possible.
3239 * Report available size in *freesize
3240 * If space cannot be found, fail.
3241 */
3242 struct dl *dl;
3243 struct ddf_super *ddf = st->sb;
3244 int cnt = 0;
3245
3246 for (dl = ddf->dlist; dl ; dl=dl->next) {
3247 dl->raiddisk = -1;
3248 dl->esize = 0;
3249 }
3250 /* Now find largest extent on each device */
3251 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3252 struct extent *e = get_extents(ddf, dl);
3253 unsigned long long pos = 0;
3254 int i = 0;
3255 int found = 0;
3256 unsigned long long minsize = size;
3257
3258 if (size == 0)
3259 minsize = chunk;
3260
3261 if (!e)
3262 continue;
3263 do {
3264 unsigned long long esize;
3265 esize = e[i].start - pos;
3266 if (esize >= minsize) {
3267 found = 1;
3268 minsize = esize;
3269 }
3270 pos = e[i].start + e[i].size;
3271 i++;
3272 } while (e[i-1].size);
3273 if (found) {
3274 cnt++;
3275 dl->esize = minsize;
3276 }
3277 free(e);
3278 }
3279 if (cnt < raiddisks) {
3280 pr_err("not enough devices with space to create array.\n");
3281 return 0; /* No enough free spaces large enough */
3282 }
3283 if (size == 0) {
3284 /* choose the largest size of which there are at least 'raiddisk' */
3285 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3286 struct dl *dl2;
3287 if (dl->esize <= size)
3288 continue;
3289 /* This is bigger than 'size', see if there are enough */
3290 cnt = 0;
3291 for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next)
3292 if (dl2->esize >= dl->esize)
3293 cnt++;
3294 if (cnt >= raiddisks)
3295 size = dl->esize;
3296 }
3297 if (chunk) {
3298 size = size / chunk;
3299 size *= chunk;
3300 }
3301 *freesize = size;
3302 if (size < 32) {
3303 pr_err("not enough spare devices to create array.\n");
3304 return 0;
3305 }
3306 }
3307 /* We have a 'size' of which there are enough spaces.
3308 * We simply do a first-fit */
3309 cnt = 0;
3310 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
3311 if (dl->esize < size)
3312 continue;
3313
3314 dl->raiddisk = cnt;
3315 cnt++;
3316 }
3317 return 1;
3318 }
3319
3320 static int validate_geometry_ddf(struct supertype *st,
3321 int level, int layout, int raiddisks,
3322 int *chunk, unsigned long long size,
3323 unsigned long long data_offset,
3324 char *dev, unsigned long long *freesize,
3325 int verbose)
3326 {
3327 int fd;
3328 struct mdinfo *sra;
3329 int cfd;
3330
3331 /* ddf potentially supports lots of things, but it depends on
3332 * what devices are offered (and maybe kernel version?)
3333 * If given unused devices, we will make a container.
3334 * If given devices in a container, we will make a BVD.
3335 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3336 */
3337
3338 if (*chunk == UnSet)
3339 *chunk = DEFAULT_CHUNK;
3340
3341 if (level == LEVEL_NONE)
3342 level = LEVEL_CONTAINER;
3343 if (level == LEVEL_CONTAINER) {
3344 /* Must be a fresh device to add to a container */
3345 return validate_geometry_ddf_container(st, level, layout,
3346 raiddisks, *chunk,
3347 size, data_offset, dev,
3348 freesize,
3349 verbose);
3350 }
3351
3352 if (!dev) {
3353 mdu_array_info_t array = {
3354 .level = level,
3355 .layout = layout,
3356 .raid_disks = raiddisks
3357 };
3358 struct vd_config conf;
3359 if (layout_md2ddf(&array, &conf) == -1) {
3360 if (verbose)
3361 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3362 level, layout, raiddisks);
3363 return 0;
3364 }
3365 /* Should check layout? etc */
3366
3367 if (st->sb && freesize) {
3368 /* --create was given a container to create in.
3369 * So we need to check that there are enough
3370 * free spaces and return the amount of space.
3371 * We may as well remember which drives were
3372 * chosen so that add_to_super/getinfo_super
3373 * can return them.
3374 */
3375 return reserve_space(st, raiddisks, size, *chunk, freesize);
3376 }
3377 return 1;
3378 }
3379
3380 if (st->sb) {
3381 /* A container has already been opened, so we are
3382 * creating in there. Maybe a BVD, maybe an SVD.
3383 * Should make a distinction one day.
3384 */
3385 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
3386 chunk, size, data_offset, dev,
3387 freesize,
3388 verbose);
3389 }
3390 /* This is the first device for the array.
3391 * If it is a container, we read it in and do automagic allocations,
3392 * no other devices should be given.
3393 * Otherwise it must be a member device of a container, and we
3394 * do manual allocation.
3395 * Later we should check for a BVD and make an SVD.
3396 */
3397 fd = open(dev, O_RDONLY|O_EXCL, 0);
3398 if (fd >= 0) {
3399 sra = sysfs_read(fd, NULL, GET_VERSION);
3400 close(fd);
3401 if (sra && sra->array.major_version == -1 &&
3402 strcmp(sra->text_version, "ddf") == 0) {
3403 /* load super */
3404 /* find space for 'n' devices. */
3405 /* remember the devices */
3406 /* Somehow return the fact that we have enough */
3407 }
3408
3409 if (verbose)
3410 pr_err("ddf: Cannot create this array "
3411 "on device %s - a container is required.\n",
3412 dev);
3413 return 0;
3414 }
3415 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3416 if (verbose)
3417 pr_err("ddf: Cannot open %s: %s\n",
3418 dev, strerror(errno));
3419 return 0;
3420 }
3421 /* Well, it is in use by someone, maybe a 'ddf' container. */
3422 cfd = open_container(fd);
3423 if (cfd < 0) {
3424 close(fd);
3425 if (verbose)
3426 pr_err("ddf: Cannot use %s: %s\n",
3427 dev, strerror(EBUSY));
3428 return 0;
3429 }
3430 sra = sysfs_read(cfd, NULL, GET_VERSION);
3431 close(fd);
3432 if (sra && sra->array.major_version == -1 &&
3433 strcmp(sra->text_version, "ddf") == 0) {
3434 /* This is a member of a ddf container. Load the container
3435 * and try to create a bvd
3436 */
3437 struct ddf_super *ddf;
3438 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) {
3439 st->sb = ddf;
3440 strcpy(st->container_devnm, fd2devnm(cfd));
3441 close(cfd);
3442 return validate_geometry_ddf_bvd(st, level, layout,
3443 raiddisks, chunk, size,
3444 data_offset,
3445 dev, freesize,
3446 verbose);
3447 }
3448 close(cfd);
3449 } else /* device may belong to a different container */
3450 return 0;
3451
3452 return 1;
3453 }
3454
3455 static int
3456 validate_geometry_ddf_container(struct supertype *st,
3457 int level, int layout, int raiddisks,
3458 int chunk, unsigned long long size,
3459 unsigned long long data_offset,
3460 char *dev, unsigned long long *freesize,
3461 int verbose)
3462 {
3463 int fd;
3464 unsigned long long ldsize;
3465
3466 if (level != LEVEL_CONTAINER)
3467 return 0;
3468 if (!dev)
3469 return 1;
3470
3471 fd = open(dev, O_RDONLY|O_EXCL, 0);
3472 if (fd < 0) {
3473 if (verbose)
3474 pr_err("ddf: Cannot open %s: %s\n",
3475 dev, strerror(errno));
3476 return 0;
3477 }
3478 if (!get_dev_size(fd, dev, &ldsize)) {
3479 close(fd);
3480 return 0;
3481 }
3482 close(fd);
3483
3484 *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS);
3485 if (*freesize == 0)
3486 return 0;
3487
3488 return 1;
3489 }
3490
3491 static int validate_geometry_ddf_bvd(struct supertype *st,
3492 int level, int layout, int raiddisks,
3493 int *chunk, unsigned long long size,
3494 unsigned long long data_offset,
3495 char *dev, unsigned long long *freesize,
3496 int verbose)
3497 {
3498 struct stat stb;
3499 struct ddf_super *ddf = st->sb;
3500 struct dl *dl;
3501 unsigned long long pos = 0;
3502 unsigned long long maxsize;
3503 struct extent *e;
3504 int i;
3505 /* ddf/bvd supports lots of things, but not containers */
3506 if (level == LEVEL_CONTAINER) {
3507 if (verbose)
3508 pr_err("DDF cannot create a container within an container\n");
3509 return 0;
3510 }
3511 /* We must have the container info already read in. */
3512 if (!ddf)
3513 return 0;
3514
3515 if (!dev) {
3516 /* General test: make sure there is space for
3517 * 'raiddisks' device extents of size 'size'.
