]> 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
Release mdadm-3.3.1
[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/%s: ", __func__, 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("%u(s=%02x i=%02x) ", i,
549 ddf->virt->entries[i].state,
550 ddf->virt->entries[i].init_state);
551 }
552 dprintf("\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("%s: bad header magic\n", __func__);
829 return 0;
830 }
831 if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) {
832 pr_err("%s: bad CRC\n", __func__);
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("%s: header mismatch\n", __func__);
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("%s could not allocate disk info buffer\n",
1059 __func__);
1060 return 1;
1061 }
1062
1063 load_section(fd, super, &dl->disk,
1064 super->active->data_section_offset,
1065 super->active->data_section_length,
1066 0);
1067 dl->devname = devname ? xstrdup(devname) : NULL;
1068
1069 fstat(fd, &stb);
1070 dl->major = major(stb.st_rdev);
1071 dl->minor = minor(stb.st_rdev);
1072 dl->next = super->dlist;
1073 dl->fd = keep ? fd : -1;
1074
1075 dl->size = 0;
1076 if (get_dev_size(fd, devname, &dsize))
1077 dl->size = dsize >> 9;
1078 /* If the disks have different sizes, the LBAs will differ
1079 * between phys disks.
1080 * At this point here, the values in super->active must be valid
1081 * for this phys disk. */
1082 dl->primary_lba = super->active->primary_lba;
1083 dl->secondary_lba = super->active->secondary_lba;
1084 dl->workspace_lba = super->active->workspace_lba;
1085 dl->spare = NULL;
1086 for (i = 0 ; i < super->max_part ; i++)
1087 dl->vlist[i] = NULL;
1088 super->dlist = dl;
1089 dl->pdnum = -1;
1090 for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++)
1091 if (memcmp(super->phys->entries[i].guid,
1092 dl->disk.guid, DDF_GUID_LEN) == 0)
1093 dl->pdnum = i;
1094
1095 /* Now the config list. */
1096 /* 'conf' is an array of config entries, some of which are
1097 * probably invalid. Those which are good need to be copied into
1098 * the conflist
1099 */
1100
1101 conf = load_section(fd, super, super->conf,
1102 super->active->config_section_offset,
1103 super->active->config_section_length,
1104 0);
1105 super->conf = conf;
1106 vnum = 0;
1107 for (confsec = 0;
1108 confsec < be32_to_cpu(super->active->config_section_length);
1109 confsec += super->conf_rec_len) {
1110 struct vd_config *vd =
1111 (struct vd_config *)((char*)conf + confsec*512);
1112 struct vcl *vcl;
1113
1114 if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) {
1115 if (dl->spare)
1116 continue;
1117 if (posix_memalign((void**)&dl->spare, 512,
1118 super->conf_rec_len*512) != 0) {
1119 pr_err("%s could not allocate spare info buf\n",
1120 __func__);
1121 return 1;
1122 }
1123
1124 memcpy(dl->spare, vd, super->conf_rec_len*512);
1125 continue;
1126 }
1127 if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC))
1128 /* Must be vendor-unique - I cannot handle those */
1129 continue;
1130
1131 for (vcl = super->conflist; vcl; vcl = vcl->next) {
1132 if (memcmp(vcl->conf.guid,
1133 vd->guid, DDF_GUID_LEN) == 0)
1134 break;
1135 }
1136
1137 if (vcl) {
1138 dl->vlist[vnum++] = vcl;
1139 if (vcl->other_bvds != NULL &&
1140 vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) {
1141 add_other_bvd(vcl, vd, super->conf_rec_len*512);
1142 continue;
1143 }
1144 if (be32_to_cpu(vd->seqnum) <=
1145 be32_to_cpu(vcl->conf.seqnum))
1146 continue;
1147 } else {
1148 if (posix_memalign((void**)&vcl, 512,
1149 (super->conf_rec_len*512 +
1150 offsetof(struct vcl, conf))) != 0) {
1151 pr_err("%s could not allocate vcl buf\n",
1152 __func__);
1153 return 1;
1154 }
1155 vcl->next = super->conflist;
1156 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1157 vcl->conf.sec_elmnt_count = vd->sec_elmnt_count;
1158 if (alloc_other_bvds(super, vcl) != 0) {
1159 pr_err("%s could not allocate other bvds\n",
1160 __func__);
1161 free(vcl);
1162 return 1;
1163 };
1164 super->conflist = vcl;
1165 dl->vlist[vnum++] = vcl;
1166 }
1167 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
1168 for (i=0; i < max_virt_disks ; i++)
1169 if (memcmp(super->virt->entries[i].guid,
1170 vcl->conf.guid, DDF_GUID_LEN)==0)
1171 break;
1172 if (i < max_virt_disks)
1173 vcl->vcnum = i;
1174 }
1175
1176 return 0;
1177 }
1178
1179 static int load_super_ddf(struct supertype *st, int fd,
1180 char *devname)
1181 {
1182 unsigned long long dsize;
1183 struct ddf_super *super;
1184 int rv;
1185
1186 if (get_dev_size(fd, devname, &dsize) == 0)
1187 return 1;
1188
1189 if (test_partition(fd))
1190 /* DDF is not allowed on partitions */
1191 return 1;
1192
1193 /* 32M is a lower bound */
1194 if (dsize <= 32*1024*1024) {
1195 if (devname)
1196 pr_err("%s is too small for ddf: "
1197 "size is %llu sectors.\n",
1198 devname, dsize>>9);
1199 return 1;
1200 }
1201 if (dsize & 511) {
1202 if (devname)
1203 pr_err("%s is an odd size for ddf: "
1204 "size is %llu bytes.\n",
1205 devname, dsize);
1206 return 1;
1207 }
1208
1209 free_super_ddf(st);
1210
1211 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
1212 pr_err("malloc of %zu failed.\n",
1213 sizeof(*super));
1214 return 1;
1215 }
1216 memset(super, 0, sizeof(*super));
1217
1218 rv = load_ddf_headers(fd, super, devname);
1219 if (rv) {
1220 free(super);
1221 return rv;
1222 }
1223
1224 /* Have valid headers and have chosen the best. Let's read in the rest*/
1225
1226 rv = load_ddf_global(fd, super, devname);
1227
1228 if (rv) {
1229 if (devname)
1230 pr_err("Failed to load all information "
1231 "sections on %s\n", devname);
1232 free(super);
1233 return rv;
1234 }
1235
1236 rv = load_ddf_local(fd, super, devname, 0);
1237
1238 if (rv) {
1239 if (devname)
1240 pr_err("Failed to load all information "
1241 "sections on %s\n", devname);
1242 free(super);
1243 return rv;
1244 }
1245
1246 /* Should possibly check the sections .... */
1247
1248 st->sb = super;
1249 if (st->ss == NULL) {
1250 st->ss = &super_ddf;
1251 st->minor_version = 0;
1252 st->max_devs = 512;
1253 }
1254 return 0;
1255
1256 }
1257
1258 static void free_super_ddf(struct supertype *st)
1259 {
1260 struct ddf_super *ddf = st->sb;
1261 if (ddf == NULL)
1262 return;
1263 free(ddf->phys);
1264 free(ddf->virt);
1265 free(ddf->conf);
1266 while (ddf->conflist) {
1267 struct vcl *v = ddf->conflist;
1268 ddf->conflist = v->next;
1269 if (v->block_sizes)
1270 free(v->block_sizes);
1271 if (v->other_bvds)
1272 /*
1273 v->other_bvds[0] points to beginning of buffer,
1274 see alloc_other_bvds()
1275 */
1276 free(v->other_bvds[0]);
1277 free(v);
1278 }
1279 while (ddf->dlist) {
1280 struct dl *d = ddf->dlist;
1281 ddf->dlist = d->next;
1282 if (d->fd >= 0)
1283 close(d->fd);
1284 if (d->spare)
1285 free(d->spare);
1286 free(d);
1287 }
1288 while (ddf->add_list) {
1289 struct dl *d = ddf->add_list;
1290 ddf->add_list = d->next;
1291 if (d->fd >= 0)
1292 close(d->fd);
1293 if (d->spare)
1294 free(d->spare);
1295 free(d);
1296 }
1297 free(ddf);
1298 st->sb = NULL;
1299 }
1300
1301 static struct supertype *match_metadata_desc_ddf(char *arg)
1302 {
1303 /* 'ddf' only supports containers */
1304 struct supertype *st;
1305 if (strcmp(arg, "ddf") != 0 &&
1306 strcmp(arg, "default") != 0
1307 )
1308 return NULL;
1309
1310 st = xcalloc(1, sizeof(*st));
1311 st->ss = &super_ddf;
1312 st->max_devs = 512;
1313 st->minor_version = 0;
1314 st->sb = NULL;
1315 return st;
1316 }
1317
1318 #ifndef MDASSEMBLE
1319
1320 static mapping_t ddf_state[] = {
1321 { "Optimal", 0},
1322 { "Degraded", 1},
1323 { "Deleted", 2},
1324 { "Missing", 3},
1325 { "Failed", 4},
1326 { "Partially Optimal", 5},
1327 { "-reserved-", 6},
1328 { "-reserved-", 7},
1329 { NULL, 0}
1330 };
1331
1332 static mapping_t ddf_init_state[] = {
1333 { "Not Initialised", 0},
1334 { "QuickInit in Progress", 1},
1335 { "Fully Initialised", 2},
1336 { "*UNKNOWN*", 3},
1337 { NULL, 0}
1338 };
1339 static mapping_t ddf_access[] = {
1340 { "Read/Write", 0},
1341 { "Reserved", 1},
1342 { "Read Only", 2},
1343 { "Blocked (no access)", 3},
1344 { NULL ,0}
1345 };
1346
1347 static mapping_t ddf_level[] = {
1348 { "RAID0", DDF_RAID0},
1349 { "RAID1", DDF_RAID1},
1350 { "RAID3", DDF_RAID3},
1351 { "RAID4", DDF_RAID4},
1352 { "RAID5", DDF_RAID5},
1353 { "RAID1E",DDF_RAID1E},
1354 { "JBOD", DDF_JBOD},
1355 { "CONCAT",DDF_CONCAT},
1356 { "RAID5E",DDF_RAID5E},
1357 { "RAID5EE",DDF_RAID5EE},
1358 { "RAID6", DDF_RAID6},
1359 { NULL, 0}
1360 };
1361 static mapping_t ddf_sec_level[] = {
1362 { "Striped", DDF_2STRIPED},
1363 { "Mirrored", DDF_2MIRRORED},
1364 { "Concat", DDF_2CONCAT},
1365 { "Spanned", DDF_2SPANNED},
1366 { NULL, 0}
1367 };
1368 #endif
1369
1370 static int all_ff(const char *guid)
1371 {
1372 int i;
1373 for (i = 0; i < DDF_GUID_LEN; i++)
1374 if (guid[i] != (char)0xff)
1375 return 0;
1376 return 1;
1377 }
1378
1379 static const char *guid_str(const char *guid)
1380 {
1381 static char buf[DDF_GUID_LEN*2+1];
1382 int i;
1383 char *p = buf;
1384 for (i = 0; i < DDF_GUID_LEN; i++) {
1385 unsigned char c = guid[i];
1386 if (c >= 32 && c < 127)
1387 p += sprintf(p, "%c", c);
1388 else
1389 p += sprintf(p, "%02x", c);
1390 }
1391 *p = '\0';
1392 return (const char *) buf;
1393 }
1394
1395 #ifndef MDASSEMBLE
1396 static void print_guid(char *guid, int tstamp)
1397 {
1398 /* A GUIDs are part (or all) ASCII and part binary.
1399 * They tend to be space padded.
1400 * We print the GUID in HEX, then in parentheses add
1401 * any initial ASCII sequence, and a possible
1402 * time stamp from bytes 16-19
1403 */
1404 int l = DDF_GUID_LEN;
1405 int i;
1406
1407 for (i=0 ; i<DDF_GUID_LEN ; i++) {
1408 if ((i&3)==0 && i != 0) printf(":");
1409 printf("%02X", guid[i]&255);
1410 }
1411
1412 printf("\n (");
1413 while (l && guid[l-1] == ' ')
1414 l--;
1415 for (i=0 ; i<l ; i++) {
1416 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1417 fputc(guid[i], stdout);
1418 else
1419 break;
1420 }
1421 if (tstamp) {
1422 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1423 char tbuf[100];
1424 struct tm *tm;
1425 tm = localtime(&then);
1426 strftime(tbuf, 100, " %D %T",tm);
1427 fputs(tbuf, stdout);
1428 }
1429 printf(")");
1430 }
1431
1432 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1433 {
1434 int crl = sb->conf_rec_len;
1435 struct vcl *vcl;
1436
1437 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1438 unsigned int i;
1439 struct vd_config *vc = &vcl->conf;
1440
1441 if (!be32_eq(calc_crc(vc, crl*512), vc->crc))
1442 continue;
1443 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1444 continue;
1445
1446 /* Ok, we know about this VD, let's give more details */
1447 printf(" Raid Devices[%d] : %d (", n,
1448 be16_to_cpu(vc->prim_elmnt_count));
1449 for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) {
1450 int j;
1451 int cnt = be16_to_cpu(sb->phys->max_pdes);
1452 for (j=0; j<cnt; j++)
1453 if (be32_eq(vc->phys_refnum[i],
1454 sb->phys->entries[j].refnum))
1455 break;
1456 if (i) printf(" ");
1457 if (j < cnt)
1458 printf("%d", j);
1459 else
1460 printf("--");
1461 printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb, vc)[i])/2);
1462 }
1463 printf(")\n");
1464 if (vc->chunk_shift != 255)
1465 printf(" Chunk Size[%d] : %d sectors\n", n,
1466 1 << vc->chunk_shift);
1467 printf(" Raid Level[%d] : %s\n", n,
1468 map_num(ddf_level, vc->prl)?:"-unknown-");
1469 if (vc->sec_elmnt_count != 1) {
1470 printf(" Secondary Position[%d] : %d of %d\n", n,
1471 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1472 printf(" Secondary Level[%d] : %s\n", n,
1473 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1474 }
1475 printf(" Device Size[%d] : %llu\n", n,
1476 be64_to_cpu(vc->blocks)/2);
1477 printf(" Array Size[%d] : %llu\n", n,
1478 be64_to_cpu(vc->array_blocks)/2);
1479 }
1480 }
1481
1482 static void examine_vds(struct ddf_super *sb)
1483 {
1484 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1485 unsigned int i;
1486 printf(" Virtual Disks : %d\n", cnt);
1487
1488 for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) {
1489 struct virtual_entry *ve = &sb->virt->entries[i];
1490 if (all_ff(ve->guid))
1491 continue;
1492 printf("\n");
1493 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1494 printf("\n");
1495 printf(" unit[%d] : %d\n", i, be16_to_cpu(ve->unit));
1496 printf(" state[%d] : %s, %s%s\n", i,
1497 map_num(ddf_state, ve->state & 7),
1498 (ve->state & DDF_state_morphing) ? "Morphing, ": "",
1499 (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent");
1500 printf(" init state[%d] : %s\n", i,
1501 map_num(ddf_init_state, ve->init_state&DDF_initstate_mask));
1502 printf(" access[%d] : %s\n", i,
1503 map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6));
1504 printf(" Name[%d] : %.16s\n", i, ve->name);
1505 examine_vd(i, sb, ve->guid);
1506 }
1507 if (cnt) printf("\n");
1508 }
1509
1510 static void examine_pds(struct ddf_super *sb)
1511 {
1512 int cnt = be16_to_cpu(sb->phys->max_pdes);
1513 int i;
1514 struct dl *dl;
1515 int unlisted = 0;
1516 printf(" Physical Disks : %d\n", cnt);
1517 printf(" Number RefNo Size Device Type/State\n");
1518
1519 for (dl = sb->dlist; dl; dl = dl->next)
1520 dl->displayed = 0;
1521
1522 for (i=0 ; i<cnt ; i++) {
1523 struct phys_disk_entry *pd = &sb->phys->entries[i];
1524 int type = be16_to_cpu(pd->type);
1525 int state = be16_to_cpu(pd->state);
1526
1527 if (be32_to_cpu(pd->refnum) == 0xffffffff)
1528 /* Not in use */
1529 continue;
1530 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1531 //printf("\n");
1532 printf(" %3d %08x ", i,
1533 be32_to_cpu(pd->refnum));
1534 printf("%8lluK ",
1535 be64_to_cpu(pd->config_size)>>1);
1536 for (dl = sb->dlist; dl ; dl = dl->next) {
1537 if (be32_eq(dl->disk.refnum, pd->refnum)) {
1538 char *dv = map_dev(dl->major, dl->minor, 0);
1539 if (dv) {
1540 printf("%-15s", dv);
1541 break;
1542 }
1543 }
1544 }
1545 if (!dl)
1546 printf("%15s","");
1547 else
1548 dl->displayed = 1;
1549 printf(" %s%s%s%s%s",
1550 (type&2) ? "active":"",
1551 (type&4) ? "Global-Spare":"",
1552 (type&8) ? "spare" : "",
1553 (type&16)? ", foreign" : "",
1554 (type&32)? "pass-through" : "");
1555 if (state & DDF_Failed)
1556 /* This over-rides these three */
1557 state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition);
1558 printf("/%s%s%s%s%s%s%s",
1559 (state&1)? "Online": "Offline",
1560 (state&2)? ", Failed": "",
1561 (state&4)? ", Rebuilding": "",
1562 (state&8)? ", in-transition": "",
1563 (state&16)? ", SMART-errors": "",
1564 (state&32)? ", Unrecovered-Read-Errors": "",
1565 (state&64)? ", Missing" : "");
1566 printf("\n");
1567 }
1568 for (dl = sb->dlist; dl; dl = dl->next) {
1569 char *dv;
1570 if (dl->displayed)
1571 continue;
1572 if (!unlisted)
1573 printf(" Physical disks not in metadata!:\n");
1574 unlisted = 1;
1575 dv = map_dev(dl->major, dl->minor, 0);
1576 printf(" %08x %s\n", be32_to_cpu(dl->disk.refnum),
1577 dv ? dv : "-unknown-");
1578 }
1579 if (unlisted)
1580 printf("\n");
1581 }
1582
1583 static void examine_super_ddf(struct supertype *st, char *homehost)
1584 {
1585 struct ddf_super *sb = st->sb;
1586
1587 printf(" Magic : %08x\n", be32_to_cpu(sb->anchor.magic));
1588 printf(" Version : %.8s\n", sb->anchor.revision);
1589 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1590 printf("\n");
1591 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1592 printf("\n");
1593 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1594 printf(" Redundant hdr : %s\n", (be32_eq(sb->secondary.magic,
1595 DDF_HEADER_MAGIC)
1596 ?"yes" : "no"));
1597 examine_vds(sb);
1598 examine_pds(sb);
1599 }
1600
1601 static unsigned int get_vd_num_of_subarray(struct supertype *st)
1602 {
1603 /*
1604 * Figure out the VD number for this supertype.
1605 * Returns DDF_CONTAINER for the container itself,
1606 * and DDF_NOTFOUND on error.
