imsm: pass subarray id to kill_subarray function
[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 static int load_super_ddf_all(struct supertype *st, int fd,
504 void **sbp, char *devname);
505 static int get_svd_state(const struct ddf_super *, const struct vcl *);
506 static int
507 validate_geometry_ddf_container(struct supertype *st,
508 int level, int layout, int raiddisks,
509 int chunk, unsigned long long size,
510 unsigned long long data_offset,
511 char *dev, unsigned long long *freesize,
512 int verbose);
513
514 static int validate_geometry_ddf_bvd(struct supertype *st,
515 int level, int layout, int raiddisks,
516 int *chunk, unsigned long long size,
517 unsigned long long data_offset,
518 char *dev, unsigned long long *freesize,
519 int verbose);
520
521 static void free_super_ddf(struct supertype *st);
522 static int all_ff(const char *guid);
523 static unsigned int get_pd_index_from_refnum(const struct vcl *vc,
524 be32 refnum, unsigned int nmax,
525 const struct vd_config **bvd,
526 unsigned int *idx);
527 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map);
528 static void uuid_from_ddf_guid(const char *guid, int uuid[4]);
529 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
530 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i);
531 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map);
532 static int init_super_ddf_bvd(struct supertype *st,
533 mdu_array_info_t *info,
534 unsigned long long size,
535 char *name, char *homehost,
536 int *uuid, unsigned long long data_offset);
537
538 #if DEBUG
539 static void pr_state(struct ddf_super *ddf, const char *msg)
540 {
541 unsigned int i;
542 dprintf("%s: ", msg);
543 for (i = 0; i < be16_to_cpu(ddf->active->max_vd_entries); i++) {
544 if (all_ff(ddf->virt->entries[i].guid))
545 continue;
546 dprintf_cont("%u(s=%02x i=%02x) ", i,
547 ddf->virt->entries[i].state,
548 ddf->virt->entries[i].init_state);
549 }
550 dprintf_cont("\n");
551 }
552 #else
553 static void pr_state(const struct ddf_super *ddf, const char *msg) {}
554 #endif
555
556 static void _ddf_set_updates_pending(struct ddf_super *ddf, struct vd_config *vc,
557 const char *func)
558 {
559 if (vc) {
560 vc->timestamp = cpu_to_be32(time(0)-DECADE);
561 vc->seqnum = cpu_to_be32(be32_to_cpu(vc->seqnum) + 1);
562 }
563 if (ddf->updates_pending)
564 return;
565 ddf->updates_pending = 1;
566 ddf->active->seq = cpu_to_be32((be32_to_cpu(ddf->active->seq)+1));
567 pr_state(ddf, func);
568 }
569
570 #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__)
571
572 static be32 calc_crc(void *buf, int len)
573 {
574 /* crcs are always at the same place as in the ddf_header */
575 struct ddf_header *ddf = buf;
576 be32 oldcrc = ddf->crc;
577 __u32 newcrc;
578 ddf->crc = cpu_to_be32(0xffffffff);
579
580 newcrc = crc32(0, buf, len);
581 ddf->crc = oldcrc;
582 /* The crc is stored (like everything) bigendian, so convert
583 * here for simplicity
584 */
585 return cpu_to_be32(newcrc);
586 }
587
588 #define DDF_INVALID_LEVEL 0xff
589 #define DDF_NO_SECONDARY 0xff
590 static int err_bad_md_layout(const mdu_array_info_t *array)
591 {
592 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
593 array->level, array->layout, array->raid_disks);
594 return -1;
595 }
596
597 static int layout_md2ddf(const mdu_array_info_t *array,
598 struct vd_config *conf)
599 {
600 be16 prim_elmnt_count = cpu_to_be16(array->raid_disks);
601 __u8 prl = DDF_INVALID_LEVEL, rlq = 0;
602 __u8 sec_elmnt_count = 1;
603 __u8 srl = DDF_NO_SECONDARY;
604
605 switch (array->level) {
606 case LEVEL_LINEAR:
607 prl = DDF_CONCAT;
608 break;
609 case 0:
610 rlq = DDF_RAID0_SIMPLE;
611 prl = DDF_RAID0;
612 break;
613 case 1:
614 switch (array->raid_disks) {
615 case 2:
616 rlq = DDF_RAID1_SIMPLE;
617 break;
618 case 3:
619 rlq = DDF_RAID1_MULTI;
620 break;
621 default:
622 return err_bad_md_layout(array);
623 }
624 prl = DDF_RAID1;
625 break;
626 case 4:
627 if (array->layout != 0)
628 return err_bad_md_layout(array);
629 rlq = DDF_RAID4_N;
630 prl = DDF_RAID4;
631 break;
632 case 5:
633 switch (array->layout) {