3518 */
3519 unsigned long long minsize = size;
3520 int dcnt = 0;
3521 if (minsize == 0)
3522 minsize = 8;
3523 for (dl = ddf->dlist; dl ; dl = dl->next) {
3524 int found = 0;
3525 pos = 0;
3526
3527 i = 0;
3528 e = get_extents(ddf, dl);
3529 if (!e) continue;
3530 do {
3531 unsigned long long esize;
3532 esize = e[i].start - pos;
3533 if (esize >= minsize)
3534 found = 1;
3535 pos = e[i].start + e[i].size;
3536 i++;
3537 } while (e[i-1].size);
3538 if (found)
3539 dcnt++;
3540 free(e);
3541 }
3542 if (dcnt < raiddisks) {
3543 if (verbose)
3544 pr_err("ddf: Not enough devices with "
3545 "space for this array (%d < %d)\n",
3546 dcnt, raiddisks);
3547 return 0;
3548 }
3549 return 1;
3550 }
3551 /* This device must be a member of the set */
3552 if (stat(dev, &stb) < 0)
3553 return 0;
3554 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3555 return 0;
3556 for (dl = ddf->dlist ; dl ; dl = dl->next) {
3557 if (dl->major == (int)major(stb.st_rdev) &&
3558 dl->minor == (int)minor(stb.st_rdev))
3559 break;
3560 }
3561 if (!dl) {
3562 if (verbose)
3563 pr_err("ddf: %s is not in the "
3564 "same DDF set\n",
3565 dev);
3566 return 0;
3567 }
3568 e = get_extents(ddf, dl);
3569 maxsize = 0;
3570 i = 0;
3571 if (e)
3572 do {
3573 unsigned long long esize;
3574 esize = e[i].start - pos;
3575 if (esize >= maxsize)
3576 maxsize = esize;
3577 pos = e[i].start + e[i].size;
3578 i++;
3579 } while (e[i-1].size);
3580 *freesize = maxsize;
3581 // FIXME here I am
3582
3583 return 1;
3584 }
3585
3586 static int load_super_ddf_all(struct supertype *st, int fd,
3587 void **sbp, char *devname)
3588 {
3589 struct mdinfo *sra;
3590 struct ddf_super *super;
3591 struct mdinfo *sd, *best = NULL;
3592 int bestseq = 0;
3593 int seq;
3594 char nm[20];
3595 int dfd;
3596
3597 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
3598 if (!sra)
3599 return 1;
3600 if (sra->array.major_version != -1 ||
3601 sra->array.minor_version != -2 ||
3602 strcmp(sra->text_version, "ddf") != 0)
3603 return 1;
3604
3605 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
3606 return 1;
3607 memset(super, 0, sizeof(*super));
3608
3609 /* first, try each device, and choose the best ddf */
3610 for (sd = sra->devs ; sd ; sd = sd->next) {
3611 int rv;
3612 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3613 dfd = dev_open(nm, O_RDONLY);
3614 if (dfd < 0)
3615 return 2;
3616 rv = load_ddf_headers(dfd, super, NULL);
3617 close(dfd);
3618 if (rv == 0) {
3619 seq = be32_to_cpu(super->active->seq);
3620 if (super->active->openflag)
3621 seq--;
3622 if (!best || seq > bestseq) {
3623 bestseq = seq;
3624 best = sd;
3625 }
3626 }
3627 }
3628 if (!best)
3629 return 1;
3630 /* OK, load this ddf */
3631 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
3632 dfd = dev_open(nm, O_RDONLY);
3633 if (dfd < 0)
3634 return 1;
3635 load_ddf_headers(dfd, super, NULL);
3636 load_ddf_global(dfd, super, NULL);
3637 close(dfd);
3638 /* Now we need the device-local bits */
3639 for (sd = sra->devs ; sd ; sd = sd->next) {
3640 int rv;
3641
3642 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3643 dfd = dev_open(nm, O_RDWR);
3644 if (dfd < 0)
3645 return 2;
3646 rv = load_ddf_headers(dfd, super, NULL);
3647 if (rv == 0)
3648 rv = load_ddf_local(dfd, super, NULL, 1);
3649 if (rv)
3650 return 1;
3651 }
3652
3653 *sbp = super;
3654 if (st->ss == NULL) {
3655 st->ss = &super_ddf;
3656 st->minor_version = 0;
3657 st->max_devs = 512;
3658 }
3659 strcpy(st->container_devnm, fd2devnm(fd));
3660 return 0;
3661 }
3662
3663 static int load_container_ddf(struct supertype *st, int fd,
3664 char *devname)
3665 {
3666 return load_super_ddf_all(st, fd, &st->sb, devname);
3667 }
3668
3669 #endif /* MDASSEMBLE */
3670
3671 static int check_secondary(const struct vcl *vc)
3672 {
3673 const struct vd_config *conf = &vc->conf;
3674 int i;
3675
3676 /* The only DDF secondary RAID level md can support is
3677 * RAID 10, if the stripe sizes and Basic volume sizes
3678 * are all equal.
3679 * Other configurations could in theory be supported by exposing
3680 * the BVDs to user space and using device mapper for the secondary
3681 * mapping. So far we don't support that.
3682 */
3683
3684 __u64 sec_elements[4] = {0, 0, 0, 0};
3685 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3686 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3687
3688 if (vc->other_bvds == NULL) {
3689 pr_err("No BVDs for secondary RAID found\n");
3690 return -1;
3691 }
3692 if (conf->prl != DDF_RAID1) {
3693 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3694 return -1;
3695 }
3696 if (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED) {
3697 pr_err("Secondary RAID level %d is unsupported\n",
3698 conf->srl);
3699 return -1;
3700 }
3701 __set_sec_seen(conf->sec_elmnt_seq);
3702 for (i = 0; i < conf->sec_elmnt_count-1; i++) {
3703 const struct vd_config *bvd = vc->other_bvds[i];
3704 if (bvd->sec_elmnt_seq == DDF_UNUSED_BVD)
3705 continue;
3706 if (bvd->srl != conf->srl) {
3707 pr_err("Inconsistent secondary RAID level across BVDs\n");
3708 return -1;
3709 }
3710 if (bvd->prl != conf->prl) {
3711 pr_err("Different RAID levels for BVDs are unsupported\n");
3712 return -1;
3713 }
3714 if (!be16_eq(bvd->prim_elmnt_count, conf->prim_elmnt_count)) {
3715 pr_err("All BVDs must have the same number of primary elements\n");
3716 return -1;
3717 }
3718 if (bvd->chunk_shift != conf->chunk_shift) {
3719 pr_err("Different strip sizes for BVDs are unsupported\n");
3720 return -1;
3721 }
3722 if (!be64_eq(bvd->array_blocks, conf->array_blocks)) {
3723 pr_err("Different BVD sizes are unsupported\n");
3724 return -1;
3725 }
3726 __set_sec_seen(bvd->sec_elmnt_seq);
3727 }
3728 for (i = 0; i < conf->sec_elmnt_count; i++) {
3729 if (!__was_sec_seen(i)) {
3730 pr_err("BVD %d is missing\n", i);
3731 return -1;
3732 }
3733 }
3734 return 0;
3735 }
3736
3737 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
3738 be32 refnum, unsigned int nmax,
3739 const struct vd_config **bvd,
3740 unsigned int *idx)
3741 {
3742 unsigned int i, j, n, sec, cnt;
3743
3744 cnt = be16_to_cpu(vc->conf.prim_elmnt_count);
3745 sec = (vc->conf.sec_elmnt_count == 1 ? 0 : vc->conf.sec_elmnt_seq);
3746
3747 for (i = 0, j = 0 ; i < nmax ; i++) {
3748 /* j counts valid entries for this BVD */
3749 if (be32_eq(vc->conf.phys_refnum[i], refnum)) {
3750 *bvd = &vc->conf;
3751 *idx = i;
3752 return sec * cnt + j;
3753 }
3754 if (be32_to_cpu(vc->conf.phys_refnum[i]) != 0xffffffff)
3755 j++;
3756 }
3757 if (vc->other_bvds == NULL)
3758 goto bad;
3759
3760 for (n = 1; n < vc->conf.sec_elmnt_count; n++) {
3761 struct vd_config *vd = vc->other_bvds[n-1];
3762 sec = vd->sec_elmnt_seq;
3763 if (sec == DDF_UNUSED_BVD)
3764 continue;
3765 for (i = 0, j = 0 ; i < nmax ; i++) {
3766 if (be32_eq(vd->phys_refnum[i], refnum)) {
3767 *bvd = vd;
3768 *idx = i;
3769 return sec * cnt + j;
3770 }
3771 if (be32_to_cpu(vd->phys_refnum[i]) != 0xffffffff)
3772 j++;
3773 }
3774 }
3775 bad:
3776 *bvd = NULL;
3777 return DDF_NOTFOUND;
3778 }
3779
3780 static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray)
3781 {
3782 /* Given a container loaded by load_super_ddf_all,
3783 * extract information about all the arrays into
3784 * an mdinfo tree.
3785 *
3786 * For each vcl in conflist: create an mdinfo, fill it in,
3787 * then look for matching devices (phys_refnum) in dlist
3788 * and create appropriate device mdinfo.