1607 */
1608 struct ddf_super *ddf = st->sb;
1609 struct mdinfo *sra;
1610 char *sub, *end;
1611 unsigned int vcnum;
1612
1613 if (*st->container_devnm == '\0')
1614 return DDF_CONTAINER;
1615
1616 sra = sysfs_read(-1, st->devnm, GET_VERSION);
1617 if (!sra || sra->array.major_version != -1 ||
1618 sra->array.minor_version != -2 ||
1619 !is_subarray(sra->text_version))
1620 return DDF_NOTFOUND;
1621
1622 sub = strchr(sra->text_version + 1, '/');
1623 if (sub != NULL)
1624 vcnum = strtoul(sub + 1, &end, 10);
1625 if (sub == NULL || *sub == '\0' || *end != '\0' ||
1626 vcnum >= be16_to_cpu(ddf->active->max_vd_entries))
1627 return DDF_NOTFOUND;
1628
1629 return vcnum;
1630 }
1631
1632 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1633 {
1634 /* We just write a generic DDF ARRAY entry
1635 */
1636 struct mdinfo info;
1637 char nbuf[64];
1638 getinfo_super_ddf(st, &info, NULL);
1639 fname_from_uuid(st, &info, nbuf, ':');
1640
1641 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1642 }
1643
1644 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1645 {
1646 /* We write a DDF ARRAY member entry for each vd, identifying container
1647 * by uuid and member by unit number and uuid.
1648 */
1649 struct ddf_super *ddf = st->sb;
1650 struct mdinfo info;
1651 unsigned int i;
1652 char nbuf[64];
1653 getinfo_super_ddf(st, &info, NULL);
1654 fname_from_uuid(st, &info, nbuf, ':');
1655
1656 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
1657 struct virtual_entry *ve = &ddf->virt->entries[i];
1658 struct vcl vcl;
1659 char nbuf1[64];
1660 char namebuf[17];
1661 if (all_ff(ve->guid))
1662 continue;
1663 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1664 ddf->currentconf =&vcl;
1665 vcl.vcnum = i;
1666 uuid_from_super_ddf(st, info.uuid);
1667 fname_from_uuid(st, &info, nbuf1, ':');
1668 _ddf_array_name(namebuf, ddf, i);
1669 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1670 namebuf[0] == '\0' ? "" : " /dev/md/", namebuf,
1671 nbuf+5, i, nbuf1+5);
1672 }
1673 }
1674
1675 static void export_examine_super_ddf(struct supertype *st)
1676 {
1677 struct mdinfo info;
1678 char nbuf[64];
1679 getinfo_super_ddf(st, &info, NULL);
1680 fname_from_uuid(st, &info, nbuf, ':');
1681 printf("MD_METADATA=ddf\n");
1682 printf("MD_LEVEL=container\n");
1683 printf("MD_UUID=%s\n", nbuf+5);
1684 printf("MD_DEVICES=%u\n",
1685 be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes));
1686 }
1687
1688 static int copy_metadata_ddf(struct supertype *st, int from, int to)
1689 {
1690 void *buf;
1691 unsigned long long dsize, offset;
1692 int bytes;
1693 struct ddf_header *ddf;
1694 int written = 0;
1695
1696 /* The meta consists of an anchor, a primary, and a secondary.
1697 * This all lives at the end of the device.
1698 * So it is easiest to find the earliest of primary and
1699 * secondary, and copy everything from there.
1700 *
1701 * Anchor is 512 from end. It contains primary_lba and secondary_lba
1702 * we choose one of those
1703 */
1704
1705 if (posix_memalign(&buf, 4096, 4096) != 0)
1706 return 1;
1707
1708 if (!get_dev_size(from, NULL, &dsize))
1709 goto err;
1710
1711 if (lseek64(from, dsize-512, 0) < 0)
1712 goto err;
1713 if (read(from, buf, 512) != 512)
1714 goto err;
1715 ddf = buf;
1716 if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) ||
1717 !be32_eq(calc_crc(ddf, 512), ddf->crc) ||
1718 (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 &&
1719 memcmp(ddf->revision, DDF_REVISION_2, 8) != 0))
1720 goto err;
1721
1722 offset = dsize - 512;
1723 if ((be64_to_cpu(ddf->primary_lba) << 9) < offset)
1724 offset = be64_to_cpu(ddf->primary_lba) << 9;
1725 if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset)
1726 offset = be64_to_cpu(ddf->secondary_lba) << 9;
1727
1728 bytes = dsize - offset;
1729
1730 if (lseek64(from, offset, 0) < 0 ||
1731 lseek64(to, offset, 0) < 0)
1732 goto err;
1733 while (written < bytes) {
1734 int n = bytes - written;
1735 if (n > 4096)
1736 n = 4096;
1737 if (read(from, buf, n) != n)
1738 goto err;
1739 if (write(to, buf, n) != n)
1740 goto err;
1741 written += n;
1742 }
1743 free(buf);
1744 return 0;
1745 err:
1746 free(buf);
1747 return 1;
1748 }
1749
1750 static void detail_super_ddf(struct supertype *st, char *homehost)
1751 {
1752 struct ddf_super *sb = st->sb;
1753 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1754
1755 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1756 printf("\n");
1757 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1758 printf(" Virtual Disks : %d\n", cnt);
1759 printf("\n");
1760 }
1761 #endif
1762
1763 static const char *vendors_with_variable_volume_UUID[] = {
1764 "LSI ",
1765 };
1766
1767 static int volume_id_is_reliable(const struct ddf_super *ddf)
1768 {
1769 int n = ARRAY_SIZE(vendors_with_variable_volume_UUID);
1770 int i;
1771 for (i = 0; i < n; i++)
1772 if (!memcmp(ddf->controller.guid,
1773 vendors_with_variable_volume_UUID[i], 8))
1774 return 0;
1775 return 1;
1776 }
1777
1778 static void uuid_of_ddf_subarray(const struct ddf_super *ddf,
1779 unsigned int vcnum, int uuid[4])
1780 {
1781 char buf[DDF_GUID_LEN+18], sha[20], *p;
1782 struct sha1_ctx ctx;
1783 if (volume_id_is_reliable(ddf)) {
1784 uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid);
1785 return;
1786 }
1787 /*
1788 * Some fake RAID BIOSes (in particular, LSI ones) change the
1789 * VD GUID at every boot. These GUIDs are not suitable for
1790 * identifying an array. Luckily the header GUID appears to
1791 * remain constant.
1792 * We construct a pseudo-UUID from the header GUID and those
1793 * properties of the subarray that we expect to remain constant.
1794 */
1795 memset(buf, 0, sizeof(buf));
1796 p = buf;
1797 memcpy(p, ddf->anchor.guid, DDF_GUID_LEN);
1798 p += DDF_GUID_LEN;
1799 memcpy(p, ddf->virt->entries[vcnum].name, 16);
1800 p += 16;
1801 *((__u16 *) p) = vcnum;
1802 sha1_init_ctx(&ctx);
1803 sha1_process_bytes(buf, sizeof(buf), &ctx);
1804 sha1_finish_ctx(&ctx, sha);
1805 memcpy(uuid, sha, 4*4);
1806 }
1807
1808 #ifndef MDASSEMBLE
1809 static void brief_detail_super_ddf(struct supertype *st)
1810 {
1811 struct mdinfo info;
1812 char nbuf[64];
1813 struct ddf_super *ddf = st->sb;
1814 unsigned int vcnum = get_vd_num_of_subarray(st);
1815 if (vcnum == DDF_CONTAINER)
1816 uuid_from_super_ddf(st, info.uuid);
1817 else if (vcnum == DDF_NOTFOUND)
1818 return;
1819 else
1820 uuid_of_ddf_subarray(ddf, vcnum, info.uuid);
1821 fname_from_uuid(st, &info, nbuf,':');
1822 printf(" UUID=%s", nbuf + 5);
1823 }
1824 #endif
1825
1826 static int match_home_ddf(struct supertype *st, char *homehost)
1827 {
1828 /* It matches 'this' host if the controller is a
1829 * Linux-MD controller with vendor_data matching
1830 * the hostname. It would be nice if we could
1831 * test against controller found in /sys or somewhere...
1832 */
1833 struct ddf_super *ddf = st->sb;
1834 unsigned int len;
1835
1836 if (!homehost)
1837 return 0;
1838 len = strlen(homehost);
1839
1840 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1841 len < sizeof(ddf->controller.vendor_data) &&
1842 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1843 ddf->controller.vendor_data[len] == 0);
1844 }
1845
1846 #ifndef MDASSEMBLE
1847 static int find_index_in_bvd(const struct ddf_super *ddf,
1848 const struct vd_config *conf, unsigned int n,
1849 unsigned int *n_bvd)
1850 {
1851 /*
1852 * Find the index of the n-th valid physical disk in this BVD.
1853 * Unused entries can be sprinkled in with the used entries,
1854 * but don't count.
1855 */
1856 unsigned int i, j;
1857 for (i = 0, j = 0;
1858 i < ddf->mppe && j < be16_to_cpu(conf->prim_elmnt_count);
1859 i++) {
1860 if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) {
1861 if (n == j) {
1862 *n_bvd = i;
1863 return 1;
1864 }
1865 j++;
1866 }
1867 }
1868 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1869 __func__, n, be16_to_cpu(conf->prim_elmnt_count));
1870 return 0;
1871 }
1872
1873 /* Given a member array instance number, and a raid disk within that instance,
1874 * find the vd_config structure. The offset of the given disk in the phys_refnum
1875 * table is returned in n_bvd.
1876 * For two-level members with a secondary raid level the vd_config for
1877 * the appropriate BVD is returned.
1878 * The return value is always &vlc->conf, where vlc is returned in last pointer.
1879 */
1880 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst,
1881 unsigned int n,
1882 unsigned int *n_bvd, struct vcl **vcl)
1883 {
1884 struct vcl *v;
1885
1886 for (v = ddf->conflist; v; v = v->next) {
1887 unsigned int nsec, ibvd = 0;
1888 struct vd_config *conf;
1889 if (inst != v->vcnum)
1890 continue;
1891 conf = &v->conf;
1892 if (conf->sec_elmnt_count == 1) {
1893 if (find_index_in_bvd(ddf, conf, n, n_bvd)) {
1894 *vcl = v;
1895 return conf;
1896 } else
1897 goto bad;
1898 }
1899 if (v->other_bvds == NULL) {
1900 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1901 __func__, conf->sec_elmnt_count);
1902 goto bad;
1903 }
1904 nsec = n / be16_to_cpu(conf->prim_elmnt_count);
1905 if (conf->sec_elmnt_seq != nsec) {
1906 for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) {
1907 if (v->other_bvds[ibvd-1]->sec_elmnt_seq
1908 == nsec)
1909 break;
1910 }
1911 if (ibvd == conf->sec_elmnt_count)
1912 goto bad;
1913 conf = v->other_bvds[ibvd-1];
1914 }
1915 if (!find_index_in_bvd(ddf, conf,
1916 n - nsec*conf->sec_elmnt_count, n_bvd))
1917 goto bad;
1918 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1919 , __func__, n, *n_bvd, ibvd, inst);
1920 *vcl = v;
1921 return conf;
1922 }
1923 bad:
1924 pr_err("%s: Could't find disk %d in array %u\n", __func__, n, inst);
1925 return NULL;
1926 }
1927 #endif
1928
1929 static int find_phys(const struct ddf_super *ddf, be32 phys_refnum)
1930 {
1931 /* Find the entry in phys_disk which has the given refnum
1932 * and return it's index
1933 */
1934 unsigned int i;
1935 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++)
1936 if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum))
1937 return i;
1938 return -1;
1939 }
1940
1941 static void uuid_from_ddf_guid(const char *guid, int uuid[4])
1942 {
1943 char buf[20];
1944 struct sha1_ctx ctx;
1945 sha1_init_ctx(&ctx);
1946 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1947 sha1_finish_ctx(&ctx, buf);
1948 memcpy(uuid, buf, 4*4);
1949 }
1950
1951 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1952 {
1953 /* The uuid returned here is used for:
1954 * uuid to put into bitmap file (Create, Grow)
1955 * uuid for backup header when saving critical section (Grow)
1956 * comparing uuids when re-adding a device into an array
1957 * In these cases the uuid required is that of the data-array,
1958 * not the device-set.
1959 * uuid to recognise same set when adding a missing device back
1960 * to an array. This is a uuid for the device-set.
1961 *
1962 * For each of these we can make do with a truncated
1963 * or hashed uuid rather than the original, as long as
1964 * everyone agrees.
1965 * In the case of SVD we assume the BVD is of interest,
1966 * though that might be the case if a bitmap were made for
1967 * a mirrored SVD - worry about that later.
1968 * So we need to find the VD configuration record for the
1969 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1970 * The first 16 bytes of the sha1 of these is used.
1971 */
1972 struct ddf_super *ddf = st->sb;
1973 struct vcl *vcl = ddf->currentconf;
1974
1975 if (vcl)
1976 uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid);
1977 else
1978 uuid_from_ddf_guid(ddf->anchor.guid, uuid);
1979 }
1980
1981 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1982 {
1983 struct ddf_super *ddf = st->sb;
1984 int map_disks = info->array.raid_disks;
1985 __u32 *cptr;
1986
1987 if (ddf->currentconf) {
1988 getinfo_super_ddf_bvd(st, info, map);
1989 return;
1990 }
1991 memset(info, 0, sizeof(*info));
1992
1993 info->array.raid_disks = be16_to_cpu(ddf->phys->used_pdes);
1994 info->array.level = LEVEL_CONTAINER;
1995 info->array.layout = 0;
1996 info->array.md_minor = -1;
1997 cptr = (__u32 *)(ddf->anchor.guid + 16);
1998 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
1999
2000 info->array.chunk_size = 0;
2001 info->container_enough = 1;
2002
2003 info->disk.major = 0;
2004 info->disk.minor = 0;
2005 if (ddf->dlist) {
2006 struct phys_disk_entry *pde = NULL;
2007 info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum);
2008 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
2009
2010 info->data_offset = be64_to_cpu(ddf->phys->
2011 entries[info->disk.raid_disk].
2012 config_size);
2013 info->component_size = ddf->dlist->size - info->data_offset;
2014 if (info->disk.raid_disk >= 0)
2015 pde = ddf->phys->entries + info->disk.raid_disk;
2016 if (pde &&
2017 !(be16_to_cpu(pde->state) & DDF_Failed) &&
2018 !(be16_to_cpu(pde->state) & DDF_Missing))
2019 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2020 else
2021 info->disk.state = 1 << MD_DISK_FAULTY;
2022
2023 } else {
2024 /* There should always be a dlist, but just in case...*/
2025 info->disk.number = -1;
2026 info->disk.raid_disk = -1;
2027 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2028 }
2029 info->events = be32_to_cpu(ddf->active->seq);
2030 info->array.utime = DECADE + be32_to_cpu(ddf->active->timestamp);
2031
2032 info->recovery_start = MaxSector;
2033 info->reshape_active = 0;
2034 info->recovery_blocked = 0;
2035 info->name[0] = 0;
2036
2037 info->array.major_version = -1;
2038 info->array.minor_version = -2;
2039 strcpy(info->text_version, "ddf");
2040 info->safe_mode_delay = 0;
2041
2042 uuid_from_super_ddf(st, info->uuid);
2043
2044 if (map) {
2045 int i, e = 0;
2046 int max = be16_to_cpu(ddf->phys->max_pdes);
2047 for (i = e = 0 ; i < map_disks ; i++, e++) {
2048 while (e < max &&
2049 be32_to_cpu(ddf->phys->entries[e].refnum) == 0xffffffff)
2050 e++;
2051 if (i < info->array.raid_disks && e < max &&
2052 !(be16_to_cpu(ddf->phys->entries[e].state)
2053 & DDF_Failed))
2054 map[i] = 1;
2055 else
2056 map[i] = 0;
2057 }
2058 }
2059 }
2060
2061 /* size of name must be at least 17 bytes! */
2062 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i)
2063 {
2064 int j;
2065 memcpy(name, ddf->virt->entries[i].name, 16);
2066 name[16] = 0;
2067 for(j = 0; j < 16; j++)
2068 if (name[j] == ' ')
2069 name[j] = 0;
2070 }
2071
2072 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
2073 {
2074 struct ddf_super *ddf = st->sb;
2075 struct vcl *vc = ddf->currentconf;
2076 int cd = ddf->currentdev;
2077 int n_prim;
2078 int j;
2079 struct dl *dl = NULL;
2080 int map_disks = info->array.raid_disks;
2081 __u32 *cptr;
2082 struct vd_config *conf;
2083
2084 memset(info, 0, sizeof(*info));
2085 if (layout_ddf2md(&vc->conf, &info->array) == -1)
2086 return;
2087 info->array.md_minor = -1;
2088 cptr = (__u32 *)(vc->conf.guid + 16);
2089 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
2090 info->array.utime = DECADE + be32_to_cpu(vc->conf.timestamp);
2091 info->array.chunk_size = 512 << vc->conf.chunk_shift;
2092 info->custom_array_size = be64_to_cpu(vc->conf.array_blocks);
2093
2094 conf = &vc->conf;
2095 n_prim = be16_to_cpu(conf->prim_elmnt_count);
2096 if (conf->sec_elmnt_count > 1 && cd >= n_prim) {
2097 int ibvd = cd / n_prim - 1;
2098 cd %= n_prim;
2099 conf = vc->other_bvds[ibvd];
2100 }
2101
2102 if (cd >= 0 && (unsigned)cd < ddf->mppe) {
2103 info->data_offset =
2104 be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]);
2105 if (vc->block_sizes)
2106 info->component_size = vc->block_sizes[cd];
2107 else
2108 info->component_size = be64_to_cpu(conf->blocks);
2109
2110 for (dl = ddf->dlist; dl ; dl = dl->next)
2111 if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd]))
2112 break;
2113 }
2114
2115 info->disk.major = 0;
2116 info->disk.minor = 0;
2117 info->disk.state = 0;
2118 if (dl && dl->pdnum >= 0) {
2119 info->disk.major = dl->major;
2120 info->disk.minor = dl->minor;
2121 info->disk.raid_disk = cd + conf->sec_elmnt_seq
2122 * be16_to_cpu(conf->prim_elmnt_count);
2123 info->disk.number = dl->pdnum;
2124 info->disk.state = 0;
2125 if (info->disk.number >= 0 &&
2126 (be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Online) &&
2127 !(be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Failed))
2128 info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2129 info->events = be32_to_cpu(ddf->active->seq);
2130 }
2131
2132 info->container_member = ddf->currentconf->vcnum;
2133
2134 info->recovery_start = MaxSector;
2135 info->resync_start = 0;
2136 info->reshape_active = 0;
2137 info->recovery_blocked = 0;
2138 if (!(ddf->virt->entries[info->container_member].state
2139 & DDF_state_inconsistent) &&
2140 (ddf->virt->entries[info->container_member].init_state
2141 & DDF_initstate_mask)
2142 == DDF_init_full)
2143 info->resync_start = MaxSector;
2144
2145 uuid_from_super_ddf(st, info->uuid);
2146
2147 info->array.major_version = -1;
2148 info->array.minor_version = -2;
2149 sprintf(info->text_version, "/%s/%d",
2150 st->container_devnm,
2151 info->container_member);
2152 info->safe_mode_delay = DDF_SAFE_MODE_DELAY;
2153
2154 _ddf_array_name(info->name, ddf, info->container_member);
2155
2156 if (map)
2157 for (j = 0; j < map_disks; j++) {
2158 map[j] = 0;
2159 if (j < info->array.raid_disks) {
2160 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
2161 if (i >= 0 &&
2162 (be16_to_cpu(ddf->phys->entries[i].state)
2163 & DDF_Online) &&
2164 !(be16_to_cpu(ddf->phys->entries[i].state)
2165 & DDF_Failed))
2166 map[i] = 1;
2167 }
2168 }
2169 }
2170
2171 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
2172 char *update,
2173 char *devname, int verbose,
2174 int uuid_set, char *homehost)
2175 {
2176 /* For 'assemble' and 'force' we need to return non-zero if any
2177 * change was made. For others, the return value is ignored.
