]> 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
Consistently print program Name and __func__ in debug messages.
[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 #include <stddef.h>
34
35 /* a non-official T10 name for creation GUIDs */
36 static char T10[] = "Linux-MD";
37
38 /* DDF timestamps are 1980 based, so we need to add
39 * second-in-decade-of-seventies to convert to linux timestamps.
40 * 10 years with 2 leap years.
41 */
42 #define DECADE (3600*24*(365*10+2))
43 unsigned long crc32(
44 unsigned long crc,
45 const unsigned char *buf,
46 unsigned len);
47
48 #define DDF_NOTFOUND (~0U)
49 #define DDF_CONTAINER (DDF_NOTFOUND-1)
50
51 /* Default for safe_mode_delay. Same value as for IMSM.
52 */
53 static const int DDF_SAFE_MODE_DELAY = 4000;
54
55 /* The DDF metadata handling.
56 * DDF metadata lives at the end of the device.
57 * The last 512 byte block provides an 'anchor' which is used to locate
58 * the rest of the metadata which usually lives immediately behind the anchor.
59 *
60 * Note:
61 * - all multibyte numeric fields are bigendian.
62 * - all strings are space padded.
63 *
64 */
65
66 typedef struct __be16 {
67 __u16 _v16;
68 } be16;
69 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
70 #define be16_and(x, y) ((x)._v16 & (y)._v16)
71 #define be16_or(x, y) ((x)._v16 | (y)._v16)
72 #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16)
73 #define be16_set(x, y) ((x)._v16 |= (y)._v16)
74
75 typedef struct __be32 {
76 __u32 _v32;
77 } be32;
78 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
79
80 typedef struct __be64 {
81 __u64 _v64;
82 } be64;
83 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
84
85 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
86 static inline be16 cpu_to_be16(__u16 x)
87 {
88 be16 be = { ._v16 = __cpu_to_be16(x) };
89 return be;
90 }
91
92 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
93 static inline be32 cpu_to_be32(__u32 x)
94 {
95 be32 be = { ._v32 = __cpu_to_be32(x) };
96 return be;
97 }
98
99 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
100 static inline be64 cpu_to_be64(__u64 x)
101 {
102 be64 be = { ._v64 = __cpu_to_be64(x) };
103 return be;
104 }
105
106 /* Primary Raid Level (PRL) */
107 #define DDF_RAID0 0x00
108 #define DDF_RAID1 0x01
109 #define DDF_RAID3 0x03
110 #define DDF_RAID4 0x04
111 #define DDF_RAID5 0x05
112 #define DDF_RAID1E 0x11
113 #define DDF_JBOD 0x0f
114 #define DDF_CONCAT 0x1f
115 #define DDF_RAID5E 0x15
116 #define DDF_RAID5EE 0x25
117 #define DDF_RAID6 0x06
118
119 /* Raid Level Qualifier (RLQ) */
120 #define DDF_RAID0_SIMPLE 0x00
121 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
122 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
123 #define DDF_RAID3_0 0x00 /* parity in first extent */
124 #define DDF_RAID3_N 0x01 /* parity in last extent */
125 #define DDF_RAID4_0 0x00 /* parity in first extent */
126 #define DDF_RAID4_N 0x01 /* parity in last extent */
127 /* these apply to raid5e and raid5ee as well */
128 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
129 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
130 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
131 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
132
133 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
134 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
135
136 /* Secondary RAID Level (SRL) */
137 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
138 #define DDF_2MIRRORED 0x01
139 #define DDF_2CONCAT 0x02
140 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
141
142 /* Magic numbers */
143 #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11)
144 #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111)
145 #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222)
146 #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333)
147 #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD)
148 #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE)
149 #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555)
150 #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888)
151 #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0)
152 #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C)
153
154 #define DDF_GUID_LEN 24
155 #define DDF_REVISION_0 "01.00.00"
156 #define DDF_REVISION_2 "01.02.00"
157
158 struct ddf_header {
159 be32 magic; /* DDF_HEADER_MAGIC */
160 be32 crc;
161 char guid[DDF_GUID_LEN];
162 char revision[8]; /* 01.02.00 */
163 be32 seq; /* starts at '1' */
164 be32 timestamp;
165 __u8 openflag;
166 __u8 foreignflag;
167 __u8 enforcegroups;
168 __u8 pad0; /* 0xff */
169 __u8 pad1[12]; /* 12 * 0xff */
170 /* 64 bytes so far */
171 __u8 header_ext[32]; /* reserved: fill with 0xff */
172 be64 primary_lba;
173 be64 secondary_lba;
174 __u8 type;
175 __u8 pad2[3]; /* 0xff */
176 be32 workspace_len; /* sectors for vendor space -
177 * at least 32768(sectors) */
178 be64 workspace_lba;
179 be16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
180 be16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
181 be16 max_partitions; /* i.e. max num of configuration
182 record entries per disk */
183 be16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
184 *12/512) */
185 be16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
186 __u8 pad3[54]; /* 0xff */
187 /* 192 bytes so far */
188 be32 controller_section_offset;
189 be32 controller_section_length;
190 be32 phys_section_offset;
191 be32 phys_section_length;
192 be32 virt_section_offset;
193 be32 virt_section_length;
194 be32 config_section_offset;
195 be32 config_section_length;
196 be32 data_section_offset;
197 be32 data_section_length;
198 be32 bbm_section_offset;
199 be32 bbm_section_length;
200 be32 diag_space_offset;
201 be32 diag_space_length;
202 be32 vendor_offset;
203 be32 vendor_length;
204 /* 256 bytes so far */
205 __u8 pad4[256]; /* 0xff */
206 };
207
208 /* type field */
209 #define DDF_HEADER_ANCHOR 0x00
210 #define DDF_HEADER_PRIMARY 0x01
211 #define DDF_HEADER_SECONDARY 0x02
212
213 /* The content of the 'controller section' - global scope */
214 struct ddf_controller_data {
215 be32 magic; /* DDF_CONTROLLER_MAGIC */
216 be32 crc;
217 char guid[DDF_GUID_LEN];
218 struct controller_type {
219 be16 vendor_id;
220 be16 device_id;
221 be16 sub_vendor_id;
222 be16 sub_device_id;
223 } type;
224 char product_id[16];
225 __u8 pad[8]; /* 0xff */
226 __u8 vendor_data[448];
227 };
228
229 /* The content of phys_section - global scope */
230 struct phys_disk {
231 be32 magic; /* DDF_PHYS_RECORDS_MAGIC */
232 be32 crc;
233 be16 used_pdes; /* This is a counter, not a max - the list
234 * of used entries may not be dense */
235 be16 max_pdes;
236 __u8 pad[52];
237 struct phys_disk_entry {
238 char guid[DDF_GUID_LEN];
239 be32 refnum;
240 be16 type;
241 be16 state;
242 be64 config_size; /* DDF structures must be after here */
243 char path[18]; /* Another horrible structure really
244 * but is "used for information
245 * purposes only" */
246 __u8 pad[6];
247 } entries[0];
248 };
249
250 /* phys_disk_entry.type is a bitmap - bigendian remember */
251 #define DDF_Forced_PD_GUID 1
252 #define DDF_Active_in_VD 2
253 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
254 #define DDF_Spare 8 /* overrides Global_spare */
255 #define DDF_Foreign 16
256 #define DDF_Legacy 32 /* no DDF on this device */
257
258 #define DDF_Interface_mask 0xf00
259 #define DDF_Interface_SCSI 0x100
260 #define DDF_Interface_SAS 0x200
261 #define DDF_Interface_SATA 0x300
262 #define DDF_Interface_FC 0x400
263
264 /* phys_disk_entry.state is a bigendian bitmap */
265 #define DDF_Online 1
266 #define DDF_Failed 2 /* overrides 1,4,8 */
267 #define DDF_Rebuilding 4
268 #define DDF_Transition 8
269 #define DDF_SMART 16
270 #define DDF_ReadErrors 32
271 #define DDF_Missing 64
272
273 /* The content of the virt_section global scope */
274 struct virtual_disk {
275 be32 magic; /* DDF_VIRT_RECORDS_MAGIC */
276 be32 crc;
277 be16 populated_vdes;
278 be16 max_vdes;
279 __u8 pad[52];
280 struct virtual_entry {
281 char guid[DDF_GUID_LEN];
282 be16 unit;
283 __u16 pad0; /* 0xffff */
284 be16 guid_crc;
285 be16 type;
286 __u8 state;
287 __u8 init_state;
288 __u8 pad1[14];
289 char name[16];
290 } entries[0];
291 };
292
293 /* virtual_entry.type is a bitmap - bigendian */
294 #define DDF_Shared 1
295 #define DDF_Enforce_Groups 2
296 #define DDF_Unicode 4
297 #define DDF_Owner_Valid 8
298
299 /* virtual_entry.state is a bigendian bitmap */
300 #define DDF_state_mask 0x7
301 #define DDF_state_optimal 0x0
302 #define DDF_state_degraded 0x1
303 #define DDF_state_deleted 0x2
304 #define DDF_state_missing 0x3
305 #define DDF_state_failed 0x4
306 #define DDF_state_part_optimal 0x5
307
308 #define DDF_state_morphing 0x8
309 #define DDF_state_inconsistent 0x10
310
311 /* virtual_entry.init_state is a bigendian bitmap */
312 #define DDF_initstate_mask 0x03
313 #define DDF_init_not 0x00
314 #define DDF_init_quick 0x01 /* initialisation is progress.
315 * i.e. 'state_inconsistent' */
316 #define DDF_init_full 0x02
317
318 #define DDF_access_mask 0xc0
319 #define DDF_access_rw 0x00
320 #define DDF_access_ro 0x80
321 #define DDF_access_blocked 0xc0
322
323 /* The content of the config_section - local scope
324 * It has multiple records each config_record_len sectors
325 * They can be vd_config or spare_assign
326 */
327
328 struct vd_config {
329 be32 magic; /* DDF_VD_CONF_MAGIC */
330 be32 crc;
331 char guid[DDF_GUID_LEN];
332 be32 timestamp;
333 be32 seqnum;
334 __u8 pad0[24];
335 be16 prim_elmnt_count;
336 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
337 __u8 prl;
338 __u8 rlq;
339 __u8 sec_elmnt_count;
340 __u8 sec_elmnt_seq;
341 __u8 srl;
342 be64 blocks; /* blocks per component could be different
343 * on different component devices...(only
344 * for concat I hope) */
345 be64 array_blocks; /* blocks in array */
346 __u8 pad1[8];
347 be32 spare_refs[8]; /* This is used to detect missing spares.
348 * As we don't have an interface for that
349 * the values are ignored.
350 */
351 __u8 cache_pol[8];
352 __u8 bg_rate;
353 __u8 pad2[3];
354 __u8 pad3[52];
355 __u8 pad4[192];
356 __u8 v0[32]; /* reserved- 0xff */
357 __u8 v1[32]; /* reserved- 0xff */
358 __u8 v2[16]; /* reserved- 0xff */
359 __u8 v3[16]; /* reserved- 0xff */
360 __u8 vendor[32];
361 be32 phys_refnum[0]; /* refnum of each disk in sequence */
362 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
363 bvd are always the same size */
364 };
365 #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe])
366
367 /* vd_config.cache_pol[7] is a bitmap */
368 #define DDF_cache_writeback 1 /* else writethrough */
369 #define DDF_cache_wadaptive 2 /* only applies if writeback */
370 #define DDF_cache_readahead 4
371 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
372 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
373 #define DDF_cache_wallowed 32 /* enable write caching */
374 #define DDF_cache_rallowed 64 /* enable read caching */
375
376 struct spare_assign {
377 be32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
378 be32 crc;
379 be32 timestamp;
380 __u8 reserved[7];
381 __u8 type;
382 be16 populated; /* SAEs used */
383 be16 max; /* max SAEs */
384 __u8 pad[8];
385 struct spare_assign_entry {
386 char guid[DDF_GUID_LEN];
387 be16 secondary_element;
388 __u8 pad[6];
389 } spare_ents[0];
390 };
391 /* spare_assign.type is a bitmap */
392 #define DDF_spare_dedicated 0x1 /* else global */
393 #define DDF_spare_revertible 0x2 /* else committable */
394 #define DDF_spare_active 0x4 /* else not active */
395 #define DDF_spare_affinity 0x8 /* enclosure affinity */
396
397 /* The data_section contents - local scope */
398 struct disk_data {
399 be32 magic; /* DDF_PHYS_DATA_MAGIC */
400 be32 crc;
401 char guid[DDF_GUID_LEN];
402 be32 refnum; /* crc of some magic drive data ... */
403 __u8 forced_ref; /* set when above was not result of magic */
404 __u8 forced_guid; /* set if guid was forced rather than magic */
405 __u8 vendor[32];
406 __u8 pad[442];
407 };
408
409 /* bbm_section content */
410 struct bad_block_log {
411 be32 magic;
412 be32 crc;
413 be16 entry_count;
414 be32 spare_count;
415 __u8 pad[10];
416 be64 first_spare;
417 struct mapped_block {
418 be64 defective_start;
419 be32 replacement_start;
420 be16 remap_count;
421 __u8 pad[2];
422 } entries[0];
423 };
424
425 /* Struct for internally holding ddf structures */
426 /* The DDF structure stored on each device is potentially
427 * quite different, as some data is global and some is local.
428 * The global data is:
429 * - ddf header
430 * - controller_data
431 * - Physical disk records
432 * - Virtual disk records
433 * The local data is:
434 * - Configuration records
435 * - Physical Disk data section
436 * ( and Bad block and vendor which I don't care about yet).
437 *
438 * The local data is parsed into separate lists as it is read
439 * and reconstructed for writing. This means that we only need
440 * to make config changes once and they are automatically
441 * propagated to all devices.
442 * The global (config and disk data) records are each in a list
443 * of separate data structures. When writing we find the entry
444 * or entries applicable to the particular device.
445 */
446 struct ddf_super {
447 struct ddf_header anchor, primary, secondary;
448 struct ddf_controller_data controller;
449 struct ddf_header *active;
450 struct phys_disk *phys;
451 struct virtual_disk *virt;
452 char *conf;
453 int pdsize, vdsize;
454 unsigned int max_part, mppe, conf_rec_len;
455 int currentdev;
456 int updates_pending;
457 struct vcl {
458 union {
459 char space[512];
460 struct {
461 struct vcl *next;
462 unsigned int vcnum; /* index into ->virt */
463 /* For an array with a secondary level there are
464 * multiple vd_config structures, all with the same
465 * guid but with different sec_elmnt_seq.
466 * One of these structures is in 'conf' below.
467 * The others are in other_bvds, not in any
468 * particular order.
469 */
470 struct vd_config **other_bvds;
471 __u64 *block_sizes; /* NULL if all the same */
472 };
473 };
474 struct vd_config conf;
475 } *conflist, *currentconf;
476 struct dl {
477 union {
478 char space[512];
479 struct {
480 struct dl *next;
481 int major, minor;
482 char *devname;
483 int fd;
484 unsigned long long size; /* sectors */
485 be64 primary_lba; /* sectors */
486 be64 secondary_lba; /* sectors */
487 be64 workspace_lba; /* sectors */
488 int pdnum; /* index in ->phys */
489 struct spare_assign *spare;
490 void *mdupdate; /* hold metadata update */
491
492 /* These fields used by auto-layout */
493 int raiddisk; /* slot to fill in autolayout */
494 __u64 esize;
495 int displayed;
496 };
497 };
498 struct disk_data disk;
499 struct vcl *vlist[0]; /* max_part in size */
500 } *dlist, *add_list;
501 };
502
503 #ifndef MDASSEMBLE
504 static int load_super_ddf_all(struct supertype *st, int fd,
505 void **sbp, char *devname);
506 static int get_svd_state(const struct ddf_super *, const struct vcl *);
507 static int
508 validate_geometry_ddf_container(struct supertype *st,
509 int level, int layout, int raiddisks,
510 int chunk, unsigned long long size,
511 unsigned long long data_offset,
512 char *dev, unsigned long long *freesize,
513 int verbose);
514
515 static int validate_geometry_ddf_bvd(struct supertype *st,
516 int level, int layout, int raiddisks,
517 int *chunk, unsigned long long size,
518 unsigned long long data_offset,
519 char *dev, unsigned long long *freesize,
520 int verbose);
521 #endif
522
523 static void free_super_ddf(struct supertype *st);
524 static int all_ff(const char *guid);
525 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
526 be32 refnum, unsigned int nmax,
527 const struct vd_config **bvd,
528 unsigned int *idx);
529 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map);
530 static void uuid_from_ddf_guid(const char *guid, int uuid[4]);
531 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
532 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i);
533 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map);
534 static int init_super_ddf_bvd(struct supertype *st,
535 mdu_array_info_t *info,
536 unsigned long long size,
537 char *name, char *homehost,
538 int *uuid, unsigned long long data_offset);
539
540 #if DEBUG
541 static void pr_state(struct ddf_super *ddf, const char *msg)
542 {
543 unsigned int i;
544 dprintf("%s: ", msg);
545 for (i = 0; i < be16_to_cpu(ddf->active->max_vd_entries); i++) {
546 if (all_ff(ddf->virt->entries[i].guid))
547 continue;
548 dprintf_cont("%u(s=%02x i=%02x) ", i,
549 ddf->virt->entries[i].state,
550 ddf->virt->entries[i].init_state);
551 }
552 dprintf_cont("\n");
553 }
554 #else
555 static void pr_state(const struct ddf_super *ddf, const char *msg) {}
556 #endif
557
558 static void _ddf_set_updates_pending(struct ddf_super *ddf, struct vd_config *vc,
559 const char *func)
560 {
561 if (vc) {
562 vc->timestamp = cpu_to_be32(time(0)-DECADE);
563 vc->seqnum = cpu_to_be32(be32_to_cpu(vc->seqnum) + 1);
564 }
565 if (ddf->updates_pending)
566 return;
567 ddf->updates_pending = 1;
568 ddf->active->seq = cpu_to_be32((be32_to_cpu(ddf->active->seq)+1));
569 pr_state(ddf, func);
570 }
571
572 #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__)
573
574 static be32 calc_crc(void *buf, int len)
575 {
576 /* crcs are always at the same place as in the ddf_header */
577 struct ddf_header *ddf = buf;
578 be32 oldcrc = ddf->crc;
579 __u32 newcrc;
580 ddf->crc = cpu_to_be32(0xffffffff);
581
582 newcrc = crc32(0, buf, len);
583 ddf->crc = oldcrc;
584 /* The crc is stored (like everything) bigendian, so convert
585 * here for simplicity
586 */
587 return cpu_to_be32(newcrc);
588 }
589
590 #define DDF_INVALID_LEVEL 0xff
591 #define DDF_NO_SECONDARY 0xff
592 static int err_bad_md_layout(const mdu_array_info_t *array)
593 {
594 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
595 array->level, array->layout, array->raid_disks);
596 return -1;
597 }
598
599 static int layout_md2ddf(const mdu_array_info_t *array,
600 struct vd_config *conf)
601 {
602 be16 prim_elmnt_count = cpu_to_be16(array->raid_disks);
603 __u8 prl = DDF_INVALID_LEVEL, rlq = 0;
604 __u8 sec_elmnt_count = 1;
605 __u8 srl = DDF_NO_SECONDARY;
606
607 switch (array->level) {
608 case LEVEL_LINEAR:
609 prl = DDF_CONCAT;
610 break;
611 case 0:
612 rlq = DDF_RAID0_SIMPLE;
613 prl = DDF_RAID0;
614 break;
615 case 1:
616 switch (array->raid_disks) {
617 case 2:
618 rlq = DDF_RAID1_SIMPLE;
619 break;
620 case 3:
621 rlq = DDF_RAID1_MULTI;
622 break;
623 default:
624 return err_bad_md_layout(array);
625 }
626 prl = DDF_RAID1;
627 break;
628 case 4:
629 if (array->layout != 0)
630 return err_bad_md_layout(array);
631 rlq = DDF_RAID4_N;
632 prl = DDF_RAID4;
633 break;
634 case 5:
635 switch (array->layout) {
636 case ALGORITHM_LEFT_ASYMMETRIC:
637 rlq = DDF_RAID5_N_RESTART;
638 break;
639 case ALGORITHM_RIGHT_ASYMMETRIC:
640 rlq = DDF_RAID5_0_RESTART;
641 break;
642 case ALGORITHM_LEFT_SYMMETRIC:
643 rlq = DDF_RAID5_N_CONTINUE;
644 break;
645 case ALGORITHM_RIGHT_SYMMETRIC:
646 /* not mentioned in standard */
647 default:
648 return err_bad_md_layout(array);
649 }
650 prl = DDF_RAID5;
651 break;
652 case 6:
653 switch (array->layout) {
654 case ALGORITHM_ROTATING_N_RESTART:
655 rlq = DDF_RAID5_N_RESTART;
656 break;
657 case ALGORITHM_ROTATING_ZERO_RESTART:
658 rlq = DDF_RAID6_0_RESTART;
659 break;
660 case ALGORITHM_ROTATING_N_CONTINUE:
661 rlq = DDF_RAID5_N_CONTINUE;
662 break;
663 default:
664 return err_bad_md_layout(array);
665 }
666 prl = DDF_RAID6;
667 break;
668 case 10:
669 if (array->raid_disks % 2 == 0 && array->layout == 0x102) {
670 rlq = DDF_RAID1_SIMPLE;
671 prim_elmnt_count = cpu_to_be16(2);
672 sec_elmnt_count = array->raid_disks / 2;
673 srl = DDF_2SPANNED;
674 prl = DDF_RAID1;
675 } else if (array->raid_disks % 3 == 0
676 && array->layout == 0x103) {
677 rlq = DDF_RAID1_MULTI;
678 prim_elmnt_count = cpu_to_be16(3);
679 sec_elmnt_count = array->raid_disks / 3;
680 srl = DDF_2SPANNED;
681 prl = DDF_RAID1;
682 } else if (array->layout == 0x201) {
683 prl = DDF_RAID1E;
684 rlq = DDF_RAID1E_OFFSET;
685 } else if (array->layout == 0x102) {
686 prl = DDF_RAID1E;
687 rlq = DDF_RAID1E_ADJACENT;
688 } else
689 return err_bad_md_layout(array);
690 break;
691 default:
692 return err_bad_md_layout(array);
693 }
694 conf->prl = prl;
695 conf->prim_elmnt_count = prim_elmnt_count;
696 conf->rlq = rlq;
697 conf->srl = srl;
698 conf->sec_elmnt_count = sec_elmnt_count;
699 return 0;
700 }
701
702 static int err_bad_ddf_layout(const struct vd_config *conf)
703 {
704 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
705 conf->prl, conf->rlq, be16_to_cpu(conf->prim_elmnt_count));
706 return -1;
707 }
708
709 static int layout_ddf2md(const struct vd_config *conf,
710 mdu_array_info_t *array)
711 {
712 int level = LEVEL_UNSUPPORTED;
713 int layout = 0;
714 int raiddisks = be16_to_cpu(conf->prim_elmnt_count);
715
716 if (conf->sec_elmnt_count > 1) {
717 /* see also check_secondary() */
718 if (conf->prl != DDF_RAID1 ||
719 (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED)) {
720 pr_err("Unsupported secondary RAID level %u/%u\n",
721 conf->prl, conf->srl);
722 return -1;
723 }
724 if (raiddisks == 2 && conf->rlq == DDF_RAID1_SIMPLE)
725 layout = 0x102;
726 else if (raiddisks == 3 && conf->rlq == DDF_RAID1_MULTI)
727 layout = 0x103;
728 else
729 return err_bad_ddf_layout(conf);
730 raiddisks *= conf->sec_elmnt_count;
731 level = 10;
732 goto good;
733 }
734
735 switch (conf->prl) {
736 case DDF_CONCAT:
737 level = LEVEL_LINEAR;
738 break;
739 case DDF_RAID0:
740 if (conf->rlq != DDF_RAID0_SIMPLE)
741 return err_bad_ddf_layout(conf);
742 level = 0;
743 break;
744 case DDF_RAID1:
745 if (!((conf->rlq == DDF_RAID1_SIMPLE && raiddisks == 2) ||
746 (conf->rlq == DDF_RAID1_MULTI && raiddisks == 3)))
747 return err_bad_ddf_layout(conf);
748 level = 1;
749 break;
750 case DDF_RAID1E:
751 if (conf->rlq == DDF_RAID1E_ADJACENT)
752 layout = 0x102;
753 else if (conf->rlq == DDF_RAID1E_OFFSET)
754 layout = 0x201;
755 else
756 return err_bad_ddf_layout(conf);
757 level = 10;
758 break;
759 case DDF_RAID4:
760 if (conf->rlq != DDF_RAID4_N)
761 return err_bad_ddf_layout(conf);
762 level = 4;
763 break;
764 case DDF_RAID5:
765 switch (conf->rlq) {
766 case DDF_RAID5_N_RESTART:
767 layout = ALGORITHM_LEFT_ASYMMETRIC;
768 break;
769 case DDF_RAID5_0_RESTART:
770 layout = ALGORITHM_RIGHT_ASYMMETRIC;
771 break;
772 case DDF_RAID5_N_CONTINUE:
773 layout = ALGORITHM_LEFT_SYMMETRIC;
774 break;
775 default:
776 return err_bad_ddf_layout(conf);
777 }
778 level = 5;
779 break;
780 case DDF_RAID6:
781 switch (conf->rlq) {
782 case DDF_RAID5_N_RESTART:
783 layout = ALGORITHM_ROTATING_N_RESTART;
784 break;
785 case DDF_RAID6_0_RESTART:
786 layout = ALGORITHM_ROTATING_ZERO_RESTART;
787 break;
788 case DDF_RAID5_N_CONTINUE:
789 layout = ALGORITHM_ROTATING_N_CONTINUE;
790 break;
791 default:
792 return err_bad_ddf_layout(conf);
793 }
794 level = 6;
795 break;
796 default:
797 return err_bad_ddf_layout(conf);
798 };
799
800 good:
801 array->level = level;
802 array->layout = layout;
803 array->raid_disks = raiddisks;
804 return 0;
805 }
806
807 static int load_ddf_header(int fd, unsigned long long lba,
808 unsigned long long size,
809 int type,
810 struct ddf_header *hdr, struct ddf_header *anchor)
811 {
812 /* read a ddf header (primary or secondary) from fd/lba
813 * and check that it is consistent with anchor
814 * Need to check:
815 * magic, crc, guid, rev, and LBA's header_type, and
816 * everything after header_type must be the same
817 */
818 if (lba >= size-1)
819 return 0;
820
821 if (lseek64(fd, lba<<9, 0) < 0)
822 return 0;
823
824 if (read(fd, hdr, 512) != 512)
825 return 0;
826
827 if (!be32_eq(hdr->magic, DDF_HEADER_MAGIC)) {
828 pr_err("bad header magic\n");
829 return 0;
830 }
831 if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) {
832 pr_err("bad CRC\n");
833 return 0;
834 }
835 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
836 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
837 !be64_eq(anchor->primary_lba, hdr->primary_lba) ||
838 !be64_eq(anchor->secondary_lba, hdr->secondary_lba) ||
839 hdr->type != type ||
840 memcmp(anchor->pad2, hdr->pad2, 512 -
841 offsetof(struct ddf_header, pad2)) != 0) {
842 pr_err("header mismatch\n");
843 return 0;
844 }
845
846 /* Looks good enough to me... */
847 return 1;
848 }
849
850 static void *load_section(int fd, struct ddf_super *super, void *buf,
851 be32 offset_be, be32 len_be, int check)
852 {
853 unsigned long long offset = be32_to_cpu(offset_be);
854 unsigned long long len = be32_to_cpu(len_be);
855 int dofree = (buf == NULL);
856
857 if (check)
858 if (len != 2 && len != 8 && len != 32
859 && len != 128 && len != 512)
860 return NULL;
861
862 if (len > 1024)
863 return NULL;
864 if (!buf && posix_memalign(&buf, 512, len<<9) != 0)
865 buf = NULL;
866
867 if (!buf)
868 return NULL;
869
870 if (super->active->type == 1)
871 offset += be64_to_cpu(super->active->primary_lba);
872 else
873 offset += be64_to_cpu(super->active->secondary_lba);
874
875 if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
876 if (dofree)
877 free(buf);
878 return NULL;
879 }
880 if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
881 if (dofree)
882 free(buf);
883 return NULL;
884 }
885 return buf;
886 }
887
888 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
889 {
890 unsigned long long dsize;
891
892 get_dev_size(fd, NULL, &dsize);
893
894 if (lseek64(fd, dsize-512, 0) < 0) {
895 if (devname)
896 pr_err("Cannot seek to anchor block on %s: %s\n",
897 devname, strerror(errno));
898 return 1;
899 }
900 if (read(fd, &super->anchor, 512) != 512) {
901 if (devname)
902 pr_err("Cannot read anchor block on %s: %s\n",
903 devname, strerror(errno));
904 return 1;
905 }
906 if (!be32_eq(super->anchor.magic, DDF_HEADER_MAGIC)) {
907 if (devname)
908 pr_err("no DDF anchor found on %s\n",
909 devname);
910 return 2;
911 }
912 if (!be32_eq(calc_crc(&super->anchor, 512), super->anchor.crc)) {
913 if (devname)
914 pr_err("bad CRC on anchor on %s\n",
915 devname);
916 return 2;
917 }
918 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
919 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
920 if (devname)
921 pr_err("can only support super revision"
922 " %.8s and earlier, not %.8s on %s\n",
923 DDF_REVISION_2, super->anchor.revision,devname);
924 return 2;
925 }
926 super->active = NULL;
927 if (load_ddf_header(fd, be64_to_cpu(super->anchor.primary_lba),
928 dsize >> 9, 1,
929 &super->primary, &super->anchor) == 0) {
930 if (devname)
931 pr_err("Failed to load primary DDF header "
932 "on %s\n", devname);
933 } else
934 super->active = &super->primary;
935
936 if (load_ddf_header(fd, be64_to_cpu(super->anchor.secondary_lba),
937 dsize >> 9, 2,
938 &super->secondary, &super->anchor)) {
939 if (super->active == NULL
940 || (be32_to_cpu(super->primary.seq)
941 < be32_to_cpu(super->secondary.seq) &&
942 !super->secondary.openflag)
943 || (be32_to_cpu(super->primary.seq)
944 == be32_to_cpu(super->secondary.seq) &&
945 super->primary.openflag && !super->secondary.openflag)
946 )
947 super->active = &super->secondary;
948 } else if (devname &&
949 be64_to_cpu(super->anchor.secondary_lba) != ~(__u64)0)
950 pr_err("Failed to load secondary DDF header on %s\n",
951 devname);
952 if (super->active == NULL)
953 return 2;
954 return 0;
955 }
956
957 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
958 {
959 void *ok;
960 ok = load_section(fd, super, &super->controller,
961 super->active->controller_section_offset,
962 super->active->controller_section_length,
963 0);
964 super->phys = load_section(fd, super, NULL,
965 super->active->phys_section_offset,
966 super->active->phys_section_length,
967 1);
968 super->pdsize = be32_to_cpu(super->active->phys_section_length) * 512;
969
970 super->virt = load_section(fd, super, NULL,
971 super->active->virt_section_offset,
972 super->active->virt_section_length,
973 1);
974 super->vdsize = be32_to_cpu(super->active->virt_section_length) * 512;
975 if (!ok ||
976 !super->phys ||
977 !super->virt) {
978 free(super->phys);
979 free(super->virt);
980 super->phys = NULL;
981 super->virt = NULL;
982 return 2;
983 }
984 super->conflist = NULL;
985 super->dlist = NULL;
986
987 super->max_part = be16_to_cpu(super->active->max_partitions);
988 super->mppe = be16_to_cpu(super->active->max_primary_element_entries);
989 super->conf_rec_len = be16_to_cpu(super->active->config_record_len);
990 return 0;
991 }
992
993 #define DDF_UNUSED_BVD 0xff
994 static int alloc_other_bvds(const struct ddf_super *ddf, struct vcl *vcl)
995 {
996 unsigned int n_vds = vcl->conf.sec_elmnt_count - 1;
997 unsigned int i, vdsize;
998 void *p;
999 if (n_vds == 0) {
1000 vcl->other_bvds = NULL;
1001 return 0;
1002 }
1003 vdsize = ddf->conf_rec_len * 512;
1004 if (posix_memalign(&p, 512, n_vds *
1005 (vdsize + sizeof(struct vd_config *))) != 0)
1006 return -1;
1007 vcl->other_bvds = (struct vd_config **) (p + n_vds * vdsize);
1008 for (i = 0; i < n_vds; i++) {
1009 vcl->other_bvds[i] = p + i * vdsize;
1010 memset(vcl->other_bvds[i], 0, vdsize);
1011 vcl->other_bvds[i]->sec_elmnt_seq = DDF_UNUSED_BVD;
1012 }
1013 return 0;
1014 }
1015
1016 static void add_other_bvd(struct vcl *vcl, struct vd_config *vd,
1017 unsigned int len)
1018 {
1019 int i;
1020 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1021 if (vcl->other_bvds[i]->sec_elmnt_seq == vd->sec_elmnt_seq)
1022 break;
1023
1024 if (i < vcl->conf.sec_elmnt_count-1) {
1025 if (be32_to_cpu(vd->seqnum) <=
1026 be32_to_cpu(vcl->other_bvds[i]->seqnum))
1027 return;
1028 } else {
1029 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1030 if (vcl->other_bvds[i]->sec_elmnt_seq == DDF_UNUSED_BVD)
1031 break;
1032 if (i == vcl->conf.sec_elmnt_count-1) {
1033 pr_err("no space for sec level config %u, count is %u\n",
1034 vd->sec_elmnt_seq, vcl->conf.sec_elmnt_count);
1035 return;
1036 }
1037 }
1038 memcpy(vcl->other_bvds[i], vd, len);
1039 }
1040
1041 static int load_ddf_local(int fd, struct ddf_super *super,
1042 char *devname, int keep)
1043 {
1044 struct dl *dl;
1045 struct stat stb;
1046 char *conf;
1047 unsigned int i;
1048 unsigned int confsec;
1049 int vnum;
1050 unsigned int max_virt_disks =
1051 be16_to_cpu(super->active->max_vd_entries);
1052 unsigned long long dsize;
1053
1054 /* First the local disk info */
1055 if (posix_memalign((void**)&dl, 512,
1056 sizeof(*dl) +
1057 (super->max_part) * sizeof(dl->vlist[0])) != 0) {
1058 pr_err("could not allocate disk info buffer\n");
1059 return 1;
1060 }
1061
1062 load_section(fd, super, &dl->disk,
1063 super->active->data_section_offset,
1064 super->active->data_section_length,
1065 0);
1066 dl->devname = devname ? xstrdup(devname) : NULL;
1067
1068 fstat(fd, &stb);
1069 dl->major = major(stb.st_rdev);
1070 dl->minor = minor(stb.st_rdev);
1071 dl->next = super->dlist;
1072 dl->fd = keep ? fd : -1;
1073
1074 dl->size = 0;
1075 if (get_dev_size(fd, devname, &dsize))
1076 dl->size = dsize >> 9;
1077 /* If the disks have different sizes, the LBAs will differ
1078 * between phys disks.
1079 * At this point here, the values in super->active must be valid
1080 * for this phys disk. */
1081 dl->primary_lba = super->active->primary_lba;
1082 dl->secondary_lba = super->active->secondary_lba;
1083 dl->workspace_lba = super->active->workspace_lba;
1084 dl->spare = NULL;
1085 for (i = 0 ; i < super->max_part ; i++)
1086 dl->vlist[i] = NULL;
1087 super->dlist = dl;
1088 dl->pdnum = -1;
1089 for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++)
1090 if (memcmp(super->phys->entries[i].guid,
1091 dl->disk.guid, DDF_GUID_LEN) == 0)
1092 dl->pdnum = i;
1093
1094 /* Now the config list. */
1095 /* 'conf' is an array of config entries, some of which are
1096 * probably invalid. Those which are good need to be copied into
1097 * the conflist
1098 */
1099
1100 conf = load_section(fd, super, super->conf,
1101 super->active->config_section_offset,
1102 super->active->config_section_length,
1103 0);
1104 super->conf = conf;
1105 vnum = 0;
1106 for (confsec = 0;
1107 confsec < be32_to_cpu(super->active->config_section_length);
1108 confsec += super->conf_rec_len) {
1109 struct vd_config *vd =
1110 (struct vd_config *)((char*)conf + confsec*512);
1111 struct vcl *vcl;
1112
1113 if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) {
1114 if (dl->spare)
1115 continue;
1116 if (posix_memalign((void**)&dl->spare, 512,
1117 super->conf_rec_len*512) != 0) {
1118 pr_err("could not allocate spare info buf\n");
1119 return 1;
1120 }
1121
1122 memcpy(dl->spare, vd, super->conf_rec_len*512);
1123 continue;
1124 }
1125 if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC))
1126 /* Must be vendor-unique - I cannot handle those */
1127 continue;
1128
1129 for (vcl = super->conflist; vcl; vcl = vcl->next) {
1130 if (memcmp(vcl->conf.guid,
1131 vd->guid, DDF_GUID_LEN) == 0)
1132 break;
1133 }
1134
1135 if (vcl) {
1136 dl->vlist[vnum++] = vcl;
1137 if (vcl->other_bvds != NULL &&
1138 vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) {
1139 add_other_bvd(vcl, vd, super->conf_rec_len*512);
1140 continue;
1141 }
1142 if (be32_to_cpu(vd->seqnum) <=
1143 be32_to_cpu(vcl->conf.seqnum))
1144 continue;
1145 } else {
1146 if (posix_memalign((void**)&vcl, 512,
1147 (super->conf_rec_len*512 +
1148 offsetof(struct vcl, conf))) != 0) {
1149 pr_err("could not allocate vcl buf\n");
1150 return 1;
1151 }
1152 vcl->next = super->conflist;
1153 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1154 vcl->conf.sec_elmnt_count = vd->sec_elmnt_count;
1155 if (alloc_other_bvds(super, vcl) != 0) {
1156 pr_err("could not allocate other bvds\n");
1157 free(vcl);
1158 return 1;
1159 };
1160 super->conflist = vcl;
1161 dl->vlist[vnum++] = vcl;
1162 }
1163 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
1164 for (i=0; i < max_virt_disks ; i++)
1165 if (memcmp(super->virt->entries[i].guid,
1166 vcl->conf.guid, DDF_GUID_LEN)==0)
1167 break;
1168 if (i < max_virt_disks)
1169 vcl->vcnum = i;
1170 }
1171
1172 return 0;
1173 }
1174
1175 static int load_super_ddf(struct supertype *st, int fd,
1176 char *devname)
1177 {
1178 unsigned long long dsize;
1179 struct ddf_super *super;
1180 int rv;
1181
1182 if (get_dev_size(fd, devname, &dsize) == 0)
1183 return 1;
1184
1185 if (test_partition(fd))
1186 /* DDF is not allowed on partitions */
1187 return 1;
1188
1189 /* 32M is a lower bound */
1190 if (dsize <= 32*1024*1024) {
1191 if (devname)
1192 pr_err("%s is too small for ddf: "
1193 "size is %llu sectors.\n",
1194 devname, dsize>>9);
1195 return 1;
1196 }
1197 if (dsize & 511) {
1198 if (devname)
1199 pr_err("%s is an odd size for ddf: "
1200 "size is %llu bytes.\n",
1201 devname, dsize);
1202 return 1;
1203 }
1204
1205 free_super_ddf(st);
1206
1207 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
1208 pr_err("malloc of %zu failed.\n",
1209 sizeof(*super));
1210 return 1;
1211 }
1212 memset(super, 0, sizeof(*super));
1213
1214 rv = load_ddf_headers(fd, super, devname);
1215 if (rv) {
1216 free(super);
1217 return rv;
1218 }
1219
1220 /* Have valid headers and have chosen the best. Let's read in the rest*/
1221
1222 rv = load_ddf_global(fd, super, devname);
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 rv = load_ddf_local(fd, super, devname, 0);
1233
1234 if (rv) {
1235 if (devname)
1236 pr_err("Failed to load all information "
1237 "sections on %s\n", devname);
1238 free(super);
1239 return rv;
1240 }
1241
1242 /* Should possibly check the sections .... */
1243
1244 st->sb = super;
1245 if (st->ss == NULL) {
1246 st->ss = &super_ddf;
1247 st->minor_version = 0;
1248 st->max_devs = 512;
1249 }
1250 return 0;
1251
1252 }
1253
1254 static void free_super_ddf(struct supertype *st)
1255 {
1256 struct ddf_super *ddf = st->sb;
1257 if (ddf == NULL)
1258 return;
1259 free(ddf->phys);
1260 free(ddf->virt);
1261 free(ddf->conf);
1262 while (ddf->conflist) {
1263 struct vcl *v = ddf->conflist;
1264 ddf->conflist = v->next;
1265 if (v->block_sizes)
1266 free(v->block_sizes);
1267 if (v->other_bvds)
1268 /*
1269 v->other_bvds[0] points to beginning of buffer,
1270 see alloc_other_bvds()
1271 */
1272 free(v->other_bvds[0]);
1273 free(v);
1274 }
1275 while (ddf->dlist) {
1276 struct dl *d = ddf->dlist;
1277 ddf->dlist = d->next;
1278 if (d->fd >= 0)
1279 close(d->fd);
1280 if (d->spare)
1281 free(d->spare);
1282 free(d);
1283 }
1284 while (ddf->add_list) {
1285 struct dl *d = ddf->add_list;
1286 ddf->add_list = d->next;
1287 if (d->fd >= 0)
1288 close(d->fd);
1289 if (d->spare)
1290 free(d->spare);
1291 free(d);
1292 }
1293 free(ddf);
1294 st->sb = NULL;
1295 }
1296
1297 static struct supertype *match_metadata_desc_ddf(char *arg)
1298 {
1299 /* 'ddf' only supports containers */
1300 struct supertype *st;
1301 if (strcmp(arg, "ddf") != 0 &&
1302 strcmp(arg, "default") != 0
1303 )
1304 return NULL;
1305
1306 st = xcalloc(1, sizeof(*st));
1307 st->ss = &super_ddf;
1308 st->max_devs = 512;
1309 st->minor_version = 0;
1310 st->sb = NULL;
1311 return st;
1312 }
1313
1314 #ifndef MDASSEMBLE
1315
1316 static mapping_t ddf_state[] = {
1317 { "Optimal", 0},
1318 { "Degraded", 1},
1319 { "Deleted", 2},
1320 { "Missing", 3},
1321 { "Failed", 4},
1322 { "Partially Optimal", 5},
1323 { "-reserved-", 6},
1324 { "-reserved-", 7},
1325 { NULL, 0}
1326 };
1327
1328 static mapping_t ddf_init_state[] = {
1329 { "Not Initialised", 0},
1330 { "QuickInit in Progress", 1},
1331 { "Fully Initialised", 2},
1332 { "*UNKNOWN*", 3},
1333 { NULL, 0}
1334 };
1335 static mapping_t ddf_access[] = {
1336 { "Read/Write", 0},
1337 { "Reserved", 1},
1338 { "Read Only", 2},
1339 { "Blocked (no access)", 3},
1340 { NULL ,0}
1341 };
1342
1343 static mapping_t ddf_level[] = {
1344 { "RAID0", DDF_RAID0},
1345 { "RAID1", DDF_RAID1},
1346 { "RAID3", DDF_RAID3},
1347 { "RAID4", DDF_RAID4},
1348 { "RAID5", DDF_RAID5},
1349 { "RAID1E",DDF_RAID1E},
1350 { "JBOD", DDF_JBOD},
1351 { "CONCAT",DDF_CONCAT},
1352 { "RAID5E",DDF_RAID5E},
1353 { "RAID5EE",DDF_RAID5EE},
1354 { "RAID6", DDF_RAID6},
1355 { NULL, 0}
1356 };
1357 static mapping_t ddf_sec_level[] = {
1358 { "Striped", DDF_2STRIPED},
1359 { "Mirrored", DDF_2MIRRORED},
1360 { "Concat", DDF_2CONCAT},
1361 { "Spanned", DDF_2SPANNED},
1362 { NULL, 0}
1363 };
1364 #endif
1365
1366 static int all_ff(const char *guid)
1367 {
1368 int i;
1369 for (i = 0; i < DDF_GUID_LEN; i++)
1370 if (guid[i] != (char)0xff)
1371 return 0;
1372 return 1;
1373 }
1374
1375 static const char *guid_str(const char *guid)
1376 {
1377 static char buf[DDF_GUID_LEN*2+1];
1378 int i;
1379 char *p = buf;
1380 for (i = 0; i < DDF_GUID_LEN; i++) {
1381 unsigned char c = guid[i];
1382 if (c >= 32 && c < 127)
1383 p += sprintf(p, "%c", c);
1384 else
1385 p += sprintf(p, "%02x", c);
1386 }
1387 *p = '\0';
1388 return (const char *) buf;
1389 }
1390
1391 #ifndef MDASSEMBLE
1392 static void print_guid(char *guid, int tstamp)
1393 {
1394 /* A GUIDs are part (or all) ASCII and part binary.
1395 * They tend to be space padded.