3789 */
3790 struct ddf_super *ddf = st->sb;
3791 struct mdinfo *rest = NULL;
3792 struct vcl *vc;
3793
3794 for (vc = ddf->conflist ; vc ; vc=vc->next) {
3795 unsigned int i;
3796 struct mdinfo *this;
3797 char *ep;
3798 __u32 *cptr;
3799 unsigned int pd;
3800
3801 if (subarray &&
3802 (strtoul(subarray, &ep, 10) != vc->vcnum ||
3803 *ep != '\0'))
3804 continue;
3805
3806 if (vc->conf.sec_elmnt_count > 1) {
3807 if (check_secondary(vc) != 0)
3808 continue;
3809 }
3810
3811 this = xcalloc(1, sizeof(*this));
3812 this->next = rest;
3813 rest = this;
3814
3815 if (layout_ddf2md(&vc->conf, &this->array))
3816 continue;
3817 this->array.md_minor = -1;
3818 this->array.major_version = -1;
3819 this->array.minor_version = -2;
3820 this->safe_mode_delay = DDF_SAFE_MODE_DELAY;
3821 cptr = (__u32 *)(vc->conf.guid + 16);
3822 this->array.ctime = DECADE + __be32_to_cpu(*cptr);
3823 this->array.utime = DECADE +
3824 be32_to_cpu(vc->conf.timestamp);
3825 this->array.chunk_size = 512 << vc->conf.chunk_shift;
3826
3827 i = vc->vcnum;
3828 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
3829 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
3830 DDF_init_full) {
3831 this->array.state = 0;
3832 this->resync_start = 0;
3833 } else {
3834 this->array.state = 1;
3835 this->resync_start = MaxSector;
3836 }
3837 _ddf_array_name(this->name, ddf, i);
3838 memset(this->uuid, 0, sizeof(this->uuid));
3839 this->component_size = be64_to_cpu(vc->conf.blocks);
3840 this->array.size = this->component_size / 2;
3841 this->container_member = i;
3842
3843 ddf->currentconf = vc;
3844 uuid_from_super_ddf(st, this->uuid);
3845 if (!subarray)
3846 ddf->currentconf = NULL;
3847
3848 sprintf(this->text_version, "/%s/%d",
3849 st->container_devnm, this->container_member);
3850
3851 for (pd = 0; pd < be16_to_cpu(ddf->phys->max_pdes); pd++) {
3852 struct mdinfo *dev;
3853 struct dl *d;
3854 const struct vd_config *bvd;
3855 unsigned int iphys;
3856 int stt;
3857
3858 if (be32_to_cpu(ddf->phys->entries[pd].refnum)
3859 == 0xFFFFFFFF)
3860 continue;
3861
3862 stt = be16_to_cpu(ddf->phys->entries[pd].state);
3863 if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3864 != DDF_Online)
3865 continue;
3866
3867 i = get_pd_index_from_refnum(
3868 vc, ddf->phys->entries[pd].refnum,
3869 ddf->mppe, &bvd, &iphys);
3870 if (i == DDF_NOTFOUND)
3871 continue;
3872
3873 this->array.working_disks++;
3874
3875 for (d = ddf->dlist; d ; d=d->next)
3876 if (be32_eq(d->disk.refnum,
3877 ddf->phys->entries[pd].refnum))
3878 break;
3879 if (d == NULL)
3880 /* Haven't found that one yet, maybe there are others */
3881 continue;
3882
3883 dev = xcalloc(1, sizeof(*dev));
3884 dev->next = this->devs;
3885 this->devs = dev;
3886
3887 dev->disk.number = be32_to_cpu(d->disk.refnum);
3888 dev->disk.major = d->major;
3889 dev->disk.minor = d->minor;
3890 dev->disk.raid_disk = i;
3891 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
3892 dev->recovery_start = MaxSector;
3893
3894 dev->events = be32_to_cpu(ddf->active->seq);
3895 dev->data_offset =
3896 be64_to_cpu(LBA_OFFSET(ddf, bvd)[iphys]);
3897 dev->component_size = be64_to_cpu(bvd->blocks);
3898 if (d->devname)
3899 strcpy(dev->name, d->devname);
3900 }
3901 }
3902 return rest;
3903 }
3904
3905 static int store_super_ddf(struct supertype *st, int fd)
3906 {
3907 struct ddf_super *ddf = st->sb;
3908 unsigned long long dsize;
3909 void *buf;
3910 int rc;
3911
3912 if (!ddf)
3913 return 1;
3914
3915 if (!get_dev_size(fd, NULL, &dsize))
3916 return 1;
3917
3918 if (ddf->dlist || ddf->conflist) {
3919 struct stat sta;
3920 struct dl *dl;
3921 int ofd, ret;
3922
3923 if (fstat(fd, &sta) == -1 || !S_ISBLK(sta.st_mode)) {
3924 pr_err("%s: file descriptor for invalid device\n",
3925 __func__);
3926 return 1;
3927 }
3928 for (dl = ddf->dlist; dl; dl = dl->next)
3929 if (dl->major == (int)major(sta.st_rdev) &&
3930 dl->minor == (int)minor(sta.st_rdev))
3931 break;
3932 if (!dl) {
3933 pr_err("%s: couldn't find disk %d/%d\n", __func__,
3934 (int)major(sta.st_rdev),
3935 (int)minor(sta.st_rdev));
3936 return 1;
3937 }
3938 ofd = dl->fd;
3939 dl->fd = fd;
3940 ret = (_write_super_to_disk(ddf, dl) != 1);
3941 dl->fd = ofd;
3942 return ret;
3943 }
3944
3945 if (posix_memalign(&buf, 512, 512) != 0)
3946 return 1;
3947 memset(buf, 0, 512);
3948
3949 lseek64(fd, dsize-512, 0);
3950 rc = write(fd, buf, 512);
3951 free(buf);
3952 if (rc < 0)
3953 return 1;
3954 return 0;
3955 }
3956
3957 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3958 {
3959 /*
3960 * return:
3961 * 0 same, or first was empty, and second was copied
3962 * 1 second had wrong magic number - but that isn't possible
3963 * 2 wrong uuid
3964 * 3 wrong other info
3965 */
3966 struct ddf_super *first = st->sb;
3967 struct ddf_super *second = tst->sb;
3968 struct dl *dl1, *dl2;
3969 struct vcl *vl1, *vl2;
3970 unsigned int max_vds, max_pds, pd, vd;
3971
3972 if (!first) {
3973 st->sb = tst->sb;
3974 tst->sb = NULL;
3975 return 0;
3976 }
3977
3978 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3979 return 2;
3980
3981 /* It is only OK to compare info in the anchor. Anything else
3982 * could be changing due to a reconfig so must be ignored.
3983 * guid really should be enough anyway.
3984 */
3985
3986 if (!be32_eq(first->active->seq, second->active->seq)) {
3987 dprintf("%s: sequence number mismatch %u<->%u\n", __func__,
3988 be32_to_cpu(first->active->seq),
3989 be32_to_cpu(second->active->seq));
3990 return 0;
3991 }
3992
3993 /*
3994 * At this point we are fairly sure that the meta data matches.
3995 * But the new disk may contain additional local data.
3996 * Add it to the super block.
3997 */
3998 max_vds = be16_to_cpu(first->active->max_vd_entries);
3999 max_pds = be16_to_cpu(first->phys->max_pdes);
4000 for (vl2 = second->conflist; vl2; vl2 = vl2->next) {
4001 for (vl1 = first->conflist; vl1; vl1 = vl1->next)
4002 if (!memcmp(vl1->conf.guid, vl2->conf.guid,
4003 DDF_GUID_LEN))
4004 break;
4005 if (vl1) {
4006 if (vl1->other_bvds != NULL &&
4007 vl1->conf.sec_elmnt_seq !=
4008 vl2->conf.sec_elmnt_seq) {
4009 dprintf("%s: adding BVD %u\n", __func__,
4010 vl2->conf.sec_elmnt_seq);
4011 add_other_bvd(vl1, &vl2->conf,
4012 first->conf_rec_len*512);
4013 }
4014 continue;
4015 }
4016
4017 if (posix_memalign((void **)&vl1, 512,
4018 (first->conf_rec_len*512 +
4019 offsetof(struct vcl, conf))) != 0) {
4020 pr_err("%s could not allocate vcl buf\n",
4021 __func__);
4022 return 3;
4023 }
4024
4025 vl1->next = first->conflist;
4026 vl1->block_sizes = NULL;
4027 memcpy(&vl1->conf, &vl2->conf, first->conf_rec_len*512);
4028 if (alloc_other_bvds(first, vl1) != 0) {
4029 pr_err("%s could not allocate other bvds\n",
4030 __func__);
4031 free(vl1);
4032 return 3;
4033 }
4034 for (vd = 0; vd < max_vds; vd++)
4035 if (!memcmp(first->virt->entries[vd].guid,
4036 vl1->conf.guid, DDF_GUID_LEN))
4037 break;
4038 vl1->vcnum = vd;
4039 dprintf("%s: added config for VD %u\n", __func__, vl1->vcnum);
4040 first->conflist = vl1;
4041 }
4042
4043 for (dl2 = second->dlist; dl2; dl2 = dl2->next) {
4044 for (dl1 = first->dlist; dl1; dl1 = dl1->next)
4045 if (be32_eq(dl1->disk.refnum, dl2->disk.refnum))
4046 break;
4047 if (dl1)
4048 continue;
4049
4050 if (posix_memalign((void **)&dl1, 512,
4051 sizeof(*dl1) + (first->max_part) * sizeof(dl1->vlist[0]))
4052 != 0) {
4053 pr_err("%s could not allocate disk info buffer\n",
4054 __func__);
4055 return 3;
4056 }
4057 memcpy(dl1, dl2, sizeof(*dl1));
4058 dl1->mdupdate = NULL;
4059 dl1->next = first->dlist;
4060 dl1->fd = -1;
4061 for (pd = 0; pd < max_pds; pd++)
4062 if (be32_eq(first->phys->entries[pd].refnum,
4063 dl1->disk.refnum))
4064 break;
4065 dl1->pdnum = pd < max_pds ? (int)pd : -1;
4066 if (dl2->spare) {
4067 if (posix_memalign((void **)&dl1->spare, 512,
4068 first->conf_rec_len*512) != 0) {
4069 pr_err("%s could not allocate spare info buf\n",
4070 __func__);
4071 return 3;
4072 }
4073 memcpy(dl1->spare, dl2->spare, first->conf_rec_len*512);
4074 }
4075 for (vd = 0 ; vd < first->max_part ; vd++) {
4076 if (!dl2->vlist[vd]) {
4077 dl1->vlist[vd] = NULL;
4078 continue;
4079 }
4080 for (vl1 = first->conflist; vl1; vl1 = vl1->next) {
4081 if (!memcmp(vl1->conf.guid,
4082 dl2->vlist[vd]->conf.guid,
4083 DDF_GUID_LEN))
4084 break;
4085 dl1->vlist[vd] = vl1;
4086 }
4087 }
4088 first->dlist = dl1;
4089 dprintf("%s: added disk %d: %08x\n", __func__, dl1->pdnum,
4090 be32_to_cpu(dl1->disk.refnum));
4091 }
4092
4093 return 0;
4094 }
4095
4096 #ifndef MDASSEMBLE
4097 /*
4098 * A new array 'a' has been started which claims to be instance 'inst'
4099 * within container 'c'.