2178 * Update options are:
2179 * force-one : This device looks a bit old but needs to be included,
2180 * update age info appropriately.
2181 * assemble: clear any 'faulty' flag to allow this device to
2182 * be assembled.
2183 * force-array: Array is degraded but being forced, mark it clean
2184 * if that will be needed to assemble it.
2185 *
2186 * newdev: not used ????
2187 * grow: Array has gained a new device - this is currently for
2188 * linear only
2189 * resync: mark as dirty so a resync will happen.
2190 * uuid: Change the uuid of the array to match what is given
2191 * homehost: update the recorded homehost
2192 * name: update the name - preserving the homehost
2193 * _reshape_progress: record new reshape_progress position.
2194 *
2195 * Following are not relevant for this version:
2196 * sparc2.2 : update from old dodgey metadata
2197 * super-minor: change the preferred_minor number
2198 * summaries: update redundant counters.
2199 */
2200 int rv = 0;
2201 // struct ddf_super *ddf = st->sb;
2202 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2203 // struct virtual_entry *ve = find_ve(ddf);
2204
2205 /* we don't need to handle "force-*" or "assemble" as
2206 * there is no need to 'trick' the kernel. When the metadata is
2207 * first updated to activate the array, all the implied modifications
2208 * will just happen.
2209 */
2210
2211 if (strcmp(update, "grow") == 0) {
2212 /* FIXME */
2213 } else if (strcmp(update, "resync") == 0) {
2214 // info->resync_checkpoint = 0;
2215 } else if (strcmp(update, "homehost") == 0) {
2216 /* homehost is stored in controller->vendor_data,
2217 * or it is when we are the vendor
2218 */
2219 // if (info->vendor_is_local)
2220 // strcpy(ddf->controller.vendor_data, homehost);
2221 rv = -1;
2222 } else if (strcmp(update, "name") == 0) {
2223 /* name is stored in virtual_entry->name */
2224 // memset(ve->name, ' ', 16);
2225 // strncpy(ve->name, info->name, 16);
2226 rv = -1;
2227 } else if (strcmp(update, "_reshape_progress") == 0) {
2228 /* We don't support reshape yet */
2229 } else if (strcmp(update, "assemble") == 0 ) {
2230 /* Do nothing, just succeed */
2231 rv = 0;
2232 } else
2233 rv = -1;
2234
2235 // update_all_csum(ddf);
2236
2237 return rv;
2238 }
2239
2240 static void make_header_guid(char *guid)
2241 {
2242 be32 stamp;
2243 /* Create a DDF Header of Virtual Disk GUID */
2244
2245 /* 24 bytes of fiction required.
2246 * first 8 are a 'vendor-id' - "Linux-MD"
2247 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2248 * Remaining 8 random number plus timestamp
2249 */
2250 memcpy(guid, T10, sizeof(T10));
2251 stamp = cpu_to_be32(0xdeadbeef);
2252 memcpy(guid+8, &stamp, 4);
2253 stamp = cpu_to_be32(0);
2254 memcpy(guid+12, &stamp, 4);
2255 stamp = cpu_to_be32(time(0) - DECADE);
2256 memcpy(guid+16, &stamp, 4);
2257 stamp._v32 = random32();
2258 memcpy(guid+20, &stamp, 4);
2259 }
2260
2261 static unsigned int find_unused_vde(const struct ddf_super *ddf)
2262 {
2263 unsigned int i;
2264 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2265 if (all_ff(ddf->virt->entries[i].guid))
2266 return i;
2267 }
2268 return DDF_NOTFOUND;
2269 }
2270
2271 static unsigned int find_vde_by_name(const struct ddf_super *ddf,
2272 const char *name)
2273 {
2274 unsigned int i;
2275 if (name == NULL)
2276 return DDF_NOTFOUND;
2277 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2278 if (all_ff(ddf->virt->entries[i].guid))
2279 continue;
2280 if (!strncmp(name, ddf->virt->entries[i].name,
2281 sizeof(ddf->virt->entries[i].name)))
2282 return i;
2283 }
2284 return DDF_NOTFOUND;
2285 }
2286
2287 #ifndef MDASSEMBLE
2288 static unsigned int find_vde_by_guid(const struct ddf_super *ddf,
2289 const char *guid)
2290 {
2291 unsigned int i;
2292 if (guid == NULL || all_ff(guid))
2293 return DDF_NOTFOUND;
2294 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++)
2295 if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN))
2296 return i;
2297 return DDF_NOTFOUND;
2298 }
2299 #endif
2300
2301 static int init_super_ddf(struct supertype *st,
2302 mdu_array_info_t *info,
2303 unsigned long long size, char *name, char *homehost,
2304 int *uuid, unsigned long long data_offset)
2305 {
2306 /* This is primarily called by Create when creating a new array.
2307 * We will then get add_to_super called for each component, and then
2308 * write_init_super called to write it out to each device.
2309 * For DDF, Create can create on fresh devices or on a pre-existing
2310 * array.
2311 * To create on a pre-existing array a different method will be called.
2312 * This one is just for fresh drives.
2313 *
2314 * We need to create the entire 'ddf' structure which includes:
2315 * DDF headers - these are easy.
2316 * Controller data - a Sector describing this controller .. not that
2317 * this is a controller exactly.
2318 * Physical Disk Record - one entry per device, so
2319 * leave plenty of space.
2320 * Virtual Disk Records - again, just leave plenty of space.
2321 * This just lists VDs, doesn't give details.
2322 * Config records - describe the VDs that use this disk
2323 * DiskData - describes 'this' device.
2324 * BadBlockManagement - empty
2325 * Diag Space - empty
2326 * Vendor Logs - Could we put bitmaps here?
2327 *
2328 */
2329 struct ddf_super *ddf;
2330 char hostname[17];
2331 int hostlen;
2332 int max_phys_disks, max_virt_disks;
2333 unsigned long long sector;
2334 int clen;
2335 int i;
2336 int pdsize, vdsize;
2337 struct phys_disk *pd;
2338 struct virtual_disk *vd;
2339
2340 if (st->sb)
2341 return init_super_ddf_bvd(st, info, size, name, homehost, uuid,
2342 data_offset);
2343
2344 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
2345 pr_err("%s could not allocate superblock\n", __func__);
2346 return 0;
2347 }
2348 memset(ddf, 0, sizeof(*ddf));
2349 st->sb = ddf;
2350
2351 if (info == NULL) {
2352 /* zeroing superblock */
2353 return 0;
2354 }
2355
2356 /* At least 32MB *must* be reserved for the ddf. So let's just
2357 * start 32MB from the end, and put the primary header there.
2358 * Don't do secondary for now.
2359 * We don't know exactly where that will be yet as it could be
2360 * different on each device. So just set up the lengths.
2361 */
2362
2363 ddf->anchor.magic = DDF_HEADER_MAGIC;
2364 make_header_guid(ddf->anchor.guid);
2365
2366 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
2367 ddf->anchor.seq = cpu_to_be32(1);
2368 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
2369 ddf->anchor.openflag = 0xFF;
2370 ddf->anchor.foreignflag = 0;
2371 ddf->anchor.enforcegroups = 0; /* Is this best?? */
2372 ddf->anchor.pad0 = 0xff;
2373 memset(ddf->anchor.pad1, 0xff, 12);
2374 memset(ddf->anchor.header_ext, 0xff, 32);
2375 ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0);
2376 ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0);
2377 ddf->anchor.type = DDF_HEADER_ANCHOR;
2378 memset(ddf->anchor.pad2, 0xff, 3);
2379 ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */
2380 /* Put this at bottom of 32M reserved.. */
2381 ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0);
2382 max_phys_disks = 1023; /* Should be enough, 4095 is also allowed */
2383 ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks);
2384 max_virt_disks = 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2385 ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks);
2386 ddf->max_part = 64;
2387 ddf->anchor.max_partitions = cpu_to_be16(ddf->max_part);
2388 ddf->mppe = 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2389 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
2390 ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len);
2391 ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe);
2392 memset(ddf->anchor.pad3, 0xff, 54);
2393 /* Controller section is one sector long immediately
2394 * after the ddf header */
2395 sector = 1;
2396 ddf->anchor.controller_section_offset = cpu_to_be32(sector);
2397 ddf->anchor.controller_section_length = cpu_to_be32(1);
2398 sector += 1;
2399
2400 /* phys is 8 sectors after that */
2401 pdsize = ROUND_UP(sizeof(struct phys_disk) +
2402 sizeof(struct phys_disk_entry)*max_phys_disks,
2403 512);
2404 switch(pdsize/512) {
2405 case 2: case 8: case 32: case 128: case 512: break;
2406 default: abort();
2407 }
2408 ddf->anchor.phys_section_offset = cpu_to_be32(sector);
2409 ddf->anchor.phys_section_length =
2410 cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
2411 sector += pdsize/512;
2412
2413 /* virt is another 32 sectors */
2414 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
2415 sizeof(struct virtual_entry) * max_virt_disks,
2416 512);
2417 switch(vdsize/512) {
2418 case 2: case 8: case 32: case 128: case 512: break;
2419 default: abort();
2420 }
2421 ddf->anchor.virt_section_offset = cpu_to_be32(sector);
2422 ddf->anchor.virt_section_length =
2423 cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
2424 sector += vdsize/512;
2425
2426 clen = ddf->conf_rec_len * (ddf->max_part+1);
2427 ddf->anchor.config_section_offset = cpu_to_be32(sector);
2428 ddf->anchor.config_section_length = cpu_to_be32(clen);
2429 sector += clen;
2430
2431 ddf->anchor.data_section_offset = cpu_to_be32(sector);
2432 ddf->anchor.data_section_length = cpu_to_be32(1);
2433 sector += 1;
2434
2435 ddf->anchor.bbm_section_length = cpu_to_be32(0);
2436 ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF);
2437 ddf->anchor.diag_space_length = cpu_to_be32(0);
2438 ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF);
2439 ddf->anchor.vendor_length = cpu_to_be32(0);
2440 ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF);
2441
2442 memset(ddf->anchor.pad4, 0xff, 256);
2443
2444 memcpy(&ddf->primary, &ddf->anchor, 512);
2445 memcpy(&ddf->secondary, &ddf->anchor, 512);
2446
2447 ddf->primary.openflag = 1; /* I guess.. */
2448 ddf->primary.type = DDF_HEADER_PRIMARY;
2449
2450 ddf->secondary.openflag = 1; /* I guess.. */
2451 ddf->secondary.type = DDF_HEADER_SECONDARY;
2452
2453 ddf->active = &ddf->primary;
2454
2455 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
2456
2457 /* 24 more bytes of fiction required.
2458 * first 8 are a 'vendor-id' - "Linux-MD"
2459 * Remaining 16 are serial number.... maybe a hostname would do?
2460 */
2461 memcpy(ddf->controller.guid, T10, sizeof(T10));
2462 gethostname(hostname, sizeof(hostname));
2463 hostname[sizeof(hostname) - 1] = 0;
2464 hostlen = strlen(hostname);
2465 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
2466 for (i = strlen(T10) ; i+hostlen < 24; i++)
2467 ddf->controller.guid[i] = ' ';
2468
2469 ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD);
2470 ddf->controller.type.device_id = cpu_to_be16(0xBEEF);
2471 ddf->controller.type.sub_vendor_id = cpu_to_be16(0);
2472 ddf->controller.type.sub_device_id = cpu_to_be16(0);
2473 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
2474 memset(ddf->controller.pad, 0xff, 8);
2475 memset(ddf->controller.vendor_data, 0xff, 448);
2476 if (homehost && strlen(homehost) < 440)
2477 strcpy((char*)ddf->controller.vendor_data, homehost);
2478
2479 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
2480 pr_err("%s could not allocate pd\n", __func__);
2481 return 0;
2482 }
2483 ddf->phys = pd;
2484 ddf->pdsize = pdsize;
2485
2486 memset(pd, 0xff, pdsize);
2487 memset(pd, 0, sizeof(*pd));
2488 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2489 pd->used_pdes = cpu_to_be16(0);
2490 pd->max_pdes = cpu_to_be16(max_phys_disks);
2491 memset(pd->pad, 0xff, 52);
2492 for (i = 0; i < max_phys_disks; i++)
2493 memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN);
2494
2495 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
2496 pr_err("%s could not allocate vd\n", __func__);
2497 return 0;
2498 }
2499 ddf->virt = vd;
2500 ddf->vdsize = vdsize;
2501 memset(vd, 0, vdsize);
2502 vd->magic = DDF_VIRT_RECORDS_MAGIC;
2503 vd->populated_vdes = cpu_to_be16(0);
2504 vd->max_vdes = cpu_to_be16(max_virt_disks);
2505 memset(vd->pad, 0xff, 52);
2506
2507 for (i=0; i<max_virt_disks; i++)
2508 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
2509
2510 st->sb = ddf;
2511 ddf_set_updates_pending(ddf, NULL);
2512 return 1;
2513 }
2514
2515 static int chunk_to_shift(int chunksize)
2516 {
2517 return ffs(chunksize/512)-1;
2518 }
2519
2520 #ifndef MDASSEMBLE
2521 struct extent {
2522 unsigned long long start, size;
2523 };
2524 static int cmp_extent(const void *av, const void *bv)
2525 {
2526 const struct extent *a = av;
2527 const struct extent *b = bv;
2528 if (a->start < b->start)
2529 return -1;
2530 if (a->start > b->start)
2531 return 1;
2532 return 0;
2533 }
2534
2535 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
2536 {
2537 /* Find a list of used extents on the given physical device
2538 * (dnum) of the given ddf.
2539 * Return a malloced array of 'struct extent'
2540 */
2541 struct extent *rv;
2542 int n = 0;
2543 unsigned int i;
2544 __u16 state;
2545
2546 if (dl->pdnum < 0)
2547 return NULL;
2548 state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state);
2549
2550 if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online)
2551 return NULL;
2552
2553 rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2));
2554
2555 for (i = 0; i < ddf->max_part; i++) {
2556 const struct vd_config *bvd;
2557 unsigned int ibvd;
2558 struct vcl *v = dl->vlist[i];
2559 if (v == NULL ||
2560 get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe,
2561 &bvd, &ibvd) == DDF_NOTFOUND)
2562 continue;
2563 rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]);
2564 rv[n].size = be64_to_cpu(bvd->blocks);
2565 n++;
2566 }
2567 qsort(rv, n, sizeof(*rv), cmp_extent);
2568
2569 rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
2570 rv[n].size = 0;
2571 return rv;
2572 }
2573
2574 static unsigned long long find_space(
2575 struct ddf_super *ddf, struct dl *dl,
2576 unsigned long long data_offset,
2577 unsigned long long *size)
2578 {
2579 /* Find if the requested amount of space is available.
2580 * If it is, return start.
2581 * If not, set *size to largest space.
2582 * If data_offset != INVALID_SECTORS, then the space must start
2583 * at this location.
2584 */
2585 struct extent *e = get_extents(ddf, dl);
2586 int i = 0;
2587 unsigned long long pos = 0;
2588 unsigned long long max_size = 0;
2589
2590 if (!e) {
2591 *size = 0;
2592 return INVALID_SECTORS;
2593 }
2594 do {
2595 unsigned long long esize = e[i].start - pos;
2596 if (data_offset != INVALID_SECTORS &&
2597 pos <= data_offset &&
2598 e[i].start > data_offset) {
2599 pos = data_offset;
2600 esize = e[i].start - pos;
2601 }
2602 if (data_offset != INVALID_SECTORS &&
2603 pos != data_offset) {
2604 i++;
2605 continue;
2606 }
2607 if (esize >= *size) {
2608 /* Found! */
2609 free(e);
2610 return pos;
2611 }
2612 if (esize > max_size)
2613 max_size = esize;
2614 pos = e[i].start + e[i].size;
2615 i++;
2616 } while (e[i-1].size);
2617 *size = max_size;
2618 free(e);
2619 return INVALID_SECTORS;
2620 }
2621 #endif
2622
2623 static int init_super_ddf_bvd(struct supertype *st,
2624 mdu_array_info_t *info,
2625 unsigned long long size,
2626 char *name, char *homehost,
2627 int *uuid, unsigned long long data_offset)
2628 {
2629 /* We are creating a BVD inside a pre-existing container.
2630 * so st->sb is already set.
2631 * We need to create a new vd_config and a new virtual_entry
2632 */
2633 struct ddf_super *ddf = st->sb;
2634 unsigned int venum, i;
2635 struct virtual_entry *ve;
2636 struct vcl *vcl;
2637 struct vd_config *vc;
2638
2639 if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) {
2640 pr_err("This ddf already has an array called %s\n", name);
2641 return 0;
2642 }
2643 venum = find_unused_vde(ddf);
2644 if (venum == DDF_NOTFOUND) {
2645 pr_err("Cannot find spare slot for virtual disk\n");
2646 return 0;
2647 }
2648 ve = &ddf->virt->entries[venum];
2649
2650 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2651 * timestamp, random number
2652 */
2653 make_header_guid(ve->guid);
2654 ve->unit = cpu_to_be16(info->md_minor);
2655 ve->pad0 = 0xFFFF;
2656 ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid,
2657 DDF_GUID_LEN);
2658 ve->type = cpu_to_be16(0);
2659 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
2660 if (info->state & 1) /* clean */
2661 ve->init_state = DDF_init_full;
2662 else
2663 ve->init_state = DDF_init_not;
2664
2665 memset(ve->pad1, 0xff, 14);
2666 memset(ve->name, ' ', 16);
2667 if (name)
2668 strncpy(ve->name, name, 16);
2669 ddf->virt->populated_vdes =
2670 cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1);
2671
2672 /* Now create a new vd_config */
2673 if (posix_memalign((void**)&vcl, 512,
2674 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
2675 pr_err("%s could not allocate vd_config\n", __func__);
2676 return 0;
2677 }
2678 vcl->vcnum = venum;
2679 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2680 vc = &vcl->conf;
2681
2682 vc->magic = DDF_VD_CONF_MAGIC;
2683 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2684 vc->timestamp = cpu_to_be32(time(0)-DECADE);
2685 vc->seqnum = cpu_to_be32(1);
2686 memset(vc->pad0, 0xff, 24);
2687 vc->chunk_shift = chunk_to_shift(info->chunk_size);
2688 if (layout_md2ddf(info, vc) == -1 ||
2689 be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) {
2690 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2691 __func__, info->level, info->layout, info->raid_disks);
2692 free(vcl);
2693 return 0;
2694 }
2695 vc->sec_elmnt_seq = 0;
2696 if (alloc_other_bvds(ddf, vcl) != 0) {
2697 pr_err("%s could not allocate other bvds\n",
2698 __func__);
2699 free(vcl);
2700 return 0;
2701 }
2702 vc->blocks = cpu_to_be64(info->size * 2);
2703 vc->array_blocks = cpu_to_be64(
2704 calc_array_size(info->level, info->raid_disks, info->layout,
2705 info->chunk_size, info->size*2));
2706 memset(vc->pad1, 0xff, 8);
2707 vc->spare_refs[0] = cpu_to_be32(0xffffffff);
2708 vc->spare_refs[1] = cpu_to_be32(0xffffffff);
2709 vc->spare_refs[2] = cpu_to_be32(0xffffffff);
2710 vc->spare_refs[3] = cpu_to_be32(0xffffffff);
2711 vc->spare_refs[4] = cpu_to_be32(0xffffffff);
2712 vc->spare_refs[5] = cpu_to_be32(0xffffffff);
2713 vc->spare_refs[6] = cpu_to_be32(0xffffffff);
2714 vc->spare_refs[7] = cpu_to_be32(0xffffffff);
2715 memset(vc->cache_pol, 0, 8);
2716 vc->bg_rate = 0x80;
2717 memset(vc->pad2, 0xff, 3);
2718 memset(vc->pad3, 0xff, 52);
2719 memset(vc->pad4, 0xff, 192);
2720 memset(vc->v0, 0xff, 32);
2721 memset(vc->v1, 0xff, 32);
2722 memset(vc->v2, 0xff, 16);
2723 memset(vc->v3, 0xff, 16);
2724 memset(vc->vendor, 0xff, 32);
2725
2726 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2727 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2728
2729 for (i = 1; i < vc->sec_elmnt_count; i++) {
2730 memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512);
2731 vcl->other_bvds[i-1]->sec_elmnt_seq = i;
2732 }
2733
2734 vcl->next = ddf->conflist;
2735 ddf->conflist = vcl;
2736 ddf->currentconf = vcl;
2737 ddf_set_updates_pending(ddf, NULL);
2738 return 1;
2739 }
2740
2741 #ifndef MDASSEMBLE
2742 static void add_to_super_ddf_bvd(struct supertype *st,
2743 mdu_disk_info_t *dk, int fd, char *devname,
2744 unsigned long long data_offset)
2745 {
2746 /* fd and devname identify a device within the ddf container (st).