634 case ALGORITHM_LEFT_ASYMMETRIC:
635 rlq = DDF_RAID5_N_RESTART;
636 break;
637 case ALGORITHM_RIGHT_ASYMMETRIC:
638 rlq = DDF_RAID5_0_RESTART;
639 break;
640 case ALGORITHM_LEFT_SYMMETRIC:
641 rlq = DDF_RAID5_N_CONTINUE;
642 break;
643 case ALGORITHM_RIGHT_SYMMETRIC:
644 /* not mentioned in standard */
645 default:
646 return err_bad_md_layout(array);
647 }
648 prl = DDF_RAID5;
649 break;
650 case 6:
651 switch (array->layout) {
652 case ALGORITHM_ROTATING_N_RESTART:
653 rlq = DDF_RAID5_N_RESTART;
654 break;
655 case ALGORITHM_ROTATING_ZERO_RESTART:
656 rlq = DDF_RAID6_0_RESTART;
657 break;
658 case ALGORITHM_ROTATING_N_CONTINUE:
659 rlq = DDF_RAID5_N_CONTINUE;
660 break;
661 default:
662 return err_bad_md_layout(array);
663 }
664 prl = DDF_RAID6;
665 break;
666 case 10:
667 if (array->raid_disks % 2 == 0 && array->layout == 0x102) {
668 rlq = DDF_RAID1_SIMPLE;
669 prim_elmnt_count = cpu_to_be16(2);
670 sec_elmnt_count = array->raid_disks / 2;
671 srl = DDF_2SPANNED;
672 prl = DDF_RAID1;
673 } else if (array->raid_disks % 3 == 0 &&
674 array->layout == 0x103) {
675 rlq = DDF_RAID1_MULTI;
676 prim_elmnt_count = cpu_to_be16(3);
677 sec_elmnt_count = array->raid_disks / 3;
678 srl = DDF_2SPANNED;
679 prl = DDF_RAID1;
680 } else if (array->layout == 0x201) {
681 prl = DDF_RAID1E;
682 rlq = DDF_RAID1E_OFFSET;
683 } else if (array->layout == 0x102) {
684 prl = DDF_RAID1E;
685 rlq = DDF_RAID1E_ADJACENT;
686 } else
687 return err_bad_md_layout(array);
688 break;
689 default:
690 return err_bad_md_layout(array);
691 }
692 conf->prl = prl;
693 conf->prim_elmnt_count = prim_elmnt_count;
694 conf->rlq = rlq;
695 conf->srl = srl;
696 conf->sec_elmnt_count = sec_elmnt_count;
697 return 0;
698 }
699
700 static int err_bad_ddf_layout(const struct vd_config *conf)
701 {
702 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
703 conf->prl, conf->rlq, be16_to_cpu(conf->prim_elmnt_count));
704 return -1;
705 }
706
707 static int layout_ddf2md(const struct vd_config *conf,
708 mdu_array_info_t *array)
709 {
710 int level = LEVEL_UNSUPPORTED;
711 int layout = 0;
712 int raiddisks = be16_to_cpu(conf->prim_elmnt_count);
713
714 if (conf->sec_elmnt_count > 1) {
715 /* see also check_secondary() */
716 if (conf->prl != DDF_RAID1 ||
717 (conf->srl != DDF_2STRIPED && conf->srl != DDF_2SPANNED)) {
718 pr_err("Unsupported secondary RAID level %u/%u\n",
719 conf->prl, conf->srl);
720 return -1;
721 }
722 if (raiddisks == 2 && conf->rlq == DDF_RAID1_SIMPLE)
723 layout = 0x102;
724 else if (raiddisks == 3 && conf->rlq == DDF_RAID1_MULTI)
725 layout = 0x103;
726 else
727 return err_bad_ddf_layout(conf);
728 raiddisks *= conf->sec_elmnt_count;
729 level = 10;
730 goto good;
731 }
732
733 switch (conf->prl) {
734 case DDF_CONCAT:
735 level = LEVEL_LINEAR;
736 break;
737 case DDF_RAID0:
738 if (conf->rlq != DDF_RAID0_SIMPLE)
739 return err_bad_ddf_layout(conf);
740 level = 0;
741 break;
742 case DDF_RAID1:
743 if (!((conf->rlq == DDF_RAID1_SIMPLE && raiddisks == 2) ||
744 (conf->rlq == DDF_RAID1_MULTI && raiddisks == 3)))
745 return err_bad_ddf_layout(conf);
746 level = 1;
747 break;
748 case DDF_RAID1E:
749 if (conf->rlq == DDF_RAID1E_ADJACENT)
750 layout = 0x102;
751 else if (conf->rlq == DDF_RAID1E_OFFSET)
752 layout = 0x201;
753 else
754 return err_bad_ddf_layout(conf);
755 level = 10;
756 break;
757 case DDF_RAID4:
758 if (conf->rlq != DDF_RAID4_N)
759 return err_bad_ddf_layout(conf);
760 level = 4;
761 break;
762 case DDF_RAID5:
763 switch (conf->rlq) {
764 case DDF_RAID5_N_RESTART:
765 layout = ALGORITHM_LEFT_ASYMMETRIC;
766 break;
767 case DDF_RAID5_0_RESTART:
768 layout = ALGORITHM_RIGHT_ASYMMETRIC;
769 break;
770 case DDF_RAID5_N_CONTINUE:
771 layout = ALGORITHM_LEFT_SYMMETRIC;
772 break;
773 default:
774 return err_bad_ddf_layout(conf);
775 }
776 level = 5;
777 break;
778 case DDF_RAID6:
779 switch (conf->rlq) {
780 case DDF_RAID5_N_RESTART:
781 layout = ALGORITHM_ROTATING_N_RESTART;
782 break;
783 case DDF_RAID6_0_RESTART:
784 layout = ALGORITHM_ROTATING_ZERO_RESTART;
785 break;
786 case DDF_RAID5_N_CONTINUE:
787 layout = ALGORITHM_ROTATING_N_CONTINUE;