4100 * We need to confirm that the array matches the metadata in 'c' so
4101 * that we don't corrupt any metadata.
4102 */
4103 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
4104 {
4105 struct ddf_super *ddf = c->sb;
4106 int n = atoi(inst);
4107 struct mdinfo *dev;
4108 struct dl *dl;
4109 static const char faulty[] = "faulty";
4110
4111 if (all_ff(ddf->virt->entries[n].guid)) {
4112 pr_err("%s: subarray %d doesn't exist\n", __func__, n);
4113 return -ENODEV;
4114 }
4115 dprintf("%s: new subarray %d, GUID: %s\n", __func__, n,
4116 guid_str(ddf->virt->entries[n].guid));
4117 for (dev = a->info.devs; dev; dev = dev->next) {
4118 for (dl = ddf->dlist; dl; dl = dl->next)
4119 if (dl->major == dev->disk.major &&
4120 dl->minor == dev->disk.minor)
4121 break;
4122 if (!dl || dl->pdnum < 0) {
4123 pr_err("%s: device %d/%d of subarray %d not found in meta data\n",
4124 __func__, dev->disk.major, dev->disk.minor, n);
4125 return -1;
4126 }
4127 if ((be16_to_cpu(ddf->phys->entries[dl->pdnum].state) &
4128 (DDF_Online|DDF_Missing|DDF_Failed)) != DDF_Online) {
4129 pr_err("%s: new subarray %d contains broken device %d/%d (%02x)\n",
4130 __func__, n, dl->major, dl->minor,
4131 be16_to_cpu(
4132 ddf->phys->entries[dl->pdnum].state));
4133 if (write(dev->state_fd, faulty, sizeof(faulty)-1) !=
4134 sizeof(faulty) - 1)
4135 pr_err("Write to state_fd failed\n");
4136 dev->curr_state = DS_FAULTY;
4137 }
4138 }
4139 a->info.container_member = n;
4140 return 0;
4141 }
4142
4143 static void handle_missing(struct ddf_super *ddf, struct active_array *a, int inst)
4144 {
4145 /* This member array is being activated. If any devices
4146 * are missing they must now be marked as failed.
4147 */
4148 struct vd_config *vc;
4149 unsigned int n_bvd;
4150 struct vcl *vcl;
4151 struct dl *dl;
4152 int pd;
4153 int n;
4154 int state;
4155
4156 for (n = 0; ; n++) {
4157 vc = find_vdcr(ddf, inst, n, &n_bvd, &vcl);
4158 if (!vc)
4159 break;
4160 for (dl = ddf->dlist; dl; dl = dl->next)
4161 if (be32_eq(dl->disk.refnum, vc->phys_refnum[n_bvd]))
4162 break;
4163 if (dl)
4164 /* Found this disk, so not missing */
4165 continue;
4166
4167 /* Mark the device as failed/missing. */
4168 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4169 if (pd >= 0 && be16_and(ddf->phys->entries[pd].state,
4170 cpu_to_be16(DDF_Online))) {
4171 be16_clear(ddf->phys->entries[pd].state,
4172 cpu_to_be16(DDF_Online));
4173 be16_set(ddf->phys->entries[pd].state,
4174 cpu_to_be16(DDF_Failed|DDF_Missing));
4175 vc->phys_refnum[n_bvd] = cpu_to_be32(0);
4176 ddf_set_updates_pending(ddf, vc);
4177 }
4178
4179 /* Mark the array as Degraded */
4180 state = get_svd_state(ddf, vcl);
4181 if (ddf->virt->entries[inst].state !=
4182 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4183 | state)) {
4184 ddf->virt->entries[inst].state =
4185 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4186 | state;
4187 a->check_degraded = 1;
4188 ddf_set_updates_pending(ddf, vc);
4189 }
4190 }
4191 }
4192
4193 /*
4194 * The array 'a' is to be marked clean in the metadata.
4195 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4196 * clean up to the point (in sectors). If that cannot be recorded in the
4197 * metadata, then leave it as dirty.
4198 *
4199 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4200 * !global! virtual_disk.virtual_entry structure.
4201 */
4202 static int ddf_set_array_state(struct active_array *a, int consistent)
4203 {
4204 struct ddf_super *ddf = a->container->sb;
4205 int inst = a->info.container_member;
4206 int old = ddf->virt->entries[inst].state;
4207 if (consistent == 2) {
4208 handle_missing(ddf, a, inst);
4209 /* Should check if a recovery should be started FIXME */
4210 consistent = 1;
4211 if (!is_resync_complete(&a->info))
4212 consistent = 0;
4213 }
4214 if (consistent)
4215 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
4216 else
4217 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
4218 if (old != ddf->virt->entries[inst].state)
4219 ddf_set_updates_pending(ddf, NULL);
4220
4221 old = ddf->virt->entries[inst].init_state;
4222 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
4223 if (is_resync_complete(&a->info))
4224 ddf->virt->entries[inst].init_state |= DDF_init_full;
4225 else if (a->info.resync_start == 0)
4226 ddf->virt->entries[inst].init_state |= DDF_init_not;
4227 else
4228 ddf->virt->entries[inst].init_state |= DDF_init_quick;
4229 if (old != ddf->virt->entries[inst].init_state)
4230 ddf_set_updates_pending(ddf, NULL);
4231
4232 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst,
4233 guid_str(ddf->virt->entries[inst].guid), a->curr_state,
4234 consistent?"clean":"dirty",
4235 a->info.resync_start);
4236 return consistent;
4237 }
4238
4239 static int get_bvd_state(const struct ddf_super *ddf,
4240 const struct vd_config *vc)
4241 {
4242 unsigned int i, n_bvd, working = 0;
4243 unsigned int n_prim = be16_to_cpu(vc->prim_elmnt_count);
4244 int pd, st, state;
4245 for (i = 0; i < n_prim; i++) {
4246 if (!find_index_in_bvd(ddf, vc, i, &n_bvd))
4247 continue;
4248 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4249 if (pd < 0)
4250 continue;
4251 st = be16_to_cpu(ddf->phys->entries[pd].state);
4252 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
4253 == DDF_Online)
4254 working++;
4255 }
4256
4257 state = DDF_state_degraded;
4258 if (working == n_prim)
4259 state = DDF_state_optimal;
4260 else
4261 switch (vc->prl) {
4262 case DDF_RAID0:
4263 case DDF_CONCAT:
4264 case DDF_JBOD:
4265 state = DDF_state_failed;
4266 break;
4267 case DDF_RAID1:
4268 if (working == 0)
4269 state = DDF_state_failed;
4270 else if (working >= 2)
4271 state = DDF_state_part_optimal;
4272 break;
4273 case DDF_RAID4:
4274 case DDF_RAID5:
4275 if (working < n_prim - 1)
4276 state = DDF_state_failed;
4277 break;
4278 case DDF_RAID6:
4279 if (working < n_prim - 2)
4280 state = DDF_state_failed;
4281 else if (working == n_prim - 1)
4282 state = DDF_state_part_optimal;
4283 break;
4284 }
4285 return state;
4286 }
4287
4288 static int secondary_state(int state, int other, int seclevel)
4289 {
4290 if (state == DDF_state_optimal && other == DDF_state_optimal)
4291 return DDF_state_optimal;
4292 if (seclevel == DDF_2MIRRORED) {
4293 if (state == DDF_state_optimal || other == DDF_state_optimal)
4294 return DDF_state_part_optimal;
4295 if (state == DDF_state_failed && other == DDF_state_failed)
4296 return DDF_state_failed;
4297 return DDF_state_degraded;
4298 } else {
4299 if (state == DDF_state_failed || other == DDF_state_failed)
4300 return DDF_state_failed;
4301 if (state == DDF_state_degraded || other == DDF_state_degraded)
4302 return DDF_state_degraded;
4303 return DDF_state_part_optimal;
4304 }
4305 }
4306
4307 static int get_svd_state(const struct ddf_super *ddf, const struct vcl *vcl)
4308 {
4309 int state = get_bvd_state(ddf, &vcl->conf);
4310 unsigned int i;
4311 for (i = 1; i < vcl->conf.sec_elmnt_count; i++) {
4312 state = secondary_state(
4313 state,
4314 get_bvd_state(ddf, vcl->other_bvds[i-1]),
4315 vcl->conf.srl);
4316 }
4317 return state;
4318 }
4319
4320 /*
4321 * The state of each disk is stored in the global phys_disk structure
4322 * in phys_disk.entries[n].state.