2747 * dk identifies a location in the new BVD.
2748 * We need to find suitable free space in that device and update
2749 * the phys_refnum and lba_offset for the newly created vd_config.
2750 * We might also want to update the type in the phys_disk
2751 * section.
2752 *
2753 * Alternately: fd == -1 and we have already chosen which device to
2754 * use and recorded in dlist->raid_disk;
2755 */
2756 struct dl *dl;
2757 struct ddf_super *ddf = st->sb;
2758 struct vd_config *vc;
2759 unsigned int i;
2760 unsigned long long blocks, pos;
2761 unsigned int raid_disk = dk->raid_disk;
2762
2763 if (fd == -1) {
2764 for (dl = ddf->dlist; dl ; dl = dl->next)
2765 if (dl->raiddisk == dk->raid_disk)
2766 break;
2767 } else {
2768 for (dl = ddf->dlist; dl ; dl = dl->next)
2769 if (dl->major == dk->major &&
2770 dl->minor == dk->minor)
2771 break;
2772 }
2773 if (!dl || dl->pdnum < 0 || ! (dk->state & (1<<MD_DISK_SYNC)))
2774 return;
2775
2776 vc = &ddf->currentconf->conf;
2777 if (vc->sec_elmnt_count > 1) {
2778 unsigned int n = be16_to_cpu(vc->prim_elmnt_count);
2779 if (raid_disk >= n)
2780 vc = ddf->currentconf->other_bvds[raid_disk / n - 1];
2781 raid_disk %= n;
2782 }
2783
2784 blocks = be64_to_cpu(vc->blocks);
2785 if (ddf->currentconf->block_sizes)
2786 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2787
2788 pos = find_space(ddf, dl, data_offset, &blocks);
2789 if (pos == INVALID_SECTORS)
2790 return;
2791
2792 ddf->currentdev = dk->raid_disk;
2793 vc->phys_refnum[raid_disk] = dl->disk.refnum;
2794 LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos);
2795
2796 for (i = 0; i < ddf->max_part ; i++)
2797 if (dl->vlist[i] == NULL)
2798 break;
2799 if (i == ddf->max_part)
2800 return;
2801 dl->vlist[i] = ddf->currentconf;
2802
2803 if (fd >= 0)
2804 dl->fd = fd;
2805 if (devname)
2806 dl->devname = devname;
2807
2808 /* Check if we can mark array as optimal yet */
2809 i = ddf->currentconf->vcnum;
2810 ddf->virt->entries[i].state =
2811 (ddf->virt->entries[i].state & ~DDF_state_mask)
2812 | get_svd_state(ddf, ddf->currentconf);
2813 be16_clear(ddf->phys->entries[dl->pdnum].type,
2814 cpu_to_be16(DDF_Global_Spare));
2815 be16_set(ddf->phys->entries[dl->pdnum].type,
2816 cpu_to_be16(DDF_Active_in_VD));
2817 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2818 __func__, dl->pdnum, be32_to_cpu(dl->disk.refnum),
2819 ddf->currentconf->vcnum, guid_str(vc->guid),
2820 dk->raid_disk);
2821 ddf_set_updates_pending(ddf, vc);
2822 }
2823
2824 static unsigned int find_unused_pde(const struct ddf_super *ddf)
2825 {
2826 unsigned int i;
2827 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) {
2828 if (all_ff(ddf->phys->entries[i].guid))
2829 return i;
2830 }
2831 return DDF_NOTFOUND;
2832 }
2833
2834 static void _set_config_size(struct phys_disk_entry *pde, const struct dl *dl)
2835 {
2836 __u64 cfs, t;
2837 cfs = min(dl->size - 32*1024*2ULL, be64_to_cpu(dl->primary_lba));
2838 t = be64_to_cpu(dl->secondary_lba);
2839 if (t != ~(__u64)0)
2840 cfs = min(cfs, t);
2841 /*
2842 * Some vendor DDF structures interpret workspace_lba
2843 * very differently than we do: Make a sanity check on the value.
2844 */
2845 t = be64_to_cpu(dl->workspace_lba);
2846 if (t < cfs) {
2847 __u64 wsp = cfs - t;
2848 if (wsp > 1024*1024*2ULL && wsp > dl->size / 16) {
2849 pr_err("%s: %x:%x: workspace size 0x%llx too big, ignoring\n",
2850 __func__, dl->major, dl->minor, wsp);
2851 } else
2852 cfs = t;
2853 }
2854 pde->config_size = cpu_to_be64(cfs);
2855 dprintf("%s: %x:%x config_size %llx, DDF structure is %llx blocks\n",
2856 __func__, dl->major, dl->minor, cfs, dl->size-cfs);
2857 }
2858
2859 /* Add a device to a container, either while creating it or while
2860 * expanding a pre-existing container
2861 */
2862 static int add_to_super_ddf(struct supertype *st,
2863 mdu_disk_info_t *dk, int fd, char *devname,
2864 unsigned long long data_offset)
2865 {
2866 struct ddf_super *ddf = st->sb;
2867 struct dl *dd;
2868 time_t now;
2869 struct tm *tm;
2870 unsigned long long size;
2871 struct phys_disk_entry *pde;
2872 unsigned int n, i;
2873 struct stat stb;
2874 __u32 *tptr;
2875
2876 if (ddf->currentconf) {
2877 add_to_super_ddf_bvd(st, dk, fd, devname, data_offset);
2878 return 0;
2879 }
2880
2881 /* This is device numbered dk->number. We need to create
2882 * a phys_disk entry and a more detailed disk_data entry.
2883 */
2884 fstat(fd, &stb);
2885 n = find_unused_pde(ddf);
2886 if (n == DDF_NOTFOUND) {
2887 pr_err("%s: No free slot in array, cannot add disk\n",
2888 __func__);
2889 return 1;
2890 }
2891 pde = &ddf->phys->entries[n];
2892 get_dev_size(fd, NULL, &size);
2893 if (size <= 32*1024*1024) {
2894 pr_err("%s: device size must be at least 32MB\n",
2895 __func__);
2896 return 1;
2897 }
2898 size >>= 9;
2899
2900 if (posix_memalign((void**)&dd, 512,
2901 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2902 pr_err("%s could allocate buffer for new disk, aborting\n",
2903 __func__);
2904 return 1;
2905 }
2906 dd->major = major(stb.st_rdev);
2907 dd->minor = minor(stb.st_rdev);
2908 dd->devname = devname;
2909 dd->fd = fd;
2910 dd->spare = NULL;
2911
2912 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2913 now = time(0);
2914 tm = localtime(&now);
2915 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2916 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2917 tptr = (__u32 *)(dd->disk.guid + 16);
2918 *tptr++ = random32();
2919 *tptr = random32();
2920
2921 do {
2922 /* Cannot be bothered finding a CRC of some irrelevant details*/
2923 dd->disk.refnum._v32 = random32();
2924 for (i = be16_to_cpu(ddf->active->max_pd_entries);
2925 i > 0; i--)
2926 if (be32_eq(ddf->phys->entries[i-1].refnum,
2927 dd->disk.refnum))
2928 break;
2929 } while (i > 0);
2930
2931 dd->disk.forced_ref = 1;
2932 dd->disk.forced_guid = 1;
2933 memset(dd->disk.vendor, ' ', 32);
2934 memcpy(dd->disk.vendor, "Linux", 5);
2935 memset(dd->disk.pad, 0xff, 442);
2936 for (i = 0; i < ddf->max_part ; i++)
2937 dd->vlist[i] = NULL;
2938
2939 dd->pdnum = n;
2940
2941 if (st->update_tail) {
2942 int len = (sizeof(struct phys_disk) +
2943 sizeof(struct phys_disk_entry));
2944 struct phys_disk *pd;
2945
2946 pd = xmalloc(len);
2947 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2948 pd->used_pdes = cpu_to_be16(n);
2949 pde = &pd->entries[0];
2950 dd->mdupdate = pd;
2951 } else
2952 ddf->phys->used_pdes = cpu_to_be16(
2953 1 + be16_to_cpu(ddf->phys->used_pdes));
2954
2955 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2956 pde->refnum = dd->disk.refnum;
2957 pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2958 pde->state = cpu_to_be16(DDF_Online);
2959 dd->size = size;
2960 /*
2961 * If there is already a device in dlist, try to reserve the same
2962 * amount of workspace. Otherwise, use 32MB.
2963 * We checked disk size above already.
2964 */
2965 #define __calc_lba(new, old, lba, mb) do { \
2966 unsigned long long dif; \
2967 if ((old) != NULL) \
2968 dif = (old)->size - be64_to_cpu((old)->lba); \
2969 else \
2970 dif = (new)->size; \
2971 if ((new)->size > dif) \
2972 (new)->lba = cpu_to_be64((new)->size - dif); \
2973 else \
2974 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2975 } while (0)
2976 __calc_lba(dd, ddf->dlist, workspace_lba, 32);
2977 __calc_lba(dd, ddf->dlist, primary_lba, 16);
2978 if (ddf->dlist == NULL ||
2979 be64_to_cpu(ddf->dlist->secondary_lba) != ~(__u64)0)
2980 __calc_lba(dd, ddf->dlist, secondary_lba, 32);
2981 _set_config_size(pde, dd);
2982
2983 sprintf(pde->path, "%17.17s","Information: nil") ;
2984 memset(pde->pad, 0xff, 6);
2985
2986 if (st->update_tail) {
2987 dd->next = ddf->add_list;
2988 ddf->add_list = dd;
2989 } else {
2990 dd->next = ddf->dlist;
2991 ddf->dlist = dd;
2992 ddf_set_updates_pending(ddf, NULL);
2993 }
2994
2995 return 0;
2996 }
2997
2998 static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk)
2999 {
3000 struct ddf_super *ddf = st->sb;
3001 struct dl *dl;
3002
3003 /* mdmon has noticed that this disk (dk->major/dk->minor) has
3004 * disappeared from the container.
3005 * We need to arrange that it disappears from the metadata and
3006 * internal data structures too.
3007 * Most of the work is done by ddf_process_update which edits
3008 * the metadata and closes the file handle and attaches the memory
3009 * where free_updates will free it.
3010 */
3011 for (dl = ddf->dlist; dl ; dl = dl->next)
3012 if (dl->major == dk->major &&
3013 dl->minor == dk->minor)
3014 break;
3015 if (!dl || dl->pdnum < 0)
3016 return -1;
3017
3018 if (st->update_tail) {
3019 int len = (sizeof(struct phys_disk) +
3020 sizeof(struct phys_disk_entry));
3021 struct phys_disk *pd;
3022
3023 pd = xmalloc(len);
3024 pd->magic = DDF_PHYS_RECORDS_MAGIC;
3025 pd->used_pdes = cpu_to_be16(dl->pdnum);
3026 pd->entries[0].state = cpu_to_be16(DDF_Missing);
3027 append_metadata_update(st, pd, len);
3028 }
3029 return 0;
3030 }
3031 #endif
3032
3033 /*
3034 * This is the write_init_super method for a ddf container. It is
3035 * called when creating a container or adding another device to a
3036 * container.
3037 */
3038
3039 static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type)
3040 {
3041 unsigned long long sector;
3042 struct ddf_header *header;
3043 int fd, i, n_config, conf_size, buf_size;
3044 int ret = 0;
3045 char *conf;
3046
3047 fd = d->fd;
3048
3049 switch (type) {
3050 case DDF_HEADER_PRIMARY:
3051 header = &ddf->primary;
3052 sector = be64_to_cpu(header->primary_lba);
3053 break;
3054 case DDF_HEADER_SECONDARY:
3055 header = &ddf->secondary;
3056 sector = be64_to_cpu(header->secondary_lba);
3057 break;
3058 default:
3059 return 0;
3060 }
3061 if (sector == ~(__u64)0)
3062 return 0;
3063
3064 header->type = type;
3065 header->openflag = 1;
3066 header->crc = calc_crc(header, 512);
3067
3068 lseek64(fd, sector<<9, 0);
3069 if (write(fd, header, 512) < 0)
3070 goto out;
3071
3072 ddf->controller.crc = calc_crc(&ddf->controller, 512);
3073 if (write(fd, &ddf->controller, 512) < 0)
3074 goto out;
3075
3076 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
3077 if (write(fd, ddf->phys, ddf->pdsize) < 0)
3078 goto out;
3079 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
3080 if (write(fd, ddf->virt, ddf->vdsize) < 0)
3081 goto out;
3082
3083 /* Now write lots of config records. */
3084 n_config = ddf->max_part;
3085 conf_size = ddf->conf_rec_len * 512;
3086 conf = ddf->conf;
3087 buf_size = conf_size * (n_config + 1);
3088 if (!conf) {
3089 if (posix_memalign((void**)&conf, 512, buf_size) != 0)
3090 goto out;
3091 ddf->conf = conf;
3092 }
3093 for (i = 0 ; i <= n_config ; i++) {
3094 struct vcl *c;
3095 struct vd_config *vdc = NULL;
3096 if (i == n_config) {
3097 c = (struct vcl *)d->spare;
3098 if (c)
3099 vdc = &c->conf;
3100 } else {
3101 unsigned int dummy;
3102 c = d->vlist[i];
3103 if (c)
3104 get_pd_index_from_refnum(
3105 c, d->disk.refnum,
3106 ddf->mppe,
3107 (const struct vd_config **)&vdc,
3108 &dummy);
3109 }
3110 if (vdc) {
3111 dprintf("writing conf record %i on disk %08x for %s/%u\n",
3112 i, be32_to_cpu(d->disk.refnum),
3113 guid_str(vdc->guid),
3114 vdc->sec_elmnt_seq);
3115 vdc->crc = calc_crc(vdc, conf_size);
3116 memcpy(conf + i*conf_size, vdc, conf_size);
3117 } else
3118 memset(conf + i*conf_size, 0xff, conf_size);
3119 }
3120 if (write(fd, conf, buf_size) != buf_size)
3121 goto out;
3122
3123 d->disk.crc = calc_crc(&d->disk, 512);
3124 if (write(fd, &d->disk, 512) < 0)
3125 goto out;
3126
3127 ret = 1;
3128 out:
3129 header->openflag = 0;
3130 header->crc = calc_crc(header, 512);
3131
3132 lseek64(fd, sector<<9, 0);
3133 if (write(fd, header, 512) < 0)
3134 ret = 0;
3135
3136 return ret;
3137 }
3138
3139 static int _write_super_to_disk(struct ddf_super *ddf, struct dl *d)
3140 {
3141 unsigned long long size;
3142 int fd = d->fd;
3143 if (fd < 0)
3144 return 0;
3145
3146 /* We need to fill in the primary, (secondary) and workspace
3147 * lba's in the headers, set their checksums,
3148 * Also checksum phys, virt....
3149 *
3150 * Then write everything out, finally the anchor is written.