788 break;
789 default:
790 return err_bad_ddf_layout(conf);
791 }
792 level = 6;
793 break;
794 default:
795 return err_bad_ddf_layout(conf);
796 };
797
798 good:
799 array->level = level;
800 array->layout = layout;
801 array->raid_disks = raiddisks;
802 return 0;
803 }
804
805 static int load_ddf_header(int fd, unsigned long long lba,
806 unsigned long long size,
807 int type,
808 struct ddf_header *hdr, struct ddf_header *anchor)
809 {
810 /* read a ddf header (primary or secondary) from fd/lba
811 * and check that it is consistent with anchor
812 * Need to check:
813 * magic, crc, guid, rev, and LBA's header_type, and
814 * everything after header_type must be the same
815 */
816 if (lba >= size-1)
817 return 0;
818
819 if (lseek64(fd, lba<<9, 0) < 0)
820 return 0;
821
822 if (read(fd, hdr, 512) != 512)
823 return 0;
824
825 if (!be32_eq(hdr->magic, DDF_HEADER_MAGIC)) {
826 pr_err("bad header magic\n");
827 return 0;
828 }
829 if (!be32_eq(calc_crc(hdr, 512), hdr->crc)) {
830 pr_err("bad CRC\n");
831 return 0;
832 }
833 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
834 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
835 !be64_eq(anchor->primary_lba, hdr->primary_lba) ||
836 !be64_eq(anchor->secondary_lba, hdr->secondary_lba) ||
837 hdr->type != type ||
838 memcmp(anchor->pad2, hdr->pad2, 512 -
839 offsetof(struct ddf_header, pad2)) != 0) {
840 pr_err("header mismatch\n");
841 return 0;
842 }
843
844 /* Looks good enough to me... */
845 return 1;
846 }
847
848 static void *load_section(int fd, struct ddf_super *super, void *buf,
849 be32 offset_be, be32 len_be, int check)
850 {
851 unsigned long long offset = be32_to_cpu(offset_be);
852 unsigned long long len = be32_to_cpu(len_be);
853 int dofree = (buf == NULL);
854
855 if (check)
856 if (len != 2 && len != 8 && len != 32 &&
857 len != 128 && len != 512)
858 return NULL;
859
860 if (len > 1024)
861 return NULL;
862 if (!buf && posix_memalign(&buf, 512, len<<9) != 0)
863 buf = NULL;
864
865 if (!buf)
866 return NULL;
867
868 if (super->active->type == 1)
869 offset += be64_to_cpu(super->active->primary_lba);
870 else
871 offset += be64_to_cpu(super->active->secondary_lba);
872
873 if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
874 if (dofree)
875 free(buf);
876 return NULL;
877 }
878 if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
879 if (dofree)
880 free(buf);
881 return NULL;
882 }
883 return buf;
884 }
885
886 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
887 {
888 unsigned long long dsize;
889
890 get_dev_size(fd, NULL, &dsize);
891
892 if (lseek64(fd, dsize-512, 0) < 0) {
893 if (devname)
894 pr_err("Cannot seek to anchor block on %s: %s\n",
895 devname, strerror(errno));
896 return 1;
897 }
898 if (read(fd, &super->anchor, 512) != 512) {
899 if (devname)
900 pr_err("Cannot read anchor block on %s: %s\n",
901 devname, strerror(errno));
902 return 1;
903 }
904 if (!be32_eq(super->anchor.magic, DDF_HEADER_MAGIC)) {
905 if (devname)
906 pr_err("no DDF anchor found on %s\n",
907 devname);
908 return 2;
909 }
910 if (!be32_eq(calc_crc(&super->anchor, 512), super->anchor.crc)) {
911 if (devname)
912 pr_err("bad CRC on anchor on %s\n",
913 devname);
914 return 2;
915 }
916 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
917 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
918 if (devname)
919 pr_err("can only support super revision %.8s and earlier, not %.8s on %s\n",
920 DDF_REVISION_2, super->anchor.revision,devname);
921 return 2;
922 }
923 super->active = NULL;
924 if (load_ddf_header(fd, be64_to_cpu(super->anchor.primary_lba),
925 dsize >> 9, 1,
926 &super->primary, &super->anchor) == 0) {
927 if (devname)
928 pr_err("Failed to load primary DDF header on %s\n", devname);
929 } else
930 super->active = &super->primary;
931
932 if (load_ddf_header(fd, be64_to_cpu(super->anchor.secondary_lba),
933 dsize >> 9, 2,
934 &super->secondary, &super->anchor)) {
935 if (super->active == NULL ||
936 (be32_to_cpu(super->primary.seq)
937 < be32_to_cpu(super->secondary.seq) &&
938 !super->secondary.openflag) ||
939 (be32_to_cpu(super->primary.seq) ==