4323 * This makes various combinations awkward.
4324 * - When a device fails in any array, it must be failed in all arrays
4325 * that include a part of this device.
4326 * - When a component is rebuilding, we cannot include it officially in the
4327 * array unless this is the only array that uses the device.
4328 *
4329 * So: when transitioning:
4330 * Online -> failed, just set failed flag. monitor will propagate
4331 * spare -> online, the device might need to be added to the array.
4332 * spare -> failed, just set failed. Don't worry if in array or not.
4333 */
4334 static void ddf_set_disk(struct active_array *a, int n, int state)
4335 {
4336 struct ddf_super *ddf = a->container->sb;
4337 unsigned int inst = a->info.container_member, n_bvd;
4338 struct vcl *vcl;
4339 struct vd_config *vc = find_vdcr(ddf, inst, (unsigned int)n,
4340 &n_bvd, &vcl);
4341 int pd;
4342 struct mdinfo *mdi;
4343 struct dl *dl;
4344 int update = 0;
4345
4346 dprintf("%s: %d to %x\n", __func__, n, state);
4347 if (vc == NULL) {
4348 dprintf("ddf: cannot find instance %d!!\n", inst);
4349 return;
4350 }
4351 /* Find the matching slot in 'info'. */
4352 for (mdi = a->info.devs; mdi; mdi = mdi->next)
4353 if (mdi->disk.raid_disk == n)
4354 break;
4355 if (!mdi) {
4356 pr_err("%s: cannot find raid disk %d\n",
4357 __func__, n);
4358 return;
4359 }
4360
4361 /* and find the 'dl' entry corresponding to that. */
4362 for (dl = ddf->dlist; dl; dl = dl->next)
4363 if (mdi->state_fd >= 0 &&
4364 mdi->disk.major == dl->major &&
4365 mdi->disk.minor == dl->minor)
4366 break;
4367 if (!dl) {
4368 pr_err("%s: cannot find raid disk %d (%d/%d)\n",
4369 __func__, n,
4370 mdi->disk.major, mdi->disk.minor);
4371 return;
4372 }
4373
4374 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4375 if (pd < 0 || pd != dl->pdnum) {
4376 /* disk doesn't currently exist or has changed.
4377 * If it is now in_sync, insert it. */
4378 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4379 __func__, dl->pdnum, dl->major, dl->minor,
4380 be32_to_cpu(dl->disk.refnum));
4381 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4382 __func__, inst, n_bvd,
4383 be32_to_cpu(vc->phys_refnum[n_bvd]), pd);
4384 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
4385 pd = dl->pdnum; /* FIXME: is this really correct ? */
4386 vc->phys_refnum[n_bvd] = dl->disk.refnum;
4387 LBA_OFFSET(ddf, vc)[n_bvd] =
4388 cpu_to_be64(mdi->data_offset);
4389 be16_clear(ddf->phys->entries[pd].type,
4390 cpu_to_be16(DDF_Global_Spare));
4391 be16_set(ddf->phys->entries[pd].type,
4392 cpu_to_be16(DDF_Active_in_VD));
4393 update = 1;
4394 }
4395 } else {
4396 be16 old = ddf->phys->entries[pd].state;
4397 if (state & DS_FAULTY)
4398 be16_set(ddf->phys->entries[pd].state,
4399 cpu_to_be16(DDF_Failed));
4400 if (state & DS_INSYNC) {
4401 be16_set(ddf->phys->entries[pd].state,
4402 cpu_to_be16(DDF_Online));
4403 be16_clear(ddf->phys->entries[pd].state,
4404 cpu_to_be16(DDF_Rebuilding));
4405 }
4406 if (!be16_eq(old, ddf->phys->entries[pd].state))
4407 update = 1;
4408 }
4409
4410 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n,
4411 be32_to_cpu(dl->disk.refnum), state,
4412 be16_to_cpu(ddf->phys->entries[pd].state));
4413
4414 /* Now we need to check the state of the array and update
4415 * virtual_disk.entries[n].state.
4416 * It needs to be one of "optimal", "degraded", "failed".
4417 * I don't understand 'deleted' or 'missing'.
4418 */
4419 state = get_svd_state(ddf, vcl);
4420
4421 if (ddf->virt->entries[inst].state !=
4422 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4423 | state)) {
4424 ddf->virt->entries[inst].state =
4425 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4426 | state;
4427 update = 1;
4428 }
4429 if (update)
4430 ddf_set_updates_pending(ddf, vc);
4431 }
4432
4433 static void ddf_sync_metadata(struct supertype *st)
4434 {
4435 /*
4436 * Write all data to all devices.
4437 * Later, we might be able to track whether only local changes
4438 * have been made, or whether any global data has been changed,
4439 * but ddf is sufficiently weird that it probably always
4440 * changes global data ....
4441 */
4442 struct ddf_super *ddf = st->sb;
4443 if (!ddf->updates_pending)
4444 return;
4445 ddf->updates_pending = 0;
4446 __write_init_super_ddf(st);
4447 dprintf("ddf: sync_metadata\n");
4448 }
4449
4450 static int del_from_conflist(struct vcl **list, const char *guid)
4451 {
4452 struct vcl **p;
4453 int found = 0;
4454 for (p = list; p && *p; p = &((*p)->next))
4455 if (!memcmp((*p)->conf.guid, guid, DDF_GUID_LEN)) {
4456 found = 1;
4457 *p = (*p)->next;
4458 }
4459 return found;
4460 }
4461
4462 static int _kill_subarray_ddf(struct ddf_super *ddf, const char *guid)
4463 {
4464 struct dl *dl;
4465 unsigned int vdnum, i;
4466 vdnum = find_vde_by_guid(ddf, guid);
4467 if (vdnum == DDF_NOTFOUND) {
4468 pr_err("%s: could not find VD %s\n", __func__,
4469 guid_str(guid));
4470 return -1;
4471 }
4472 if (del_from_conflist(&ddf->conflist, guid) == 0) {
4473 pr_err("%s: could not find conf %s\n", __func__,
4474 guid_str(guid));
4475 return -1;
4476 }
4477 for (dl = ddf->dlist; dl; dl = dl->next)
4478 for (i = 0; i < ddf->max_part; i++)
4479 if (dl->vlist[i] != NULL &&
4480 !memcmp(dl->vlist[i]->conf.guid, guid,
4481 DDF_GUID_LEN))
4482 dl->vlist[i] = NULL;
4483 memset(ddf->virt->entries[vdnum].guid, 0xff, DDF_GUID_LEN);
4484 dprintf("%s: deleted %s\n", __func__, guid_str(guid));
4485 return 0;
4486 }
4487
4488 static int kill_subarray_ddf(struct supertype *st)
4489 {
4490 struct ddf_super *ddf = st->sb;
4491 /*
4492 * currentconf is set in container_content_ddf,
4493 * called with subarray arg
4494 */
4495 struct vcl *victim = ddf->currentconf;
4496 struct vd_config *conf;
4497 unsigned int vdnum;
4498
4499 ddf->currentconf = NULL;
4500 if (!victim) {
4501 pr_err("%s: nothing to kill\n", __func__);
4502 return -1;
4503 }
4504 conf = &victim->conf;
4505 vdnum = find_vde_by_guid(ddf, conf->guid);
4506 if (vdnum == DDF_NOTFOUND) {
4507 pr_err("%s: could not find VD %s\n", __func__,
4508 guid_str(conf->guid));
4509 return -1;
4510 }
4511 if (st->update_tail) {
4512 struct virtual_disk *vd;
4513 int len = sizeof(struct virtual_disk)
4514 + sizeof(struct virtual_entry);
4515 vd = xmalloc(len);
4516 if (vd == NULL) {
4517 pr_err("%s: failed to allocate %d bytes\n", __func__,
4518 len);
4519 return -1;
4520 }
4521 memset(vd, 0 , len);
4522 vd->magic = DDF_VIRT_RECORDS_MAGIC;
4523 vd->populated_vdes = cpu_to_be16(0);
4524 memcpy(vd->entries[0].guid, conf->guid, DDF_GUID_LEN);
4525 /* we use DDF_state_deleted as marker */
4526 vd->entries[0].state = DDF_state_deleted;
4527 append_metadata_update(st, vd, len);
4528 } else {
4529 _kill_subarray_ddf(ddf, conf->guid);
4530 ddf_set_updates_pending(ddf, NULL);
4531 ddf_sync_metadata(st);
4532 }
4533 return 0;
4534 }
4535
4536 static void copy_matching_bvd(struct ddf_super *ddf,
4537 struct vd_config *conf,
4538 const struct metadata_update *update)
4539 {
4540 unsigned int mppe =
4541 be16_to_cpu(ddf->anchor.max_primary_element_entries);
4542 unsigned int len = ddf->conf_rec_len * 512;
4543 char *p;
4544 struct vd_config *vc;
4545 for (p = update->buf; p < update->buf + update->len; p += len) {
4546 vc = (struct vd_config *) p;
4547 if (vc->sec_elmnt_seq == conf->sec_elmnt_seq) {
4548 memcpy(conf->phys_refnum, vc->phys_refnum,
4549 mppe * (sizeof(__u32) + sizeof(__u64)));
4550 return;
4551 }
4552 }
4553 pr_err("%s: no match for BVD %d of %s in update\n", __func__,
4554 conf->sec_elmnt_seq, guid_str(conf->guid));
4555 }
4556
4557 static void ddf_process_update(struct supertype *st,
4558 struct metadata_update *update)
4559 {
4560 /* Apply this update to the metadata.