3151 */
3152 get_dev_size(fd, NULL, &size);
3153 size /= 512;
3154 memcpy(&ddf->anchor, ddf->active, 512);
3155 if (be64_to_cpu(d->workspace_lba) != 0ULL)
3156 ddf->anchor.workspace_lba = d->workspace_lba;
3157 else
3158 ddf->anchor.workspace_lba =
3159 cpu_to_be64(size - 32*1024*2);
3160 if (be64_to_cpu(d->primary_lba) != 0ULL)
3161 ddf->anchor.primary_lba = d->primary_lba;
3162 else
3163 ddf->anchor.primary_lba =
3164 cpu_to_be64(size - 16*1024*2);
3165 if (be64_to_cpu(d->secondary_lba) != 0ULL)
3166 ddf->anchor.secondary_lba = d->secondary_lba;
3167 else
3168 ddf->anchor.secondary_lba =
3169 cpu_to_be64(size - 32*1024*2);
3170 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
3171 memcpy(&ddf->primary, &ddf->anchor, 512);
3172 memcpy(&ddf->secondary, &ddf->anchor, 512);
3173
3174 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
3175 ddf->anchor.seq = cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3176 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
3177
3178 if (!__write_ddf_structure(d, ddf, DDF_HEADER_PRIMARY))
3179 return 0;
3180
3181 if (!__write_ddf_structure(d, ddf, DDF_HEADER_SECONDARY))
3182 return 0;
3183
3184 lseek64(fd, (size-1)*512, SEEK_SET);
3185 if (write(fd, &ddf->anchor, 512) < 0)
3186 return 0;
3187
3188 return 1;
3189 }
3190
3191 #ifndef MDASSEMBLE
3192 static int __write_init_super_ddf(struct supertype *st)
3193 {
3194 struct ddf_super *ddf = st->sb;
3195 struct dl *d;
3196 int attempts = 0;
3197 int successes = 0;
3198
3199 pr_state(ddf, __func__);
3200
3201 /* try to write updated metadata,
3202 * if we catch a failure move on to the next disk
3203 */
3204 for (d = ddf->dlist; d; d=d->next) {
3205 attempts++;
3206 successes += _write_super_to_disk(ddf, d);
3207 }
3208
3209 return attempts != successes;
3210 }
3211
3212 static int write_init_super_ddf(struct supertype *st)
3213 {
3214 struct ddf_super *ddf = st->sb;
3215 struct vcl *currentconf = ddf->currentconf;
3216
3217 /* We are done with currentconf - reset it so st refers to the container */
3218 ddf->currentconf = NULL;
3219
3220 if (st->update_tail) {
3221 /* queue the virtual_disk and vd_config as metadata updates */
3222 struct virtual_disk *vd;
3223 struct vd_config *vc;
3224 int len, tlen;
3225 unsigned int i;
3226
3227 if (!currentconf) {
3228 /* Must be adding a physical disk to the container */
3229 int len = (sizeof(struct phys_disk) +
3230 sizeof(struct phys_disk_entry));
3231
3232 /* adding a disk to the container. */
3233 if (!ddf->add_list)
3234 return 0;
3235
3236 append_metadata_update(st, ddf->add_list->mdupdate, len);
3237 ddf->add_list->mdupdate = NULL;
3238 return 0;
3239 }
3240
3241 /* Newly created VD */
3242
3243 /* First the virtual disk. We have a slightly fake header */
3244 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
3245 vd = xmalloc(len);
3246 *vd = *ddf->virt;
3247 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
3248 vd->populated_vdes = cpu_to_be16(currentconf->vcnum);
3249 append_metadata_update(st, vd, len);
3250
3251 /* Then the vd_config */
3252 len = ddf->conf_rec_len * 512;
3253 tlen = len * currentconf->conf.sec_elmnt_count;
3254 vc = xmalloc(tlen);
3255 memcpy(vc, &currentconf->conf, len);
3256 for (i = 1; i < currentconf->conf.sec_elmnt_count; i++)
3257 memcpy((char *)vc + i*len, currentconf->other_bvds[i-1],
3258 len);
3259 append_metadata_update(st, vc, tlen);
3260
3261 return 0;
3262 } else {
3263 struct dl *d;
3264 if (!currentconf)
3265 for (d = ddf->dlist; d; d=d->next)
3266 while (Kill(d->devname, NULL, 0, -1, 1) == 0);
3267 /* Note: we don't close the fd's now, but a subsequent
3268 * ->free_super() will
3269 */
3270 return __write_init_super_ddf(st);
3271 }
3272 }
3273
3274 #endif
3275
3276 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize,
3277 unsigned long long data_offset)
3278 {
3279 /* We must reserve the last 32Meg */
3280 if (devsize <= 32*1024*2)
3281 return 0;
3282 return devsize - 32*1024*2;
3283 }
3284
3285 #ifndef MDASSEMBLE
3286
3287 static int reserve_space(struct supertype *st, int raiddisks,
3288 unsigned long long size, int chunk,
3289 unsigned long long data_offset,
3290 unsigned long long *freesize)
3291 {
3292 /* Find 'raiddisks' spare extents at least 'size' big (but
3293 * only caring about multiples of 'chunk') and remember
3294 * them. If size==0, find the largest size possible.
3295 * Report available size in *freesize
3296 * If space cannot be found, fail.
3297 */
3298 struct dl *dl;
3299 struct ddf_super *ddf = st->sb;
3300 int cnt = 0;
3301
3302 for (dl = ddf->dlist; dl ; dl=dl->next) {
3303 dl->raiddisk = -1;
3304 dl->esize = 0;
3305 }
3306 /* Now find largest extent on each device */
3307 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3308 unsigned long long minsize = ULLONG_MAX;
3309
3310 find_space(ddf, dl, data_offset, &minsize);
3311 if (minsize >= size && minsize >= (unsigned)chunk) {
3312 cnt++;
3313 dl->esize = minsize;
3314 }
3315 }
3316 if (cnt < raiddisks) {
3317 pr_err("not enough devices with space to create array.\n");
3318 return 0; /* No enough free spaces large enough */
3319 }
3320 if (size == 0) {
3321 /* choose the largest size of which there are at least 'raiddisk' */
3322 for (dl = ddf->dlist ; dl ; dl=dl->next) {
3323 struct dl *dl2;
3324 if (dl->esize <= size)
3325 continue;
3326 /* This is bigger than 'size', see if there are enough */
3327 cnt = 0;
3328 for (dl2 = ddf->dlist; dl2 ; dl2=dl2->next)
3329 if (dl2->esize >= dl->esize)
3330 cnt++;
3331 if (cnt >= raiddisks)
3332 size = dl->esize;
3333 }
3334 if (chunk) {
3335 size = size / chunk;
3336 size *= chunk;
3337 }
3338 *freesize = size;
3339 if (size < 32) {
3340 pr_err("not enough spare devices to create array.\n");
3341 return 0;
3342 }
3343 }
3344 /* We have a 'size' of which there are enough spaces.
3345 * We simply do a first-fit */
3346 cnt = 0;
3347 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
3348 if (dl->esize < size)
3349 continue;
3350
3351 dl->raiddisk = cnt;
3352 cnt++;
3353 }
3354 return 1;
3355 }
3356
3357 static int validate_geometry_ddf(struct supertype *st,
3358 int level, int layout, int raiddisks,
3359 int *chunk, unsigned long long size,
3360 unsigned long long data_offset,
3361 char *dev, unsigned long long *freesize,
3362 int verbose)
3363 {
3364 int fd;
3365 struct mdinfo *sra;
3366 int cfd;
3367
3368 /* ddf potentially supports lots of things, but it depends on
3369 * what devices are offered (and maybe kernel version?)
3370 * If given unused devices, we will make a container.
3371 * If given devices in a container, we will make a BVD.
3372 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3373 */
3374
3375 if (*chunk == UnSet)
3376 *chunk = DEFAULT_CHUNK;
3377
3378 if (level == LEVEL_NONE)
3379 level = LEVEL_CONTAINER;
3380 if (level == LEVEL_CONTAINER) {
3381 /* Must be a fresh device to add to a container */
3382 return validate_geometry_ddf_container(st, level, layout,
3383 raiddisks, *chunk,
3384 size, data_offset, dev,
3385 freesize,
3386 verbose);
3387 }
3388
3389 if (!dev) {
3390 mdu_array_info_t array = {
3391 .level = level,
3392 .layout = layout,
3393 .raid_disks = raiddisks
3394 };
3395 struct vd_config conf;
3396 if (layout_md2ddf(&array, &conf) == -1) {
3397 if (verbose)
3398 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3399 level, layout, raiddisks);
3400 return 0;
3401 }
3402 /* Should check layout? etc */
3403
3404 if (st->sb && freesize) {
3405 /* --create was given a container to create in.
3406 * So we need to check that there are enough
3407 * free spaces and return the amount of space.
3408 * We may as well remember which drives were
3409 * chosen so that add_to_super/getinfo_super
3410 * can return them.
3411 */
3412 return reserve_space(st, raiddisks, size, *chunk,
3413 data_offset, freesize);
3414 }
3415 return 1;
3416 }
3417
3418 if (st->sb) {
3419 /* A container has already been opened, so we are
3420 * creating in there. Maybe a BVD, maybe an SVD.
3421 * Should make a distinction one day.
3422 */
3423 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
3424 chunk, size, data_offset, dev,
3425 freesize,
3426 verbose);
3427 }
3428 /* This is the first device for the array.
3429 * If it is a container, we read it in and do automagic allocations,
3430 * no other devices should be given.
3431 * Otherwise it must be a member device of a container, and we
3432 * do manual allocation.
3433 * Later we should check for a BVD and make an SVD.
3434 */
3435 fd = open(dev, O_RDONLY|O_EXCL, 0);
3436 if (fd >= 0) {
3437 close(fd);
3438 /* Just a bare device, no good to us */
3439 if (verbose)
3440 pr_err("ddf: Cannot create this array "
3441 "on device %s - a container is required.\n",
3442 dev);
3443 return 0;
3444 }
3445 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3446 if (verbose)
3447 pr_err("ddf: Cannot open %s: %s\n",
3448 dev, strerror(errno));
3449 return 0;
3450 }
3451 /* Well, it is in use by someone, maybe a 'ddf' container. */
3452 cfd = open_container(fd);
3453 if (cfd < 0) {
3454 close(fd);
3455 if (verbose)
3456 pr_err("ddf: Cannot use %s: %s\n",
3457 dev, strerror(EBUSY));
3458 return 0;
3459 }
3460 sra = sysfs_read(cfd, NULL, GET_VERSION);
3461 close(fd);
3462 if (sra && sra->array.major_version == -1 &&
3463 strcmp(sra->text_version, "ddf") == 0) {
3464 /* This is a member of a ddf container. Load the container
3465 * and try to create a bvd
3466 */
3467 struct ddf_super *ddf;
3468 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL) == 0) {
3469 st->sb = ddf;
3470 strcpy(st->container_devnm, fd2devnm(cfd));
3471 close(cfd);
3472 return validate_geometry_ddf_bvd(st, level, layout,
3473 raiddisks, chunk, size,
3474 data_offset,
3475 dev, freesize,
3476 verbose);
3477 }
3478 close(cfd);
3479 } else /* device may belong to a different container */
3480 return 0;
3481
3482 return 1;
3483 }
3484
3485 static int
3486 validate_geometry_ddf_container(struct supertype *st,
3487 int level, int layout, int raiddisks,
3488 int chunk, unsigned long long size,
3489 unsigned long long data_offset,
3490 char *dev, unsigned long long *freesize,
3491 int verbose)
3492 {
3493 int fd;
3494 unsigned long long ldsize;
3495
3496 if (level != LEVEL_CONTAINER)
3497 return 0;
3498 if (!dev)
3499 return 1;
3500
3501 fd = open(dev, O_RDONLY|O_EXCL, 0);
3502 if (fd < 0) {
3503 if (verbose)
3504 pr_err("ddf: Cannot open %s: %s\n",
3505 dev, strerror(errno));
3506 return 0;
3507 }
3508 if (!get_dev_size(fd, dev, &ldsize)) {
3509 close(fd);
3510 return 0;
3511 }
3512 close(fd);
3513
3514 *freesize = avail_size_ddf(st, ldsize >> 9, INVALID_SECTORS);
3515 if (*freesize == 0)
3516 return 0;
3517
3518 return 1;
3519 }
3520
3521 static int validate_geometry_ddf_bvd(struct supertype *st,
3522 int level, int layout, int raiddisks,
3523 int *chunk, unsigned long long size,
3524 unsigned long long data_offset,
3525 char *dev, unsigned long long *freesize,
3526 int verbose)
3527 {
3528 struct stat stb;
3529 struct ddf_super *ddf = st->sb;
3530 struct dl *dl;
3531 unsigned long long maxsize;
3532 /* ddf/bvd supports lots of things, but not containers */
3533 if (level == LEVEL_CONTAINER) {
3534 if (verbose)
3535 pr_err("DDF cannot create a container within an container\n");
3536 return 0;
3537 }
3538 /* We must have the container info already read in. */
3539 if (!ddf)
3540 return 0;
3541
3542 if (!dev) {
3543 /* General test: make sure there is space for
3544 * 'raiddisks' device extents of size 'size'.
3545 */
3546 unsigned long long minsize = size;
3547 int dcnt = 0;
3548 if (minsize == 0)
3549 minsize = 8;
3550 for (dl = ddf->dlist; dl ; dl = dl->next) {
3551 if (find_space(ddf, dl, data_offset, &minsize)
3552 != INVALID_SECTORS)
3553 dcnt++;
3554 }
3555 if (dcnt < raiddisks) {
3556 if (verbose)
3557 pr_err("ddf: Not enough devices with "
3558 "space for this array (%d < %d)\n",
3559 dcnt, raiddisks);
3560 return 0;
3561 }
3562 return 1;
3563 }
3564 /* This device must be a member of the set */
3565 if (stat(dev, &stb) < 0)
3566 return 0;
3567 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3568 return 0;
3569 for (dl = ddf->dlist ; dl ; dl = dl->next) {
3570 if (dl->major == (int)major(stb.st_rdev) &&
3571 dl->minor == (int)minor(stb.st_rdev))
3572 break;
3573 }
3574 if (!dl) {
3575 if (verbose)
3576 pr_err("ddf: %s is not in the "
3577 "same DDF set\n",
3578 dev);
3579 return 0;
3580 }
3581 maxsize = ULLONG_MAX;
3582 find_space(ddf, dl, data_offset, &maxsize);
3583 *freesize = maxsize;
3584
3585 return 1;
3586 }
3587
3588 static int load_super_ddf_all(struct supertype *st, int fd,
3589 void **sbp, char *devname)
3590 {
3591 struct mdinfo *sra;
3592 struct ddf_super *super;
3593 struct mdinfo *sd, *best = NULL;
3594 int bestseq = 0;
3595 int seq;
3596 char nm[20];
3597 int dfd;
3598
3599 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
3600 if (!sra)
3601 return 1;
3602 if (sra->array.major_version != -1 ||
3603 sra->array.minor_version != -2 ||
3604 strcmp(sra->text_version, "ddf") != 0)
3605 return 1;
3606
3607 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
3608 return 1;
3609 memset(super, 0, sizeof(*super));
3610
3611 /* first, try each device, and choose the best ddf */
3612 for (sd = sra->devs ; sd ; sd = sd->next) {
3613 int rv;
3614 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3615 dfd = dev_open(nm, O_RDONLY);
3616 if (dfd < 0)
3617 return 2;
3618 rv = load_ddf_headers(dfd, super, NULL);
3619 close(dfd);
3620 if (rv == 0) {
3621 seq = be32_to_cpu(super->active->seq);
3622 if (super->active->openflag)
3623 seq--;
3624 if (!best || seq > bestseq) {
3625 bestseq = seq;
3626 best = sd;
3627 }
3628 }
3629 }
3630 if (!best)
3631 return 1;
3632 /* OK, load this ddf */
3633 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
3634 dfd = dev_open(nm, O_RDONLY);
3635 if (dfd < 0)
3636 return 1;
3637 load_ddf_headers(dfd, super, NULL);
3638 load_ddf_global(dfd, super, NULL);
3639 close(dfd);
3640 /* Now we need the device-local bits */
3641 for (sd = sra->devs ; sd ; sd = sd->next) {
3642 int rv;
3643
3644 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
3645 dfd = dev_open(nm, O_RDWR);
3646 if (dfd < 0)
3647 return 2;
3648 rv = load_ddf_headers(dfd, super, NULL);
3649 if (rv == 0)
3650 rv = load_ddf_local(dfd, super, NULL, 1);
3651 if (rv)
3652 return 1;
3653 }
3654
3655 *sbp = super;
3656 if (st->ss == NULL) {
3657 st->ss = &super_ddf;
3658 st->minor_version = 0;
3659 st->max_devs = 512;
3660 }
3661 strcpy(st->container_devnm, fd2devnm(fd));
3662 return 0;
3663 }
3664
3665 static int load_container_ddf(struct supertype *st, int fd,
3666 char *devname)
3667 {
3668 return load_super_ddf_all(st, fd, &st->sb, devname);
3669 }
3670
3671 #endif /* MDASSEMBLE */
3672
3673 static int check_secondary(const struct vcl *vc)
3674 {
3675 const struct vd_config *conf = &vc->conf;
3676 int i;
3677
3678 /* The only DDF secondary RAID level md can support is
3679 * RAID 10, if the stripe sizes and Basic volume sizes
3680 * are all equal.
3681 * Other configurations could in theory be supported by exposing
3682 * the BVDs to user space and using device mapper for the secondary
3683 * mapping. So far we don't support that.
3684 */
3685
3686 __u64 sec_elements[4] = {0, 0, 0, 0};
3687 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3688 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3689
3690 if (vc->other_bvds == NULL) {
3691 pr_err("No BVDs for secondary RAID found\n");
3692 return -1;
3693 }
3694 if (conf->prl != DDF_RAID1) {
3695 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3696 return -1;
3697 }
3698 if (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED) {
3699 pr_err("Secondary RAID level %d is unsupported\n",
3700 conf->srl);
3701 return -1;
3702 }
3703 __set_sec_seen(conf->sec_elmnt_seq);
3704 for (i = 0; i < conf->sec_elmnt_count-1; i++) {
3705 const struct vd_config *bvd = vc->other_bvds[i];
3706 if (bvd->sec_elmnt_seq == DDF_UNUSED_BVD)
3707 continue;
3708 if (bvd->srl != conf->srl) {
3709 pr_err("Inconsistent secondary RAID level across BVDs\n");
3710 return -1;
3711 }
3712 if (bvd->prl != conf->prl) {
3713 pr_err("Different RAID levels for BVDs are unsupported\n");
3714 return -1;
3715 }
3716 if (!be16_eq(bvd->prim_elmnt_count, conf->prim_elmnt_count)) {
3717 pr_err("All BVDs must have the same number of primary elements\n");
3718 return -1;
3719 }
3720 if (bvd->chunk_shift != conf->chunk_shift) {
3721 pr_err("Different strip sizes for BVDs are unsupported\n");
3722 return -1;
3723 }
3724 if (!be64_eq(bvd->array_blocks, conf->array_blocks)) {
3725 pr_err("Different BVD sizes are unsupported\n");
3726 return -1;
3727 }
3728 __set_sec_seen(bvd->sec_elmnt_seq);
3729 }
3730 for (i = 0; i < conf->sec_elmnt_count; i++) {
3731 if (!__was_sec_seen(i)) {
3732 /* pr_err("BVD %d is missing\n", i); */
3733 return -1;
3734 }
3735 }
3736 return 0;
3737 }
3738
3739 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
3740 be32 refnum, unsigned int nmax,
3741 const struct vd_config **bvd,
3742 unsigned int *idx)
3743 {
3744 unsigned int i, j, n, sec, cnt;
3745
3746 cnt = be16_to_cpu(vc->conf.prim_elmnt_count);
3747 sec = (vc->conf.sec_elmnt_count == 1 ? 0 : vc->conf.sec_elmnt_seq);
3748
3749 for (i = 0, j = 0 ; i < nmax ; i++) {
3750 /* j counts valid entries for this BVD */
3751 if (be32_eq(vc->conf.phys_refnum[i], refnum)) {
3752 *bvd = &vc->conf;
3753 *idx = i;
3754 return sec * cnt + j;
3755 }
3756 if (be32_to_cpu(vc->conf.phys_refnum[i]) != 0xffffffff)
3757 j++;
3758 }
3759 if (vc->other_bvds == NULL)
3760 goto bad;
3761
3762 for (n = 1; n < vc->conf.sec_elmnt_count; n++) {
3763 struct vd_config *vd = vc->other_bvds[n-1];
3764 sec = vd->sec_elmnt_seq;
3765 if (sec == DDF_UNUSED_BVD)
3766 continue;
3767 for (i = 0, j = 0 ; i < nmax ; i++) {
3768 if (be32_eq(vd->phys_refnum[i], refnum)) {
3769 *bvd = vd;
3770 *idx = i;
3771 return sec * cnt + j;
3772 }
3773 if (be32_to_cpu(vd->phys_refnum[i]) != 0xffffffff)
3774 j++;
3775 }
3776 }
3777 bad:
3778 *bvd = NULL;
3779 return DDF_NOTFOUND;
3780 }
3781
3782 static struct mdinfo *container_content_ddf(struct supertype *st, char *subarray)
3783 {
3784 /* Given a container loaded by load_super_ddf_all,
3785 * extract information about all the arrays into
3786 * an mdinfo tree.