940 be32_to_cpu(super->secondary.seq) &&
941 super->primary.openflag && !super->secondary.openflag))
942 super->active = &super->secondary;
943 } else if (devname &&
944 be64_to_cpu(super->anchor.secondary_lba) != ~(__u64)0)
945 pr_err("Failed to load secondary DDF header on %s\n",
946 devname);
947 if (super->active == NULL)
948 return 2;
949 return 0;
950 }
951
952 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
953 {
954 void *ok;
955 ok = load_section(fd, super, &super->controller,
956 super->active->controller_section_offset,
957 super->active->controller_section_length,
958 0);
959 super->phys = load_section(fd, super, NULL,
960 super->active->phys_section_offset,
961 super->active->phys_section_length,
962 1);
963 super->pdsize = be32_to_cpu(super->active->phys_section_length) * 512;
964
965 super->virt = load_section(fd, super, NULL,
966 super->active->virt_section_offset,
967 super->active->virt_section_length,
968 1);
969 super->vdsize = be32_to_cpu(super->active->virt_section_length) * 512;
970 if (!ok ||
971 !super->phys ||
972 !super->virt) {
973 free(super->phys);
974 free(super->virt);
975 super->phys = NULL;
976 super->virt = NULL;
977 return 2;
978 }
979 super->conflist = NULL;
980 super->dlist = NULL;
981
982 super->max_part = be16_to_cpu(super->active->max_partitions);
983 super->mppe = be16_to_cpu(super->active->max_primary_element_entries);
984 super->conf_rec_len = be16_to_cpu(super->active->config_record_len);
985 return 0;
986 }
987
988 #define DDF_UNUSED_BVD 0xff
989 static int alloc_other_bvds(const struct ddf_super *ddf, struct vcl *vcl)
990 {
991 unsigned int n_vds = vcl->conf.sec_elmnt_count - 1;
992 unsigned int i, vdsize;
993 void *p;
994 if (n_vds == 0) {
995 vcl->other_bvds = NULL;
996 return 0;
997 }
998 vdsize = ddf->conf_rec_len * 512;
999 if (posix_memalign(&p, 512, n_vds *
1000 (vdsize + sizeof(struct vd_config *))) != 0)
1001 return -1;
1002 vcl->other_bvds = (struct vd_config **) (p + n_vds * vdsize);
1003 for (i = 0; i < n_vds; i++) {
1004 vcl->other_bvds[i] = p + i * vdsize;
1005 memset(vcl->other_bvds[i], 0, vdsize);
1006 vcl->other_bvds[i]->sec_elmnt_seq = DDF_UNUSED_BVD;
1007 }
1008 return 0;
1009 }
1010
1011 static void add_other_bvd(struct vcl *vcl, struct vd_config *vd,
1012 unsigned int len)
1013 {
1014 int i;
1015 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1016 if (vcl->other_bvds[i]->sec_elmnt_seq == vd->sec_elmnt_seq)
1017 break;
1018
1019 if (i < vcl->conf.sec_elmnt_count-1) {
1020 if (be32_to_cpu(vd->seqnum) <=
1021 be32_to_cpu(vcl->other_bvds[i]->seqnum))
1022 return;
1023 } else {
1024 for (i = 0; i < vcl->conf.sec_elmnt_count-1; i++)
1025 if (vcl->other_bvds[i]->sec_elmnt_seq == DDF_UNUSED_BVD)
1026 break;
1027 if (i == vcl->conf.sec_elmnt_count-1) {
1028 pr_err("no space for sec level config %u, count is %u\n",
1029 vd->sec_elmnt_seq, vcl->conf.sec_elmnt_count);
1030 return;
1031 }
1032 }
1033 memcpy(vcl->other_bvds[i], vd, len);
1034 }
1035
1036 static int load_ddf_local(int fd, struct ddf_super *super,
1037 char *devname, int keep)
1038 {
1039 struct dl *dl;
1040 struct stat stb;
1041 char *conf;
1042 unsigned int i;
1043 unsigned int confsec;
1044 int vnum;
1045 unsigned int max_virt_disks =
1046 be16_to_cpu(super->active->max_vd_entries);
1047 unsigned long long dsize;
1048
1049 /* First the local disk info */
1050 if (posix_memalign((void**)&dl, 512,
1051 sizeof(*dl) +
1052 (super->max_part) * sizeof(dl->vlist[0])) != 0) {
1053 pr_err("could not allocate disk info buffer\n");
1054 return 1;
1055 }
1056
1057 load_section(fd, super, &dl->disk,
1058 super->active->data_section_offset,
1059 super->active->data_section_length,
1060 0);
1061 dl->devname = devname ? xstrdup(devname) : NULL;
1062
1063 fstat(fd, &stb);
1064 dl->major = major(stb.st_rdev);
1065 dl->minor = minor(stb.st_rdev);
1066 dl->next = super->dlist;
1067 dl->fd = keep ? fd : -1;
1068
1069 dl->size = 0;
1070 if (get_dev_size(fd, devname, &dsize))
1071 dl->size = dsize >> 9;
1072 /* If the disks have different sizes, the LBAs will differ
1073 * between phys disks.