4561 * The first 4 bytes are a DDF_*_MAGIC which guides
4562 * our actions.
4563 * Possible update are:
4564 * DDF_PHYS_RECORDS_MAGIC
4565 * Add a new physical device or remove an old one.
4566 * Changes to this record only happen implicitly.
4567 * used_pdes is the device number.
4568 * DDF_VIRT_RECORDS_MAGIC
4569 * Add a new VD. Possibly also change the 'access' bits.
4570 * populated_vdes is the entry number.
4571 * DDF_VD_CONF_MAGIC
4572 * New or updated VD. the VIRT_RECORD must already
4573 * exist. For an update, phys_refnum and lba_offset
4574 * (at least) are updated, and the VD_CONF must
4575 * be written to precisely those devices listed with
4576 * a phys_refnum.
4577 * DDF_SPARE_ASSIGN_MAGIC
4578 * replacement Spare Assignment Record... but for which device?
4579 *
4580 * So, e.g.:
4581 * - to create a new array, we send a VIRT_RECORD and
4582 * a VD_CONF. Then assemble and start the array.
4583 * - to activate a spare we send a VD_CONF to add the phys_refnum
4584 * and offset. This will also mark the spare as active with
4585 * a spare-assignment record.
4586 */
4587 struct ddf_super *ddf = st->sb;
4588 be32 *magic = (be32 *)update->buf;
4589 struct phys_disk *pd;
4590 struct virtual_disk *vd;
4591 struct vd_config *vc;
4592 struct vcl *vcl;
4593 struct dl *dl;
4594 unsigned int ent;
4595 unsigned int pdnum, pd2, len;
4596
4597 dprintf("Process update %x\n", be32_to_cpu(*magic));
4598
4599 if (be32_eq(*magic, DDF_PHYS_RECORDS_MAGIC)) {
4600 if (update->len != (sizeof(struct phys_disk) +
4601 sizeof(struct phys_disk_entry)))
4602 return;
4603 pd = (struct phys_disk*)update->buf;
4604
4605 ent = be16_to_cpu(pd->used_pdes);
4606 if (ent >= be16_to_cpu(ddf->phys->max_pdes))
4607 return;
4608 if (be16_and(pd->entries[0].state, cpu_to_be16(DDF_Missing))) {
4609 struct dl **dlp;
4610 /* removing this disk. */
4611 be16_set(ddf->phys->entries[ent].state,
4612 cpu_to_be16(DDF_Missing));
4613 for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) {
4614 struct dl *dl = *dlp;
4615 if (dl->pdnum == (signed)ent) {
4616 close(dl->fd);
4617 dl->fd = -1;
4618 /* FIXME this doesn't free
4619 * dl->devname */
4620 update->space = dl;
4621 *dlp = dl->next;
4622 break;
4623 }
4624 }
4625 ddf_set_updates_pending(ddf, NULL);
4626 return;
4627 }
4628 if (!all_ff(ddf->phys->entries[ent].guid))
4629 return;
4630 ddf->phys->entries[ent] = pd->entries[0];
4631 ddf->phys->used_pdes = cpu_to_be16
4632 (1 + be16_to_cpu(ddf->phys->used_pdes));
4633 ddf_set_updates_pending(ddf, NULL);
4634 if (ddf->add_list) {
4635 struct active_array *a;
4636 struct dl *al = ddf->add_list;
4637 ddf->add_list = al->next;
4638
4639 al->next = ddf->dlist;
4640 ddf->dlist = al;
4641
4642 /* As a device has been added, we should check
4643 * for any degraded devices that might make
4644 * use of this spare */
4645 for (a = st->arrays ; a; a=a->next)
4646 a->check_degraded = 1;
4647 }
4648 } else if (be32_eq(*magic, DDF_VIRT_RECORDS_MAGIC)) {
4649 if (update->len != (sizeof(struct virtual_disk) +
4650 sizeof(struct virtual_entry)))
4651 return;
4652 vd = (struct virtual_disk*)update->buf;
4653
4654 if (vd->entries[0].state == DDF_state_deleted) {
4655 if (_kill_subarray_ddf(ddf, vd->entries[0].guid))
4656 return;
4657 } else {
4658 ent = find_vde_by_guid(ddf, vd->entries[0].guid);
4659 if (ent != DDF_NOTFOUND) {
4660 dprintf("%s: VD %s exists already in slot %d\n",
4661 __func__, guid_str(vd->entries[0].guid),
4662 ent);
4663 return;
4664 }
4665 ent = find_unused_vde(ddf);
4666 if (ent == DDF_NOTFOUND)
4667 return;
4668 ddf->virt->entries[ent] = vd->entries[0];
4669 ddf->virt->populated_vdes =
4670 cpu_to_be16(
4671 1 + be16_to_cpu(
4672 ddf->virt->populated_vdes));
4673 dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n",
4674 __func__, guid_str(vd->entries[0].guid), ent,
4675 ddf->virt->entries[ent].state,
4676 ddf->virt->entries[ent].init_state);
4677 }
4678 ddf_set_updates_pending(ddf, NULL);
4679 }
4680
4681 else if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4682 vc = (struct vd_config*)update->buf;
4683 len = ddf->conf_rec_len * 512;
4684 if ((unsigned int)update->len != len * vc->sec_elmnt_count) {
4685 pr_err("%s: %s: insufficient data (%d) for %u BVDs\n",
4686 __func__, guid_str(vc->guid), update->len,
4687 vc->sec_elmnt_count);
4688 return;
4689 }
4690 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
4691 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
4692 break;
4693 dprintf("%s: conf update for %s (%s)\n", __func__,
4694 guid_str(vc->guid), (vcl ? "old" : "new"));
4695 if (vcl) {
4696 /* An update, just copy the phys_refnum and lba_offset
4697 * fields
4698 */
4699 unsigned int i;
4700 unsigned int k;
4701 copy_matching_bvd(ddf, &vcl->conf, update);
4702 for (k = 0; k < be16_to_cpu(vc->prim_elmnt_count); k++)
4703 dprintf("BVD %u has %08x at %llu\n", 0,
4704 be32_to_cpu(vcl->conf.phys_refnum[k]),
4705 be64_to_cpu(LBA_OFFSET(ddf,
4706 &vcl->conf)[k]));
4707 for (i = 1; i < vc->sec_elmnt_count; i++) {
4708 copy_matching_bvd(ddf, vcl->other_bvds[i-1],
4709 update);
4710 for (k = 0; k < be16_to_cpu(
4711 vc->prim_elmnt_count); k++)
4712 dprintf("BVD %u has %08x at %llu\n", i,
4713 be32_to_cpu
4714 (vcl->other_bvds[i-1]->
4715 phys_refnum[k]),
4716 be64_to_cpu
4717 (LBA_OFFSET
4718 (ddf,
4719 vcl->other_bvds[i-1])[k]));
4720 }
4721 } else {
4722 /* A new VD_CONF */
4723 unsigned int i;
4724 if (!update->space)
4725 return;
4726 vcl = update->space;
4727 update->space = NULL;
4728 vcl->next = ddf->conflist;
4729 memcpy(&vcl->conf, vc, len);
4730 ent = find_vde_by_guid(ddf, vc->guid);
4731 if (ent == DDF_NOTFOUND)
4732 return;
4733 vcl->vcnum = ent;
4734 ddf->conflist = vcl;
4735 for (i = 1; i < vc->sec_elmnt_count; i++)
4736 memcpy(vcl->other_bvds[i-1],
4737 update->buf + len * i, len);
4738 }
4739 /* Set DDF_Transition on all Failed devices - to help
4740 * us detect those that are no longer in use
4741 */
4742 for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4743 pdnum++)
4744 if (be16_and(ddf->phys->entries[pdnum].state,
4745 cpu_to_be16(DDF_Failed)))
4746 be16_set(ddf->phys->entries[pdnum].state,
4747 cpu_to_be16(DDF_Transition));
4748 /* Now make sure vlist is correct for each dl. */
4749 for (dl = ddf->dlist; dl; dl = dl->next) {
4750 unsigned int vn = 0;
4751 int in_degraded = 0;
4752
4753 if (dl->pdnum < 0)
4754 continue;
4755 for (vcl = ddf->conflist; vcl ; vcl = vcl->next) {
4756 unsigned int dn, ibvd;
4757 const struct vd_config *conf;
4758 int vstate;
4759 dn = get_pd_index_from_refnum(vcl,
4760 dl->disk.refnum,
4761 ddf->mppe,
4762 &conf, &ibvd);
4763 if (dn == DDF_NOTFOUND)
4764 continue;
4765 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4766 dl->pdnum,
4767 be32_to_cpu(dl->disk.refnum),
4768 guid_str(conf->guid),
4769 conf->sec_elmnt_seq, vn);
4770 /* Clear the Transition flag */
4771 if (be16_and
4772 (ddf->phys->entries[dl->pdnum].state,
4773 cpu_to_be16(DDF_Failed)))
4774 be16_clear(ddf->phys
4775 ->entries[dl->pdnum].state,
4776 cpu_to_be16(DDF_Transition));
4777 dl->vlist[vn++] = vcl;
4778 vstate = ddf->virt->entries[vcl->vcnum].state
4779 & DDF_state_mask;
4780 if (vstate == DDF_state_degraded ||
4781 vstate == DDF_state_part_optimal)
4782 in_degraded = 1;
4783 }
4784 while (vn < ddf->max_part)
4785 dl->vlist[vn++] = NULL;
4786 if (dl->vlist[0]) {
4787 be16_clear(ddf->phys->entries[dl->pdnum].type,
4788 cpu_to_be16(DDF_Global_Spare));
4789 if (!be16_and(ddf->phys
4790 ->entries[dl->pdnum].type,
4791 cpu_to_be16(DDF_Active_in_VD))) {
4792 be16_set(ddf->phys
4793 ->entries[dl->pdnum].type,
4794 cpu_to_be16(DDF_Active_in_VD));
4795 if (in_degraded)
4796 be16_set(ddf->phys
4797 ->entries[dl->pdnum]
4798 .state,
4799 cpu_to_be16
4800 (DDF_Rebuilding));
4801 }
4802 }
4803 if (dl->spare) {
4804 be16_clear(ddf->phys->entries[dl->pdnum].type,
4805 cpu_to_be16(DDF_Global_Spare));
4806 be16_set(ddf->phys->entries[dl->pdnum].type,
4807 cpu_to_be16(DDF_Spare));
4808 }
4809 if (!dl->vlist[0] && !dl->spare) {
4810 be16_set(ddf->phys->entries[dl->pdnum].type,
4811 cpu_to_be16(DDF_Global_Spare));
4812 be16_clear(ddf->phys->entries[dl->pdnum].type,
4813 cpu_to_be16(DDF_Spare));
4814 be16_clear(ddf->phys->entries[dl->pdnum].type,
4815 cpu_to_be16(DDF_Active_in_VD));
4816 }
4817 }
4818
4819 /* Now remove any 'Failed' devices that are not part
4820 * of any VD. They will have the Transition flag set.