3787 *
3788 * For each vcl in conflist: create an mdinfo, fill it in,
3789 * then look for matching devices (phys_refnum) in dlist
3790 * and create appropriate device mdinfo.
3791 */
3792 struct ddf_super *ddf = st->sb;
3793 struct mdinfo *rest = NULL;
3794 struct vcl *vc;
3795
3796 for (vc = ddf->conflist ; vc ; vc=vc->next) {
3797 unsigned int i;
3798 struct mdinfo *this;
3799 char *ep;
3800 __u32 *cptr;
3801 unsigned int pd;
3802
3803 if (subarray &&
3804 (strtoul(subarray, &ep, 10) != vc->vcnum ||
3805 *ep != '\0'))
3806 continue;
3807
3808 if (vc->conf.sec_elmnt_count > 1) {
3809 if (check_secondary(vc) != 0)
3810 continue;
3811 }
3812
3813 this = xcalloc(1, sizeof(*this));
3814 this->next = rest;
3815 rest = this;
3816
3817 if (layout_ddf2md(&vc->conf, &this->array))
3818 continue;
3819 this->array.md_minor = -1;
3820 this->array.major_version = -1;
3821 this->array.minor_version = -2;
3822 this->safe_mode_delay = DDF_SAFE_MODE_DELAY;
3823 cptr = (__u32 *)(vc->conf.guid + 16);
3824 this->array.ctime = DECADE + __be32_to_cpu(*cptr);
3825 this->array.utime = DECADE +
3826 be32_to_cpu(vc->conf.timestamp);
3827 this->array.chunk_size = 512 << vc->conf.chunk_shift;
3828
3829 i = vc->vcnum;
3830 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
3831 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
3832 DDF_init_full) {
3833 this->array.state = 0;
3834 this->resync_start = 0;
3835 } else {
3836 this->array.state = 1;
3837 this->resync_start = MaxSector;
3838 }
3839 _ddf_array_name(this->name, ddf, i);
3840 memset(this->uuid, 0, sizeof(this->uuid));
3841 this->component_size = be64_to_cpu(vc->conf.blocks);
3842 this->array.size = this->component_size / 2;
3843 this->container_member = i;
3844
3845 ddf->currentconf = vc;
3846 uuid_from_super_ddf(st, this->uuid);
3847 if (!subarray)
3848 ddf->currentconf = NULL;
3849
3850 sprintf(this->text_version, "/%s/%d",
3851 st->container_devnm, this->container_member);
3852
3853 for (pd = 0; pd < be16_to_cpu(ddf->phys->max_pdes); pd++) {
3854 struct mdinfo *dev;
3855 struct dl *d;
3856 const struct vd_config *bvd;
3857 unsigned int iphys;
3858 int stt;
3859
3860 if (be32_to_cpu(ddf->phys->entries[pd].refnum)
3861 == 0xFFFFFFFF)
3862 continue;
3863
3864 stt = be16_to_cpu(ddf->phys->entries[pd].state);
3865 if ((stt & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3866 != DDF_Online)
3867 continue;
3868
3869 i = get_pd_index_from_refnum(
3870 vc, ddf->phys->entries[pd].refnum,
3871 ddf->mppe, &bvd, &iphys);
3872 if (i == DDF_NOTFOUND)
3873 continue;
3874
3875 this->array.working_disks++;
3876
3877 for (d = ddf->dlist; d ; d=d->next)
3878 if (be32_eq(d->disk.refnum,
3879 ddf->phys->entries[pd].refnum))
3880 break;
3881 if (d == NULL)
3882 /* Haven't found that one yet, maybe there are others */
3883 continue;
3884
3885 dev = xcalloc(1, sizeof(*dev));
3886 dev->next = this->devs;
3887 this->devs = dev;
3888
3889 dev->disk.number = be32_to_cpu(d->disk.refnum);
3890 dev->disk.major = d->major;
3891 dev->disk.minor = d->minor;
3892 dev->disk.raid_disk = i;
3893 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
3894 dev->recovery_start = MaxSector;
3895
3896 dev->events = be32_to_cpu(ddf->active->seq);
3897 dev->data_offset =
3898 be64_to_cpu(LBA_OFFSET(ddf, bvd)[iphys]);
3899 dev->component_size = be64_to_cpu(bvd->blocks);
3900 if (d->devname)
3901 strcpy(dev->name, d->devname);
3902 }
3903 }
3904 return rest;
3905 }
3906
3907 static int store_super_ddf(struct supertype *st, int fd)
3908 {
3909 struct ddf_super *ddf = st->sb;
3910 unsigned long long dsize;
3911 void *buf;
3912 int rc;
3913
3914 if (!ddf)
3915 return 1;
3916
3917 if (!get_dev_size(fd, NULL, &dsize))
3918 return 1;
3919
3920 if (ddf->dlist || ddf->conflist) {
3921 struct stat sta;
3922 struct dl *dl;
3923 int ofd, ret;
3924
3925 if (fstat(fd, &sta) == -1 || !S_ISBLK(sta.st_mode)) {
3926 pr_err("%s: file descriptor for invalid device\n",
3927 __func__);
3928 return 1;
3929 }
3930 for (dl = ddf->dlist; dl; dl = dl->next)
3931 if (dl->major == (int)major(sta.st_rdev) &&
3932 dl->minor == (int)minor(sta.st_rdev))
3933 break;
3934 if (!dl) {
3935 pr_err("%s: couldn't find disk %d/%d\n", __func__,
3936 (int)major(sta.st_rdev),
3937 (int)minor(sta.st_rdev));
3938 return 1;
3939 }
3940 ofd = dl->fd;
3941 dl->fd = fd;
3942 ret = (_write_super_to_disk(ddf, dl) != 1);
3943 dl->fd = ofd;
3944 return ret;
3945 }
3946
3947 if (posix_memalign(&buf, 512, 512) != 0)
3948 return 1;
3949 memset(buf, 0, 512);
3950
3951 lseek64(fd, dsize-512, 0);
3952 rc = write(fd, buf, 512);
3953 free(buf);
3954 if (rc < 0)
3955 return 1;
3956 return 0;
3957 }
3958
3959 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3960 {
3961 /*
3962 * return:
3963 * 0 same, or first was empty, and second was copied
3964 * 1 second had wrong magic number - but that isn't possible
3965 * 2 wrong uuid
3966 * 3 wrong other info
3967 */
3968 struct ddf_super *first = st->sb;
3969 struct ddf_super *second = tst->sb;
3970 struct dl *dl1, *dl2;
3971 struct vcl *vl1, *vl2;
3972 unsigned int max_vds, max_pds, pd, vd;
3973
3974 if (!first) {
3975 st->sb = tst->sb;
3976 tst->sb = NULL;
3977 return 0;
3978 }
3979
3980 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3981 return 2;
3982
3983 /* It is only OK to compare info in the anchor. Anything else
3984 * could be changing due to a reconfig so must be ignored.
3985 * guid really should be enough anyway.
3986 */
3987
3988 if (!be32_eq(first->active->seq, second->active->seq)) {
3989 dprintf("%s: sequence number mismatch %u<->%u\n", __func__,
3990 be32_to_cpu(first->active->seq),
3991 be32_to_cpu(second->active->seq));
3992 return 0;
3993 }
3994
3995 /*
3996 * At this point we are fairly sure that the meta data matches.
3997 * But the new disk may contain additional local data.
3998 * Add it to the super block.
3999 */
4000 max_vds = be16_to_cpu(first->active->max_vd_entries);
4001 max_pds = be16_to_cpu(first->phys->max_pdes);
4002 for (vl2 = second->conflist; vl2; vl2 = vl2->next) {
4003 for (vl1 = first->conflist; vl1; vl1 = vl1->next)
4004 if (!memcmp(vl1->conf.guid, vl2->conf.guid,
4005 DDF_GUID_LEN))
4006 break;
4007 if (vl1) {
4008 if (vl1->other_bvds != NULL &&
4009 vl1->conf.sec_elmnt_seq !=
4010 vl2->conf.sec_elmnt_seq) {
4011 dprintf("%s: adding BVD %u\n", __func__,
4012 vl2->conf.sec_elmnt_seq);
4013 add_other_bvd(vl1, &vl2->conf,
4014 first->conf_rec_len*512);
4015 }
4016 continue;
4017 }
4018
4019 if (posix_memalign((void **)&vl1, 512,
4020 (first->conf_rec_len*512 +
4021 offsetof(struct vcl, conf))) != 0) {
4022 pr_err("%s could not allocate vcl buf\n",
4023 __func__);
4024 return 3;
4025 }
4026
4027 vl1->next = first->conflist;
4028 vl1->block_sizes = NULL;
4029 memcpy(&vl1->conf, &vl2->conf, first->conf_rec_len*512);
4030 if (alloc_other_bvds(first, vl1) != 0) {
4031 pr_err("%s could not allocate other bvds\n",
4032 __func__);
4033 free(vl1);
4034 return 3;
4035 }
4036 for (vd = 0; vd < max_vds; vd++)
4037 if (!memcmp(first->virt->entries[vd].guid,
4038 vl1->conf.guid, DDF_GUID_LEN))
4039 break;
4040 vl1->vcnum = vd;
4041 dprintf("%s: added config for VD %u\n", __func__, vl1->vcnum);
4042 first->conflist = vl1;
4043 }
4044
4045 for (dl2 = second->dlist; dl2; dl2 = dl2->next) {
4046 for (dl1 = first->dlist; dl1; dl1 = dl1->next)
4047 if (be32_eq(dl1->disk.refnum, dl2->disk.refnum))
4048 break;
4049 if (dl1)
4050 continue;
4051
4052 if (posix_memalign((void **)&dl1, 512,
4053 sizeof(*dl1) + (first->max_part) * sizeof(dl1->vlist[0]))
4054 != 0) {
4055 pr_err("%s could not allocate disk info buffer\n",
4056 __func__);
4057 return 3;
4058 }
4059 memcpy(dl1, dl2, sizeof(*dl1));
4060 dl1->mdupdate = NULL;
4061 dl1->next = first->dlist;
4062 dl1->fd = -1;
4063 for (pd = 0; pd < max_pds; pd++)
4064 if (be32_eq(first->phys->entries[pd].refnum,
4065 dl1->disk.refnum))
4066 break;
4067 dl1->pdnum = pd < max_pds ? (int)pd : -1;
4068 if (dl2->spare) {
4069 if (posix_memalign((void **)&dl1->spare, 512,
4070 first->conf_rec_len*512) != 0) {
4071 pr_err("%s could not allocate spare info buf\n",
4072 __func__);
4073 return 3;
4074 }
4075 memcpy(dl1->spare, dl2->spare, first->conf_rec_len*512);
4076 }
4077 for (vd = 0 ; vd < first->max_part ; vd++) {
4078 if (!dl2->vlist[vd]) {
4079 dl1->vlist[vd] = NULL;
4080 continue;
4081 }
4082 for (vl1 = first->conflist; vl1; vl1 = vl1->next) {
4083 if (!memcmp(vl1->conf.guid,
4084 dl2->vlist[vd]->conf.guid,
4085 DDF_GUID_LEN))
4086 break;
4087 dl1->vlist[vd] = vl1;
4088 }
4089 }
4090 first->dlist = dl1;
4091 dprintf("%s: added disk %d: %08x\n", __func__, dl1->pdnum,
4092 be32_to_cpu(dl1->disk.refnum));
4093 }
4094
4095 return 0;
4096 }
4097
4098 #ifndef MDASSEMBLE
4099 /*
4100 * A new array 'a' has been started which claims to be instance 'inst'
4101 * within container 'c'.
4102 * We need to confirm that the array matches the metadata in 'c' so
4103 * that we don't corrupt any metadata.
4104 */
4105 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
4106 {
4107 struct ddf_super *ddf = c->sb;
4108 int n = atoi(inst);
4109 struct mdinfo *dev;
4110 struct dl *dl;
4111 static const char faulty[] = "faulty";
4112
4113 if (all_ff(ddf->virt->entries[n].guid)) {
4114 pr_err("%s: subarray %d doesn't exist\n", __func__, n);
4115 return -ENODEV;
4116 }
4117 dprintf("%s: new subarray %d, GUID: %s\n", __func__, n,
4118 guid_str(ddf->virt->entries[n].guid));
4119 for (dev = a->info.devs; dev; dev = dev->next) {
4120 for (dl = ddf->dlist; dl; dl = dl->next)
4121 if (dl->major == dev->disk.major &&
4122 dl->minor == dev->disk.minor)
4123 break;
4124 if (!dl || dl->pdnum < 0) {
4125 pr_err("%s: device %d/%d of subarray %d not found in meta data\n",
4126 __func__, dev->disk.major, dev->disk.minor, n);
4127 return -1;
4128 }
4129 if ((be16_to_cpu(ddf->phys->entries[dl->pdnum].state) &
4130 (DDF_Online|DDF_Missing|DDF_Failed)) != DDF_Online) {
4131 pr_err("%s: new subarray %d contains broken device %d/%d (%02x)\n",
4132 __func__, n, dl->major, dl->minor,
4133 be16_to_cpu(
4134 ddf->phys->entries[dl->pdnum].state));
4135 if (write(dev->state_fd, faulty, sizeof(faulty)-1) !=
4136 sizeof(faulty) - 1)
4137 pr_err("Write to state_fd failed\n");
4138 dev->curr_state = DS_FAULTY;
4139 }
4140 }
4141 a->info.container_member = n;
4142 return 0;
4143 }
4144
4145 static void handle_missing(struct ddf_super *ddf, struct active_array *a, int inst)
4146 {
4147 /* This member array is being activated. If any devices
4148 * are missing they must now be marked as failed.
4149 */
4150 struct vd_config *vc;
4151 unsigned int n_bvd;
4152 struct vcl *vcl;
4153 struct dl *dl;
4154 int pd;
4155 int n;
4156 int state;
4157
4158 for (n = 0; ; n++) {
4159 vc = find_vdcr(ddf, inst, n, &n_bvd, &vcl);
4160 if (!vc)
4161 break;
4162 for (dl = ddf->dlist; dl; dl = dl->next)
4163 if (be32_eq(dl->disk.refnum, vc->phys_refnum[n_bvd]))
4164 break;
4165 if (dl)
4166 /* Found this disk, so not missing */
4167 continue;
4168
4169 /* Mark the device as failed/missing. */
4170 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4171 if (pd >= 0 && be16_and(ddf->phys->entries[pd].state,
4172 cpu_to_be16(DDF_Online))) {
4173 be16_clear(ddf->phys->entries[pd].state,
4174 cpu_to_be16(DDF_Online));
4175 be16_set(ddf->phys->entries[pd].state,
4176 cpu_to_be16(DDF_Failed|DDF_Missing));
4177 vc->phys_refnum[n_bvd] = cpu_to_be32(0);
4178 ddf_set_updates_pending(ddf, vc);
4179 }
4180
4181 /* Mark the array as Degraded */
4182 state = get_svd_state(ddf, vcl);
4183 if (ddf->virt->entries[inst].state !=
4184 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4185 | state)) {
4186 ddf->virt->entries[inst].state =
4187 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4188 | state;
4189 a->check_degraded = 1;
4190 ddf_set_updates_pending(ddf, vc);
4191 }
4192 }
4193 }
4194
4195 /*
4196 * The array 'a' is to be marked clean in the metadata.
4197 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4198 * clean up to the point (in sectors). If that cannot be recorded in the
4199 * metadata, then leave it as dirty.
4200 *
4201 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4202 * !global! virtual_disk.virtual_entry structure.
4203 */
4204 static int ddf_set_array_state(struct active_array *a, int consistent)
4205 {
4206 struct ddf_super *ddf = a->container->sb;
4207 int inst = a->info.container_member;
4208 int old = ddf->virt->entries[inst].state;
4209 if (consistent == 2) {
4210 handle_missing(ddf, a, inst);
4211 consistent = 1;
4212 if (!is_resync_complete(&a->info))
4213 consistent = 0;
4214 }
4215 if (consistent)
4216 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
4217 else
4218 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
4219 if (old != ddf->virt->entries[inst].state)
4220 ddf_set_updates_pending(ddf, NULL);
4221
4222 old = ddf->virt->entries[inst].init_state;
4223 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
4224 if (is_resync_complete(&a->info))
4225 ddf->virt->entries[inst].init_state |= DDF_init_full;
4226 else if (a->info.resync_start == 0)
4227 ddf->virt->entries[inst].init_state |= DDF_init_not;
4228 else
4229 ddf->virt->entries[inst].init_state |= DDF_init_quick;
4230 if (old != ddf->virt->entries[inst].init_state)
4231 ddf_set_updates_pending(ddf, NULL);
4232
4233 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst,
4234 guid_str(ddf->virt->entries[inst].guid), a->curr_state,
4235 consistent?"clean":"dirty",
4236 a->info.resync_start);
4237 return consistent;
4238 }
4239
4240 static int get_bvd_state(const struct ddf_super *ddf,
4241 const struct vd_config *vc)
4242 {
4243 unsigned int i, n_bvd, working = 0;
4244 unsigned int n_prim = be16_to_cpu(vc->prim_elmnt_count);
4245 int pd, st, state;
4246 char *avail = xcalloc(1, n_prim);
4247 mdu_array_info_t array;
4248
4249 layout_ddf2md(vc, &array);
4250
4251 for (i = 0; i < n_prim; i++) {
4252 if (!find_index_in_bvd(ddf, vc, i, &n_bvd))
4253 continue;
4254 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4255 if (pd < 0)
4256 continue;
4257 st = be16_to_cpu(ddf->phys->entries[pd].state);
4258 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
4259 == DDF_Online) {
4260 working++;
4261 avail[i] = 1;
4262 }
4263 }
4264
4265 state = DDF_state_degraded;
4266 if (working == n_prim)
4267 state = DDF_state_optimal;
4268 else
4269 switch (vc->prl) {
4270 case DDF_RAID0:
4271 case DDF_CONCAT:
4272 case DDF_JBOD:
4273 state = DDF_state_failed;
4274 break;
4275 case DDF_RAID1:
4276 if (working == 0)
4277 state = DDF_state_failed;
4278 else if (working >= 2)
4279 state = DDF_state_part_optimal;
4280 break;
4281 case DDF_RAID1E:
4282 if (!enough(10, n_prim, array.layout, 1, avail))
4283 state = DDF_state_failed;
4284 break;
4285 case DDF_RAID4:
4286 case DDF_RAID5:
4287 if (working < n_prim - 1)
4288 state = DDF_state_failed;
4289 break;
4290 case DDF_RAID6:
4291 if (working < n_prim - 2)
4292 state = DDF_state_failed;
4293 else if (working == n_prim - 1)
4294 state = DDF_state_part_optimal;
4295 break;
4296 }
4297 return state;
4298 }
4299
4300 static int secondary_state(int state, int other, int seclevel)
4301 {
4302 if (state == DDF_state_optimal && other == DDF_state_optimal)
4303 return DDF_state_optimal;
4304 if (seclevel == DDF_2MIRRORED) {
4305 if (state == DDF_state_optimal || other == DDF_state_optimal)
4306 return DDF_state_part_optimal;
4307 if (state == DDF_state_failed && other == DDF_state_failed)
4308 return DDF_state_failed;
4309 return DDF_state_degraded;
4310 } else {
4311 if (state == DDF_state_failed || other == DDF_state_failed)
4312 return DDF_state_failed;
4313 if (state == DDF_state_degraded || other == DDF_state_degraded)
4314 return DDF_state_degraded;
4315 return DDF_state_part_optimal;
4316 }
4317 }
4318
4319 static int get_svd_state(const struct ddf_super *ddf, const struct vcl *vcl)
4320 {
4321 int state = get_bvd_state(ddf, &vcl->conf);
4322 unsigned int i;
4323 for (i = 1; i < vcl->conf.sec_elmnt_count; i++) {
4324 state = secondary_state(
4325 state,
4326 get_bvd_state(ddf, vcl->other_bvds[i-1]),
4327 vcl->conf.srl);
4328 }
4329 return state;
4330 }
4331
4332 /*
4333 * The state of each disk is stored in the global phys_disk structure
4334 * in phys_disk.entries[n].state.