1074 * At this point here, the values in super->active must be valid
1075 * for this phys disk. */
1076 dl->primary_lba = super->active->primary_lba;
1077 dl->secondary_lba = super->active->secondary_lba;
1078 dl->workspace_lba = super->active->workspace_lba;
1079 dl->spare = NULL;
1080 for (i = 0 ; i < super->max_part ; i++)
1081 dl->vlist[i] = NULL;
1082 super->dlist = dl;
1083 dl->pdnum = -1;
1084 for (i = 0; i < be16_to_cpu(super->active->max_pd_entries); i++)
1085 if (memcmp(super->phys->entries[i].guid,
1086 dl->disk.guid, DDF_GUID_LEN) == 0)
1087 dl->pdnum = i;
1088
1089 /* Now the config list. */
1090 /* 'conf' is an array of config entries, some of which are
1091 * probably invalid. Those which are good need to be copied into
1092 * the conflist
1093 */
1094
1095 conf = load_section(fd, super, super->conf,
1096 super->active->config_section_offset,
1097 super->active->config_section_length,
1098 0);
1099 super->conf = conf;
1100 vnum = 0;
1101 for (confsec = 0;
1102 confsec < be32_to_cpu(super->active->config_section_length);
1103 confsec += super->conf_rec_len) {
1104 struct vd_config *vd =
1105 (struct vd_config *)((char*)conf + confsec*512);
1106 struct vcl *vcl;
1107
1108 if (be32_eq(vd->magic, DDF_SPARE_ASSIGN_MAGIC)) {
1109 if (dl->spare)
1110 continue;
1111 if (posix_memalign((void**)&dl->spare, 512,
1112 super->conf_rec_len*512) != 0) {
1113 pr_err("could not allocate spare info buf\n");
1114 return 1;
1115 }
1116
1117 memcpy(dl->spare, vd, super->conf_rec_len*512);
1118 continue;
1119 }
1120 if (!be32_eq(vd->magic, DDF_VD_CONF_MAGIC))
1121 /* Must be vendor-unique - I cannot handle those */
1122 continue;
1123
1124 for (vcl = super->conflist; vcl; vcl = vcl->next) {
1125 if (memcmp(vcl->conf.guid,
1126 vd->guid, DDF_GUID_LEN) == 0)
1127 break;
1128 }
1129
1130 if (vcl) {
1131 dl->vlist[vnum++] = vcl;
1132 if (vcl->other_bvds != NULL &&
1133 vcl->conf.sec_elmnt_seq != vd->sec_elmnt_seq) {
1134 add_other_bvd(vcl, vd, super->conf_rec_len*512);
1135 continue;
1136 }
1137 if (be32_to_cpu(vd->seqnum) <=
1138 be32_to_cpu(vcl->conf.seqnum))
1139 continue;
1140 } else {
1141 if (posix_memalign((void**)&vcl, 512,
1142 (super->conf_rec_len*512 +
1143 offsetof(struct vcl, conf))) != 0) {
1144 pr_err("could not allocate vcl buf\n");
1145 return 1;
1146 }
1147 vcl->next = super->conflist;
1148 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1149 vcl->conf.sec_elmnt_count = vd->sec_elmnt_count;
1150 if (alloc_other_bvds(super, vcl) != 0) {
1151 pr_err("could not allocate other bvds\n");
1152 free(vcl);
1153 return 1;
1154 };
1155 super->conflist = vcl;
1156 dl->vlist[vnum++] = vcl;
1157 }
1158 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
1159 for (i=0; i < max_virt_disks ; i++)
1160 if (memcmp(super->virt->entries[i].guid,
1161 vcl->conf.guid, DDF_GUID_LEN)==0)
1162 break;
1163 if (i < max_virt_disks)
1164 vcl->vcnum = i;
1165 }
1166
1167 return 0;
1168 }
1169
1170 static int load_super_ddf(struct supertype *st, int fd,
1171 char *devname)
1172 {
1173 unsigned long long dsize;
1174 struct ddf_super *super;
1175 int rv;
1176
1177 if (get_dev_size(fd, devname, &dsize) == 0)
1178 return 1;
1179
1180 if (test_partition(fd))
1181 /* DDF is not allowed on partitions */
1182 return 1;
1183
1184 /* 32M is a lower bound */
1185 if (dsize <= 32*1024*1024) {
1186 if (devname)
1187 pr_err("%s is too small for ddf: size is %llu sectors.\n",
1188 devname, dsize>>9);
1189 return 1;
1190 }
1191 if (dsize & 511) {
1192 if (devname)
1193 pr_err("%s is an odd size for ddf: size is %llu bytes.\n",
1194 devname, dsize);
1195 return 1;
1196 }
1197
1198 free_super_ddf(st);
1199
1200 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
1201 pr_err("malloc of %zu failed.\n",
1202 sizeof(*super));
1203 return 1;
1204 }
1205 memset(super, 0, sizeof(*super));
1206
1207 rv = load_ddf_headers(fd, super, devname);
1208 if (rv) {
1209 free(super);
1210 return rv;
1211 }