4821 * Once done, we need to update all dl->pdnum numbers.
4822 */
4823 pd2 = 0;
4824 for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4825 pdnum++) {
4826 if (be32_to_cpu(ddf->phys->entries[pdnum].refnum) ==
4827 0xFFFFFFFF)
4828 continue;
4829 if (be16_and(ddf->phys->entries[pdnum].state,
4830 cpu_to_be16(DDF_Failed))
4831 && be16_and(ddf->phys->entries[pdnum].state,
4832 cpu_to_be16(DDF_Transition))) {
4833 /* skip this one unless in dlist*/
4834 for (dl = ddf->dlist; dl; dl = dl->next)
4835 if (dl->pdnum == (int)pdnum)
4836 break;
4837 if (!dl)
4838 continue;
4839 }
4840 if (pdnum == pd2)
4841 pd2++;
4842 else {
4843 ddf->phys->entries[pd2] =
4844 ddf->phys->entries[pdnum];
4845 for (dl = ddf->dlist; dl; dl = dl->next)
4846 if (dl->pdnum == (int)pdnum)
4847 dl->pdnum = pd2;
4848 pd2++;
4849 }
4850 }
4851 ddf->phys->used_pdes = cpu_to_be16(pd2);
4852 while (pd2 < pdnum) {
4853 memset(ddf->phys->entries[pd2].guid, 0xff,
4854 DDF_GUID_LEN);
4855 pd2++;
4856 }
4857
4858 ddf_set_updates_pending(ddf, vc);
4859 }
4860 /* case DDF_SPARE_ASSIGN_MAGIC */
4861 }
4862
4863 static void ddf_prepare_update(struct supertype *st,
4864 struct metadata_update *update)
4865 {
4866 /* This update arrived at managemon.
4867 * We are about to pass it to monitor.
4868 * If a malloc is needed, do it here.
4869 */
4870 struct ddf_super *ddf = st->sb;
4871 be32 *magic = (be32 *)update->buf;
4872 if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4873 struct vcl *vcl;
4874 struct vd_config *conf = (struct vd_config *) update->buf;
4875 if (posix_memalign(&update->space, 512,
4876 offsetof(struct vcl, conf)
4877 + ddf->conf_rec_len * 512) != 0) {
4878 update->space = NULL;
4879 return;
4880 }
4881 vcl = update->space;
4882 vcl->conf.sec_elmnt_count = conf->sec_elmnt_count;
4883 if (alloc_other_bvds(ddf, vcl) != 0) {
4884 free(update->space);
4885 update->space = NULL;
4886 }
4887 }
4888 }
4889
4890 /*
4891 * Check degraded state of a RAID10.
4892 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4893 */
4894 static int raid10_degraded(struct mdinfo *info)
4895 {
4896 int n_prim, n_bvds;
4897 int i;
4898 struct mdinfo *d;
4899 char *found;
4900 int ret = -1;
4901
4902 n_prim = info->array.layout & ~0x100;
4903 n_bvds = info->array.raid_disks / n_prim;
4904 found = xmalloc(n_bvds);
4905 if (found == NULL)
4906 return ret;
4907 memset(found, 0, n_bvds);
4908 for (d = info->devs; d; d = d->next) {
4909 i = d->disk.raid_disk / n_prim;
4910 if (i >= n_bvds) {
4911 pr_err("%s: BUG: invalid raid disk\n", __func__);
4912 goto out;
4913 }
4914 if (d->state_fd > 0)
4915 found[i]++;
4916 }
4917 ret = 2;
4918 for (i = 0; i < n_bvds; i++)
4919 if (!found[i]) {
4920 dprintf("%s: BVD %d/%d failed\n", __func__, i, n_bvds);
4921 ret = 0;
4922 goto out;
4923 } else if (found[i] < n_prim) {
4924 dprintf("%s: BVD %d/%d degraded\n", __func__, i,
4925 n_bvds);
4926 ret = 1;
4927 }
4928 out:
4929 free(found);
4930 return ret;
4931 }
4932
4933 /*
4934 * Check if the array 'a' is degraded but not failed.
4935 * If it is, find as many spares as are available and needed and
4936 * arrange for their inclusion.
4937 * We only choose devices which are not already in the array,
4938 * and prefer those with a spare-assignment to this array.
4939 * Otherwise we choose global spares - assuming always that
4940 * there is enough room.
4941 * For each spare that we assign, we return an 'mdinfo' which
4942 * describes the position for the device in the array.
4943 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4944 * the new phys_refnum and lba_offset values.
4945 *
4946 * Only worry about BVDs at the moment.
4947 */
4948 static struct mdinfo *ddf_activate_spare(struct active_array *a,
4949 struct metadata_update **updates)
4950 {
4951 int working = 0;
4952 struct mdinfo *d;
4953 struct ddf_super *ddf = a->container->sb;
4954 int global_ok = 0;
4955 struct mdinfo *rv = NULL;
4956 struct mdinfo *di;
4957 struct metadata_update *mu;
4958 struct dl *dl;
4959 int i;
4960 unsigned int j;
4961 struct vcl *vcl;
4962 struct vd_config *vc;
4963 unsigned int n_bvd;
4964
4965 for (d = a->info.devs ; d ; d = d->next) {
4966 if ((d->curr_state & DS_FAULTY) &&
4967 d->state_fd >= 0)
4968 /* wait for Removal to happen */
4969 return NULL;
4970 if (d->state_fd >= 0)
4971 working ++;
4972 }
4973
4974 dprintf("%s: working=%d (%d) level=%d\n", __func__, working,
4975 a->info.array.raid_disks,
4976 a->info.array.level);
4977 if (working == a->info.array.raid_disks)
4978 return NULL; /* array not degraded */
4979 switch (a->info.array.level) {
4980 case 1:
4981 if (working == 0)
4982 return NULL; /* failed */
4983 break;
4984 case 4:
4985 case 5:
4986 if (working < a->info.array.raid_disks - 1)
4987 return NULL; /* failed */
4988 break;
4989 case 6:
4990 if (working < a->info.array.raid_disks - 2)
4991 return NULL; /* failed */
4992 break;
4993 case 10:
4994 if (raid10_degraded(&a->info) < 1)
4995 return NULL;
4996 break;
4997 default: /* concat or stripe */
4998 return NULL; /* failed */
4999 }
5000
5001 /* For each slot, if it is not working, find a spare */
5002 dl = ddf->dlist;
5003 for (i = 0; i < a->info.array.raid_disks; i++) {
5004 for (d = a->info.devs ; d ; d = d->next)
5005 if (d->disk.raid_disk == i)
5006 break;
5007 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
5008 if (d && (d->state_fd >= 0))
5009 continue;
5010
5011 /* OK, this device needs recovery. Find a spare */
5012 again:
5013 for ( ; dl ; dl = dl->next) {
5014 unsigned long long esize;
5015 unsigned long long pos;
5016 struct mdinfo *d2;
5017 int is_global = 0;
5018 int is_dedicated = 0;
5019 struct extent *ex;
5020 unsigned int j;
5021 be16 state;
5022
5023 if (dl->pdnum < 0)
5024 continue;
5025 state = ddf->phys->entries[dl->pdnum].state;
5026 if (be16_and(state,
5027 cpu_to_be16(DDF_Failed|DDF_Missing)) ||
5028 !be16_and(state,
5029 cpu_to_be16(DDF_Online)))
5030 continue;
5031
5032 /* If in this array, skip */
5033 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
5034 if (d2->state_fd >= 0 &&
5035 d2->disk.major == dl->major &&
5036 d2->disk.minor == dl->minor) {
5037 dprintf("%x:%x (%08x) already in array\n",
5038 dl->major, dl->minor,
5039 be32_to_cpu(dl->disk.refnum));
5040 break;
5041 }
5042 if (d2)
5043 continue;
5044 if (be16_and(ddf->phys->entries[dl->pdnum].type,
5045 cpu_to_be16(DDF_Spare))) {
5046 /* Check spare assign record */
5047 if (dl->spare) {
5048 if (dl->spare->type & DDF_spare_dedicated) {
5049 /* check spare_ents for guid */
5050 for (j = 0 ;
5051 j < be16_to_cpu
5052 (dl->spare
5053 ->populated);
5054 j++) {
5055 if (memcmp(dl->spare->spare_ents[j].guid,
5056 ddf->virt->entries[a->info.container_member].guid,
5057 DDF_GUID_LEN) == 0)
5058 is_dedicated = 1;
5059 }
5060 } else
5061 is_global = 1;
5062 }
5063 } else if (be16_and(ddf->phys->entries[dl->pdnum].type,
5064 cpu_to_be16(DDF_Global_Spare))) {
5065 is_global = 1;
5066 } else if (!be16_and(ddf->phys
5067 ->entries[dl->pdnum].state,
5068 cpu_to_be16(DDF_Failed))) {
5069 /* we can possibly use some of this */
5070 is_global = 1;
5071 }
5072 if ( ! (is_dedicated ||
5073 (is_global && global_ok))) {
5074 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
5075 is_dedicated, is_global);
5076 continue;
5077 }
5078
5079 /* We are allowed to use this device - is there space?