4335 * This makes various combinations awkward.
4336 * - When a device fails in any array, it must be failed in all arrays
4337 * that include a part of this device.
4338 * - When a component is rebuilding, we cannot include it officially in the
4339 * array unless this is the only array that uses the device.
4340 *
4341 * So: when transitioning:
4342 * Online -> failed, just set failed flag. monitor will propagate
4343 * spare -> online, the device might need to be added to the array.
4344 * spare -> failed, just set failed. Don't worry if in array or not.
4345 */
4346 static void ddf_set_disk(struct active_array *a, int n, int state)
4347 {
4348 struct ddf_super *ddf = a->container->sb;
4349 unsigned int inst = a->info.container_member, n_bvd;
4350 struct vcl *vcl;
4351 struct vd_config *vc = find_vdcr(ddf, inst, (unsigned int)n,
4352 &n_bvd, &vcl);
4353 int pd;
4354 struct mdinfo *mdi;
4355 struct dl *dl;
4356 int update = 0;
4357
4358 dprintf("%s: %d to %x\n", __func__, n, state);
4359 if (vc == NULL) {
4360 dprintf("ddf: cannot find instance %d!!\n", inst);
4361 return;
4362 }
4363 /* Find the matching slot in 'info'. */
4364 for (mdi = a->info.devs; mdi; mdi = mdi->next)
4365 if (mdi->disk.raid_disk == n)
4366 break;
4367 if (!mdi) {
4368 pr_err("%s: cannot find raid disk %d\n",
4369 __func__, n);
4370 return;
4371 }
4372
4373 /* and find the 'dl' entry corresponding to that. */
4374 for (dl = ddf->dlist; dl; dl = dl->next)
4375 if (mdi->state_fd >= 0 &&
4376 mdi->disk.major == dl->major &&
4377 mdi->disk.minor == dl->minor)
4378 break;
4379 if (!dl) {
4380 pr_err("%s: cannot find raid disk %d (%d/%d)\n",
4381 __func__, n,
4382 mdi->disk.major, mdi->disk.minor);
4383 return;
4384 }
4385
4386 pd = find_phys(ddf, vc->phys_refnum[n_bvd]);
4387 if (pd < 0 || pd != dl->pdnum) {
4388 /* disk doesn't currently exist or has changed.
4389 * If it is now in_sync, insert it. */
4390 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4391 __func__, dl->pdnum, dl->major, dl->minor,
4392 be32_to_cpu(dl->disk.refnum));
4393 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4394 __func__, inst, n_bvd,
4395 be32_to_cpu(vc->phys_refnum[n_bvd]), pd);
4396 if ((state & DS_INSYNC) && ! (state & DS_FAULTY) &&
4397 dl->pdnum >= 0) {
4398 pd = dl->pdnum;
4399 vc->phys_refnum[n_bvd] = dl->disk.refnum;
4400 LBA_OFFSET(ddf, vc)[n_bvd] =
4401 cpu_to_be64(mdi->data_offset);
4402 be16_clear(ddf->phys->entries[pd].type,
4403 cpu_to_be16(DDF_Global_Spare));
4404 be16_set(ddf->phys->entries[pd].type,
4405 cpu_to_be16(DDF_Active_in_VD));
4406 update = 1;
4407 }
4408 } else {
4409 be16 old = ddf->phys->entries[pd].state;
4410 if (state & DS_FAULTY)
4411 be16_set(ddf->phys->entries[pd].state,
4412 cpu_to_be16(DDF_Failed));
4413 if (state & DS_INSYNC) {
4414 be16_set(ddf->phys->entries[pd].state,
4415 cpu_to_be16(DDF_Online));
4416 be16_clear(ddf->phys->entries[pd].state,
4417 cpu_to_be16(DDF_Rebuilding));
4418 }
4419 if (!be16_eq(old, ddf->phys->entries[pd].state))
4420 update = 1;
4421 }
4422
4423 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n,
4424 be32_to_cpu(dl->disk.refnum), state,
4425 be16_to_cpu(ddf->phys->entries[pd].state));
4426
4427 /* Now we need to check the state of the array and update
4428 * virtual_disk.entries[n].state.
4429 * It needs to be one of "optimal", "degraded", "failed".
4430 * I don't understand 'deleted' or 'missing'.
4431 */
4432 state = get_svd_state(ddf, vcl);
4433
4434 if (ddf->virt->entries[inst].state !=
4435 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
4436 | state)) {
4437 ddf->virt->entries[inst].state =
4438 (ddf->virt->entries[inst].state & ~DDF_state_mask)
4439 | state;
4440 update = 1;
4441 }
4442 if (update)
4443 ddf_set_updates_pending(ddf, vc);
4444 }
4445
4446 static void ddf_sync_metadata(struct supertype *st)
4447 {
4448 /*
4449 * Write all data to all devices.
4450 * Later, we might be able to track whether only local changes
4451 * have been made, or whether any global data has been changed,
4452 * but ddf is sufficiently weird that it probably always
4453 * changes global data ....
4454 */
4455 struct ddf_super *ddf = st->sb;
4456 if (!ddf->updates_pending)
4457 return;
4458 ddf->updates_pending = 0;
4459 __write_init_super_ddf(st);
4460 dprintf("ddf: sync_metadata\n");
4461 }
4462
4463 static int del_from_conflist(struct vcl **list, const char *guid)
4464 {
4465 struct vcl **p;
4466 int found = 0;
4467 for (p = list; p && *p; p = &((*p)->next))
4468 if (!memcmp((*p)->conf.guid, guid, DDF_GUID_LEN)) {
4469 found = 1;
4470 *p = (*p)->next;
4471 }
4472 return found;
4473 }
4474
4475 static int _kill_subarray_ddf(struct ddf_super *ddf, const char *guid)
4476 {
4477 struct dl *dl;
4478 unsigned int vdnum, i;
4479 vdnum = find_vde_by_guid(ddf, guid);
4480 if (vdnum == DDF_NOTFOUND) {
4481 pr_err("%s: could not find VD %s\n", __func__,
4482 guid_str(guid));
4483 return -1;
4484 }
4485 if (del_from_conflist(&ddf->conflist, guid) == 0) {
4486 pr_err("%s: could not find conf %s\n", __func__,
4487 guid_str(guid));
4488 return -1;
4489 }
4490 for (dl = ddf->dlist; dl; dl = dl->next)
4491 for (i = 0; i < ddf->max_part; i++)
4492 if (dl->vlist[i] != NULL &&
4493 !memcmp(dl->vlist[i]->conf.guid, guid,
4494 DDF_GUID_LEN))
4495 dl->vlist[i] = NULL;
4496 memset(ddf->virt->entries[vdnum].guid, 0xff, DDF_GUID_LEN);
4497 dprintf("%s: deleted %s\n", __func__, guid_str(guid));
4498 return 0;
4499 }
4500
4501 static int kill_subarray_ddf(struct supertype *st)
4502 {
4503 struct ddf_super *ddf = st->sb;
4504 /*
4505 * currentconf is set in container_content_ddf,
4506 * called with subarray arg
4507 */
4508 struct vcl *victim = ddf->currentconf;
4509 struct vd_config *conf;
4510 unsigned int vdnum;
4511
4512 ddf->currentconf = NULL;
4513 if (!victim) {
4514 pr_err("%s: nothing to kill\n", __func__);
4515 return -1;
4516 }
4517 conf = &victim->conf;
4518 vdnum = find_vde_by_guid(ddf, conf->guid);
4519 if (vdnum == DDF_NOTFOUND) {
4520 pr_err("%s: could not find VD %s\n", __func__,
4521 guid_str(conf->guid));
4522 return -1;
4523 }
4524 if (st->update_tail) {
4525 struct virtual_disk *vd;
4526 int len = sizeof(struct virtual_disk)
4527 + sizeof(struct virtual_entry);
4528 vd = xmalloc(len);
4529 if (vd == NULL) {
4530 pr_err("%s: failed to allocate %d bytes\n", __func__,
4531 len);
4532 return -1;
4533 }
4534 memset(vd, 0 , len);
4535 vd->magic = DDF_VIRT_RECORDS_MAGIC;
4536 vd->populated_vdes = cpu_to_be16(0);
4537 memcpy(vd->entries[0].guid, conf->guid, DDF_GUID_LEN);
4538 /* we use DDF_state_deleted as marker */
4539 vd->entries[0].state = DDF_state_deleted;
4540 append_metadata_update(st, vd, len);
4541 } else {
4542 _kill_subarray_ddf(ddf, conf->guid);
4543 ddf_set_updates_pending(ddf, NULL);
4544 ddf_sync_metadata(st);
4545 }
4546 return 0;
4547 }
4548
4549 static void copy_matching_bvd(struct ddf_super *ddf,
4550 struct vd_config *conf,
4551 const struct metadata_update *update)
4552 {
4553 unsigned int mppe =
4554 be16_to_cpu(ddf->anchor.max_primary_element_entries);
4555 unsigned int len = ddf->conf_rec_len * 512;
4556 char *p;
4557 struct vd_config *vc;
4558 for (p = update->buf; p < update->buf + update->len; p += len) {
4559 vc = (struct vd_config *) p;
4560 if (vc->sec_elmnt_seq == conf->sec_elmnt_seq) {
4561 memcpy(conf->phys_refnum, vc->phys_refnum,
4562 mppe * (sizeof(__u32) + sizeof(__u64)));
4563 return;
4564 }
4565 }
4566 pr_err("%s: no match for BVD %d of %s in update\n", __func__,
4567 conf->sec_elmnt_seq, guid_str(conf->guid));
4568 }
4569
4570 static void ddf_process_phys_update(struct supertype *st,
4571 struct metadata_update *update)
4572 {
4573 struct ddf_super *ddf = st->sb;
4574 struct phys_disk *pd;
4575 unsigned int ent;
4576
4577 pd = (struct phys_disk*)update->buf;
4578 ent = be16_to_cpu(pd->used_pdes);
4579 if (ent >= be16_to_cpu(ddf->phys->max_pdes))
4580 return;
4581 if (be16_and(pd->entries[0].state, cpu_to_be16(DDF_Missing))) {
4582 struct dl **dlp;
4583 /* removing this disk. */
4584 be16_set(ddf->phys->entries[ent].state,
4585 cpu_to_be16(DDF_Missing));
4586 for (dlp = &ddf->dlist; *dlp; dlp = &(*dlp)->next) {
4587 struct dl *dl = *dlp;
4588 if (dl->pdnum == (signed)ent) {
4589 close(dl->fd);
4590 dl->fd = -1;
4591 *dlp = dl->next;
4592 update->space = dl->devname;
4593 *(void**)dl = update->space_list;
4594 update->space_list = (void**)dl;
4595 break;
4596 }
4597 }
4598 ddf_set_updates_pending(ddf, NULL);
4599 return;
4600 }
4601 if (!all_ff(ddf->phys->entries[ent].guid))
4602 return;
4603 ddf->phys->entries[ent] = pd->entries[0];
4604 ddf->phys->used_pdes = cpu_to_be16
4605 (1 + be16_to_cpu(ddf->phys->used_pdes));
4606 ddf_set_updates_pending(ddf, NULL);
4607 if (ddf->add_list) {
4608 struct active_array *a;
4609 struct dl *al = ddf->add_list;
4610 ddf->add_list = al->next;
4611
4612 al->next = ddf->dlist;
4613 ddf->dlist = al;
4614
4615 /* As a device has been added, we should check
4616 * for any degraded devices that might make
4617 * use of this spare */
4618 for (a = st->arrays ; a; a=a->next)
4619 a->check_degraded = 1;
4620 }
4621 }
4622
4623 static void ddf_process_virt_update(struct supertype *st,
4624 struct metadata_update *update)
4625 {
4626 struct ddf_super *ddf = st->sb;
4627 struct virtual_disk *vd;
4628 unsigned int ent;
4629
4630 vd = (struct virtual_disk*)update->buf;
4631
4632 if (vd->entries[0].state == DDF_state_deleted) {
4633 if (_kill_subarray_ddf(ddf, vd->entries[0].guid))
4634 return;
4635 } else {
4636 ent = find_vde_by_guid(ddf, vd->entries[0].guid);
4637 if (ent != DDF_NOTFOUND) {
4638 dprintf("%s: VD %s exists already in slot %d\n",
4639 __func__, guid_str(vd->entries[0].guid),
4640 ent);
4641 return;
4642 }
4643 ent = find_unused_vde(ddf);
4644 if (ent == DDF_NOTFOUND)
4645 return;
4646 ddf->virt->entries[ent] = vd->entries[0];
4647 ddf->virt->populated_vdes =
4648 cpu_to_be16(
4649 1 + be16_to_cpu(
4650 ddf->virt->populated_vdes));
4651 dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n",
4652 __func__, guid_str(vd->entries[0].guid), ent,
4653 ddf->virt->entries[ent].state,
4654 ddf->virt->entries[ent].init_state);
4655 }
4656 ddf_set_updates_pending(ddf, NULL);
4657 }
4658
4659 static void ddf_remove_failed(struct ddf_super *ddf)
4660 {
4661 /* Now remove any 'Failed' devices that are not part
4662 * of any VD. They will have the Transition flag set.
4663 * Once done, we need to update all dl->pdnum numbers.
4664 */
4665 unsigned int pdnum;
4666 unsigned int pd2 = 0;
4667 struct dl *dl;
4668
4669 for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4670 pdnum++) {
4671 if (be32_to_cpu(ddf->phys->entries[pdnum].refnum) ==
4672 0xFFFFFFFF)
4673 continue;
4674 if (be16_and(ddf->phys->entries[pdnum].state,
4675 cpu_to_be16(DDF_Failed))
4676 && be16_and(ddf->phys->entries[pdnum].state,
4677 cpu_to_be16(DDF_Transition))) {
4678 /* skip this one unless in dlist*/
4679 for (dl = ddf->dlist; dl; dl = dl->next)
4680 if (dl->pdnum == (int)pdnum)
4681 break;
4682 if (!dl)
4683 continue;
4684 }
4685 if (pdnum == pd2)
4686 pd2++;
4687 else {
4688 ddf->phys->entries[pd2] =
4689 ddf->phys->entries[pdnum];
4690 for (dl = ddf->dlist; dl; dl = dl->next)
4691 if (dl->pdnum == (int)pdnum)
4692 dl->pdnum = pd2;
4693 pd2++;
4694 }
4695 }
4696 ddf->phys->used_pdes = cpu_to_be16(pd2);
4697 while (pd2 < pdnum) {
4698 memset(ddf->phys->entries[pd2].guid, 0xff,
4699 DDF_GUID_LEN);
4700 pd2++;
4701 }
4702 }
4703
4704 static void ddf_update_vlist(struct ddf_super *ddf, struct dl *dl)
4705 {
4706 struct vcl *vcl;
4707 unsigned int vn = 0;
4708 int in_degraded = 0;
4709
4710 if (dl->pdnum < 0)
4711 return;
4712 for (vcl = ddf->conflist; vcl ; vcl = vcl->next) {
4713 unsigned int dn, ibvd;
4714 const struct vd_config *conf;
4715 int vstate;
4716 dn = get_pd_index_from_refnum(vcl,
4717 dl->disk.refnum,
4718 ddf->mppe,
4719 &conf, &ibvd);
4720 if (dn == DDF_NOTFOUND)
4721 continue;
4722 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4723 dl->pdnum,
4724 be32_to_cpu(dl->disk.refnum),
4725 guid_str(conf->guid),
4726 conf->sec_elmnt_seq, vn);
4727 /* Clear the Transition flag */
4728 if (be16_and
4729 (ddf->phys->entries[dl->pdnum].state,
4730 cpu_to_be16(DDF_Failed)))
4731 be16_clear(ddf->phys
4732 ->entries[dl->pdnum].state,
4733 cpu_to_be16(DDF_Transition));
4734 dl->vlist[vn++] = vcl;
4735 vstate = ddf->virt->entries[vcl->vcnum].state
4736 & DDF_state_mask;
4737 if (vstate == DDF_state_degraded ||
4738 vstate == DDF_state_part_optimal)
4739 in_degraded = 1;
4740 }
4741 while (vn < ddf->max_part)
4742 dl->vlist[vn++] = NULL;
4743 if (dl->vlist[0]) {
4744 be16_clear(ddf->phys->entries[dl->pdnum].type,
4745 cpu_to_be16(DDF_Global_Spare));
4746 if (!be16_and(ddf->phys
4747 ->entries[dl->pdnum].type,
4748 cpu_to_be16(DDF_Active_in_VD))) {
4749 be16_set(ddf->phys
4750 ->entries[dl->pdnum].type,
4751 cpu_to_be16(DDF_Active_in_VD));
4752 if (in_degraded)
4753 be16_set(ddf->phys
4754 ->entries[dl->pdnum]
4755 .state,
4756 cpu_to_be16
4757 (DDF_Rebuilding));
4758 }
4759 }
4760 if (dl->spare) {
4761 be16_clear(ddf->phys->entries[dl->pdnum].type,
4762 cpu_to_be16(DDF_Global_Spare));
4763 be16_set(ddf->phys->entries[dl->pdnum].type,
4764 cpu_to_be16(DDF_Spare));
4765 }
4766 if (!dl->vlist[0] && !dl->spare) {
4767 be16_set(ddf->phys->entries[dl->pdnum].type,
4768 cpu_to_be16(DDF_Global_Spare));
4769 be16_clear(ddf->phys->entries[dl->pdnum].type,
4770 cpu_to_be16(DDF_Spare));
4771 be16_clear(ddf->phys->entries[dl->pdnum].type,
4772 cpu_to_be16(DDF_Active_in_VD));
4773 }
4774 }
4775
4776 static void ddf_process_conf_update(struct supertype *st,
4777 struct metadata_update *update)
4778 {
4779 struct ddf_super *ddf = st->sb;
4780 struct vd_config *vc;
4781 struct vcl *vcl;
4782 struct dl *dl;
4783 unsigned int ent;
4784 unsigned int pdnum, len;
4785
4786 vc = (struct vd_config*)update->buf;
4787 len = ddf->conf_rec_len * 512;
4788 if ((unsigned int)update->len != len * vc->sec_elmnt_count) {
4789 pr_err("%s: %s: insufficient data (%d) for %u BVDs\n",
4790 __func__, guid_str(vc->guid), update->len,
4791 vc->sec_elmnt_count);
4792 return;
4793 }
4794 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
4795 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
4796 break;
4797 dprintf("%s: conf update for %s (%s)\n", __func__,
4798 guid_str(vc->guid), (vcl ? "old" : "new"));
4799 if (vcl) {
4800 /* An update, just copy the phys_refnum and lba_offset
4801 * fields
4802 */
4803 unsigned int i;
4804 unsigned int k;
4805 copy_matching_bvd(ddf, &vcl->conf, update);
4806 for (k = 0; k < be16_to_cpu(vc->prim_elmnt_count); k++)
4807 dprintf("BVD %u has %08x at %llu\n", 0,
4808 be32_to_cpu(vcl->conf.phys_refnum[k]),
4809 be64_to_cpu(LBA_OFFSET(ddf,
4810 &vcl->conf)[k]));
4811 for (i = 1; i < vc->sec_elmnt_count; i++) {
4812 copy_matching_bvd(ddf, vcl->other_bvds[i-1],
4813 update);
4814 for (k = 0; k < be16_to_cpu(
4815 vc->prim_elmnt_count); k++)
4816 dprintf("BVD %u has %08x at %llu\n", i,
4817 be32_to_cpu
4818 (vcl->other_bvds[i-1]->
4819 phys_refnum[k]),
4820 be64_to_cpu
4821 (LBA_OFFSET
4822 (ddf,
4823 vcl->other_bvds[i-1])[k]));
4824 }
4825 } else {
4826 /* A new VD_CONF */
4827 unsigned int i;
4828 if (!update->space)
4829 return;
4830 vcl = update->space;
4831 update->space = NULL;
4832 vcl->next = ddf->conflist;
4833 memcpy(&vcl->conf, vc, len);
4834 ent = find_vde_by_guid(ddf, vc->guid);
4835 if (ent == DDF_NOTFOUND)
4836 return;
4837 vcl->vcnum = ent;
4838 ddf->conflist = vcl;
4839 for (i = 1; i < vc->sec_elmnt_count; i++)
4840 memcpy(vcl->other_bvds[i-1],
4841 update->buf + len * i, len);
4842 }
4843 /* Set DDF_Transition on all Failed devices - to help
4844 * us detect those that are no longer in use
4845 */
4846 for (pdnum = 0; pdnum < be16_to_cpu(ddf->phys->max_pdes);
4847 pdnum++)
4848 if (be16_and(ddf->phys->entries[pdnum].state,
4849 cpu_to_be16(DDF_Failed)))
4850 be16_set(ddf->phys->entries[pdnum].state,
4851 cpu_to_be16(DDF_Transition));
4852
4853 /* Now make sure vlist is correct for each dl. */
4854 for (dl = ddf->dlist; dl; dl = dl->next)
4855 ddf_update_vlist(ddf, dl);
4856 ddf_remove_failed(ddf);
4857
4858 ddf_set_updates_pending(ddf, vc);
4859 }
4860
4861 static void ddf_process_update(struct supertype *st,
4862 struct metadata_update *update)
4863 {
4864 /* Apply this update to the metadata.