1212
1213 /* Have valid headers and have chosen the best. Let's read in the rest*/
1214
1215 rv = load_ddf_global(fd, super, devname);
1216
1217 if (rv) {
1218 if (devname)
1219 pr_err("Failed to load all information sections on %s\n", devname);
1220 free(super);
1221 return rv;
1222 }
1223
1224 rv = load_ddf_local(fd, super, devname, 0);
1225
1226 if (rv) {
1227 if (devname)
1228 pr_err("Failed to load all information sections on %s\n", devname);
1229 free(super);
1230 return rv;
1231 }
1232
1233 /* Should possibly check the sections .... */
1234
1235 st->sb = super;
1236 if (st->ss == NULL) {
1237 st->ss = &super_ddf;
1238 st->minor_version = 0;
1239 st->max_devs = 512;
1240 }
1241 return 0;
1242
1243 }
1244
1245 static void free_super_ddf(struct supertype *st)
1246 {
1247 struct ddf_super *ddf = st->sb;
1248 if (ddf == NULL)
1249 return;
1250 free(ddf->phys);
1251 free(ddf->virt);
1252 free(ddf->conf);
1253 while (ddf->conflist) {
1254 struct vcl *v = ddf->conflist;
1255 ddf->conflist = v->next;
1256 if (v->block_sizes)
1257 free(v->block_sizes);
1258 if (v->other_bvds)
1259 /*
1260 v->other_bvds[0] points to beginning of buffer,
1261 see alloc_other_bvds()
1262 */
1263 free(v->other_bvds[0]);
1264 free(v);
1265 }
1266 while (ddf->dlist) {
1267 struct dl *d = ddf->dlist;
1268 ddf->dlist = d->next;
1269 if (d->fd >= 0)
1270 close(d->fd);
1271 if (d->spare)
1272 free(d->spare);
1273 free(d);
1274 }
1275 while (ddf->add_list) {
1276 struct dl *d = ddf->add_list;
1277 ddf->add_list = d->next;
1278 if (d->fd >= 0)
1279 close(d->fd);
1280 if (d->spare)
1281 free(d->spare);
1282 free(d);
1283 }
1284 free(ddf);
1285 st->sb = NULL;
1286 }
1287
1288 static struct supertype *match_metadata_desc_ddf(char *arg)
1289 {
1290 /* 'ddf' only supports containers */
1291 struct supertype *st;
1292 if (strcmp(arg, "ddf") != 0 &&
1293 strcmp(arg, "default") != 0
1294 )
1295 return NULL;
1296
1297 st = xcalloc(1, sizeof(*st));
1298 st->ss = &super_ddf;
1299 st->max_devs = 512;
1300 st->minor_version = 0;
1301 st->sb = NULL;
1302 return st;
1303 }
1304
1305 static mapping_t ddf_state[] = {
1306 { "Optimal", 0},
1307 { "Degraded", 1},
1308 { "Deleted", 2},
1309 { "Missing", 3},
1310 { "Failed", 4},
1311 { "Partially Optimal", 5},
1312 { "-reserved-", 6},
1313 { "-reserved-", 7},
1314 { NULL, 0}
1315 };
1316
1317 static mapping_t ddf_init_state[] = {
1318 { "Not Initialised", 0},
1319 { "QuickInit in Progress", 1},
1320 { "Fully Initialised", 2},
1321 { "*UNKNOWN*", 3},
1322 { NULL, 0}
1323 };
1324 static mapping_t ddf_access[] = {
1325 { "Read/Write", 0},
1326 { "Reserved", 1},
1327 { "Read Only", 2},
1328 { "Blocked (no access)", 3},
1329 { NULL ,0}
1330 };
1331
1332 static mapping_t ddf_level[] = {
1333 { "RAID0", DDF_RAID0},
1334 { "RAID1", DDF_RAID1},
1335 { "RAID3", DDF_RAID3},
1336 { "RAID4", DDF_RAID4},
1337 { "RAID5", DDF_RAID5},
1338 { "RAID1E",DDF_RAID1E},
1339 { "JBOD", DDF_JBOD},
1340 { "CONCAT",DDF_CONCAT},
1341 { "RAID5E",DDF_RAID5E},
1342 { "RAID5EE",DDF_RAID5EE},
1343 { "RAID6", DDF_RAID6},
1344 { NULL, 0}
1345 };
1346 static mapping_t ddf_sec_level[] = {
1347 { "Striped", DDF_2STRIPED},
1348 { "Mirrored", DDF_2MIRRORED},
1349 { "Concat", DDF_2CONCAT},
1350 { "Spanned", DDF_2SPANNED},
1351 { NULL, 0}
1352 };
1353
1354 static int all_ff(const char *guid)
1355 {
1356 int i;
1357 for (i = 0; i < DDF_GUID_LEN; i++)
1358 if (guid[i] != (char)0xff)
1359 return 0;
1360 return 1;
1361 }
1362
1363 static const char *guid_str(const char *guid)
1364 {
1365 static char buf[DDF_GUID_LEN*2+1];
1366 int i;
1367 char *p = buf;
1368 for (i = 0; i < DDF_GUID_LEN; i++) {
1369 unsigned char c = guid[i];
1370 if (c >= 32 && c < 127)
1371 p += sprintf(p, "%c", c);
1372 else
1373 p += sprintf(p, "%02x", c);
1374 }
1375 *p = '\0';
1376 return (const char *) buf;
1377 }
1378
1379 static void print_guid(char *guid, int tstamp)
1380 {
1381 /* A GUIDs are part (or all) ASCII and part binary.