1396 * We print the GUID in HEX, then in parentheses add
1397 * any initial ASCII sequence, and a possible
1398 * time stamp from bytes 16-19
1399 */
1400 int l = DDF_GUID_LEN;
1401 int i;
1402
1403 for (i=0 ; i<DDF_GUID_LEN ; i++) {
1404 if ((i&3)==0 && i != 0) printf(":");
1405 printf("%02X", guid[i]&255);
1406 }
1407
1408 printf("\n (");
1409 while (l && guid[l-1] == ' ')
1410 l--;
1411 for (i=0 ; i<l ; i++) {
1412 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1413 fputc(guid[i], stdout);
1414 else
1415 break;
1416 }
1417 if (tstamp) {
1418 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1419 char tbuf[100];
1420 struct tm *tm;
1421 tm = localtime(&then);
1422 strftime(tbuf, 100, " %D %T",tm);
1423 fputs(tbuf, stdout);
1424 }
1425 printf(")");
1426 }
1427
1428 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1429 {
1430 int crl = sb->conf_rec_len;
1431 struct vcl *vcl;
1432
1433 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1434 unsigned int i;
1435 struct vd_config *vc = &vcl->conf;
1436
1437 if (!be32_eq(calc_crc(vc, crl*512), vc->crc))
1438 continue;
1439 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1440 continue;
1441
1442 /* Ok, we know about this VD, let's give more details */
1443 printf(" Raid Devices[%d] : %d (", n,
1444 be16_to_cpu(vc->prim_elmnt_count));
1445 for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) {
1446 int j;
1447 int cnt = be16_to_cpu(sb->phys->max_pdes);
1448 for (j=0; j<cnt; j++)
1449 if (be32_eq(vc->phys_refnum[i],
1450 sb->phys->entries[j].refnum))
1451 break;
1452 if (i) printf(" ");
1453 if (j < cnt)
1454 printf("%d", j);
1455 else
1456 printf("--");
1457 printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb, vc)[i])/2);
1458 }
1459 printf(")\n");
1460 if (vc->chunk_shift != 255)
1461 printf(" Chunk Size[%d] : %d sectors\n", n,
1462 1 << vc->chunk_shift);
1463 printf(" Raid Level[%d] : %s\n", n,
1464 map_num(ddf_level, vc->prl)?:"-unknown-");
1465 if (vc->sec_elmnt_count != 1) {
1466 printf(" Secondary Position[%d] : %d of %d\n", n,
1467 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1468 printf(" Secondary Level[%d] : %s\n", n,
1469 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1470 }
1471 printf(" Device Size[%d] : %llu\n", n,
1472 be64_to_cpu(vc->blocks)/2);
1473 printf(" Array Size[%d] : %llu\n", n,
1474 be64_to_cpu(vc->array_blocks)/2);
1475 }
1476 }
1477
1478 static void examine_vds(struct ddf_super *sb)
1479 {
1480 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1481 unsigned int i;
1482 printf(" Virtual Disks : %d\n", cnt);
1483
1484 for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) {
1485 struct virtual_entry *ve = &sb->virt->entries[i];
1486 if (all_ff(ve->guid))
1487 continue;
1488 printf("\n");
1489 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1490 printf("\n");
1491 printf(" unit[%d] : %d\n", i, be16_to_cpu(ve->unit));
1492 printf(" state[%d] : %s, %s%s\n", i,
1493 map_num(ddf_state, ve->state & 7),
1494 (ve->state & DDF_state_morphing) ? "Morphing, ": "",
1495 (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent");
1496 printf(" init state[%d] : %s\n", i,
1497 map_num(ddf_init_state, ve->init_state&DDF_initstate_mask));
1498 printf(" access[%d] : %s\n", i,
1499 map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6));
1500 printf(" Name[%d] : %.16s\n", i, ve->name);
1501 examine_vd(i, sb, ve->guid);
1502 }
1503 if (cnt) printf("\n");
1504 }
1505
1506 static void examine_pds(struct ddf_super *sb)
1507 {
1508 int cnt = be16_to_cpu(sb->phys->max_pdes);
1509 int i;
1510 struct dl *dl;
1511 int unlisted = 0;
1512 printf(" Physical Disks : %d\n", cnt);
1513 printf(" Number RefNo Size Device Type/State\n");
1514
1515 for (dl = sb->dlist; dl; dl = dl->next)
1516 dl->displayed = 0;
1517
1518 for (i=0 ; i<cnt ; i++) {
1519 struct phys_disk_entry *pd = &sb->phys->entries[i];
1520 int type = be16_to_cpu(pd->type);
1521 int state = be16_to_cpu(pd->state);
1522
1523 if (be32_to_cpu(pd->refnum) == 0xffffffff)
1524 /* Not in use */
1525 continue;
1526 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1527 //printf("\n");
1528 printf(" %3d %08x ", i,
1529 be32_to_cpu(pd->refnum));
1530 printf("%8lluK ",
1531 be64_to_cpu(pd->config_size)>>1);
1532 for (dl = sb->dlist; dl ; dl = dl->next) {
1533 if (be32_eq(dl->disk.refnum, pd->refnum)) {
1534 char *dv = map_dev(dl->major, dl->minor, 0);
1535 if (dv) {
1536 printf("%-15s", dv);
1537 break;
1538 }
1539 }
1540 }
1541 if (!dl)
1542 printf("%15s","");
1543 else
1544 dl->displayed = 1;
1545 printf(" %s%s%s%s%s",
1546 (type&2) ? "active":"",
1547 (type&4) ? "Global-Spare":"",
1548 (type&8) ? "spare" : "",
1549 (type&16)? ", foreign" : "",
1550 (type&32)? "pass-through" : "");
1551 if (state & DDF_Failed)
1552 /* This over-rides these three */
1553 state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition);
1554 printf("/%s%s%s%s%s%s%s",
1555 (state&1)? "Online": "Offline",
1556 (state&2)? ", Failed": "",
1557 (state&4)? ", Rebuilding": "",
1558 (state&8)? ", in-transition": "",
1559 (state&16)? ", SMART-errors": "",
1560 (state&32)? ", Unrecovered-Read-Errors": "",
1561 (state&64)? ", Missing" : "");
1562 printf("\n");
1563 }
1564 for (dl = sb->dlist; dl; dl = dl->next) {
1565 char *dv;
1566 if (dl->displayed)
1567 continue;
1568 if (!unlisted)
1569 printf(" Physical disks not in metadata!:\n");
1570 unlisted = 1;
1571 dv = map_dev(dl->major, dl->minor, 0);
1572 printf(" %08x %s\n", be32_to_cpu(dl->disk.refnum),
1573 dv ? dv : "-unknown-");
1574 }
1575 if (unlisted)
1576 printf("\n");
1577 }
1578
1579 static void examine_super_ddf(struct supertype *st, char *homehost)
1580 {
1581 struct ddf_super *sb = st->sb;
1582
1583 printf(" Magic : %08x\n", be32_to_cpu(sb->anchor.magic));
1584 printf(" Version : %.8s\n", sb->anchor.revision);
1585 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1586 printf("\n");
1587 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1588 printf("\n");
1589 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1590 printf(" Redundant hdr : %s\n", (be32_eq(sb->secondary.magic,
1591 DDF_HEADER_MAGIC)
1592 ?"yes" : "no"));
1593 examine_vds(sb);
1594 examine_pds(sb);
1595 }
1596
1597 static unsigned int get_vd_num_of_subarray(struct supertype *st)
1598 {
1599 /*
1600 * Figure out the VD number for this supertype.
1601 * Returns DDF_CONTAINER for the container itself,
1602 * and DDF_NOTFOUND on error.
1603 */
1604 struct ddf_super *ddf = st->sb;
1605 struct mdinfo *sra;
1606 char *sub, *end;
1607 unsigned int vcnum;
1608
1609 if (*st->container_devnm == '\0')
1610 return DDF_CONTAINER;
1611
1612 sra = sysfs_read(-1, st->devnm, GET_VERSION);
1613 if (!sra || sra->array.major_version != -1 ||
1614 sra->array.minor_version != -2 ||
1615 !is_subarray(sra->text_version))
1616 return DDF_NOTFOUND;
1617
1618 sub = strchr(sra->text_version + 1, '/');
1619 if (sub != NULL)
1620 vcnum = strtoul(sub + 1, &end, 10);
1621 if (sub == NULL || *sub == '\0' || *end != '\0' ||
1622 vcnum >= be16_to_cpu(ddf->active->max_vd_entries))
1623 return DDF_NOTFOUND;
1624
1625 return vcnum;
1626 }
1627
1628 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1629 {
1630 /* We just write a generic DDF ARRAY entry
1631 */
1632 struct mdinfo info;
1633 char nbuf[64];
1634 getinfo_super_ddf(st, &info, NULL);
1635 fname_from_uuid(st, &info, nbuf, ':');
1636
1637 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1638 }
1639
1640 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1641 {
1642 /* We write a DDF ARRAY member entry for each vd, identifying container
1643 * by uuid and member by unit number and uuid.
1644 */
1645 struct ddf_super *ddf = st->sb;
1646 struct mdinfo info;
1647 unsigned int i;
1648 char nbuf[64];
1649 getinfo_super_ddf(st, &info, NULL);
1650 fname_from_uuid(st, &info, nbuf, ':');
1651
1652 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
1653 struct virtual_entry *ve = &ddf->virt->entries[i];
1654 struct vcl vcl;
1655 char nbuf1[64];
1656 char namebuf[17];
1657 if (all_ff(ve->guid))
1658 continue;
1659 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1660 ddf->currentconf =&vcl;
1661 vcl.vcnum = i;
1662 uuid_from_super_ddf(st, info.uuid);
1663 fname_from_uuid(st, &info, nbuf1, ':');
1664 _ddf_array_name(namebuf, ddf, i);
1665 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1666 namebuf[0] == '\0' ? "" : " /dev/md/", namebuf,
1667 nbuf+5, i, nbuf1+5);
1668 }
1669 }
1670
1671 static void export_examine_super_ddf(struct supertype *st)
1672 {
1673 struct mdinfo info;
1674 char nbuf[64];
1675 getinfo_super_ddf(st, &info, NULL);
1676 fname_from_uuid(st, &info, nbuf, ':');
1677 printf("MD_METADATA=ddf\n");
1678 printf("MD_LEVEL=container\n");
1679 printf("MD_UUID=%s\n", nbuf+5);
1680 printf("MD_DEVICES=%u\n",
1681 be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes));
1682 }
1683
1684 static int copy_metadata_ddf(struct supertype *st, int from, int to)
1685 {
1686 void *buf;
1687 unsigned long long dsize, offset;
1688 int bytes;
1689 struct ddf_header *ddf;
1690 int written = 0;
1691
1692 /* The meta consists of an anchor, a primary, and a secondary.
1693 * This all lives at the end of the device.
1694 * So it is easiest to find the earliest of primary and
1695 * secondary, and copy everything from there.
1696 *
1697 * Anchor is 512 from end. It contains primary_lba and secondary_lba
1698 * we choose one of those
1699 */
1700
1701 if (posix_memalign(&buf, 4096, 4096) != 0)
1702 return 1;
1703
1704 if (!get_dev_size(from, NULL, &dsize))
1705 goto err;
1706
1707 if (lseek64(from, dsize-512, 0) < 0)
1708 goto err;
1709 if (read(from, buf, 512) != 512)
1710 goto err;
1711 ddf = buf;
1712 if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) ||
1713 !be32_eq(calc_crc(ddf, 512), ddf->crc) ||
1714 (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 &&
1715 memcmp(ddf->revision, DDF_REVISION_2, 8) != 0))
1716 goto err;
1717
1718 offset = dsize - 512;
1719 if ((be64_to_cpu(ddf->primary_lba) << 9) < offset)
1720 offset = be64_to_cpu(ddf->primary_lba) << 9;
1721 if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset)
1722 offset = be64_to_cpu(ddf->secondary_lba) << 9;
1723
1724 bytes = dsize - offset;
1725
1726 if (lseek64(from, offset, 0) < 0 ||
1727 lseek64(to, offset, 0) < 0)
1728 goto err;
1729 while (written < bytes) {
1730 int n = bytes - written;
1731 if (n > 4096)
1732 n = 4096;
1733 if (read(from, buf, n) != n)
1734 goto err;
1735 if (write(to, buf, n) != n)
1736 goto err;
1737 written += n;
1738 }
1739 free(buf);
1740 return 0;
1741 err:
1742 free(buf);
1743 return 1;
1744 }
1745
1746 static void detail_super_ddf(struct supertype *st, char *homehost)
1747 {
1748 struct ddf_super *sb = st->sb;
1749 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1750
1751 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1752 printf("\n");
1753 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1754 printf(" Virtual Disks : %d\n", cnt);
1755 printf("\n");
1756 }
1757 #endif
1758
1759 static const char *vendors_with_variable_volume_UUID[] = {
1760 "LSI ",
1761 };
1762
1763 static int volume_id_is_reliable(const struct ddf_super *ddf)
1764 {
1765 int n = ARRAY_SIZE(vendors_with_variable_volume_UUID);
1766 int i;
1767 for (i = 0; i < n; i++)
1768 if (!memcmp(ddf->controller.guid,
1769 vendors_with_variable_volume_UUID[i], 8))
1770 return 0;
1771 return 1;
1772 }
1773
1774 static void uuid_of_ddf_subarray(const struct ddf_super *ddf,
1775 unsigned int vcnum, int uuid[4])
1776 {
1777 char buf[DDF_GUID_LEN+18], sha[20], *p;
1778 struct sha1_ctx ctx;
1779 if (volume_id_is_reliable(ddf)) {
1780 uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid);
1781 return;
1782 }
1783 /*
1784 * Some fake RAID BIOSes (in particular, LSI ones) change the
1785 * VD GUID at every boot. These GUIDs are not suitable for
1786 * identifying an array. Luckily the header GUID appears to
1787 * remain constant.
1788 * We construct a pseudo-UUID from the header GUID and those
1789 * properties of the subarray that we expect to remain constant.
1790 */
1791 memset(buf, 0, sizeof(buf));
1792 p = buf;
1793 memcpy(p, ddf->anchor.guid, DDF_GUID_LEN);
1794 p += DDF_GUID_LEN;
1795 memcpy(p, ddf->virt->entries[vcnum].name, 16);
1796 p += 16;
1797 *((__u16 *) p) = vcnum;
1798 sha1_init_ctx(&ctx);
1799 sha1_process_bytes(buf, sizeof(buf), &ctx);
1800 sha1_finish_ctx(&ctx, sha);
1801 memcpy(uuid, sha, 4*4);
1802 }
1803
1804 #ifndef MDASSEMBLE
1805 static void brief_detail_super_ddf(struct supertype *st)
1806 {
1807 struct mdinfo info;
1808 char nbuf[64];
1809 struct ddf_super *ddf = st->sb;
1810 unsigned int vcnum = get_vd_num_of_subarray(st);
1811 if (vcnum == DDF_CONTAINER)
1812 uuid_from_super_ddf(st, info.uuid);
1813 else if (vcnum == DDF_NOTFOUND)
1814 return;
1815 else
1816 uuid_of_ddf_subarray(ddf, vcnum, info.uuid);
1817 fname_from_uuid(st, &info, nbuf,':');
1818 printf(" UUID=%s", nbuf + 5);
1819 }
1820 #endif
1821
1822 static int match_home_ddf(struct supertype *st, char *homehost)
1823 {
1824 /* It matches 'this' host if the controller is a
1825 * Linux-MD controller with vendor_data matching
1826 * the hostname. It would be nice if we could
1827 * test against controller found in /sys or somewhere...
1828 */
1829 struct ddf_super *ddf = st->sb;
1830 unsigned int len;
1831
1832 if (!homehost)
1833 return 0;
1834 len = strlen(homehost);
1835
1836 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1837 len < sizeof(ddf->controller.vendor_data) &&
1838 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1839 ddf->controller.vendor_data[len] == 0);
1840 }
1841
1842 #ifndef MDASSEMBLE
1843 static int find_index_in_bvd(const struct ddf_super *ddf,
1844 const struct vd_config *conf, unsigned int n,
1845 unsigned int *n_bvd)
1846 {
1847 /*
1848 * Find the index of the n-th valid physical disk in this BVD.
1849 * Unused entries can be sprinkled in with the used entries,
1850 * but don't count.
1851 */
1852 unsigned int i, j;
1853 for (i = 0, j = 0;
1854 i < ddf->mppe && j < be16_to_cpu(conf->prim_elmnt_count);
1855 i++) {
1856 if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) {
1857 if (n == j) {
1858 *n_bvd = i;
1859 return 1;
1860 }
1861 j++;
1862 }
1863 }
1864 dprintf("couldn't find BVD member %u (total %u)\n",
1865 n, be16_to_cpu(conf->prim_elmnt_count));
1866 return 0;
1867 }
1868
1869 /* Given a member array instance number, and a raid disk within that instance,
1870 * find the vd_config structure. The offset of the given disk in the phys_refnum
1871 * table is returned in n_bvd.
1872 * For two-level members with a secondary raid level the vd_config for
1873 * the appropriate BVD is returned.
1874 * The return value is always &vlc->conf, where vlc is returned in last pointer.
1875 */
1876 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst,
1877 unsigned int n,
1878 unsigned int *n_bvd, struct vcl **vcl)
1879 {
1880 struct vcl *v;
1881
1882 for (v = ddf->conflist; v; v = v->next) {
1883 unsigned int nsec, ibvd = 0;
1884 struct vd_config *conf;
1885 if (inst != v->vcnum)
1886 continue;
1887 conf = &v->conf;
1888 if (conf->sec_elmnt_count == 1) {
1889 if (find_index_in_bvd(ddf, conf, n, n_bvd)) {
1890 *vcl = v;
1891 return conf;
1892 } else
1893 goto bad;
1894 }
1895 if (v->other_bvds == NULL) {
1896 pr_err("BUG: other_bvds is NULL, nsec=%u\n",
1897 conf->sec_elmnt_count);
1898 goto bad;
1899 }
1900 nsec = n / be16_to_cpu(conf->prim_elmnt_count);
1901 if (conf->sec_elmnt_seq != nsec) {
1902 for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) {
1903 if (v->other_bvds[ibvd-1]->sec_elmnt_seq
1904 == nsec)
1905 break;
1906 }
1907 if (ibvd == conf->sec_elmnt_count)
1908 goto bad;
1909 conf = v->other_bvds[ibvd-1];
1910 }
1911 if (!find_index_in_bvd(ddf, conf,
1912 n - nsec*conf->sec_elmnt_count, n_bvd))
1913 goto bad;
1914 dprintf("found disk %u as member %u in bvd %d of array %u\n",
1915 n, *n_bvd, ibvd, inst);
1916 *vcl = v;
1917 return conf;
1918 }
1919 bad:
1920 pr_err("Could't find disk %d in array %u\n", n, inst);
1921 return NULL;
1922 }
1923 #endif
1924
1925 static int find_phys(const struct ddf_super *ddf, be32 phys_refnum)
1926 {
1927 /* Find the entry in phys_disk which has the given refnum
1928 * and return it's index
1929 */
1930 unsigned int i;
1931 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++)
1932 if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum))
1933 return i;
1934 return -1;
1935 }
1936
1937 static void uuid_from_ddf_guid(const char *guid, int uuid[4])
1938 {
1939 char buf[20];
1940 struct sha1_ctx ctx;
1941 sha1_init_ctx(&ctx);
1942 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1943 sha1_finish_ctx(&ctx, buf);
1944 memcpy(uuid, buf, 4*4);
1945 }
1946
1947 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1948 {
1949 /* The uuid returned here is used for:
1950 * uuid to put into bitmap file (Create, Grow)
1951 * uuid for backup header when saving critical section (Grow)
1952 * comparing uuids when re-adding a device into an array
1953 * In these cases the uuid required is that of the data-array,
1954 * not the device-set.
1955 * uuid to recognise same set when adding a missing device back
1956 * to an array. This is a uuid for the device-set.
1957 *
1958 * For each of these we can make do with a truncated
1959 * or hashed uuid rather than the original, as long as
1960 * everyone agrees.
1961 * In the case of SVD we assume the BVD is of interest,
1962 * though that might be the case if a bitmap were made for
1963 * a mirrored SVD - worry about that later.
1964 * So we need to find the VD configuration record for the
1965 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1966 * The first 16 bytes of the sha1 of these is used.
1967 */
1968 struct ddf_super *ddf = st->sb;
1969 struct vcl *vcl = ddf->currentconf;
1970
1971 if (vcl)
1972 uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid);
1973 else
1974 uuid_from_ddf_guid(ddf->anchor.guid, uuid);
1975 }
1976
1977 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1978 {
1979 struct ddf_super *ddf = st->sb;
1980 int map_disks = info->array.raid_disks;
1981 __u32 *cptr;
1982
1983 if (ddf->currentconf) {
1984 getinfo_super_ddf_bvd(st, info, map);
1985 return;
1986 }
1987 memset(info, 0, sizeof(*info));
1988
1989 info->array.raid_disks = be16_to_cpu(ddf->phys->used_pdes);
1990 info->array.level = LEVEL_CONTAINER;
1991 info->array.layout = 0;
1992 info->array.md_minor = -1;
1993 cptr = (__u32 *)(ddf->anchor.guid + 16);
1994 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
1995
1996 info->array.chunk_size = 0;
1997 info->container_enough = 1;
1998
1999 info->disk.major = 0;
2000 info->disk.minor = 0;
2001 if (ddf->dlist) {
2002 struct phys_disk_entry *pde = NULL;
2003 info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum);
2004 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
2005
2006 info->data_offset = be64_to_cpu(ddf->phys->
2007 entries[info->disk.raid_disk].