5080 * We need a->info.component_size sectors */
5081 ex = get_extents(ddf, dl);
5082 if (!ex) {
5083 dprintf("cannot get extents\n");
5084 continue;
5085 }
5086 j = 0; pos = 0;
5087 esize = 0;
5088
5089 do {
5090 esize = ex[j].start - pos;
5091 if (esize >= a->info.component_size)
5092 break;
5093 pos = ex[j].start + ex[j].size;
5094 j++;
5095 } while (ex[j-1].size);
5096
5097 free(ex);
5098 if (esize < a->info.component_size) {
5099 dprintf("%x:%x has no room: %llu %llu\n",
5100 dl->major, dl->minor,
5101 esize, a->info.component_size);
5102 /* No room */
5103 continue;
5104 }
5105
5106 /* Cool, we have a device with some space at pos */
5107 di = xcalloc(1, sizeof(*di));
5108 di->disk.number = i;
5109 di->disk.raid_disk = i;
5110 di->disk.major = dl->major;
5111 di->disk.minor = dl->minor;
5112 di->disk.state = 0;
5113 di->recovery_start = 0;
5114 di->data_offset = pos;
5115 di->component_size = a->info.component_size;
5116 di->next = rv;
5117 rv = di;
5118 dprintf("%x:%x (%08x) to be %d at %llu\n",
5119 dl->major, dl->minor,
5120 be32_to_cpu(dl->disk.refnum), i, pos);
5121
5122 break;
5123 }
5124 if (!dl && ! global_ok) {
5125 /* not enough dedicated spares, try global */
5126 global_ok = 1;
5127 dl = ddf->dlist;
5128 goto again;
5129 }
5130 }
5131
5132 if (!rv)
5133 /* No spares found */
5134 return rv;
5135 /* Now 'rv' has a list of devices to return.
5136 * Create a metadata_update record to update the
5137 * phys_refnum and lba_offset values
5138 */
5139 vc = find_vdcr(ddf, a->info.container_member, rv->disk.raid_disk,
5140 &n_bvd, &vcl);
5141 if (vc == NULL)
5142 return NULL;
5143
5144 mu = xmalloc(sizeof(*mu));
5145 if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
5146 free(mu);
5147 mu = NULL;
5148 }
5149
5150 mu->len = ddf->conf_rec_len * 512 * vcl->conf.sec_elmnt_count;
5151 mu->buf = xmalloc(mu->len);
5152 mu->space = NULL;
5153 mu->space_list = NULL;
5154 mu->next = *updates;
5155 memcpy(mu->buf, &vcl->conf, ddf->conf_rec_len * 512);
5156 for (j = 1; j < vcl->conf.sec_elmnt_count; j++)
5157 memcpy(mu->buf + j * ddf->conf_rec_len * 512,
5158 vcl->other_bvds[j-1], ddf->conf_rec_len * 512);
5159
5160 vc = (struct vd_config*)mu->buf;
5161 for (di = rv ; di ; di = di->next) {
5162 unsigned int i_sec, i_prim;
5163 i_sec = di->disk.raid_disk
5164 / be16_to_cpu(vcl->conf.prim_elmnt_count);
5165 i_prim = di->disk.raid_disk
5166 % be16_to_cpu(vcl->conf.prim_elmnt_count);
5167 vc = (struct vd_config *)(mu->buf
5168 + i_sec * ddf->conf_rec_len * 512);
5169 for (dl = ddf->dlist; dl; dl = dl->next)
5170 if (dl->major == di->disk.major
5171 && dl->minor == di->disk.minor)
5172 break;
5173 if (!dl || dl->pdnum < 0) {
5174 pr_err("%s: BUG: can't find disk %d (%d/%d)\n",
5175 __func__, di->disk.raid_disk,
5176 di->disk.major, di->disk.minor);
5177 return NULL;
5178 }
5179 vc->phys_refnum[i_prim] = ddf->phys->entries[dl->pdnum].refnum;
5180 LBA_OFFSET(ddf, vc)[i_prim] = cpu_to_be64(di->data_offset);
5181 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec, i_prim,
5182 be32_to_cpu(vc->phys_refnum[i_prim]),
5183 be64_to_cpu(LBA_OFFSET(ddf, vc)[i_prim]));
5184 }
5185 *updates = mu;
5186 return rv;
5187 }
5188 #endif /* MDASSEMBLE */
5189
5190 static int ddf_level_to_layout(int level)
5191 {
5192 switch(level) {
5193 case 0:
5194 case 1:
5195 return 0;
5196 case 5:
5197 return ALGORITHM_LEFT_SYMMETRIC;
5198 case 6:
5199 return ALGORITHM_ROTATING_N_CONTINUE;
5200 case 10:
5201 return 0x102;
5202 default:
5203 return UnSet;
5204 }
5205 }
5206
5207 static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk)
5208 {
5209 if (level && *level == UnSet)
5210 *level = LEVEL_CONTAINER;
5211
5212 if (level && layout && *layout == UnSet)
5213 *layout = ddf_level_to_layout(*level);
5214 }
5215
5216 struct superswitch super_ddf = {
5217 #ifndef MDASSEMBLE
5218 .examine_super = examine_super_ddf,
5219 .brief_examine_super = brief_examine_super_ddf,
5220 .brief_examine_subarrays = brief_examine_subarrays_ddf,
5221 .export_examine_super = export_examine_super_ddf,
5222 .detail_super = detail_super_ddf,
5223 .brief_detail_super = brief_detail_super_ddf,
5224 .validate_geometry = validate_geometry_ddf,
5225 .write_init_super = write_init_super_ddf,
5226 .add_to_super = add_to_super_ddf,
5227 .remove_from_super = remove_from_super_ddf,
5228 .load_container = load_container_ddf,
5229 .copy_metadata = copy_metadata_ddf,
5230 .kill_subarray = kill_subarray_ddf,
5231 #endif
5232 .match_home = match_home_ddf,
5233 .uuid_from_super= uuid_from_super_ddf,
5234 .getinfo_super = getinfo_super_ddf,
5235 .update_super = update_super_ddf,
5236
5237 .avail_size = avail_size_ddf,
5238
5239 .compare_super = compare_super_ddf,
5240
5241 .load_super = load_super_ddf,
5242 .init_super = init_super_ddf,
5243 .store_super = store_super_ddf,
5244 .free_super = free_super_ddf,
5245 .match_metadata_desc = match_metadata_desc_ddf,
5246 .container_content = container_content_ddf,
5247 .default_geometry = default_geometry_ddf,
5248
5249 .external = 1,
5250
5251 #ifndef MDASSEMBLE
5252 /* for mdmon */
5253 .open_new = ddf_open_new,
5254 .set_array_state= ddf_set_array_state,
5255 .set_disk = ddf_set_disk,
5256 .sync_metadata = ddf_sync_metadata,
5257 .process_update = ddf_process_update,
5258 .prepare_update = ddf_prepare_update,
5259 .activate_spare = ddf_activate_spare,
5260 #endif
5261 .name = "ddf",
5262 };
Unnamed repository; edit this file 'description' to name the repository.