1382 * They tend to be space padded.
1383 * We print the GUID in HEX, then in parentheses add
1384 * any initial ASCII sequence, and a possible
1385 * time stamp from bytes 16-19
1386 */
1387 int l = DDF_GUID_LEN;
1388 int i;
1389
1390 for (i=0 ; i<DDF_GUID_LEN ; i++) {
1391 if ((i&3)==0 && i != 0) printf(":");
1392 printf("%02X", guid[i]&255);
1393 }
1394
1395 printf("\n (");
1396 while (l && guid[l-1] == ' ')
1397 l--;
1398 for (i=0 ; i<l ; i++) {
1399 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1400 fputc(guid[i], stdout);
1401 else
1402 break;
1403 }
1404 if (tstamp) {
1405 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1406 char tbuf[100];
1407 struct tm *tm;
1408 tm = localtime(&then);
1409 strftime(tbuf, 100, " %D %T",tm);
1410 fputs(tbuf, stdout);
1411 }
1412 printf(")");
1413 }
1414
1415 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1416 {
1417 int crl = sb->conf_rec_len;
1418 struct vcl *vcl;
1419
1420 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1421 unsigned int i;
1422 struct vd_config *vc = &vcl->conf;
1423
1424 if (!be32_eq(calc_crc(vc, crl*512), vc->crc))
1425 continue;
1426 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1427 continue;
1428
1429 /* Ok, we know about this VD, let's give more details */
1430 printf(" Raid Devices[%d] : %d (", n,
1431 be16_to_cpu(vc->prim_elmnt_count));
1432 for (i = 0; i < be16_to_cpu(vc->prim_elmnt_count); i++) {
1433 int j;
1434 int cnt = be16_to_cpu(sb->phys->max_pdes);
1435 for (j=0; j<cnt; j++)
1436 if (be32_eq(vc->phys_refnum[i],
1437 sb->phys->entries[j].refnum))
1438 break;
1439 if (i) printf(" ");
1440 if (j < cnt)
1441 printf("%d", j);
1442 else
1443 printf("--");
1444 printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb, vc)[i])/2);
1445 }
1446 printf(")\n");
1447 if (vc->chunk_shift != 255)
1448 printf(" Chunk Size[%d] : %d sectors\n", n,
1449 1 << vc->chunk_shift);
1450 printf(" Raid Level[%d] : %s\n", n,
1451 map_num(ddf_level, vc->prl)?:"-unknown-");
1452 if (vc->sec_elmnt_count != 1) {
1453 printf(" Secondary Position[%d] : %d of %d\n", n,
1454 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1455 printf(" Secondary Level[%d] : %s\n", n,
1456 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1457 }
1458 printf(" Device Size[%d] : %llu\n", n,
1459 be64_to_cpu(vc->blocks)/2);
1460 printf(" Array Size[%d] : %llu\n", n,
1461 be64_to_cpu(vc->array_blocks)/2);
1462 }
1463 }
1464
1465 static void examine_vds(struct ddf_super *sb)
1466 {
1467 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1468 unsigned int i;
1469 printf(" Virtual Disks : %d\n", cnt);
1470
1471 for (i = 0; i < be16_to_cpu(sb->virt->max_vdes); i++) {
1472 struct virtual_entry *ve = &sb->virt->entries[i];
1473 if (all_ff(ve->guid))
1474 continue;
1475 printf("\n");
1476 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1477 printf("\n");
1478 printf(" unit[%d] : %d\n", i, be16_to_cpu(ve->unit));
1479 printf(" state[%d] : %s, %s%s\n", i,
1480 map_num(ddf_state, ve->state & 7),
1481 (ve->state & DDF_state_morphing) ? "Morphing, ": "",
1482 (ve->state & DDF_state_inconsistent)? "Not Consistent" : "Consistent");
1483 printf(" init state[%d] : %s\n", i,
1484 map_num(ddf_init_state, ve->init_state&DDF_initstate_mask));
1485 printf(" access[%d] : %s\n", i,
1486 map_num(ddf_access, (ve->init_state & DDF_access_mask) >> 6));
1487 printf(" Name[%d] : %.16s\n", i, ve->name);
1488 examine_vd(i, sb, ve->guid);
1489 }
1490 if (cnt) printf("\n");
1491 }
1492
1493 static void examine_pds(struct ddf_super *sb)
1494 {
1495 int cnt = be16_to_cpu(sb->phys->max_pdes);
1496 int i;