2008 config_size);
2009 info->component_size = ddf->dlist->size - info->data_offset;
2010 if (info->disk.raid_disk >= 0)
2011 pde = ddf->phys->entries + info->disk.raid_disk;
2012 if (pde &&
2013 !(be16_to_cpu(pde->state) & DDF_Failed) &&
2014 !(be16_to_cpu(pde->state) & DDF_Missing))
2015 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2016 else
2017 info->disk.state = 1 << MD_DISK_FAULTY;
2018
2019 } else {
2020 /* There should always be a dlist, but just in case...*/
2021 info->disk.number = -1;
2022 info->disk.raid_disk = -1;
2023 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2024 }
2025 info->events = be32_to_cpu(ddf->active->seq);
2026 info->array.utime = DECADE + be32_to_cpu(ddf->active->timestamp);
2027
2028 info->recovery_start = MaxSector;
2029 info->reshape_active = 0;
2030 info->recovery_blocked = 0;
2031 info->name[0] = 0;
2032
2033 info->array.major_version = -1;
2034 info->array.minor_version = -2;
2035 strcpy(info->text_version, "ddf");
2036 info->safe_mode_delay = 0;
2037
2038 uuid_from_super_ddf(st, info->uuid);
2039
2040 if (map) {
2041 int i, e = 0;
2042 int max = be16_to_cpu(ddf->phys->max_pdes);
2043 for (i = e = 0 ; i < map_disks ; i++, e++) {
2044 while (e < max &&
2045 be32_to_cpu(ddf->phys->entries[e].refnum) == 0xffffffff)
2046 e++;
2047 if (i < info->array.raid_disks && e < max &&
2048 !(be16_to_cpu(ddf->phys->entries[e].state)
2049 & DDF_Failed))
2050 map[i] = 1;
2051 else
2052 map[i] = 0;
2053 }
2054 }
2055 }
2056
2057 /* size of name must be at least 17 bytes! */
2058 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i)
2059 {
2060 int j;
2061 memcpy(name, ddf->virt->entries[i].name, 16);
2062 name[16] = 0;
2063 for(j = 0; j < 16; j++)
2064 if (name[j] == ' ')
2065 name[j] = 0;
2066 }
2067
2068 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
2069 {
2070 struct ddf_super *ddf = st->sb;
2071 struct vcl *vc = ddf->currentconf;
2072 int cd = ddf->currentdev;
2073 int n_prim;
2074 int j;
2075 struct dl *dl = NULL;
2076 int map_disks = info->array.raid_disks;
2077 __u32 *cptr;
2078 struct vd_config *conf;
2079
2080 memset(info, 0, sizeof(*info));
2081 if (layout_ddf2md(&vc->conf, &info->array) == -1)
2082 return;
2083 info->array.md_minor = -1;
2084 cptr = (__u32 *)(vc->conf.guid + 16);
2085 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
2086 info->array.utime = DECADE + be32_to_cpu(vc->conf.timestamp);
2087 info->array.chunk_size = 512 << vc->conf.chunk_shift;
2088 info->custom_array_size = be64_to_cpu(vc->conf.array_blocks);
2089
2090 conf = &vc->conf;
2091 n_prim = be16_to_cpu(conf->prim_elmnt_count);
2092 if (conf->sec_elmnt_count > 1 && cd >= n_prim) {
2093 int ibvd = cd / n_prim - 1;
2094 cd %= n_prim;
2095 conf = vc->other_bvds[ibvd];
2096 }
2097
2098 if (cd >= 0 && (unsigned)cd < ddf->mppe) {
2099 info->data_offset =
2100 be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]);
2101 if (vc->block_sizes)
2102 info->component_size = vc->block_sizes[cd];
2103 else
2104 info->component_size = be64_to_cpu(conf->blocks);
2105
2106 for (dl = ddf->dlist; dl ; dl = dl->next)
2107 if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd]))
2108 break;
2109 }
2110
2111 info->disk.major = 0;
2112 info->disk.minor = 0;
2113 info->disk.state = 0;
2114 if (dl && dl->pdnum >= 0) {
2115 info->disk.major = dl->major;
2116 info->disk.minor = dl->minor;
2117 info->disk.raid_disk = cd + conf->sec_elmnt_seq
2118 * be16_to_cpu(conf->prim_elmnt_count);
2119 info->disk.number = dl->pdnum;
2120 info->disk.state = 0;
2121 if (info->disk.number >= 0 &&
2122 (be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Online) &&
2123 !(be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Failed))
2124 info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2125 info->events = be32_to_cpu(ddf->active->seq);
2126 }
2127
2128 info->container_member = ddf->currentconf->vcnum;
2129
2130 info->recovery_start = MaxSector;
2131 info->resync_start = 0;
2132 info->reshape_active = 0;
2133 info->recovery_blocked = 0;
2134 if (!(ddf->virt->entries[info->container_member].state
2135 & DDF_state_inconsistent) &&
2136 (ddf->virt->entries[info->container_member].init_state
2137 & DDF_initstate_mask)
2138 == DDF_init_full)
2139 info->resync_start = MaxSector;
2140
2141 uuid_from_super_ddf(st, info->uuid);
2142
2143 info->array.major_version = -1;
2144 info->array.minor_version = -2;
2145 sprintf(info->text_version, "/%s/%d",
2146 st->container_devnm,
2147 info->container_member);
2148 info->safe_mode_delay = DDF_SAFE_MODE_DELAY;
2149
2150 _ddf_array_name(info->name, ddf, info->container_member);
2151
2152 if (map)
2153 for (j = 0; j < map_disks; j++) {
2154 map[j] = 0;
2155 if (j < info->array.raid_disks) {
2156 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
2157 if (i >= 0 &&
2158 (be16_to_cpu(ddf->phys->entries[i].state)
2159 & DDF_Online) &&
2160 !(be16_to_cpu(ddf->phys->entries[i].state)
2161 & DDF_Failed))
2162 map[i] = 1;
2163 }
2164 }
2165 }
2166
2167 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
2168 char *update,
2169 char *devname, int verbose,
2170 int uuid_set, char *homehost)
2171 {
2172 /* For 'assemble' and 'force' we need to return non-zero if any
2173 * change was made. For others, the return value is ignored.
2174 * Update options are:
2175 * force-one : This device looks a bit old but needs to be included,
2176 * update age info appropriately.
2177 * assemble: clear any 'faulty' flag to allow this device to
2178 * be assembled.
2179 * force-array: Array is degraded but being forced, mark it clean
2180 * if that will be needed to assemble it.
2181 *
2182 * newdev: not used ????
2183 * grow: Array has gained a new device - this is currently for
2184 * linear only
2185 * resync: mark as dirty so a resync will happen.
2186 * uuid: Change the uuid of the array to match what is given
2187 * homehost: update the recorded homehost
2188 * name: update the name - preserving the homehost
2189 * _reshape_progress: record new reshape_progress position.
2190 *
2191 * Following are not relevant for this version:
2192 * sparc2.2 : update from old dodgey metadata
2193 * super-minor: change the preferred_minor number
2194 * summaries: update redundant counters.
2195 */
2196 int rv = 0;
2197 // struct ddf_super *ddf = st->sb;
2198 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2199 // struct virtual_entry *ve = find_ve(ddf);
2200
2201 /* we don't need to handle "force-*" or "assemble" as
2202 * there is no need to 'trick' the kernel. When the metadata is
2203 * first updated to activate the array, all the implied modifications
2204 * will just happen.
2205 */
2206
2207 if (strcmp(update, "grow") == 0) {
2208 /* FIXME */
2209 } else if (strcmp(update, "resync") == 0) {
2210 // info->resync_checkpoint = 0;
2211 } else if (strcmp(update, "homehost") == 0) {
2212 /* homehost is stored in controller->vendor_data,
2213 * or it is when we are the vendor
2214 */
2215 // if (info->vendor_is_local)
2216 // strcpy(ddf->controller.vendor_data, homehost);
2217 rv = -1;
2218 } else if (strcmp(update, "name") == 0) {
2219 /* name is stored in virtual_entry->name */
2220 // memset(ve->name, ' ', 16);
2221 // strncpy(ve->name, info->name, 16);
2222 rv = -1;
2223 } else if (strcmp(update, "_reshape_progress") == 0) {
2224 /* We don't support reshape yet */
2225 } else if (strcmp(update, "assemble") == 0 ) {
2226 /* Do nothing, just succeed */
2227 rv = 0;
2228 } else
2229 rv = -1;
2230
2231 // update_all_csum(ddf);
2232
2233 return rv;
2234 }
2235
2236 static void make_header_guid(char *guid)
2237 {
2238 be32 stamp;
2239 /* Create a DDF Header of Virtual Disk GUID */
2240
2241 /* 24 bytes of fiction required.
2242 * first 8 are a 'vendor-id' - "Linux-MD"
2243 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2244 * Remaining 8 random number plus timestamp
2245 */
2246 memcpy(guid, T10, sizeof(T10));
2247 stamp = cpu_to_be32(0xdeadbeef);
2248 memcpy(guid+8, &stamp, 4);
2249 stamp = cpu_to_be32(0);
2250 memcpy(guid+12, &stamp, 4);
2251 stamp = cpu_to_be32(time(0) - DECADE);
2252 memcpy(guid+16, &stamp, 4);
2253 stamp._v32 = random32();
2254 memcpy(guid+20, &stamp, 4);
2255 }
2256
2257 static unsigned int find_unused_vde(const struct ddf_super *ddf)
2258 {
2259 unsigned int i;
2260 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2261 if (all_ff(ddf->virt->entries[i].guid))
2262 return i;
2263 }
2264 return DDF_NOTFOUND;
2265 }
2266
2267 static unsigned int find_vde_by_name(const struct ddf_super *ddf,
2268 const char *name)
2269 {
2270 unsigned int i;
2271 if (name == NULL)
2272 return DDF_NOTFOUND;
2273 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2274 if (all_ff(ddf->virt->entries[i].guid))
2275 continue;
2276 if (!strncmp(name, ddf->virt->entries[i].name,
2277 sizeof(ddf->virt->entries[i].name)))
2278 return i;
2279 }
2280 return DDF_NOTFOUND;
2281 }
2282
2283 #ifndef MDASSEMBLE
2284 static unsigned int find_vde_by_guid(const struct ddf_super *ddf,
2285 const char *guid)
2286 {
2287 unsigned int i;
2288 if (guid == NULL || all_ff(guid))
2289 return DDF_NOTFOUND;
2290 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++)
2291 if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN))
2292 return i;
2293 return DDF_NOTFOUND;
2294 }
2295 #endif
2296
2297 static int init_super_ddf(struct supertype *st,
2298 mdu_array_info_t *info,
2299 unsigned long long size, char *name, char *homehost,
2300 int *uuid, unsigned long long data_offset)
2301 {
2302 /* This is primarily called by Create when creating a new array.
2303 * We will then get add_to_super called for each component, and then
2304 * write_init_super called to write it out to each device.
2305 * For DDF, Create can create on fresh devices or on a pre-existing
2306 * array.
2307 * To create on a pre-existing array a different method will be called.
2308 * This one is just for fresh drives.
2309 *
2310 * We need to create the entire 'ddf' structure which includes:
2311 * DDF headers - these are easy.
2312 * Controller data - a Sector describing this controller .. not that
2313 * this is a controller exactly.
2314 * Physical Disk Record - one entry per device, so
2315 * leave plenty of space.
2316 * Virtual Disk Records - again, just leave plenty of space.
2317 * This just lists VDs, doesn't give details.
2318 * Config records - describe the VDs that use this disk
2319 * DiskData - describes 'this' device.
2320 * BadBlockManagement - empty
2321 * Diag Space - empty
2322 * Vendor Logs - Could we put bitmaps here?
2323 *
2324 */
2325 struct ddf_super *ddf;
2326 char hostname[17];
2327 int hostlen;
2328 int max_phys_disks, max_virt_disks;
2329 unsigned long long sector;
2330 int clen;
2331 int i;
2332 int pdsize, vdsize;
2333 struct phys_disk *pd;
2334 struct virtual_disk *vd;
2335
2336 if (st->sb)
2337 return init_super_ddf_bvd(st, info, size, name, homehost, uuid,
2338 data_offset);
2339
2340 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
2341 pr_err("could not allocate superblock\n");
2342 return 0;
2343 }
2344 memset(ddf, 0, sizeof(*ddf));
2345 st->sb = ddf;
2346
2347 if (info == NULL) {
2348 /* zeroing superblock */
2349 return 0;
2350 }
2351
2352 /* At least 32MB *must* be reserved for the ddf. So let's just
2353 * start 32MB from the end, and put the primary header there.
2354 * Don't do secondary for now.
2355 * We don't know exactly where that will be yet as it could be
2356 * different on each device. So just set up the lengths.
2357 */
2358
2359 ddf->anchor.magic = DDF_HEADER_MAGIC;
2360 make_header_guid(ddf->anchor.guid);
2361
2362 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
2363 ddf->anchor.seq = cpu_to_be32(1);
2364 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
2365 ddf->anchor.openflag = 0xFF;
2366 ddf->anchor.foreignflag = 0;
2367 ddf->anchor.enforcegroups = 0; /* Is this best?? */
2368 ddf->anchor.pad0 = 0xff;
2369 memset(ddf->anchor.pad1, 0xff, 12);
2370 memset(ddf->anchor.header_ext, 0xff, 32);
2371 ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0);
2372 ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0);
2373 ddf->anchor.type = DDF_HEADER_ANCHOR;
2374 memset(ddf->anchor.pad2, 0xff, 3);
2375 ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */
2376 /* Put this at bottom of 32M reserved.. */
2377 ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0);
2378 max_phys_disks = 1023; /* Should be enough, 4095 is also allowed */
2379 ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks);
2380 max_virt_disks = 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2381 ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks);
2382 ddf->max_part = 64;
2383 ddf->anchor.max_partitions = cpu_to_be16(ddf->max_part);
2384 ddf->mppe = 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2385 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
2386 ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len);
2387 ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe);
2388 memset(ddf->anchor.pad3, 0xff, 54);
2389 /* Controller section is one sector long immediately
2390 * after the ddf header */
2391 sector = 1;
2392 ddf->anchor.controller_section_offset = cpu_to_be32(sector);
2393 ddf->anchor.controller_section_length = cpu_to_be32(1);
2394 sector += 1;
2395
2396 /* phys is 8 sectors after that */
2397 pdsize = ROUND_UP(sizeof(struct phys_disk) +
2398 sizeof(struct phys_disk_entry)*max_phys_disks,
2399 512);
2400 switch(pdsize/512) {
2401 case 2: case 8: case 32: case 128: case 512: break;
2402 default: abort();
2403 }
2404 ddf->anchor.phys_section_offset = cpu_to_be32(sector);
2405 ddf->anchor.phys_section_length =
2406 cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
2407 sector += pdsize/512;
2408
2409 /* virt is another 32 sectors */
2410 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
2411 sizeof(struct virtual_entry) * max_virt_disks,
2412 512);
2413 switch(vdsize/512) {
2414 case 2: case 8: case 32: case 128: case 512: break;
2415 default: abort();
2416 }
2417 ddf->anchor.virt_section_offset = cpu_to_be32(sector);
2418 ddf->anchor.virt_section_length =
2419 cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
2420 sector += vdsize/512;
2421
2422 clen = ddf->conf_rec_len * (ddf->max_part+1);
2423 ddf->anchor.config_section_offset = cpu_to_be32(sector);
2424 ddf->anchor.config_section_length = cpu_to_be32(clen);
2425 sector += clen;
2426
2427 ddf->anchor.data_section_offset = cpu_to_be32(sector);
2428 ddf->anchor.data_section_length = cpu_to_be32(1);
2429 sector += 1;
2430
2431 ddf->anchor.bbm_section_length = cpu_to_be32(0);
2432 ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF);
2433 ddf->anchor.diag_space_length = cpu_to_be32(0);
2434 ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF);
2435 ddf->anchor.vendor_length = cpu_to_be32(0);
2436 ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF);
2437
2438 memset(ddf->anchor.pad4, 0xff, 256);
2439
2440 memcpy(&ddf->primary, &ddf->anchor, 512);
2441 memcpy(&ddf->secondary, &ddf->anchor, 512);
2442
2443 ddf->primary.openflag = 1; /* I guess.. */
2444 ddf->primary.type = DDF_HEADER_PRIMARY;
2445
2446 ddf->secondary.openflag = 1; /* I guess.. */
2447 ddf->secondary.type = DDF_HEADER_SECONDARY;
2448
2449 ddf->active = &ddf->primary;
2450
2451 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
2452
2453 /* 24 more bytes of fiction required.
2454 * first 8 are a 'vendor-id' - "Linux-MD"
2455 * Remaining 16 are serial number.... maybe a hostname would do?
2456 */
2457 memcpy(ddf->controller.guid, T10, sizeof(T10));
2458 gethostname(hostname, sizeof(hostname));
2459 hostname[sizeof(hostname) - 1] = 0;
2460 hostlen = strlen(hostname);
2461 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
2462 for (i = strlen(T10) ; i+hostlen < 24; i++)
2463 ddf->controller.guid[i] = ' ';
2464
2465 ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD);
2466 ddf->controller.type.device_id = cpu_to_be16(0xBEEF);
2467 ddf->controller.type.sub_vendor_id = cpu_to_be16(0);
2468 ddf->controller.type.sub_device_id = cpu_to_be16(0);
2469 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
2470 memset(ddf->controller.pad, 0xff, 8);
2471 memset(ddf->controller.vendor_data, 0xff, 448);
2472 if (homehost && strlen(homehost) < 440)
2473 strcpy((char*)ddf->controller.vendor_data, homehost);
2474
2475 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
2476 pr_err("could not allocate pd\n");
2477 return 0;
2478 }
2479 ddf->phys = pd;
2480 ddf->pdsize = pdsize;
2481
2482 memset(pd, 0xff, pdsize);
2483 memset(pd, 0, sizeof(*pd));
2484 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2485 pd->used_pdes = cpu_to_be16(0);
2486 pd->max_pdes = cpu_to_be16(max_phys_disks);
2487 memset(pd->pad, 0xff, 52);
2488 for (i = 0; i < max_phys_disks; i++)
2489 memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN);
2490
2491 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
2492 pr_err("could not allocate vd\n");
2493 return 0;
2494 }
2495 ddf->virt = vd;
2496 ddf->vdsize = vdsize;
2497 memset(vd, 0, vdsize);
2498 vd->magic = DDF_VIRT_RECORDS_MAGIC;
2499 vd->populated_vdes = cpu_to_be16(0);
2500 vd->max_vdes = cpu_to_be16(max_virt_disks);
2501 memset(vd->pad, 0xff, 52);
2502
2503 for (i=0; i<max_virt_disks; i++)
2504 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
2505
2506 st->sb = ddf;
2507 ddf_set_updates_pending(ddf, NULL);
2508 return 1;
2509 }
2510
2511 static int chunk_to_shift(int chunksize)
2512 {
2513 return ffs(chunksize/512)-1;
2514 }
2515
2516 #ifndef MDASSEMBLE
2517 struct extent {
2518 unsigned long long start, size;
2519 };
2520 static int cmp_extent(const void *av, const void *bv)
2521 {
2522 const struct extent *a = av;
2523 const struct extent *b = bv;
2524 if (a->start < b->start)
2525 return -1;
2526 if (a->start > b->start)
2527 return 1;
2528 return 0;
2529 }
2530
2531 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
2532 {
2533 /* Find a list of used extents on the given physical device
2534 * (dnum) of the given ddf.
2535 * Return a malloced array of 'struct extent'
2536 */
2537 struct extent *rv;
2538 int n = 0;
2539 unsigned int i;
2540 __u16 state;
2541
2542 if (dl->pdnum < 0)
2543 return NULL;
2544 state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state);
2545
2546 if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online)
2547 return NULL;
2548
2549 rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2));
2550
2551 for (i = 0; i < ddf->max_part; i++) {
2552 const struct vd_config *bvd;
2553 unsigned int ibvd;
2554 struct vcl *v = dl->vlist[i];
2555 if (v == NULL ||
2556 get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe,
2557 &bvd, &ibvd) == DDF_NOTFOUND)
2558 continue;
2559 rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]);
2560 rv[n].size = be64_to_cpu(bvd->blocks);
2561 n++;
2562 }
2563 qsort(rv, n, sizeof(*rv), cmp_extent);
2564
2565 rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
2566 rv[n].size = 0;
2567 return rv;
2568 }
2569
2570 static unsigned long long find_space(
2571 struct ddf_super *ddf, struct dl *dl,
2572 unsigned long long data_offset,
2573 unsigned long long *size)
2574 {
2575 /* Find if the requested amount of space is available.
2576 * If it is, return start.
2577 * If not, set *size to largest space.
2578 * If data_offset != INVALID_SECTORS, then the space must start
2579 * at this location.
2580 */
2581 struct extent *e = get_extents(ddf, dl);
2582 int i = 0;
2583 unsigned long long pos = 0;
2584 unsigned long long max_size = 0;
2585
2586 if (!e) {
2587 *size = 0;
2588 return INVALID_SECTORS;
2589 }
2590 do {
2591 unsigned long long esize = e[i].start - pos;
2592 if (data_offset != INVALID_SECTORS &&
2593 pos <= data_offset &&
2594 e[i].start > data_offset) {
2595 pos = data_offset;
2596 esize = e[i].start - pos;
2597 }
2598 if (data_offset != INVALID_SECTORS &&
2599 pos != data_offset) {
2600 i++;
2601 continue;
2602 }
2603 if (esize >= *size) {
2604 /* Found! */
2605 free(e);
2606 return pos;
2607 }
2608 if (esize > max_size)
2609 max_size = esize;
2610 pos = e[i].start + e[i].size;
2611 i++;
2612 } while (e[i-1].size);
2613 *size = max_size;
2614 free(e);
2615 return INVALID_SECTORS;
2616 }
2617 #endif
2618
2619 static int init_super_ddf_bvd(struct supertype *st,
2620 mdu_array_info_t *info,
2621 unsigned long long size,
2622 char *name, char *homehost,
2623 int *uuid, unsigned long long data_offset)
2624 {
2625 /* We are creating a BVD inside a pre-existing container.
2626 * so st->sb is already set.
2627 * We need to create a new vd_config and a new virtual_entry
2628 */
2629 struct ddf_super *ddf = st->sb;
2630 unsigned int venum, i;
2631 struct virtual_entry *ve;
2632 struct vcl *vcl;
2633 struct vd_config *vc;
2634
2635 if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) {
2636 pr_err("This ddf already has an array called %s\n", name);
2637 return 0;
2638 }
2639 venum = find_unused_vde(ddf);
2640 if (venum == DDF_NOTFOUND) {
2641 pr_err("Cannot find spare slot for virtual disk\n");
2642 return 0;
2643 }
2644 ve = &ddf->virt->entries[venum];
2645
2646 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2647 * timestamp, random number
2648 */
2649 make_header_guid(ve->guid);
2650 ve->unit = cpu_to_be16(info->md_minor);
2651 ve->pad0 = 0xFFFF;
2652 ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid,
2653 DDF_GUID_LEN);
2654 ve->type = cpu_to_be16(0);
2655 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
2656 if (info->state & 1) /* clean */
2657 ve->init_state = DDF_init_full;
2658 else
2659 ve->init_state = DDF_init_not;
2660
2661 memset(ve->pad1, 0xff, 14);
2662 memset(ve->name, ' ', 16);
2663 if (name)
2664 strncpy(ve->name, name, 16);
2665 ddf->virt->populated_vdes =
2666 cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1);
2667
2668 /* Now create a new vd_config */
2669 if (posix_memalign((void**)&vcl, 512,
2670 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
2671 pr_err("could not allocate vd_config\n");
2672 return 0;
2673 }
2674 vcl->vcnum = venum;
2675 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2676 vc = &vcl->conf;
2677
2678 vc->magic = DDF_VD_CONF_MAGIC;
2679 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2680 vc->timestamp = cpu_to_be32(time(0)-DECADE);
2681 vc->seqnum = cpu_to_be32(1);
2682 memset(vc->pad0, 0xff, 24);
2683 vc->chunk_shift = chunk_to_shift(info->chunk_size);
2684 if (layout_md2ddf(info, vc) == -1 ||
2685 be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) {
2686 pr_err("unsupported RAID level/layout %d/%d with %d disks\n",
2687 info->level, info->layout, info->raid_disks);
2688 free(vcl);
2689 return 0;
2690 }
2691 vc->sec_elmnt_seq = 0;
2692 if (alloc_other_bvds(ddf, vcl) != 0) {
2693 pr_err("could not allocate other bvds\n");
2694 free(vcl);
2695 return 0;
2696 }
2697 vc->blocks = cpu_to_be64(info->size * 2);
2698 vc->array_blocks = cpu_to_be64(
2699 calc_array_size(info->level, info->raid_disks, info->layout,
2700 info->chunk_size, info->size*2));
2701 memset(vc->pad1, 0xff, 8);
2702 vc->spare_refs[0] = cpu_to_be32(0xffffffff);
2703 vc->spare_refs[1] = cpu_to_be32(0xffffffff);
2704 vc->spare_refs[2] = cpu_to_be32(0xffffffff);
2705 vc->spare_refs[3] = cpu_to_be32(0xffffffff);
2706 vc->spare_refs[4] = cpu_to_be32(0xffffffff);
2707 vc->spare_refs[5] = cpu_to_be32(0xffffffff);
2708 vc->spare_refs[6] = cpu_to_be32(0xffffffff);
2709 vc->spare_refs[7] = cpu_to_be32(0xffffffff);
2710 memset(vc->cache_pol, 0, 8);
2711 vc->bg_rate = 0x80;
2712 memset(vc->pad2, 0xff, 3);
2713 memset(vc->pad3, 0xff, 52);
2714 memset(vc->pad4, 0xff, 192);
2715 memset(vc->v0, 0xff, 32);
2716 memset(vc->v1, 0xff, 32);
2717 memset(vc->v2, 0xff, 16);
2718 memset(vc->v3, 0xff, 16);
2719 memset(vc->vendor, 0xff, 32);
2720
2721 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2722 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2723
2724 for (i = 1; i < vc->sec_elmnt_count; i++) {
2725 memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512);
2726 vcl->other_bvds[i-1]->sec_elmnt_seq = i;
2727 }
2728
2729 vcl->next = ddf->conflist;
2730 ddf->conflist = vcl;
2731 ddf->currentconf = vcl;
2732 ddf_set_updates_pending(ddf, NULL);
2733 return 1;
2734 }
2735
2736 #ifndef MDASSEMBLE
2737 static void add_to_super_ddf_bvd(struct supertype *st,
2738 mdu_disk_info_t *dk, int fd, char *devname,
2739 unsigned long long data_offset)
2740 {
2741 /* fd and devname identify a device within the ddf container (st).