4865 * The first 4 bytes are a DDF_*_MAGIC which guides
4866 * our actions.
4867 * Possible update are:
4868 * DDF_PHYS_RECORDS_MAGIC
4869 * Add a new physical device or remove an old one.
4870 * Changes to this record only happen implicitly.
4871 * used_pdes is the device number.
4872 * DDF_VIRT_RECORDS_MAGIC
4873 * Add a new VD. Possibly also change the 'access' bits.
4874 * populated_vdes is the entry number.
4875 * DDF_VD_CONF_MAGIC
4876 * New or updated VD. the VIRT_RECORD must already
4877 * exist. For an update, phys_refnum and lba_offset
4878 * (at least) are updated, and the VD_CONF must
4879 * be written to precisely those devices listed with
4880 * a phys_refnum.
4881 * DDF_SPARE_ASSIGN_MAGIC
4882 * replacement Spare Assignment Record... but for which device?
4883 *
4884 * So, e.g.:
4885 * - to create a new array, we send a VIRT_RECORD and
4886 * a VD_CONF. Then assemble and start the array.
4887 * - to activate a spare we send a VD_CONF to add the phys_refnum
4888 * and offset. This will also mark the spare as active with
4889 * a spare-assignment record.
4890 */
4891 be32 *magic = (be32 *)update->buf;
4892
4893 dprintf("Process update %x\n", be32_to_cpu(*magic));
4894
4895 if (be32_eq(*magic, DDF_PHYS_RECORDS_MAGIC)) {
4896 if (update->len == (sizeof(struct phys_disk) +
4897 sizeof(struct phys_disk_entry)))
4898 ddf_process_phys_update(st, update);
4899 } else if (be32_eq(*magic, DDF_VIRT_RECORDS_MAGIC)) {
4900 if (update->len == (sizeof(struct virtual_disk) +
4901 sizeof(struct virtual_entry)))
4902 ddf_process_virt_update(st, update);
4903 } else if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4904 ddf_process_conf_update(st, update);
4905 }
4906 /* case DDF_SPARE_ASSIGN_MAGIC */
4907 }
4908
4909 static void ddf_prepare_update(struct supertype *st,
4910 struct metadata_update *update)
4911 {
4912 /* This update arrived at managemon.
4913 * We are about to pass it to monitor.
4914 * If a malloc is needed, do it here.
4915 */
4916 struct ddf_super *ddf = st->sb;
4917 be32 *magic = (be32 *)update->buf;
4918 if (be32_eq(*magic, DDF_VD_CONF_MAGIC)) {
4919 struct vcl *vcl;
4920 struct vd_config *conf = (struct vd_config *) update->buf;
4921 if (posix_memalign(&update->space, 512,
4922 offsetof(struct vcl, conf)
4923 + ddf->conf_rec_len * 512) != 0) {
4924 update->space = NULL;
4925 return;
4926 }
4927 vcl = update->space;
4928 vcl->conf.sec_elmnt_count = conf->sec_elmnt_count;
4929 if (alloc_other_bvds(ddf, vcl) != 0) {
4930 free(update->space);
4931 update->space = NULL;
4932 }
4933 }
4934 }
4935
4936 /*
4937 * Check degraded state of a RAID10.
4938 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4939 */
4940 static int raid10_degraded(struct mdinfo *info)
4941 {
4942 int n_prim, n_bvds;
4943 int i;
4944 struct mdinfo *d;
4945 char *found;
4946 int ret = -1;
4947
4948 n_prim = info->array.layout & ~0x100;
4949 n_bvds = info->array.raid_disks / n_prim;
4950 found = xmalloc(n_bvds);
4951 if (found == NULL)
4952 return ret;
4953 memset(found, 0, n_bvds);
4954 for (d = info->devs; d; d = d->next) {
4955 i = d->disk.raid_disk / n_prim;
4956 if (i >= n_bvds) {
4957 pr_err("%s: BUG: invalid raid disk\n", __func__);
4958 goto out;
4959 }
4960 if (d->state_fd > 0)
4961 found[i]++;
4962 }
4963 ret = 2;
4964 for (i = 0; i < n_bvds; i++)
4965 if (!found[i]) {
4966 dprintf("%s: BVD %d/%d failed\n", __func__, i, n_bvds);
4967 ret = 0;
4968 goto out;
4969 } else if (found[i] < n_prim) {
4970 dprintf("%s: BVD %d/%d degraded\n", __func__, i,
4971 n_bvds);
4972 ret = 1;
4973 }
4974 out:
4975 free(found);
4976 return ret;
4977 }
4978
4979 /*
4980 * Check if the array 'a' is degraded but not failed.
4981 * If it is, find as many spares as are available and needed and
4982 * arrange for their inclusion.
4983 * We only choose devices which are not already in the array,
4984 * and prefer those with a spare-assignment to this array.
4985 * Otherwise we choose global spares - assuming always that
4986 * there is enough room.
4987 * For each spare that we assign, we return an 'mdinfo' which
4988 * describes the position for the device in the array.
4989 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4990 * the new phys_refnum and lba_offset values.
4991 *
4992 * Only worry about BVDs at the moment.
4993 */
4994 static struct mdinfo *ddf_activate_spare(struct active_array *a,
4995 struct metadata_update **updates)
4996 {
4997 int working = 0;
4998 struct mdinfo *d;
4999 struct ddf_super *ddf = a->container->sb;
5000 int global_ok = 0;
5001 struct mdinfo *rv = NULL;
5002 struct mdinfo *di;
5003 struct metadata_update *mu;
5004 struct dl *dl;
5005 int i;
5006 unsigned int j;
5007 struct vcl *vcl;
5008 struct vd_config *vc;
5009 unsigned int n_bvd;
5010
5011 for (d = a->info.devs ; d ; d = d->next) {
5012 if ((d->curr_state & DS_FAULTY) &&
5013 d->state_fd >= 0)
5014 /* wait for Removal to happen */
5015 return NULL;
5016 if (d->state_fd >= 0)
5017 working ++;
5018 }
5019
5020 dprintf("%s: working=%d (%d) level=%d\n", __func__, working,
5021 a->info.array.raid_disks,
5022 a->info.array.level);
5023 if (working == a->info.array.raid_disks)
5024 return NULL; /* array not degraded */
5025 switch (a->info.array.level) {
5026 case 1:
5027 if (working == 0)
5028 return NULL; /* failed */
5029 break;
5030 case 4:
5031 case 5:
5032 if (working < a->info.array.raid_disks - 1)
5033 return NULL; /* failed */
5034 break;
5035 case 6:
5036 if (working < a->info.array.raid_disks - 2)
5037 return NULL; /* failed */
5038 break;
5039 case 10:
5040 if (raid10_degraded(&a->info) < 1)
5041 return NULL;
5042 break;
5043 default: /* concat or stripe */
5044 return NULL; /* failed */
5045 }
5046
5047 /* For each slot, if it is not working, find a spare */
5048 dl = ddf->dlist;
5049 for (i = 0; i < a->info.array.raid_disks; i++) {
5050 for (d = a->info.devs ; d ; d = d->next)
5051 if (d->disk.raid_disk == i)
5052 break;
5053 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
5054 if (d && (d->state_fd >= 0))
5055 continue;
5056
5057 /* OK, this device needs recovery. Find a spare */
5058 again:
5059 for ( ; dl ; dl = dl->next) {
5060 unsigned long long esize;
5061 unsigned long long pos;
5062 struct mdinfo *d2;
5063 int is_global = 0;
5064 int is_dedicated = 0;
5065 be16 state;
5066
5067 if (dl->pdnum < 0)
5068 continue;
5069 state = ddf->phys->entries[dl->pdnum].state;
5070 if (be16_and(state,
5071 cpu_to_be16(DDF_Failed|DDF_Missing)) ||
5072 !be16_and(state,
5073 cpu_to_be16(DDF_Online)))
5074 continue;
5075
5076 /* If in this array, skip */
5077 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
5078 if (d2->state_fd >= 0 &&
5079 d2->disk.major == dl->major &&
5080 d2->disk.minor == dl->minor) {
5081 dprintf("%x:%x (%08x) already in array\n",
5082 dl->major, dl->minor,
5083 be32_to_cpu(dl->disk.refnum));
5084 break;
5085 }
5086 if (d2)
5087 continue;
5088 if (be16_and(ddf->phys->entries[dl->pdnum].type,
5089 cpu_to_be16(DDF_Spare))) {
5090 /* Check spare assign record */
5091 if (dl->spare) {
5092 if (dl->spare->type & DDF_spare_dedicated) {
5093 /* check spare_ents for guid */
5094 unsigned int j;
5095 for (j = 0 ;
5096 j < be16_to_cpu
5097 (dl->spare
5098 ->populated);
5099 j++) {
5100 if (memcmp(dl->spare->spare_ents[j].guid,
5101 ddf->virt->entries[a->info.container_member].guid,
5102 DDF_GUID_LEN) == 0)
5103 is_dedicated = 1;
5104 }
5105 } else
5106 is_global = 1;
5107 }
5108 } else if (be16_and(ddf->phys->entries[dl->pdnum].type,
5109 cpu_to_be16(DDF_Global_Spare))) {
5110 is_global = 1;
5111 } else if (!be16_and(ddf->phys
5112 ->entries[dl->pdnum].state,
5113 cpu_to_be16(DDF_Failed))) {
5114 /* we can possibly use some of this */
5115 is_global = 1;
5116 }
5117 if ( ! (is_dedicated ||
5118 (is_global && global_ok))) {
5119 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
5120 is_dedicated, is_global);
5121 continue;
5122 }
5123
5124 /* We are allowed to use this device - is there space?
5125 * We need a->info.component_size sectors */
5126 esize = a->info.component_size;
5127 pos = find_space(ddf, dl, INVALID_SECTORS, &esize);
5128
5129 if (esize < a->info.component_size) {
5130 dprintf("%x:%x has no room: %llu %llu\n",
5131 dl->major, dl->minor,
5132 esize, a->info.component_size);
5133 /* No room */
5134 continue;
5135 }
5136
5137 /* Cool, we have a device with some space at pos */
5138 di = xcalloc(1, sizeof(*di));
5139 di->disk.number = i;
5140 di->disk.raid_disk = i;
5141 di->disk.major = dl->major;
5142 di->disk.minor = dl->minor;
5143 di->disk.state = 0;
5144 di->recovery_start = 0;
5145 di->data_offset = pos;
5146 di->component_size = a->info.component_size;
5147 di->next = rv;
5148 rv = di;
5149 dprintf("%x:%x (%08x) to be %d at %llu\n",
5150 dl->major, dl->minor,
5151 be32_to_cpu(dl->disk.refnum), i, pos);
5152
5153 break;
5154 }
5155 if (!dl && ! global_ok) {
5156 /* not enough dedicated spares, try global */
5157 global_ok = 1;
5158 dl = ddf->dlist;
5159 goto again;
5160 }
5161 }
5162
5163 if (!rv)
5164 /* No spares found */
5165 return rv;
5166 /* Now 'rv' has a list of devices to return.
5167 * Create a metadata_update record to update the
5168 * phys_refnum and lba_offset values
5169 */
5170 vc = find_vdcr(ddf, a->info.container_member, rv->disk.raid_disk,
5171 &n_bvd, &vcl);
5172 if (vc == NULL)
5173 return NULL;
5174
5175 mu = xmalloc(sizeof(*mu));
5176 if (posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
5177 free(mu);
5178 mu = NULL;
5179 }
5180
5181 mu->len = ddf->conf_rec_len * 512 * vcl->conf.sec_elmnt_count;
5182 mu->buf = xmalloc(mu->len);
5183 mu->space = NULL;
5184 mu->space_list = NULL;
5185 mu->next = *updates;
5186 memcpy(mu->buf, &vcl->conf, ddf->conf_rec_len * 512);
5187 for (j = 1; j < vcl->conf.sec_elmnt_count; j++)
5188 memcpy(mu->buf + j * ddf->conf_rec_len * 512,
5189 vcl->other_bvds[j-1], ddf->conf_rec_len * 512);
5190
5191 vc = (struct vd_config*)mu->buf;
5192 for (di = rv ; di ; di = di->next) {
5193 unsigned int i_sec, i_prim;
5194 i_sec = di->disk.raid_disk
5195 / be16_to_cpu(vcl->conf.prim_elmnt_count);
5196 i_prim = di->disk.raid_disk
5197 % be16_to_cpu(vcl->conf.prim_elmnt_count);
5198 vc = (struct vd_config *)(mu->buf
5199 + i_sec * ddf->conf_rec_len * 512);
5200 for (dl = ddf->dlist; dl; dl = dl->next)
5201 if (dl->major == di->disk.major
5202 && dl->minor == di->disk.minor)
5203 break;
5204 if (!dl || dl->pdnum < 0) {
5205 pr_err("%s: BUG: can't find disk %d (%d/%d)\n",
5206 __func__, di->disk.raid_disk,
5207 di->disk.major, di->disk.minor);
5208 return NULL;
5209 }
5210 vc->phys_refnum[i_prim] = ddf->phys->entries[dl->pdnum].refnum;
5211 LBA_OFFSET(ddf, vc)[i_prim] = cpu_to_be64(di->data_offset);
5212 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec, i_prim,
5213 be32_to_cpu(vc->phys_refnum[i_prim]),
5214 be64_to_cpu(LBA_OFFSET(ddf, vc)[i_prim]));
5215 }
5216 *updates = mu;
5217 return rv;
5218 }
5219 #endif /* MDASSEMBLE */
5220
5221 static int ddf_level_to_layout(int level)
5222 {
5223 switch(level) {
5224 case 0:
5225 case 1:
5226 return 0;
5227 case 5:
5228 return ALGORITHM_LEFT_SYMMETRIC;
5229 case 6:
5230 return ALGORITHM_ROTATING_N_CONTINUE;
5231 case 10:
5232 return 0x102;
5233 default:
5234 return UnSet;
5235 }
5236 }
5237
5238 static void default_geometry_ddf(struct supertype *st, int *level, int *layout, int *chunk)
5239 {
5240 if (level && *level == UnSet)
5241 *level = LEVEL_CONTAINER;
5242
5243 if (level && layout && *layout == UnSet)
5244 *layout = ddf_level_to_layout(*level);
5245 }
5246
5247 struct superswitch super_ddf = {
5248 #ifndef MDASSEMBLE
5249 .examine_super = examine_super_ddf,
5250 .brief_examine_super = brief_examine_super_ddf,
5251 .brief_examine_subarrays = brief_examine_subarrays_ddf,
5252 .export_examine_super = export_examine_super_ddf,
5253 .detail_super = detail_super_ddf,
5254 .brief_detail_super = brief_detail_super_ddf,
5255 .validate_geometry = validate_geometry_ddf,
5256 .write_init_super = write_init_super_ddf,
5257 .add_to_super = add_to_super_ddf,
5258 .remove_from_super = remove_from_super_ddf,
5259 .load_container = load_container_ddf,
5260 .copy_metadata = copy_metadata_ddf,
5261 .kill_subarray = kill_subarray_ddf,
5262 #endif
5263 .match_home = match_home_ddf,
5264 .uuid_from_super= uuid_from_super_ddf,
5265 .getinfo_super = getinfo_super_ddf,
5266 .update_super = update_super_ddf,
5267
5268 .avail_size = avail_size_ddf,
5269
5270 .compare_super = compare_super_ddf,
5271
5272 .load_super = load_super_ddf,
5273 .init_super = init_super_ddf,
5274 .store_super = store_super_ddf,
5275 .free_super = free_super_ddf,
5276 .match_metadata_desc = match_metadata_desc_ddf,
5277 .container_content = container_content_ddf,
5278 .default_geometry = default_geometry_ddf,
5279
5280 .external = 1,
5281
5282 #ifndef MDASSEMBLE
5283 /* for mdmon */
5284 .open_new = ddf_open_new,
5285 .set_array_state= ddf_set_array_state,
5286 .set_disk = ddf_set_disk,
5287 .sync_metadata = ddf_sync_metadata,
5288 .process_update = ddf_process_update,
5289 .prepare_update = ddf_prepare_update,
5290 .activate_spare = ddf_activate_spare,
5291 #endif
5292 .name = "ddf",
5293 };
Unnamed repository; edit this file 'description' to name the repository.