2742 * dk identifies a location in the new BVD.
2743 * We need to find suitable free space in that device and update
2744 * the phys_refnum and lba_offset for the newly created vd_config.
2745 * We might also want to update the type in the phys_disk
2746 * section.
2747 *
2748 * Alternately: fd == -1 and we have already chosen which device to
2749 * use and recorded in dlist->raid_disk;
2750 */
2751 struct dl *dl;
2752 struct ddf_super *ddf = st->sb;
2753 struct vd_config *vc;
2754 unsigned int i;
2755 unsigned long long blocks, pos;
2756 unsigned int raid_disk = dk->raid_disk;
2757
2758 if (fd == -1) {
2759 for (dl = ddf->dlist; dl ; dl = dl->next)
2760 if (dl->raiddisk == dk->raid_disk)
2761 break;
2762 } else {
2763 for (dl = ddf->dlist; dl ; dl = dl->next)
2764 if (dl->major == dk->major &&
2765 dl->minor == dk->minor)
2766 break;
2767 }
2768 if (!dl || dl->pdnum < 0 || ! (dk->state & (1<<MD_DISK_SYNC)))
2769 return;
2770
2771 vc = &ddf->currentconf->conf;
2772 if (vc->sec_elmnt_count > 1) {
2773 unsigned int n = be16_to_cpu(vc->prim_elmnt_count);
2774 if (raid_disk >= n)
2775 vc = ddf->currentconf->other_bvds[raid_disk / n - 1];
2776 raid_disk %= n;
2777 }
2778
2779 blocks = be64_to_cpu(vc->blocks);
2780 if (ddf->currentconf->block_sizes)
2781 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2782
2783 pos = find_space(ddf, dl, data_offset, &blocks);
2784 if (pos == INVALID_SECTORS)
2785 return;
2786
2787 ddf->currentdev = dk->raid_disk;
2788 vc->phys_refnum[raid_disk] = dl->disk.refnum;
2789 LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos);
2790
2791 for (i = 0; i < ddf->max_part ; i++)
2792 if (dl->vlist[i] == NULL)
2793 break;
2794 if (i == ddf->max_part)
2795 return;
2796 dl->vlist[i] = ddf->currentconf;
2797
2798 if (fd >= 0)
2799 dl->fd = fd;
2800 if (devname)
2801 dl->devname = devname;
2802
2803 /* Check if we can mark array as optimal yet */
2804 i = ddf->currentconf->vcnum;
2805 ddf->virt->entries[i].state =
2806 (ddf->virt->entries[i].state & ~DDF_state_mask)
2807 | get_svd_state(ddf, ddf->currentconf);
2808 be16_clear(ddf->phys->entries[dl->pdnum].type,
2809 cpu_to_be16(DDF_Global_Spare));
2810 be16_set(ddf->phys->entries[dl->pdnum].type,
2811 cpu_to_be16(DDF_Active_in_VD));
2812 dprintf("added disk %d/%08x to VD %d/%s as disk %d\n",
2813 dl->pdnum, be32_to_cpu(dl->disk.refnum),
2814 ddf->currentconf->vcnum, guid_str(vc->guid),
2815 dk->raid_disk);
2816 ddf_set_updates_pending(ddf, vc);
2817 }
2818
2819 static unsigned int find_unused_pde(const struct ddf_super *ddf)
2820 {
2821 unsigned int i;
2822 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) {
2823 if (all_ff(ddf->phys->entries[i].guid))
2824 return i;
2825 }
2826 return DDF_NOTFOUND;
2827 }
2828
2829 static void _set_config_size(struct phys_disk_entry *pde, const struct dl *dl)
2830 {
2831 __u64 cfs, t;
2832 cfs = min(dl->size - 32*1024*2ULL, be64_to_cpu(dl->primary_lba));
2833 t = be64_to_cpu(dl->secondary_lba);
2834 if (t != ~(__u64)0)
2835 cfs = min(cfs, t);
2836 /*
2837 * Some vendor DDF structures interpret workspace_lba
2838 * very differently than we do: Make a sanity check on the value.
2839 */
2840 t = be64_to_cpu(dl->workspace_lba);
2841 if (t < cfs) {
2842 __u64 wsp = cfs - t;
2843 if (wsp > 1024*1024*2ULL && wsp > dl->size / 16) {
2844 pr_err("%x:%x: workspace size 0x%llx too big, ignoring\n",
2845 dl->major, dl->minor, (unsigned long long)wsp);
2846 } else
2847 cfs = t;
2848 }
2849 pde->config_size = cpu_to_be64(cfs);
2850 dprintf("%x:%x config_size %llx, DDF structure is %llx blocks\n",
2851 dl->major, dl->minor,
2852 (unsigned long long)cfs, (unsigned long long)(dl->size-cfs));
2853 }
2854
2855 /* Add a device to a container, either while creating it or while
2856 * expanding a pre-existing container
2857 */
2858 static int add_to_super_ddf(struct supertype *st,
2859 mdu_disk_info_t *dk, int fd, char *devname,
2860 unsigned long long data_offset)
2861 {
2862 struct ddf_super *ddf = st->sb;
2863 struct dl *dd;
2864 time_t now;
2865 struct tm *tm;
2866 unsigned long long size;
2867 struct phys_disk_entry *pde;
2868 unsigned int n, i;
2869 struct stat stb;
2870 __u32 *tptr;
2871
2872 if (ddf->currentconf) {
2873 add_to_super_ddf_bvd(st, dk, fd, devname, data_offset);
2874 return 0;
2875 }
2876
2877 /* This is device numbered dk->number. We need to create
2878 * a phys_disk entry and a more detailed disk_data entry.
2879 */
2880 fstat(fd, &stb);
2881 n = find_unused_pde(ddf);
2882 if (n == DDF_NOTFOUND) {
2883 pr_err("No free slot in array, cannot add disk\n");
2884 return 1;
2885 }
2886 pde = &ddf->phys->entries[n];
2887 get_dev_size(fd, NULL, &size);
2888 if (size <= 32*1024*1024) {
2889 pr_err("device size must be at least 32MB\n");
2890 return 1;
2891 }
2892 size >>= 9;
2893
2894 if (posix_memalign((void**)&dd, 512,
2895 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2896 pr_err("could allocate buffer for new disk, aborting\n");
2897 return 1;
2898 }
2899 dd->major = major(stb.st_rdev);
2900 dd->minor = minor(stb.st_rdev);
2901 dd->devname = devname;
2902 dd->fd = fd;
2903 dd->spare = NULL;
2904
2905 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2906 now = time(0);
2907 tm = localtime(&now);
2908 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2909 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2910 tptr = (__u32 *)(dd->disk.guid + 16);
2911 *tptr++ = random32();
2912 *tptr = random32();
2913
2914 do {
2915 /* Cannot be bothered finding a CRC of some irrelevant details*/
2916 dd->disk.refnum._v32 = random32();
2917 for (i = be16_to_cpu(ddf->active->max_pd_entries);
2918 i > 0; i--)
2919 if (be32_eq(ddf->phys->entries[i-1].refnum,
2920 dd->disk.refnum))
2921 break;
2922 } while (i > 0);
2923
2924 dd->disk.forced_ref = 1;
2925 dd->disk.forced_guid = 1;
2926 memset(dd->disk.vendor, ' ', 32);
2927 memcpy(dd->disk.vendor, "Linux", 5);
2928 memset(dd->disk.pad, 0xff, 442);
2929 for (i = 0; i < ddf->max_part ; i++)
2930 dd->vlist[i] = NULL;
2931
2932 dd->pdnum = n;
2933
2934 if (st->update_tail) {
2935 int len = (sizeof(struct phys_disk) +
2936 sizeof(struct phys_disk_entry));
2937 struct phys_disk *pd;
2938
2939 pd = xmalloc(len);
2940 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2941 pd->used_pdes = cpu_to_be16(n);
2942 pde = &pd->entries[0];
2943 dd->mdupdate = pd;
2944 } else
2945 ddf->phys->used_pdes = cpu_to_be16(
2946 1 + be16_to_cpu(ddf->phys->used_pdes));
2947
2948 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2949 pde->refnum = dd->disk.refnum;
2950 pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2951 pde->state = cpu_to_be16(DDF_Online);
2952 dd->size = size;
2953 /*
2954 * If there is already a device in dlist, try to reserve the same
2955 * amount of workspace. Otherwise, use 32MB.
2956 * We checked disk size above already.
2957 */
2958 #define __calc_lba(new, old, lba, mb) do { \
2959 unsigned long long dif; \
2960 if ((old) != NULL) \
2961 dif = (old)->size - be64_to_cpu((old)->lba); \
2962 else \
2963 dif = (new)->size; \
2964 if ((new)->size > dif) \
2965 (new)->lba = cpu_to_be64((new)->size - dif); \
2966 else \
2967 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2968 } while (0)
2969 __calc_lba(dd, ddf->dlist, workspace_lba, 32);
2970 __calc_lba(dd, ddf->dlist, primary_lba, 16);
2971 if (ddf->dlist == NULL ||
2972 be64_to_cpu(ddf->dlist->secondary_lba) != ~(__u64)0)
2973 __calc_lba(dd, ddf->dlist, secondary_lba, 32);
2974 _set_config_size(pde, dd);
2975
2976 sprintf(pde->path, "%17.17s","Information: nil") ;
2977 memset(pde->pad, 0xff, 6);
2978
2979 if (st->update_tail) {
2980 dd->next = ddf->add_list;
2981 ddf->add_list = dd;
2982 } else {
2983 dd->next = ddf->dlist;
2984 ddf->dlist = dd;
2985 ddf_set_updates_pending(ddf, NULL);
2986 }
2987
2988 return 0;
2989 }
2990
2991 static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk)
2992 {
2993 struct ddf_super *ddf = st->sb;
2994 struct dl *dl;
2995
2996 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2997 * disappeared from the container.
2998 * We need to arrange that it disappears from the metadata and
2999 * internal data structures too.
3000 * Most of the work is done by ddf_process_update which edits
3001 * the metadata and closes the file handle and attaches the memory
3002 * where free_updates will free it.
3003 */
3004 for (dl = ddf->dlist; dl ; dl = dl->next)
3005 if (dl->major == dk->major &&
3006 dl->minor == dk->minor)
3007 break;
3008 if (!dl || dl->pdnum < 0)
3009 return -1;
3010
3011 if (st->update_tail) {
3012 int len = (sizeof(struct phys_disk) +
3013 sizeof(struct phys_disk_entry));
3014 struct phys_disk *pd;
3015
3016 pd = xmalloc(len);
3017 pd->magic = DDF_PHYS_RECORDS_MAGIC;
3018 pd->used_pdes = cpu_to_be16(dl->pdnum);
3019 pd->entries[0].state = cpu_to_be16(DDF_Missing);
3020 append_metadata_update(st, pd, len);
3021 }
3022 return 0;
3023 }
3024 #endif
3025
3026 /*
3027 * This is the write_init_super method for a ddf container. It is
3028 * called when creating a container or adding another device to a
3029 * container.
3030 */
3031
3032 static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type)
3033 {
3034 unsigned long long sector;
3035 struct ddf_header *header;
3036 int fd, i, n_config, conf_size, buf_size;
3037 int ret = 0;
3038 char *conf;
3039
3040 fd = d->fd;
3041
3042 switch (type) {
3043 case DDF_HEADER_PRIMARY:
3044 header = &ddf->primary;
3045 sector = be64_to_cpu(header->primary_lba);
3046 break;
3047 case DDF_HEADER_SECONDARY:
3048 header = &ddf->secondary;
3049 sector = be64_to_cpu(header->secondary_lba);
3050 break;
3051 default:
3052 return 0;
3053 }
3054 if (sector == ~(__u64)0)
3055 return 0;
3056
3057 header->type = type;
3058 header->openflag = 1;
3059 header->crc = calc_crc(header, 512);
3060
3061 lseek64(fd, sector<<9, 0);
3062 if (write(fd, header, 512) < 0)
3063 goto out;
3064
3065 ddf->controller.crc = calc_crc(&ddf->controller, 512);
3066 if (write(fd, &ddf->controller, 512) < 0)
3067 goto out;
3068
3069 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
3070 if (write(fd, ddf->phys, ddf->pdsize) < 0)
3071 goto out;
3072 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
3073 if (write(fd, ddf->virt, ddf->vdsize) < 0)
3074 goto out;
3075
3076 /* Now write lots of config records. */
3077 n_config = ddf->max_part;
3078 conf_size = ddf->conf_rec_len * 512;
3079 conf = ddf->conf;
3080 buf_size = conf_size * (n_config + 1);
3081 if (!conf) {
3082 if (posix_memalign((void**)&conf, 512, buf_size) != 0)
3083 goto out;
3084 ddf->conf = conf;
3085 }
3086 for (i = 0 ; i <= n_config ; i++) {
3087 struct vcl *c;
3088 struct vd_config *vdc = NULL;
3089 if (i == n_config) {
3090 c = (struct vcl *)d->spare;
3091 if (c)
3092 vdc = &c->conf;
3093 } else {
3094 unsigned int dummy;
3095 c = d->vlist[i];
3096 if (c)
3097 get_pd_index_from_refnum(
3098 c, d->disk.refnum,
3099 ddf->mppe,
3100 (const struct vd_config **)&vdc,
3101 &dummy);
3102 }
3103 if (vdc) {
3104 dprintf("writing conf record %i on disk %08x for %s/%u\n",
3105 i, be32_to_cpu(d->disk.refnum),
3106 guid_str(vdc->guid),
3107 vdc->sec_elmnt_seq);
3108 vdc->crc = calc_crc(vdc, conf_size);
3109 memcpy(conf + i*conf_size, vdc, conf_size);
3110 } else
3111 memset(conf + i*conf_size, 0xff, conf_size);
3112 }
3113 if (write(fd, conf, buf_size) != buf_size)
3114 goto out;
3115
3116 d->disk.crc = calc_crc(&d->disk, 512);
3117 if (write(fd, &d->disk, 512) < 0)
3118 goto out;
3119
3120 ret = 1;
3121 out:
3122 header->openflag = 0;
3123 header->crc = calc_crc(header, 512);
3124
3125 lseek64(fd, sector<<9, 0);
3126 if (write(fd, header, 512) < 0)
3127 ret = 0;
3128
3129 return ret;
3130 }
3131
3132 static int _write_super_to_disk(struct ddf_super *ddf, struct dl *d)
3133 {
3134 unsigned long long size;
3135 int fd = d->fd;
3136 if (fd < 0)
3137 return 0;
3138
3139 /* We need to fill in the primary, (secondary) and workspace
3140 * lba's in the headers, set their checksums,
3141 * Also checksum phys, virt....
3142 *
3143 * Then write everything out, finally the anchor is written.
3144 */
3145 get_dev_size(fd, NULL, &size);
3146 size /= 512;
3147 memcpy(&ddf->anchor, ddf->active, 512);
3148 if (be64_to_cpu(d->workspace_lba) != 0ULL)
3149 ddf->anchor.workspace_lba = d->workspace_lba;
3150 else
3151 ddf->anchor.workspace_lba =
3152 cpu_to_be64(size - 32*1024*2);
3153 if (be64_to_cpu(d->primary_lba) != 0ULL)
3154 ddf->anchor.primary_lba = d->primary_lba;
3155 else
3156 ddf->anchor.primary_lba =
3157 cpu_to_be64(size - 16*1024*2);
3158 if (be64_to_cpu(d->secondary_lba) != 0ULL)
3159 ddf->anchor.secondary_lba = d->secondary_lba;
3160 else
3161 ddf->anchor.secondary_lba =
3162 cpu_to_be64(size - 32*1024*2);
3163 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
3164 memcpy(&ddf->primary, &ddf->anchor, 512);
3165 memcpy(&ddf->secondary, &ddf->anchor, 512);
3166
3167 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
3168 ddf->anchor.seq = cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3169 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
3170
3171 if (!__write_ddf_structure(d, ddf, DDF_HEADER_PRIMARY))
3172 return 0;
3173
3174 if (!__write_ddf_structure(d, ddf, DDF_HEADER_SECONDARY))
3175 return 0;
3176
3177 lseek64(fd, (size-1)*512, SEEK_SET);
3178 if (write(fd, &ddf->anchor, 512) < 0)
3179 return 0;
3180
3181 return 1;
3182 }
3183
3184 #ifndef MDASSEMBLE
3185 static int __write_init_super_ddf(struct supertype *st)
3186 {
3187 struct ddf_super *ddf = st->sb;
3188 struct dl *d;
3189 int attempts = 0;
3190 int successes = 0;
3191
3192 pr_state(ddf, __func__);
3193
3194 /* try to write updated metadata,
3195 * if we catch a failure move on to the next disk
3196 */
3197 for (d = ddf->dlist; d; d=d->next) {
3198 attempts++;
3199 successes += _write_super_to_disk(ddf, d);
3200 }
3201
3202 return attempts != successes;
3203 }
3204
3205 static int write_init_super_ddf(struct supertype *st)
3206 {
3207 struct ddf_super *ddf = st->sb;
3208 struct vcl *currentconf = ddf->currentconf;
3209
3210 /* We are done with currentconf - reset it so st refers to the container */
3211 ddf->currentconf = NULL;
3212
3213 if (st->update_tail) {
3214 /* queue the virtual_disk and vd_config as metadata updates */
3215 struct virtual_disk *vd;
3216 struct vd_config *vc;
3217 int len, tlen;
3218 unsigned int i;
3219
3220 if (!currentconf) {
3221 /* Must be adding a physical disk to the container */
3222 int len = (sizeof(struct phys_disk) +
3223 sizeof(struct phys_disk_entry));
3224
3225 /* adding a disk to the container. */
3226 if (!ddf->add_list)
3227 return 0;
3228
3229 append_metadata_update(st, ddf->add_list->mdupdate, len);
3230 ddf->add_list->mdupdate = NULL;
3231 return 0;
3232 }
3233
3234 /* Newly created VD */
3235
3236 /* First the virtual disk. We have a slightly fake header */
3237 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
3238 vd = xmalloc(len);
3239 *vd = *ddf->virt;
3240 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
3241 vd->populated_vdes = cpu_to_be16(currentconf->vcnum);
3242 append_metadata_update(st, vd, len);
3243
3244 /* Then the vd_config */
3245 len = ddf->conf_rec_len * 512;
3246 tlen = len * currentconf->conf.sec_elmnt_count;
3247 vc = xmalloc(tlen);
3248 memcpy(vc, &currentconf->conf, len);
3249 for (i = 1; i < currentconf->conf.sec_elmnt_count; i++)
3250 memcpy((char *)vc + i*len, currentconf->other_bvds[i-1],
3251 len);
3252 append_metadata_update(st, vc, tlen);
3253
3254 return 0;
3255 } else {
3256 struct dl *d;
3257 if (!currentconf)
3258 for (d = ddf->dlist; d; d=d->next)
3259 while (Kill(d->devname, NULL, 0, -1, 1) == 0);
3260 /* Note: we don't close the fd's now, but a subsequent
3261 * ->free_super() will
3262 */
3263 return __write_init_super_ddf(st);
3264 }
3265 }
3266
3267 #endif
3268
3269 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize,
3270 unsigned long long data_offset)
3271 {
3272 /* We must reserve the last 32Meg */
3273 if (devsize <= 32*1024*2)
3274 return 0;
3275 return devsize - 32*1024*2;
3276 }
3277
3278 #ifndef MDASSEMBLE
3279
3280 static int reserve_space(struct supertype *st, int raiddisks,
3281 unsigned long long size, int chunk,
3282 unsigned long long data_offset,
3283 unsigned long long *freesize)
3284 {
3285 /* Find 'raiddisks' spare extents at least 'size' big (but
3286 * only caring about multiples of 'chunk') and remember
3287 * them. If size==0, find the largest size possible.
3288 * Report available size in *freesize
3289 * If space cannot be found, fail.
3290 */
3291 struct dl *dl;
3292 struct ddf_super *ddf = st->sb;
3293 int cnt = 0;
3294
3295 for (dl = ddf->dlist; dl ; dl=dl->next) {
3296 dl->raiddisk = -1;
3297 dl->esize = 0;
3298 }
3299 /* Now find largest extent on each device */
3300 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3301 unsigned long long minsize = ULLONG_MAX;
3302
3303 find_space(ddf, dl, data_offset, &minsize);
3304 if (minsize >= size && minsize >= (unsigned)chunk) {
3305 cnt++;
3306 dl->esize = minsize;
3307 }
3308 }
3309 if (cnt < raiddisks) {
3310 pr_err("not enough devices with space to create array.\n");
3311 return 0; /* No enough free spaces large enough */
3312 }
3313 if (size == 0) {
3314 /* choose the largest size of which there are at least 'raiddisk' */
3315 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3316 struct dl *dl2;
3317 if (dl->esize <= size)
3318 continue;
3319 /* This is bigger than 'size', see if there are enough */
3320 cnt = 0;
3321 for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next)
3322 if (dl2->esize >= dl->esize)
3323 cnt++;
3324 if (cnt >= raiddisks)
3325 size = dl->esize;
3326 }
3327 if (chunk) {
3328 size = size / chunk;
3329 size *= chunk;
3330 }
3331 *freesize = size;
3332 if (size < 32) {
3333 pr_err("not enough spare devices to create array.\n");
3334 return 0;
3335 }
3336 }
3337 /* We have a 'size' of which there are enough spaces.
3338 * We simply do a first-fit */
3339 cnt = 0;
3340 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
3341 if (dl->esize < size)
3342 continue;
3343
3344 dl->raiddisk = cnt;
3345 cnt++;
3346 }
3347 return 1;
3348 }
3349
3350 static int validate_geometry_ddf(struct supertype *st,
3351 int level, int layout, int raiddisks,
3352 int *chunk, unsigned long long size,
3353 unsigned long long data_offset,
3354 char *dev, unsigned long long *freesize,
3355 int verbose)
3356 {
3357 int fd;
3358 struct mdinfo *sra;
3359 int cfd;
3360
3361 /* ddf potentially supports lots of things, but it depends on
3362 * what devices are offered (and maybe kernel version?)
3363 * If given unused devices, we will make a container.
3364 * If given devices in a container, we will make a BVD.
3365 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3366 */
3367
3368 if (*chunk == UnSet)
3369 *chunk = DEFAULT_CHUNK;
3370
3371 if (level == LEVEL_NONE)
3372 level = LEVEL_CONTAINER;
3373 if (level == LEVEL_CONTAINER) {
3374 /* Must be a fresh device to add to a container */
3375 return validate_geometry_ddf_container(st, level, layout,
3376 raiddisks, *chunk,
3377 size, data_offset, dev,
3378 freesize,
3379 verbose);
3380 }
3381
3382 if (!dev) {
3383 mdu_array_info_t array = {
3384 .level = level,
3385 .layout = layout,
3386 .raid_disks = raiddisks
3387 };
3388 struct vd_config conf;
3389 if (layout_md2ddf(&array, &conf) == -1) {
3390 if (verbose)
3391 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3392 level, layout, raiddisks);
3393 return 0;
3394 }
3395 /* Should check layout? etc */
3396
3397 if (st->sb && freesize) {
3398 /* --create was given a container to create in.
3399 * So we need to check that there are enough
3400 * free spaces and return the amount of space.
3401 * We may as well remember which drives were
3402 * chosen so that add_to_super/getinfo_super
3403 * can return them.
3404 */
3405 return reserve_space(st, raiddisks, size, *chunk,
3406 data_offset, freesize);
3407 }
3408 return 1;
3409 }
3410
3411 if (st->sb) {
3412 /* A container has already been opened, so we are
3413 * creating in there. Maybe a BVD, maybe an SVD.
3414 * Should make a distinction one day.
3415 */
3416 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
3417 chunk, size, data_offset, dev,
3418 freesize,
3419 verbose);
3420 }
3421 /* This is the first device for the array.
3422 * If it is a container, we read it in and do automagic allocations,
3423 * no other devices should be given.
3424 * Otherwise it must be a member device of a container, and we
3425 * do manual allocation.
3426 * Later we should check for a BVD and make an SVD.
3427 */
3428 fd = open(dev, O_RDONLY|O_EXCL, 0);
3429 if (fd >= 0) {
3430 close(fd);
3431 /* Just a bare device, no good to us */
3432 if (verbose)
3433 pr_err("ddf: Cannot create this array "
3434 "on device %s - a container is required.\n",
3435 dev);
3436 return 0;
3437 }
3438 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3439 if (verbose)
3440 pr_err("ddf: Cannot open %s: %s\n",
3441 dev, strerror(errno));
3442 return 0;
3443 }
3444 /* Well, it is in use by someone, maybe a 'ddf' container. */
3445 cfd = open_container(fd);
3446 if (cfd < 0) {
3447 close(fd);
3448 if (verbose)
3449 pr_err("ddf: Cannot use %s: %s\n",
3450 dev, strerror(EBUSY));
3451 return 0;
3452 }
3453 sra = sysfs_read(cfd, NULL, GET_VERSION);
3454 close(fd);
3455 if (sra && sra->array.major_version == -1 &&
3456 strcmp(sra->text_version, "ddf") == 0) {
3457 /* This is a member of a ddf container. Load the container
3458 * and try to create a bvd
3459 */
3460 struct ddf_super *ddf;
3461 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) {
3462 st->sb = ddf;
3463 strcpy(st->container_devnm, fd2devnm(cfd));
3464 close(cfd);
3465 return validate_geometry_ddf_bvd(st, level, layout,
3466 raiddisks, chunk, size,
3467 data_offset,
3468 dev, freesize,
3469 verbose);
3470 }
3471 close(cfd);
3472 } else /* device may belong to a different container */
3473 return 0;
3474
3475 return 1;
3476 }
3477
3478 static int
3479 validate_geometry_ddf_container(struct supertype *st,
3480 int level, int layout, int raiddisks,
3481 int chunk, unsigned long long size,
3482 unsigned long long data_offset,
3483 char *dev, unsigned long long *freesize,
3484 int verbose)
3485 {
3486 int fd;
3487 unsigned long long ldsize;
3488
3489 if (level != LEVEL_CONTAINER)
3490 return 0;
3491 if (!dev)
3492 return 1;
3493
3494 fd = open(dev, O_RDONLY|O_EXCL, 0);
3495 if (fd < 0) {
3496 if (verbose)
3497 pr_err("ddf: Cannot open %s: %s\n",
3498 dev, strerror(errno));
3499 return 0;
3500 }
3501 if (!get_dev_size(fd, dev, &ldsize)) {
3502 close(fd);
3503 return 0;
3504 }
3505 close(fd);
3506
3507 *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS);
3508 if (*freesize == 0)
3509 return 0;
3510
3511 return 1;
3512 }
3513
3514 static int validate_geometry_ddf_bvd(struct supertype *st,
3515 int level, int layout, int raiddisks,
3516 int *chunk, unsigned long long size,
3517 unsigned long long data_offset,
3518 char *dev, unsigned long long *freesize,
3519 int verbose)
3520 {
3521 struct stat stb;
3522 struct ddf_super *ddf = st->sb;
3523 struct dl *dl;
3524 unsigned long long maxsize;
3525 /* ddf/bvd supports lots of things, but not containers */
3526 if (level == LEVEL_CONTAINER) {
3527 if (verbose)
3528 pr_err("DDF cannot create a container within an container\n");
3529 return 0;
3530 }
3531 /* We must have the container info already read in. */
3532 if (!ddf)
3533 return 0;
3534
3535 if (!dev) {
3536 /* General test: make sure there is space for
3537 * 'raiddisks' device extents of size 'size'.
3538 */
3539 unsigned long long minsize = size;
3540 int dcnt = 0;
3541 if (minsize == 0)
3542 minsize = 8;
3543 for (dl = ddf->dlist; dl ; dl = dl->next) {
3544 if (find_space(ddf, dl, data_offset, &minsize)
3545 != INVALID_SECTORS)
3546 dcnt++;
3547 }
3548 if (dcnt < raiddisks) {
3549 if (verbose)
3550 pr_err("ddf: Not enough devices with "
3551 "space for this array (%d < %d)\n",
3552 dcnt, raiddisks);
3553 return 0;
3554 }
3555 return 1;
3556 }
3557 /* This device must be a member of the set */
3558 if (stat(dev, &stb) < 0)
3559 return 0;
3560 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3561 return 0;
3562 for (dl = ddf->dlist ; dl ; dl = dl->next) {
3563 if (dl->major == (int)major(stb.st_rdev) &&
3564 dl->minor == (int)minor(stb.st_rdev))
3565 break;
3566 }
3567 if (!dl) {
3568 if (verbose)
3569 pr_err("ddf: %s is not in the "
3570 "same DDF set\n",
3571 dev);
3572 return 0;
3573 }
3574 maxsize = ULLONG_MAX;
3575 find_space(ddf, dl, data_offset, &maxsize);
3576 *freesize = maxsize;
3577
3578 return 1;
3579 }
3580
3581 static int load_super_ddf_all(struct supertype *st, int fd,
3582 void **sbp, char *devname)
3583 {
3584 struct mdinfo *sra;
3585 struct ddf_super *super;
3586 struct mdinfo *sd, *best = NULL;
3587 int bestseq = 0;
3588 int seq;
3589 char nm[20];
3590 int dfd;
3591
3592 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
3593 if (!sra)
3594 return 1;
3595 if (sra->array.major_version != -1 ||
3596 sra->array.minor_version != -2 ||
3597 strcmp(sra->text_version, "ddf") != 0)
3598 return 1;
3599
3600 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
3601 return 1;
3602 memset(super, 0, sizeof(*super));
3603
3604 /* first, try each device, and choose the best ddf */
3605 for (sd = sra->devs ; sd ; sd = sd->next) {
3606 int rv;
3607 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3608 dfd = dev_open(nm, O_RDONLY);
3609 if (dfd < 0)
3610 return 2;
3611 rv = load_ddf_headers(dfd, super, NULL);
3612 close(dfd);
3613 if (rv == 0) {
3614 seq = be32_to_cpu(super->active->seq);
3615 if (super->active->openflag)
3616 seq--;
3617 if (!best || seq > bestseq) {
3618 bestseq = seq;
3619 best = sd;
3620 }
3621 }
3622 }
3623 if (!best)
3624 return 1;
3625 /* OK, load this ddf */
3626 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
3627 dfd = dev_open(nm, O_RDONLY);
3628 if (dfd < 0)
3629 return 1;
3630 load_ddf_headers(dfd, super, NULL);
3631 load_ddf_global(dfd, super, NULL);
3632 close(dfd);
3633 /* Now we need the device-local bits */
3634 for (sd = sra->devs ; sd ; sd = sd->next) {
3635 int rv;
3636
3637 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3638 dfd = dev_open(nm, O_RDWR);
3639 if (dfd < 0)
3640 return 2;
3641 rv = load_ddf_headers(dfd, super, NULL);
3642 if (rv == 0)
3643 rv = load_ddf_local(dfd, super, NULL, 1);
3644 if (rv)
3645 return 1;
3646 }
3647
3648 *sbp = super;
3649 if (st->ss == NULL) {
3650 st->ss = &super_ddf;
3651 st->minor_version = 0;
3652 st->max_devs = 512;
3653 }
3654 strcpy(st->container_devnm, fd2devnm(fd));
3655 return 0;
3656 }
3657
3658 static int load_container_ddf(struct supertype *st, int fd,
3659 char *devname)
3660 {
3661 return load_super_ddf_all(st, fd, &st->sb, devname);
3662 }
3663
3664 #endif /* MDASSEMBLE */
3665
3666 static int check_secondary(const struct vcl *vc)
3667 {
3668 const struct vd_config *conf = &vc->conf;
3669 int i;
3670
3671 /* The only DDF secondary RAID level md can support is
3672 * RAID 10, if the stripe sizes and Basic volume sizes
3673 * are all equal.
3674 * Other configurations could in theory be supported by exposing
3675 * the BVDs to user space and using device mapper for the secondary
3676 * mapping. So far we don't support that.
3677 */
3678
3679 __u64 sec_elements[4] = {0, 0, 0, 0};
3680 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3681 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3682
3683 if (vc->other_bvds == NULL) {
3684 pr_err("No BVDs for secondary RAID found\n");
3685 return -1;
3686 }
3687 if (conf->prl != DDF_RAID1) {
3688 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3689 return -1;
3690 }
3691 if (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED) {
3692 pr_err("Secondary RAID level %d is unsupported\n",
3693 conf->srl);
3694 return -1;
3695 }
3696 __set_sec_seen(conf->sec_elmnt_seq);
3697 for (i = 0; i < conf->sec_elmnt_count-1; i++) {
3698 const struct vd_config *bvd = vc->other_bvds[i];
3699 if (bvd->sec_elmnt_seq == DDF_UNUSED_BVD)
3700 continue;
3701 if (bvd->srl != conf->srl) {
3702 pr_err("Inconsistent secondary RAID level across BVDs\n");
3703 return -1;
3704 }
3705 if (bvd->prl != conf->prl) {
3706 pr_err("Different RAID levels for BVDs are unsupported\n");
3707 return -1;
3708 }
3709 if (!be16_eq(bvd->prim_elmnt_count, conf->prim_elmnt_count)) {
3710 pr_err("All BVDs must have the same number of primary elements\n");
3711 return -1;
3712 }
3713 if (bvd->chunk_shift != conf->chunk_shift) {
3714 pr_err("Different strip sizes for BVDs are unsupported\n");
3715 return -1;
3716 }
3717 if (!be64_eq(bvd->array_blocks, conf->array_blocks)) {
3718 pr_err("Different BVD sizes are unsupported\n");
3719 return -1;
3720 }
3721 __set_sec_seen(bvd->sec_elmnt_seq);
3722 }
3723 for (i = 0; i < conf->sec_elmnt_count; i++) {
3724 if (!__was_sec_seen(i)) {
3725 /* pr_err("BVD %d is missing\n", i); */
3726 return -1;
3727 }
3728 }
3729 return 0;
3730 }
3731
3732 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
3733 be32 refnum, unsigned int nmax,
3734 const struct vd_config **bvd,
3735 unsigned int *idx)
3736 {
3737 unsigned int i, j, n, sec, cnt;
3738
3739 cnt = be16_to_cpu(vc->conf.prim_elmnt_count);
3740 sec = (vc->conf.sec_elmnt_count == 1 ? 0 : vc->conf.sec_elmnt_seq);
3741
3742 for (i = 0, j = 0 ; i < nmax ; i++) {
3743 /* j counts valid entries for this BVD */
3744 if (be32_eq(vc->conf.phys_refnum[i], refnum)) {
3745 *bvd = &vc->conf;
3746 *idx = i;
3747 return sec * cnt + j;
3748 }
3749 if (be32_to_cpu(vc->conf.phys_refnum[i]) != 0xffffffff)
3750 j++;
3751 }
3752 if (vc->other_bvds == NULL)
3753 goto bad;
3754
3755 for (n = 1; n < vc->conf.sec_elmnt_count; n++) {
3756 struct vd_config *vd = vc->other_bvds[n-1];
3757 sec = vd->sec_elmnt_seq;
3758 if (sec == DDF_UNUSED_BVD)
3759 continue;
3760 for (i = 0, j = 0 ; i < nmax ; i++) {
3761 if (be32_eq(vd->phys_refnum[i], refnum)) {
3762 *bvd = vd;
3763 *idx = i;
3764 return sec * cnt + j;
3765 }
3766 if (be32_to_cpu(vd->phys_refnum[i]) != 0xffffffff)
3767 j++;
3768 }
3769 }
3770 bad:
3771 *bvd = NULL;
3772 return DDF_NOTFOUND;
3773 }
3774
3775 static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray)
3776 {
3777 /* Given a container loaded by load_super_ddf_all,
3778 * extract information about all the arrays into
3779 * an mdinfo tree.
3780 *
3781 * For each vcl in conflist: create an mdinfo, fill it in,
3782 * then look for matching devices (phys_refnum) in dlist
3783 * and create appropriate device mdinfo.
3784 */
3785 struct ddf_super *ddf = st->sb;
3786 struct mdinfo *rest = NULL;
3787 struct vcl *vc;
3788
3789 for (vc = ddf->conflist ; vc ; vc=vc->next) {
3790 unsigned int i;
3791 struct mdinfo *this;
3792 char *ep;
3793 __u32 *cptr;
3794 unsigned int pd;
3795
3796 if (subarray &&
3797 (strtoul(subarray, &ep, 10) != vc->vcnum ||
3798 *ep != '\0'))
3799 continue;
3800
3801 if (vc->conf.sec_elmnt_count > 1) {
3802 if (check_secondary(vc) != 0)
3803 continue;
3804 }
3805
3806 this = xcalloc(1, sizeof(*this));
3807 this->next = rest;
3808 rest = this;
3809
3810 if (layout_ddf2md(&vc->conf, &this->array))
3811 continue;
3812 this->array.md_minor = -1;
3813 this->array.major_version = -1;
3814 this->array.minor_version = -2;
3815 this->safe_mode_delay = DDF_SAFE_MODE_DELAY;
3816 cptr = (__u32 *)(vc->conf.guid + 16);
3817 this->array.ctime = DECADE + __be32_to_cpu(*cptr);
3818 this->array.utime = DECADE +
3819 be32_to_cpu(vc->conf.timestamp);
3820 this->array.chunk_size = 512 << vc->conf.chunk_shift;
3821
3822 i = vc->vcnum;
3823 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
3824 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
3825 DDF_init_full) {
3826 this->array.state = 0;
3827 this->resync_start = 0;
3828 } else {
3829 this->array.state = 1;
3830 this->resync_start = MaxSector;
3831 }
3832 _ddf_array_name(this->name, ddf, i);
3833 memset(this->uuid, 0, sizeof(this->uuid));
3834 this->component_size = be64_to_cpu(vc->conf.blocks);
3835 this->array.size = this->component_size / 2;
3836 this->container_member = i;
3837
3838 ddf->currentconf = vc;
3839 uuid_from_super_ddf(st, this->uuid);
3840 if (!subarray)
3841 ddf->currentconf = NULL;
3842
3843 sprintf(this->text_version, "/%s/%d",
3844 st->container_devnm, this->container_member);
3845
3846 for (pd = 0; pd < be16_to_cpu(ddf->phys->max_pdes); pd++) {
3847 struct mdinfo *dev;
3848 struct dl *d;
3849 const struct vd_config *bvd;
3850 unsigned int iphys;
3851 int stt;
3852
3853 if (be32_to_cpu(ddf->phys->entries[pd].refnum)
3854 == 0xFFFFFFFF)
3855 continue;
3856
3857 stt = be16_to_cpu(ddf->phys->entries[pd].state);
3858 if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3859 != DDF_Online)
3860 continue;
3861
3862 i = get_pd_index_from_refnum(
3863 vc, ddf->phys->entries[pd].refnum,
3864 ddf->mppe, &bvd, &iphys);
3865 if (i == DDF_NOTFOUND)
3866 continue;
3867
3868 this->array.working_disks++;
3869
3870 for (d = ddf->dlist; d ; d=d->next)
3871 if (be32_eq(d->disk.refnum,
3872 ddf->phys->entries[pd].refnum))
3873 break;
3874 if (d == NULL)
3875 /* Haven't found that one yet, maybe there are others */
3876 continue;
3877
3878 dev = xcalloc(1, sizeof(*dev));
3879 dev->next = this->devs;
3880 this->devs = dev;
3881
3882 dev->disk.number = be32_to_cpu(d->disk.refnum);
3883 dev->disk.major = d->major;
3884 dev->disk.minor = d->minor;
3885 dev->disk.raid_disk = i;
3886 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
3887 dev->recovery_start = MaxSector;
3888
3889 dev->events = be32_to_cpu(ddf->active->seq);
3890 dev->data_offset =
3891 be64_to_cpu(LBA_OFFSET(ddf, bvd)[iphys]);
3892 dev->component_size = be64_to_cpu(bvd->blocks);
3893 if (d->devname)
3894 strcpy(dev->name, d->devname);
3895 }
3896 }
3897 return rest;
3898 }
3899
3900 static int store_super_ddf(struct supertype *st, int fd)
3901 {
3902 struct ddf_super *ddf = st->sb;
3903 unsigned long long dsize;
3904 void *buf;
3905 int rc;
3906
3907 if (!ddf)
3908 return 1;
3909
3910 if (!get_dev_size(fd, NULL, &dsize))
3911 return 1;
3912
3913 if (ddf->dlist || ddf->conflist) {
3914 struct stat sta;
3915 struct dl *dl;
3916 int ofd, ret;
3917
3918 if (fstat(fd, &sta) == -1 || !S_ISBLK(sta.st_mode)) {
3919 pr_err("file descriptor for invalid device\n");
3920 return 1;
3921 }
3922 for (dl = ddf->dlist; dl; dl = dl->next)
3923 if (dl->major == (int)major(sta.st_rdev) &&
3924 dl->minor == (int)minor(sta.st_rdev))
3925 break;
3926 if (!dl) {
3927 pr_err("couldn't find disk %d/%d\n",
3928 (int)major(sta.st_rdev),
3929 (int)minor(sta.st_rdev));
3930 return 1;
3931 }
3932 ofd = dl->fd;
3933 dl->fd = fd;
3934 ret = (_write_super_to_disk(ddf, dl) != 1);
3935 dl->fd = ofd;
3936 return ret;
3937 }
3938
3939 if (posix_memalign(&buf, 512, 512) != 0)
3940 return 1;
3941 memset(buf, 0, 512);
3942
3943 lseek64(fd, dsize-512, 0);
3944 rc = write(fd, buf, 512);
3945 free(buf);
3946 if (rc < 0)
3947 return 1;
3948 return 0;
3949 }
3950
3951 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3952 {
3953 /*
3954 * return:
3955 * 0 same, or first was empty, and second was copied
3956 * 1 second had wrong magic number - but that isn't possible
3957 * 2 wrong uuid
3958 * 3 wrong other info
3959 */
3960 struct ddf_super *first = st->sb;
3961 struct ddf_super *second = tst->sb;
3962 struct dl *dl1, *dl2;
3963 struct vcl *vl1, *vl2;
3964 unsigned int max_vds, max_pds, pd, vd;
3965
3966 if (!first) {
3967 st->sb = tst->sb;
3968 tst->sb = NULL;
3969 return 0;
3970 }
3971
3972 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3973 return 2;
3974
3975 /* It is only OK to compare info in the anchor. Anything else
3976 * could be changing due to a reconfig so must be ignored.
3977 * guid really should be enough anyway.
3978 */
3979
3980 if (!be32_eq(first->active->seq, second->active->seq)) {
3981 dprintf("sequence number mismatch %u<->%u\n",
3982 be32_to_cpu(first->active->seq),
3983 be32_to_cpu(second->active->seq));
3984 return 0;
3985 }
3986
3987 /*
3988 * At this point we are fairly sure that the meta data matches.
3989 * But the new disk may contain additional local data.
3990 * Add it to the super block.
3991 */
3992 max_vds = be16_to_cpu(first->active->max_vd_entries);
3993 max_pds = be16_to_cpu(first->phys->max_pdes);
3994 for (vl2 = second->conflist; vl2; vl2 = vl2->next) {
3995 for (vl1 = first->conflist; vl1; vl1 = vl1->next)
3996 if (!memcmp(vl1->conf.guid, vl2->conf.guid,
3997 DDF_GUID_LEN))
3998 break;
3999 if (vl1) {
4000 if (vl1->other_bvds != NULL &&
4001 vl1->conf.sec_elmnt_seq !=
4002 vl2->conf.sec_elmnt_seq) {
4003 dprintf("adding BVD %u\n",
4004 vl2->conf.sec_elmnt_seq);
4005 add_other_bvd(vl1, &vl2->conf,
4006 first->conf_rec_len*512);
4007 }
4008 continue;
4009 }
4010
4011 if (posix_memalign((void **)&vl1, 512,
4012 (first->conf_rec_len*512 +
4013 offsetof(struct vcl, conf))) != 0) {
4014 pr_err("could not allocate vcl buf\n");
4015 return 3;
4016 }
4017
4018 vl1->next = first->conflist;
4019 vl1->block_sizes = NULL;
4020 memcpy(&vl1->conf, &vl2->conf, first->conf_rec_len*512);
4021 if (alloc_other_bvds(first, vl1) != 0) {
4022 pr_err("could not allocate other bvds\n");
4023 free(vl1);
4024 return 3;
4025 }
4026 for (vd = 0; vd < max_vds; vd++)
4027 if (!memcmp(first->virt->entries[vd].guid,
4028 vl1->conf.guid, DDF_GUID_LEN))
4029 break;
4030 vl1->vcnum = vd;
4031 dprintf("added config for VD %u\n", vl1->vcnum);
4032 first->conflist = vl1;
4033 }
4034
4035 for (dl2 = second->dlist; dl2; dl2 = dl2->next) {
4036 for (dl1 = first->dlist; dl1; dl1 = dl1->next)
4037 if (be32_eq(dl1->disk.refnum, dl2->disk.refnum))
4038 break;
4039 if (dl1)
4040 continue;
4041
4042 if (posix_memalign((void **)&dl1, 512,
4043 sizeof(*dl1) + (first->max_part) * sizeof(dl1->vlist[0]))
4044 != 0) {
4045 pr_err("could not allocate disk info buffer\n");
4046 return 3;
4047 }
4048 memcpy(dl1, dl2, sizeof(*dl1));
4049 dl1->mdupdate = NULL;
4050 dl1->next = first->dlist;
4051 dl1->fd = -1;
4052 for (pd = 0; pd < max_pds; pd++)
4053 if (be32_eq(first->phys->entries[pd].refnum,
4054 dl1->disk.refnum))
4055 break;
4056 dl1->pdnum = pd < max_pds ? (int)pd : -1;
4057 if (dl2->spare) {
4058 if (posix_memalign((void **)&dl1->spare, 512,
4059 first->conf_rec_len*512) != 0) {
4060 pr_err("could not allocate spare info buf\n");
4061 return 3;
4062 }
4063 memcpy(dl1->spare, dl2->spare, first->conf_rec_len*512);
4064 }
4065 for (vd = 0 ; vd < first->max_part ; vd++) {
4066 if (!dl2->vlist[vd]) {
4067 dl1->vlist[vd] = NULL;
4068 continue;
4069 }
4070 for (vl1 = first->conflist; vl1; vl1 = vl1->next) {
4071 if (!memcmp(vl1->conf.guid,
4072 dl2->vlist[vd]->conf.guid,
4073 DDF_GUID_LEN))
4074 break;
4075 dl1->vlist[vd] = vl1;
4076 }
4077 }
4078 first->dlist = dl1;
4079 dprintf("added disk %d: %08x\n", dl1->pdnum,
4080 be32_to_cpu(dl1->disk.refnum));
4081 }
4082
4083 return 0;
4084 }
4085
4086 #ifndef MDASSEMBLE
4087 /*
4088 * A new array 'a' has been started which claims to be instance 'inst'
4089 * within container 'c'.
4090 * We need to confirm that the array matches the metadata in 'c' so
4091 * that we don't corrupt any metadata.
4092 */
4093 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
4094 {
4095 struct ddf_super *ddf = c->sb;
4096 int n = atoi(inst);
4097 struct mdinfo *dev;
4098 struct dl *dl;
4099 static const char faulty[] = "faulty";
4100
4101 if (all_ff(ddf->virt->entries[n].guid)) {
4102 pr_err("subarray %d doesn't exist\n", n);
4103 return -ENODEV;
4104 }
4105 dprintf("new subarray %d, GUID: %s\n", n,
4106 guid_str(ddf->virt->entries[n].guid));
4107 for (dev = a->info.devs; dev; dev = dev->next) {
4108 for (dl = ddf->dlist; dl; dl = dl->next)
4109 if (dl->major == dev->disk.major &&
4110 dl->minor == dev->disk.minor)
4111 break;
4112 if (!dl || dl->pdnum < 0) {
4113 pr_err("device %d/%d of subarray %d not found in meta data\n",
4114 dev->disk.major, dev->disk.minor, n);
4115 return -1;
4116 }
4117 if ((be16_to_cpu(ddf->phys->entries[dl->pdnum].state) &
4118 (DDF_Online|DDF_Missing|DDF_Failed)) != DDF_Online) {
4119 pr_err("new subarray %d contains broken device %d/%d (%02x)\n",
4120 n, dl->major, dl->minor,
4121 be16_to_cpu(ddf->phys->entries[dl->pdnum].state));
4122 if (write(dev->state_fd, faulty, sizeof(faulty)-1) !=
4123 sizeof(faulty) - 1)
4124 pr_err("Write to state_fd failed\n");
4125 dev->curr_state = DS_FAULTY;
4126 }
4127 }
4128 a->info.container_member = n;
4129 return 0;
4130 }
4131
4132 static void handle_missing(struct ddf_super *ddf, struct active_array *a, int inst)
4133 {
4134 /* This member array is being activated. If any devices
4135 * are missing they must now be marked as failed.
4136 */
4137 struct vd_config *vc;
4138 unsigned int n_bvd;
4139 struct vcl *vcl;
4140 struct dl *dl;
4141 int pd;
4142 int n;
4143 int state;
4144
4145 for (n = 0; ; n++) {
4146 vc = find_vdcr(ddf, inst, n, &n_bvd, &vcl);
4147 if (!vc)
4148 break;
4149 for (dl = ddf->dlist; dl; dl = dl->next)
4150 if (be32_eq(dl->disk.refnum, vc->phys_refnum[n_bvd]))
4151 break;
4152 if (dl)
4153 /* Found this disk, so not missing */
4154 continue;
4155
4156 /* Mark the device as failed/missing. */
4157 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4158 if (pd >= 0 && be16_and(ddf->phys->entries[pd].state,
4159 cpu_to_be16(DDF_Online))) {
4160 be16_clear(ddf->phys->entries[pd].state,
4161 cpu_to_be16(DDF_Online));
4162 be16_set(ddf->phys->entries[pd].state,
4163 cpu_to_be16(DDF_Failed|DDF_Missing));
4164 vc->phys_refnum[n_bvd] = cpu_to_be32(0);
4165 ddf_set_updates_pending(ddf, vc);
4166 }
4167
4168 /* Mark the array as Degraded */
4169 state = get_svd_state(ddf, vcl);
4170 if (ddf->virt->entries[inst].state !=
4171 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4172 | state)) {
4173 ddf->virt->entries[inst].state =
4174 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4175 | state;
4176 a->check_degraded = 1;
4177 ddf_set_updates_pending(ddf, vc);
4178 }
4179 }
4180 }
4181
4182 /*
4183 * The array 'a' is to be marked clean in the metadata.
4184 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4185 * clean up to the point (in sectors). If that cannot be recorded in the
4186 * metadata, then leave it as dirty.
4187 *
4188 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4189 * !global! virtual_disk.virtual_entry structure.
4190 */
4191 static int ddf_set_array_state(struct active_array *a, int consistent)
4192 {
4193 struct ddf_super *ddf = a->container->sb;
4194 int inst = a->info.container_member;
4195 int old = ddf->virt->entries[inst].state;
4196 if (consistent == 2) {
4197 handle_missing(ddf, a, inst);
4198 consistent = 1;
4199 if (!is_resync_complete(&a->info))
4200 consistent = 0;
4201 }
4202 if (consistent)
4203 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
4204 else
4205 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
4206 if (old != ddf->virt->entries[inst].state)
4207 ddf_set_updates_pending(ddf, NULL);
4208
4209 old = ddf->virt->entries[inst].init_state;
4210 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
4211 if (is_resync_complete(&a->info))
4212 ddf->virt->entries[inst].init_state |= DDF_init_full;
4213 else if (a->info.resync_start == 0)
4214 ddf->virt->entries[inst].init_state |= DDF_init_not;
4215 else
4216 ddf->virt->entries[inst].init_state |= DDF_init_quick;
4217 if (old != ddf->virt->entries[inst].init_state)
4218 ddf_set_updates_pending(ddf, NULL);
4219
4220 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst,
4221 guid_str(ddf->virt->entries[inst].guid), a->curr_state,
4222 consistent?"clean":"dirty",
4223 a->info.resync_start);
4224 return consistent;
4225 }
4226
4227 static int get_bvd_state(const struct ddf_super *ddf,
4228 const struct vd_config *vc)
4229 {
4230 unsigned int i, n_bvd, working = 0;
4231 unsigned int n_prim = be16_to_cpu(vc->prim_elmnt_count);
4232 int pd, st, state;
4233 char *avail = xcalloc(1, n_prim);
4234 mdu_array_info_t array;
4235
4236 layout_ddf2md(vc, &array);
4237
4238 for (i = 0; i < n_prim; i++) {
4239 if (!find_index_in_bvd(ddf, vc, i, &n_bvd))
4240 continue;
4241 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4242 if (pd < 0)
4243 continue;
4244 st = be16_to_cpu(ddf->phys->entries[pd].state);
4245 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
4246 == DDF_Online) {
4247 working++;
4248 avail[i] = 1;
4249 }
4250 }
4251
4252 state = DDF_state_degraded;
4253 if (working == n_prim)
4254 state = DDF_state_optimal;
4255 else
4256 switch (vc->prl) {
4257 case DDF_RAID0:
4258 case DDF_CONCAT:
4259 case DDF_JBOD:
4260 state = DDF_state_failed;
4261 break;
4262 case DDF_RAID1:
4263 if (working == 0)
4264 state = DDF_state_failed;
4265 else if (working >= 2)
4266 state = DDF_state_part_optimal;
4267 break;
4268 case DDF_RAID1E:
4269 if (!enough(10, n_prim, array.layout, 1, avail))
4270 state = DDF_state_failed;
4271 break;
4272 case DDF_RAID4:
4273 case DDF_RAID5:
4274 if (working < n_prim - 1)
4275 state = DDF_state_failed;
4276 break;
4277 case DDF_RAID6:
4278 if (working < n_prim - 2)
4279 state = DDF_state_failed;
4280 else if (working == n_prim - 1)
4281 state = DDF_state_part_optimal;
4282 break;
4283 }
4284 return state;
4285 }
4286
4287 static int secondary_state(int state, int other, int seclevel)
4288 {
4289 if (state == DDF_state_optimal && other == DDF_state_optimal)
4290 return DDF_state_optimal;
4291 if (seclevel == DDF_2MIRRORED) {
4292 if (state == DDF_state_optimal || other == DDF_state_optimal)
4293 return DDF_state_part_optimal;
4294 if (state == DDF_state_failed && other == DDF_state_failed)
4295 return DDF_state_failed;
4296 return DDF_state_degraded;
4297 } else {
4298 if (state == DDF_state_failed || other == DDF_state_failed)
4299 return DDF_state_failed;
4300 if (state == DDF_state_degraded || other == DDF_state_degraded)
4301 return DDF_state_degraded;
4302 return DDF_state_part_optimal;
4303 }
4304 }
4305
4306 static int get_svd_state(const struct ddf_super *ddf, const struct vcl *vcl)
4307 {
4308 int state = get_bvd_state(ddf, &vcl->conf);
4309 unsigned int i;
4310 for (i = 1; i < vcl->conf.sec_elmnt_count; i++) {
4311 state = secondary_state(
4312 state,
4313 get_bvd_state(ddf, vcl->other_bvds[i-1]),
4314 vcl->conf.srl);
4315 }
4316 return state;
4317 }
4318
4319 /*
4320 * The state of each disk is stored in the global phys_disk structure
4321 * in phys_disk.entries[n].state.
4322 * This makes various combinations awkward.
4323 * - When a device fails in any array, it must be failed in all arrays
4324 * that include a part of this device.
4325 * - When a component is rebuilding, we cannot include it officially in the
4326 * array unless this is the only array that uses the device.
4327 *
4328 * So: when transitioning:
4329 * Online -> failed, just set failed flag. monitor will propagate
4330 * spare -> online, the device might need to be added to the array.
4331 * spare -> failed, just set failed. Don't worry if in array or not.
4332 */
4333 static void ddf_set_disk(struct active_array *a, int n, int state)
4334 {
4335 struct ddf_super *ddf = a->container->sb;
4336 unsigned int inst = a->info.container_member, n_bvd;
4337 struct vcl *vcl;
4338 struct vd_config *vc = find_vdcr(ddf, inst, (unsigned int)n,
4339 &n_bvd, &vcl);
4340 int pd;
4341 struct mdinfo *mdi;
4342 struct dl *dl;
4343 int update = 0;
4344
4345 dprintf("%d to %x\n", n, state);
4346 if (vc == NULL) {
4347 dprintf("ddf: cannot find instance %d!!\n", inst);
4348 return;
4349 }
4350 /* Find the matching slot in 'info'. */
4351 for (mdi = a->info.devs; mdi; mdi = mdi->next)
4352 if (mdi->disk.raid_disk == n)
4353 break;
4354 if (!mdi) {
4355 pr_err("cannot find raid disk %d\n", n);
4356 return;
4357 }
4358
4359 /* and find the 'dl' entry corresponding to that. */
4360 for (dl = ddf->dlist; dl; dl = dl->next)
4361 if (mdi->state_fd >= 0 &&
4362 mdi->disk.major == dl->major &&
4363 mdi->disk.minor == dl->minor)
4364 break;
4365 if (!dl) {
4366 pr_err("cannot find raid disk %d (%d/%d)\n",
4367 n, mdi->disk.major, mdi->disk.minor);
4368 return;
4369 }
4370
4371 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4372 if (pd < 0 || pd != dl->pdnum) {
4373 /* disk doesn't currently exist or has changed.
4374 * If it is now in_sync, insert it. */
4375 dprintf("phys disk not found for %d: %d/%d ref %08x\n",
4376 dl->pdnum, dl->major, dl->minor,
4377 be32_to_cpu(dl->disk.refnum));
4378 dprintf("array %u disk %u ref %08x pd %d\n",
4379 inst, n_bvd,
4380 be32_to_cpu(vc->phys_refnum[n_bvd]), pd);
4381 if ((state & DS_INSYNC) && ! (state & DS_FAULTY) &&
4382 dl->pdnum >= 0) {
4383 pd = dl->pdnum;
4384 vc->phys_refnum[n_bvd] = dl->disk.refnum;
4385 LBA_OFFSET(ddf, vc)[n_bvd] =
4386 cpu_to_be64(mdi->data_offset);
4387 be16_clear(ddf->phys->entries[pd].type,
4388 cpu_to_be16(DDF_Global_Spare));
4389 be16_set(ddf->phys->entries[pd].type,
4390 cpu_to_be16(DDF_Active_in_VD));
4391 update = 1;
4392 }
4393 } else {
4394 be16 old = ddf->phys->entries[pd].state;
4395 if (state & DS_FAULTY)
4396 be16_set(ddf->phys->entries[pd].state,
4397 cpu_to_be16(DDF_Failed));
4398 if (state & DS_INSYNC) {
4399 be16_set(ddf->phys->entries[pd].state,
4400 cpu_to_be16(DDF_Online));
4401 be16_clear(ddf->phys->entries[pd].state,
4402 cpu_to_be16(DDF_Rebuilding));
4403 }
4404 if (!be16_eq(old, ddf->phys->entries[pd].state))
4405 update = 1;
4406 }
4407
4408 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n,
4409 be32_to_cpu(dl->disk.refnum), state,
4410 be16_to_cpu(ddf->phys->entries[pd].state));
4411
4412 /* Now we need to check the state of the array and update
4413 * virtual_disk.entries[n].state.
4414 * It needs to be one of "optimal", "degraded", "failed".
4415 * I don't understand 'deleted' or 'missing'.
4416 */
4417 state = get_svd_state(ddf, vcl);
4418
4419 if (ddf->virt->entries[inst].state !=
4420 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4421 | state)) {
4422 ddf->virt->entries[inst].state =
4423 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4424 | state;
4425 update = 1;
4426 }
4427 if (update)
4428 ddf_set_updates_pending(ddf, vc);
4429 }
4430
4431 static void ddf_sync_metadata(struct supertype *st)
4432 {
4433 /*
4434 * Write all data to all devices.
4435 * Later, we might be able to track whether only local changes
4436 * have been made, or whether any global data has been changed,
4437 * but ddf is sufficiently weird that it probably always
4438 * changes global data ....
4439 */
4440 struct ddf_super *ddf = st->sb;
4441 if (!ddf->updates_pending)
4442 return;
4443 ddf->updates_pending = 0;
4444 __write_init_super_ddf(st);
4445 dprintf("ddf: sync_metadata\n");
4446 }
4447
4448 static int del_from_conflist(struct vcl **list, const char *guid)
4449 {
4450 struct vcl **p;
4451 int found = 0;
4452 for (p = list; p && *p; p = &((*p)->next))
4453 if (!memcmp((*p)->conf.guid, guid, DDF_GUID_LEN)) {
4454 found = 1;
4455 *p = (*p)->next;
4456 }
4457 return found;
4458 }
4459
4460 static int _kill_subarray_ddf(struct ddf_super *ddf, const char *guid)
4461 {
4462 struct dl *dl;
4463 unsigned int vdnum, i;
4464 vdnum = find_vde_by_guid(ddf, guid);
4465 if (vdnum == DDF_NOTFOUND) {
4466 pr_err("could not find VD %s\n", guid_str(guid));
4467 return -1;
4468 }
4469 if (del_from_conflist(&ddf->conflist, guid) == 0) {
4470 pr_err("could not find conf %s\n", guid_str(guid));
4471 return -1;
4472 }
4473 for (dl = ddf->dlist; dl; dl = dl->next)
4474 for (i = 0; i < ddf->max_part; i++)
4475 if (dl->vlist[i] != NULL &&
4476 !memcmp(dl->vlist[i]->conf.guid, guid,
4477 DDF_GUID_LEN))
4478 dl->vlist[i] = NULL;
4479 memset(ddf->virt->entries[vdnum].guid, 0xff, DDF_GUID_LEN);
4480 dprintf("deleted %s\n", guid_str(guid));
4481 return 0;
4482 }
4483
4484 static int kill_subarray_ddf(struct supertype *st)
4485 {
4486 struct ddf_super *ddf = st->sb;
4487 /*
4488 * currentconf is set in container_content_ddf,
4489 * called with subarray arg
4490 */
4491 struct vcl *victim = ddf->currentconf;
4492 struct vd_config *conf;
4493 unsigned int vdnum;
4494
4495 ddf->currentconf = NULL;
4496 if (!victim) {
4497 pr_err("nothing to kill\n");
4498 return -1;
4499 }
4500 conf = &victim->conf;
4501 vdnum = find_vde_by_guid(ddf, conf->guid);
4502 if (vdnum == DDF_NOTFOUND) {
4503 pr_err("could not find VD %s\n", guid_str(conf->guid));
4504 return -1;
4505 }
4506 if (st->update_tail) {
4507 struct virtual_disk *vd;
4508 int len = sizeof(struct virtual_disk)
4509 + sizeof(struct virtual_entry);
4510 vd = xmalloc(len);
4511 if (vd == NULL) {
4512 pr_err("failed to allocate %d bytes\n", len);
4513 return -1;
4514 }
4515 memset(vd, 0 , len);
4516 vd->magic = DDF_VIRT_RECORDS_MAGIC;
4517 vd->populated_vdes = cpu_to_be16(0);
4518 memcpy(vd->entries[0].guid, conf->guid, DDF_GUID_LEN);
4519 /* we use DDF_state_deleted as marker */
4520 vd->entries[0].state = DDF_state_deleted;
4521 append_metadata_update(st, vd, len);
4522 } else {
4523 _kill_subarray_ddf(ddf, conf->guid);
4524 ddf_set_updates_pending(ddf, NULL);
4525 ddf_sync_metadata(st);
4526 }
4527 return 0;
4528 }
4529
4530 static void copy_matching_bvd(struct ddf_super *ddf,
4531 struct vd_config *conf,
4532 const struct metadata_update *update)
4533 {
4534 unsigned int mppe =
4535 be16_to_cpu(ddf->anchor.max_primary_element_entries);
4536 unsigned int len = ddf->conf_rec_len * 512;
4537 char *p;
4538 struct vd_config *vc;
4539 for (p = update->buf; p < update->buf + update->len; p += len) {
4540 vc = (struct vd_config *) p;
4541 if (vc->sec_elmnt_seq == conf->sec_elmnt_seq) {
4542 memcpy(conf->phys_refnum, vc->phys_refnum,
4543 mppe * (sizeof(__u32) + sizeof(__u64)));
4544 return;
4545 }
4546 }
4547 pr_err("no match for BVD %d of %s in update\n",
4548 conf->sec_elmnt_seq, guid_str(conf->guid));
4549 }
4550
4551 static void ddf_process_phys_update(struct supertype *st,
4552 struct metadata_update *update)
4553 {
4554 struct ddf_super *ddf = st->sb;
4555 struct phys_disk *pd;
4556 unsigned int ent;
4557
4558 pd = (struct phys_disk*)update->buf;
4559 ent = be16_to_cpu(pd->used_pdes);
4560 if (ent >= be16_to_cpu(ddf->phys->max_pdes))
4561 return;
4562 if (be16_and(pd->entries[0].state, cpu_to_be16(DDF_Missing))) {
4563 struct dl **dlp;
4564 /* removing this disk. */
4565 be16_set(ddf->phys->entries[ent].state,
4566 cpu_to_be16(DDF_Missing));
4567 for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) {
4568 struct dl *dl = *dlp;
4569 if (dl->pdnum == (signed)ent) {
4570 close(dl->fd);
4571 dl->fd = -1;
4572 *dlp = dl->next;
4573 update->space = dl->devname;
4574 *(void**)dl = update->space_list;
4575 update->space_list = (void**)dl;
4576 break;
4577 }
4578 }
4579 ddf_set_updates_pending(ddf, NULL);
4580 return;
4581 }
4582 if (!all_ff(ddf->phys->entries[ent].guid))
4583 return;
4584 ddf->phys->entries[ent] = pd->entries[0];
4585 ddf->phys->used_pdes = cpu_to_be16
4586 (1 + be16_to_cpu(ddf->phys->used_pdes));
4587 ddf_set_updates_pending(ddf, NULL);
4588 if (ddf->add_list) {
4589 struct active_array *a;
4590 struct dl *al = ddf->add_list;
4591 ddf->add_list = al->next;
4592
4593 al->next = ddf->dlist;
4594 ddf->dlist = al;
4595
4596 /* As a device has been added, we should check
4597 * for any degraded devices that might make
4598 * use of this spare */
4599 for (a = st->arrays ; a; a=a->next)
4600 a->check_degraded = 1;
4601 }
4602 }
4603
4604 static void ddf_process_virt_update(struct supertype *st,
4605 struct metadata_update *update)
4606 {
4607 struct ddf_super *ddf = st->sb;
4608 struct virtual_disk *vd;
4609 unsigned int ent;
4610
4611 vd = (struct virtual_disk*)update->buf;
4612
4613 if (vd->entries[0].state == DDF_state_deleted) {
4614 if (_kill_subarray_ddf(ddf, vd->entries[0].guid))
4615 return;
4616 } else {
4617 ent = find_vde_by_guid(ddf, vd->entries[0].guid);
4618 if (ent != DDF_NOTFOUND) {
4619 dprintf("VD %s exists already in slot %d\n",
4620 guid_str(vd->entries[0].guid),
4621 ent);
4622 return;
4623 }
4624 ent = find_unused_vde(ddf);
4625 if (ent == DDF_NOTFOUND)
4626 return;
4627 ddf->virt->entries[ent] = vd->entries[0];
4628 ddf->virt->populated_vdes =
4629 cpu_to_be16(
4630 1 + be16_to_cpu(
4631 ddf->virt->populated_vdes));
4632 dprintf("added VD %s in slot %d(s=%02x i=%02x)\n",
4633 guid_str(vd->entries[0].guid), ent,
4634 ddf->virt->entries[ent].state,
4635 ddf->virt->entries[ent].init_state);
4636 }
4637 ddf_set_updates_pending(ddf, NULL);
4638 }
4639
4640 static void ddf_remove_failed(struct ddf_super *ddf)
4641 {
4642 /* Now remove any 'Failed' devices that are not part
4643 * of any VD. They will have the Transition flag set.
4644 * Once done, we need to update all dl->pdnum numbers.
4645 */
4646 unsigned int pdnum;
4647 unsigned int pd2 = 0;
4648 struct dl *dl;
4649
4650 for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4651 pdnum++) {
4652 if (be32_to_cpu(ddf->phys->entries[pdnum].refnum) ==
4653 0xFFFFFFFF)
4654 continue;
4655 if (be16_and(ddf->phys->entries[pdnum].state,
4656 cpu_to_be16(DDF_Failed))
4657 && be16_and(ddf->phys->entries[pdnum].state,
4658 cpu_to_be16(DDF_Transition))) {
4659 /* skip this one unless in dlist*/
4660 for (dl = ddf->dlist; dl; dl = dl->next)
4661 if (dl->pdnum == (int)pdnum)
4662 break;
4663 if (!dl)
4664 continue;
4665 }
4666 if (pdnum == pd2)
4667 pd2++;
4668 else {
4669 ddf->phys->entries[pd2] =
4670 ddf->phys->entries[pdnum];
4671 for (dl = ddf->dlist; dl; dl = dl->next)
4672 if (dl->pdnum == (int)pdnum)
4673 dl->pdnum = pd2;
4674 pd2++;
4675 }
4676 }
4677 ddf->phys->used_pdes = cpu_to_be16(pd2);
4678 while (pd2 < pdnum) {
4679 memset(ddf->phys->entries[pd2].guid, 0xff,
4680 DDF_GUID_LEN);
4681 pd2++;
4682 }
4683 }
4684
4685 static void ddf_update_vlist(struct ddf_super *ddf, struct dl *dl)
4686 {
4687 struct vcl *vcl;
4688 unsigned int vn = 0;
4689 int in_degraded = 0;
4690
4691 if (dl->pdnum < 0)
4692 return;
4693 for (vcl = ddf->conflist; vcl ; vcl = vcl->next) {
4694 unsigned int dn, ibvd;
4695 const struct vd_config *conf;
4696 int vstate;
4697 dn = get_pd_index_from_refnum(vcl,
4698 dl->disk.refnum,
4699 ddf->mppe,
4700 &conf, &ibvd);
4701 if (dn == DDF_NOTFOUND)
4702 continue;
4703 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4704 dl->pdnum,
4705 be32_to_cpu(dl->disk.refnum),
4706 guid_str(conf->guid),
4707 conf->sec_elmnt_seq, vn);
4708 /* Clear the Transition flag */
4709 if (be16_and
4710 (ddf->phys->entries[dl->pdnum].state,
4711 cpu_to_be16(DDF_Failed)))
4712 be16_clear(ddf->phys
4713 ->entries[dl->pdnum].state,
4714 cpu_to_be16(DDF_Transition));
4715 dl->vlist[vn++] = vcl;
4716 vstate = ddf->virt->entries[vcl->vcnum].state
4717 & DDF_state_mask;
4718 if (vstate == DDF_state_degraded ||
4719 vstate == DDF_state_part_optimal)
4720 in_degraded = 1;
4721 }
4722 while (vn < ddf->max_part)
4723 dl->vlist[vn++] = NULL;
4724 if (dl->vlist[0]) {
4725 be16_clear(ddf->phys->entries[dl->pdnum].type,
4726 cpu_to_be16(DDF_Global_Spare));
4727 if (!be16_and(ddf->phys
4728 ->entries[dl->pdnum].type,
4729 cpu_to_be16(DDF_Active_in_VD))) {
4730 be16_set(ddf->phys
4731 ->entries[dl->pdnum].type,
4732 cpu_to_be16(DDF_Active_in_VD));
4733 if (in_degraded)
4734 be16_set(ddf->phys
4735 ->entries[dl->pdnum]
4736 .state,
4737 cpu_to_be16
4738 (DDF_Rebuilding));
4739 }
4740 }
4741 if (dl->spare) {
4742 be16_clear(ddf->phys->entries[dl->pdnum].type,
4743 cpu_to_be16(DDF_Global_Spare));
4744 be16_set(ddf->phys->entries[dl->pdnum].type,
4745 cpu_to_be16(DDF_Spare));
4746 }
4747 if (!dl->vlist[0] && !dl->spare) {
4748 be16_set(ddf->phys->entries[dl->pdnum].type,
4749 cpu_to_be16(DDF_Global_Spare));
4750 be16_clear(ddf->phys->entries[dl->pdnum].type,
4751 cpu_to_be16(DDF_Spare));
4752 be16_clear(ddf->phys->entries[dl->pdnum].type,
4753 cpu_to_be16(DDF_Active_in_VD));
4754 }
4755 }
4756
4757 static void ddf_process_conf_update(struct supertype *st,
4758 struct metadata_update *update)
4759 {
4760 struct ddf_super *ddf = st->sb;
4761 struct vd_config *vc;
4762 struct vcl *vcl;
4763 struct dl *dl;
4764 unsigned int ent;
4765 unsigned int pdnum, len;
4766
4767 vc = (struct vd_config*)update->buf;
4768 len = ddf->conf_rec_len * 512;
4769 if ((unsigned int)update->len != len * vc->sec_elmnt_count) {
4770 pr_err("%s: insufficient data (%d) for %u BVDs\n",
4771 guid_str(vc->guid), update->len,
4772 vc->sec_elmnt_count);
4773 return;
4774 }
4775 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
4776 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
4777 break;
4778 dprintf("conf update for %s (%s)\n",
4779 guid_str(vc->guid), (vcl ? "old" : "new"));
4780 if (vcl) {
4781 /* An update, just copy the phys_refnum and lba_offset
4782 * fields
4783 */
4784 unsigned int i;
4785 unsigned int k;
4786 copy_matching_bvd(ddf, &vcl->conf, update);
4787 for (k = 0; k < be16_to_cpu(vc->prim_elmnt_count); k++)
4788 dprintf("BVD %u has %08x at %llu\n", 0,
4789 be32_to_cpu(vcl->conf.phys_refnum[k]),
4790 be64_to_cpu(LBA_OFFSET(ddf,
4791 &vcl->conf)[k]));
4792 for (i = 1; i < vc->sec_elmnt_count; i++) {
4793 copy_matching_bvd(ddf, vcl->other_bvds[i-1],
4794 update);
4795 for (k = 0; k < be16_to_cpu(
4796 vc->prim_elmnt_count); k++)
4797 dprintf("BVD %u has %08x at %llu\n", i,
4798 be32_to_cpu
4799 (vcl->other_bvds[i-1]->
4800 phys_refnum[k]),
4801 be64_to_cpu
4802 (LBA_OFFSET
4803 (ddf,
4804 vcl->other_bvds[i-1])[k]));
4805 }
4806 } else {
4807 /* A new VD_CONF */
4808 unsigned int i;
4809 if (!update->space)
4810 return;
4811 vcl = update->space;
4812 update->space = NULL;
4813 vcl->next = ddf->conflist;
4814 memcpy(&vcl->conf, vc, len);
4815 ent = find_vde_by_guid(ddf, vc->guid);
4816 if (ent == DDF_NOTFOUND)
4817 return;
4818 vcl->vcnum = ent;
4819 ddf->conflist = vcl;
4820 for (i = 1; i < vc->sec_elmnt_count; i++)
4821 memcpy(vcl->other_bvds[i-1],
4822 update->buf + len * i, len);
4823 }
4824 /* Set DDF_Transition on all Failed devices - to help
4825 * us detect those that are no longer in use
4826 */
4827 for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4828 pdnum++)
4829 if (be16_and(ddf->phys->entries[pdnum].state,
4830 cpu_to_be16(DDF_Failed)))
4831 be16_set(ddf->phys->entries[pdnum].state,
4832 cpu_to_be16(DDF_Transition));
4833
4834 /* Now make sure vlist is correct for each dl. */
4835 for (dl = ddf->dlist; dl; dl = dl->next)
4836 ddf_update_vlist(ddf, dl);
4837 ddf_remove_failed(ddf);
4838
4839 ddf_set_updates_pending(ddf, vc);
4840 }
4841
4842 static void ddf_process_update(struct supertype *st,
4843 struct metadata_update *update)
4844 {
4845 /* Apply this update to the metadata.
4846 * The first 4 bytes are a DDF_*_MAGIC which guides
4847 * our actions.
4848 * Possible update are:
4849 * DDF_PHYS_RECORDS_MAGIC
4850 * Add a new physical device or remove an old one.
4851 * Changes to this record only happen implicitly.
4852 * used_pdes is the device number.
4853 * DDF_VIRT_RECORDS_MAGIC
4854 * Add a new VD. Possibly also change the 'access' bits.
4855 * populated_vdes is the entry number.
4856 * DDF_VD_CONF_MAGIC
4857 * New or updated VD. the VIRT_RECORD must already
4858 * exist. For an update, phys_refnum and lba_offset
4859 * (at least) are updated, and the VD_CONF must
4860 * be written to precisely those devices listed with
4861 * a phys_refnum.
4862 * DDF_SPARE_ASSIGN_MAGIC
4863 * replacement Spare Assignment Record... but for which device?
4864 *
4865 * So, e.g.:
4866 * - to create a new array, we send a VIRT_RECORD and
4867 * a VD_CONF. Then assemble and start the array.
4868 * - to activate a spare we send a VD_CONF to add the phys_refnum
4869 * and offset. This will also mark the spare as active with
4870 * a spare-assignment record.
4871 */
4872 be32 *magic = (be32 *)update->buf;
4873
4874 dprintf("Process update %x\n", be32_to_cpu(*magic));
4875
4876 if (be32_eq(*magic, DDF_PHYS_RECORDS_MAGIC)) {
4877 if (update->len == (sizeof(struct phys_disk) +
4878 sizeof(struct phys_disk_entry)))
4879 ddf_process_phys_update(st, update);
4880 } else if (be32_eq(*magic, DDF_VIRT_RECORDS_MAGIC)) {
4881 if (update->len == (sizeof(struct virtual_disk) +
4882 sizeof(struct virtual_entry)))
4883 ddf_process_virt_update(st, update);
4884 } else if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4885 ddf_process_conf_update(st, update);
4886 }
4887 /* case DDF_SPARE_ASSIGN_MAGIC */
4888 }
4889
4890 static int ddf_prepare_update(struct supertype *st,
4891 struct metadata_update *update)
4892 {
4893 /* This update arrived at managemon.
4894 * We are about to pass it to monitor.
4895 * If a malloc is needed, do it here.
4896 */
4897 struct ddf_super *ddf = st->sb;
4898 be32 *magic;
4899 if (update->len < 4)
4900 return 0;
4901 magic = (be32 *)update->buf;
4902 if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4903 struct vcl *vcl;
4904 struct vd_config *conf;
4905 if (update->len < (int)sizeof(*conf))
4906 return 0;
4907 conf = (struct vd_config *) update->buf;
4908 if (posix_memalign(&update->space, 512,
4909 offsetof(struct vcl, conf)
4910 + ddf->conf_rec_len * 512) != 0) {
4911 update->space = NULL;
4912 return 0;
4913 }
4914 vcl = update->space;
4915 vcl->conf.sec_elmnt_count = conf->sec_elmnt_count;
4916 if (alloc_other_bvds(ddf, vcl) != 0) {
4917 free(update->space);
4918 update->space = NULL;
4919 return 0;
4920 }
4921 }
4922 return 1;
4923 }
4924
4925 /*
4926 * Check degraded state of a RAID10.
4927 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4928 */
4929 static int raid10_degraded(struct mdinfo *info)
4930 {
4931 int n_prim, n_bvds;
4932 int i;
4933 struct mdinfo *d;
4934 char *found;
4935 int ret = -1;
4936
4937 n_prim = info->array.layout & ~0x100;
4938 n_bvds = info->array.raid_disks / n_prim;
4939 found = xmalloc(n_bvds);
4940 if (found == NULL)
4941 return ret;
4942 memset(found, 0, n_bvds);
4943 for (d = info->devs; d; d = d->next) {
4944 i = d->disk.raid_disk / n_prim;
4945 if (i >= n_bvds) {
4946 pr_err("BUG: invalid raid disk\n");
4947 goto out;
4948 }
4949 if (d->state_fd > 0)
4950 found[i]++;
4951 }
4952 ret = 2;
4953 for (i = 0; i < n_bvds; i++)
4954 if (!found[i]) {
4955 dprintf("BVD %d/%d failed\n", i, n_bvds);
4956 ret = 0;
4957 goto out;
4958 } else if (found[i] < n_prim) {
4959 dprintf("BVD %d/%d degraded\n", i, n_bvds);
4960 ret = 1;
4961 }
4962 out:
4963 free(found);
4964 return ret;
4965 }
4966
4967 /*
4968 * Check if the array 'a' is degraded but not failed.
4969 * If it is, find as many spares as are available and needed and
4970 * arrange for their inclusion.
4971 * We only choose devices which are not already in the array,
4972 * and prefer those with a spare-assignment to this array.
4973 * Otherwise we choose global spares - assuming always that
4974 * there is enough room.
4975 * For each spare that we assign, we return an 'mdinfo' which
4976 * describes the position for the device in the array.
4977 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4978 * the new phys_refnum and lba_offset values.
4979 *
4980 * Only worry about BVDs at the moment.
4981 */
4982 static struct mdinfo *ddf_activate_spare(struct active_array *a,
4983 struct metadata_update **updates)
4984 {
4985 int working = 0;
4986 struct mdinfo *d;
4987 struct ddf_super *ddf = a->container->sb;
4988 int global_ok = 0;
4989 struct mdinfo *rv = NULL;
4990 struct mdinfo *di;
4991 struct metadata_update *mu;
4992 struct dl *dl;
4993 int i;
4994 unsigned int j;
4995 struct vcl *vcl;
4996 struct vd_config *vc;
4997 unsigned int n_bvd;
4998
4999 for (d = a->info.devs ; d ; d = d->next) {
5000 if ((d->curr_state & DS_FAULTY) &&
5001 d->state_fd >= 0)
5002 /* wait for Removal to happen */
5003 return NULL;
5004 if (d->state_fd >= 0)
5005 working ++;
5006 }
5007
5008 dprintf("working=%d (%d) level=%d\n", working,
5009 a->info.array.raid_disks,
5010 a->info.array.level);
5011 if (working == a->info.array.raid_disks)
5012 return NULL; /* array not degraded */
5013 switch (a->info.array.level) {
5014 case 1:
5015 if (working == 0)
5016 return NULL; /* failed */
5017 break;
5018 case 4:
5019 case 5:
5020 if (working < a->info.array.raid_disks - 1)
5021 return NULL; /* failed */
5022 break;
5023 case 6:
5024 if (working < a->info.array.raid_disks - 2)
5025 return NULL; /* failed */
5026 break;
5027 case 10:
5028 if (raid10_degraded(&a->info) < 1)
5029 return NULL;
5030 break;
5031 default: /* concat or stripe */
5032 return NULL; /* failed */
5033 }
5034
5035 /* For each slot, if it is not working, find a spare */
5036 dl = ddf->dlist;
5037 for (i = 0; i < a->info.array.raid_disks; i++) {
5038 for (d = a->info.devs ; d ; d = d->next)
5039 if (d->disk.raid_disk == i)
5040 break;
5041 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
5042 if (d && (d->state_fd >= 0))
5043 continue;
5044
5045 /* OK, this device needs recovery. Find a spare */
5046 again:
5047 for ( ; dl ; dl = dl->next) {
5048 unsigned long long esize;
5049 unsigned long long pos;
5050 struct mdinfo *d2;
5051 int is_global = 0;
5052 int is_dedicated = 0;
5053 be16 state;
5054
5055 if (dl->pdnum < 0)
5056 continue;
5057 state = ddf->phys->entries[dl->pdnum].state;
5058 if (be16_and(state,
5059 cpu_to_be16(DDF_Failed|DDF_Missing)) ||
5060 !be16_and(state,
5061 cpu_to_be16(DDF_Online)))
5062 continue;
5063
5064 /* If in this array, skip */
5065 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
5066 if (d2->state_fd >= 0 &&
5067 d2->disk.major == dl->major &&
5068 d2->disk.minor == dl->minor) {
5069 dprintf("%x:%x (%08x) already in array\n",
5070 dl->major, dl->minor,
5071 be32_to_cpu(dl->disk.refnum));
5072 break;
5073 }
5074 if (d2)
5075 continue;
5076 if (be16_and(ddf->phys->entries[dl->pdnum].type,
5077 cpu_to_be16(DDF_Spare))) {
5078 /* Check spare assign record */
5079 if (dl->spare) {
5080 if (dl->spare->type & DDF_spare_dedicated) {
5081 /* check spare_ents for guid */
5082 unsigned int j;
5083 for (j = 0 ;
5084 j < be16_to_cpu
5085 (dl->spare
5086 ->populated);
5087 j++) {
5088 if (memcmp(dl->spare->spare_ents[j].guid,
5089 ddf->virt->entries[a->info.container_member].guid,
5090 DDF_GUID_LEN) == 0)
5091 is_dedicated = 1;
5092 }
5093 } else
5094 is_global = 1;
5095 }
5096 } else if (be16_and(ddf->phys->entries[dl->pdnum].type,
5097 cpu_to_be16(DDF_Global_Spare))) {
5098 is_global = 1;
5099 } else if (!be16_and(ddf->phys
5100 ->entries[dl->pdnum].state,
5101 cpu_to_be16(DDF_Failed))) {
5102 /* we can possibly use some of this */
5103 is_global = 1;
5104 }
5105 if ( ! (is_dedicated ||
5106 (is_global && global_ok))) {
5107 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
5108 is_dedicated, is_global);
5109 continue;
5110 }
5111
5112 /* We are allowed to use this device - is there space?
5113 * We need a->info.component_size sectors */
5114 esize = a->info.component_size;
5115 pos = find_space(ddf, dl, INVALID_SECTORS, &esize);
5116
5117 if (esize < a->info.component_size) {
5118 dprintf("%x:%x has no room: %llu %llu\n",
5119 dl->major, dl->minor,
5120 esize, a->info.component_size);
5121 /* No room */
5122 continue;
5123 }
5124
5125 /* Cool, we have a device with some space at pos */
5126 di = xcalloc(1, sizeof(*di));
5127 di->disk.number = i;
5128 di->disk.raid_disk = i;
5129 di->disk.major = dl->major;
5130 di->disk.minor = dl->minor;
5131 di->disk.state = 0;
5132 di->recovery_start = 0;
5133 di->data_offset = pos;
5134 di->component_size = a->info.component_size;
5135 di->next = rv;
5136 rv = di;
5137 dprintf("%x:%x (%08x) to be %d at %llu\n",
5138 dl->major, dl->minor,
5139 be32_to_cpu(dl->disk.refnum), i, pos);
5140
5141 break;
5142 }
5143 if (!dl && ! global_ok) {
5144 /* not enough dedicated spares, try global */
5145 global_ok = 1;
5146 dl = ddf->dlist;
5147 goto again;
5148 }
5149 }
5150
5151 if (!rv)
5152 /* No spares found */
5153 return rv;
5154 /* Now 'rv' has a list of devices to return.
5155 * Create a metadata_update record to update the
5156 * phys_refnum and lba_offset values
5157 */
5158 vc = find_vdcr(ddf, a->info.container_member, rv->disk.raid_disk,
5159 &n_bvd, &vcl);
5160 if (vc == NULL)
5161 return NULL;
5162
5163 mu = xmalloc(sizeof(*mu));
5164 if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
5165 free(mu);
5166 mu = NULL;
5167 }
5168
5169 mu->len = ddf->conf_rec_len * 512 * vcl->conf.sec_elmnt_count;
5170 mu->buf = xmalloc(mu->len);
5171 mu->space = NULL;
5172 mu->space_list = NULL;
5173 mu->next = *updates;
5174 memcpy(mu->buf, &vcl->conf, ddf->conf_rec_len * 512);
5175 for (j = 1; j < vcl->conf.sec_elmnt_count; j++)
5176 memcpy(mu->buf + j * ddf->conf_rec_len * 512,
5177 vcl->other_bvds[j-1], ddf->conf_rec_len * 512);
5178
5179 vc = (struct vd_config*)mu->buf;
5180 for (di = rv ; di ; di = di->next) {
5181 unsigned int i_sec, i_prim;
5182 i_sec = di->disk.raid_disk
5183 / be16_to_cpu(vcl->conf.prim_elmnt_count);
5184 i_prim = di->disk.raid_disk
5185 % be16_to_cpu(vcl->conf.prim_elmnt_count);
5186 vc = (struct vd_config *)(mu->buf
5187 + i_sec * ddf->conf_rec_len * 512);
5188 for (dl = ddf->dlist; dl; dl = dl->next)
5189 if (dl->major == di->disk.major
5190 && dl->minor == di->disk.minor)
5191 break;
5192 if (!dl || dl->pdnum < 0) {
5193 pr_err("BUG: can't find disk %d (%d/%d)\n",
5194 di->disk.raid_disk,
5195 di->disk.major, di->disk.minor);
5196 return NULL;
5197 }
5198 vc->phys_refnum[i_prim] = ddf->phys->entries[dl->pdnum].refnum;
5199 LBA_OFFSET(ddf, vc)[i_prim] = cpu_to_be64(di->data_offset);
5200 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec, i_prim,
5201 be32_to_cpu(vc->phys_refnum[i_prim]),
5202 be64_to_cpu(LBA_OFFSET(ddf, vc)[i_prim]));
5203 }
5204 *updates = mu;
5205 return rv;
5206 }
5207 #endif /* MDASSEMBLE */
5208
5209 static int ddf_level_to_layout(int level)
5210 {
5211 switch(level) {
5212 case 0:
5213 case 1:
5214 return 0;
5215 case 5:
5216 return ALGORITHM_LEFT_SYMMETRIC;
5217 case 6:
5218 return ALGORITHM_ROTATING_N_CONTINUE;
5219 case 10:
5220 return 0x102;
5221 default:
5222 return UnSet;
5223 }
5224 }
5225
5226 static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk)
5227 {
5228 if (level && *level == UnSet)
5229 *level = LEVEL_CONTAINER;
5230
5231 if (level && layout && *layout == UnSet)
5232 *layout = ddf_level_to_layout(*level);
5233 }
5234
5235 struct superswitch super_ddf = {
5236 #ifndef MDASSEMBLE
5237 .examine_super = examine_super_ddf,
5238 .brief_examine_super = brief_examine_super_ddf,
5239 .brief_examine_subarrays = brief_examine_subarrays_ddf,
5240 .export_examine_super = export_examine_super_ddf,
5241 .detail_super = detail_super_ddf,
5242 .brief_detail_super = brief_detail_super_ddf,
5243 .validate_geometry = validate_geometry_ddf,
5244 .write_init_super = write_init_super_ddf,
5245 .add_to_super = add_to_super_ddf,
5246 .remove_from_super = remove_from_super_ddf,
5247 .load_container = load_container_ddf,
5248 .copy_metadata = copy_metadata_ddf,
5249 .kill_subarray = kill_subarray_ddf,
5250 #endif
5251 .match_home = match_home_ddf,
5252 .uuid_from_super= uuid_from_super_ddf,
5253 .getinfo_super = getinfo_super_ddf,
5254 .update_super = update_super_ddf,
5255
5256 .avail_size = avail_size_ddf,
5257
5258 .compare_super = compare_super_ddf,
5259
5260 .load_super = load_super_ddf,
5261 .init_super = init_super_ddf,
5262 .store_super = store_super_ddf,
5263 .free_super = free_super_ddf,
5264 .match_metadata_desc = match_metadata_desc_ddf,
5265 .container_content = container_content_ddf,
5266 .default_geometry = default_geometry_ddf,
5267
5268 .external = 1,
5269
5270 #ifndef MDASSEMBLE
5271 /* for mdmon */
5272 .open_new = ddf_open_new,
5273 .set_array_state= ddf_set_array_state,
5274 .set_disk = ddf_set_disk,
5275 .sync_metadata = ddf_sync_metadata,
5276 .process_update = ddf_process_update,
5277 .prepare_update = ddf_prepare_update,
5278 .activate_spare = ddf_activate_spare,
5279 #endif
5280 .name = "ddf",
5281 };
Unnamed repository; edit this file 'description' to name the repository.