1497 struct dl *dl;
1498 int unlisted = 0;
1499 printf(" Physical Disks : %d\n", cnt);
1500 printf(" Number RefNo Size Device Type/State\n");
1501
1502 for (dl = sb->dlist; dl; dl = dl->next)
1503 dl->displayed = 0;
1504
1505 for (i=0 ; i<cnt ; i++) {
1506 struct phys_disk_entry *pd = &sb->phys->entries[i];
1507 int type = be16_to_cpu(pd->type);
1508 int state = be16_to_cpu(pd->state);
1509
1510 if (be32_to_cpu(pd->refnum) == 0xffffffff)
1511 /* Not in use */
1512 continue;
1513 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1514 //printf("\n");
1515 printf(" %3d %08x ", i,
1516 be32_to_cpu(pd->refnum));
1517 printf("%8lluK ",
1518 be64_to_cpu(pd->config_size)>>1);
1519 for (dl = sb->dlist; dl ; dl = dl->next) {
1520 if (be32_eq(dl->disk.refnum, pd->refnum)) {
1521 char *dv = map_dev(dl->major, dl->minor, 0);
1522 if (dv) {
1523 printf("%-15s", dv);
1524 break;
1525 }
1526 }
1527 }
1528 if (!dl)
1529 printf("%15s","");
1530 else
1531 dl->displayed = 1;
1532 printf(" %s%s%s%s%s",
1533 (type&2) ? "active":"",
1534 (type&4) ? "Global-Spare":"",
1535 (type&8) ? "spare" : "",
1536 (type&16)? ", foreign" : "",
1537 (type&32)? "pass-through" : "");
1538 if (state & DDF_Failed)
1539 /* This over-rides these three */
1540 state &= ~(DDF_Online|DDF_Rebuilding|DDF_Transition);
1541 printf("/%s%s%s%s%s%s%s",
1542 (state&1)? "Online": "Offline",
1543 (state&2)? ", Failed": "",
1544 (state&4)? ", Rebuilding": "",
1545 (state&8)? ", in-transition": "",
1546 (state&16)? ", SMART-errors": "",
1547 (state&32)? ", Unrecovered-Read-Errors": "",
1548 (state&64)? ", Missing" : "");
1549 printf("\n");
1550 }
1551 for (dl = sb->dlist; dl; dl = dl->next) {
1552 char *dv;
1553 if (dl->displayed)
1554 continue;
1555 if (!unlisted)
1556 printf(" Physical disks not in metadata!:\n");
1557 unlisted = 1;
1558 dv = map_dev(dl->major, dl->minor, 0);
1559 printf(" %08x %s\n", be32_to_cpu(dl->disk.refnum),
1560 dv ? dv : "-unknown-");
1561 }
1562 if (unlisted)
1563 printf("\n");
1564 }
1565
1566 static void examine_super_ddf(struct supertype *st, char *homehost)
1567 {
1568 struct ddf_super *sb = st->sb;
1569
1570 printf(" Magic : %08x\n", be32_to_cpu(sb->anchor.magic));
1571 printf(" Version : %.8s\n", sb->anchor.revision);
1572 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1573 printf("\n");
1574 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1575 printf("\n");
1576 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1577 printf(" Redundant hdr : %s\n", (be32_eq(sb->secondary.magic,
1578 DDF_HEADER_MAGIC)
1579 ?"yes" : "no"));
1580 examine_vds(sb);
1581 examine_pds(sb);
1582 }
1583
1584 static unsigned int get_vd_num_of_subarray(struct supertype *st)
1585 {
1586 /*
1587 * Figure out the VD number for this supertype.
1588 * Returns DDF_CONTAINER for the container itself,
1589 * and DDF_NOTFOUND on error.
1590 */
1591 struct ddf_super *ddf = st->sb;
1592 struct mdinfo *sra;
1593 char *sub, *end;
1594 unsigned int vcnum;
1595
1596 if (*st->container_devnm == '\0')
1597 return DDF_CONTAINER;
1598
1599 sra = sysfs_read(-1, st->devnm, GET_VERSION);
1600 if (!sra || sra->array.major_version != -1 ||
1601 sra->array.minor_version != -2 ||
1602 !is_subarray(sra->text_version))
1603 return DDF_NOTFOUND;
1604
1605 sub = strchr(sra->text_version + 1, '/');
1606 if (sub != NULL)
1607 vcnum = strtoul(sub + 1, &end, 10);
1608 if (sub == NULL || *sub == '\0' || *end != '\0' ||
1609 vcnum >= be16_to_cpu(ddf->active->max_vd_entries))
1610 return DDF_NOTFOUND;
1611
1612 return vcnum;
1613 }
1614
1615 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1616 {
1617 /* We just write a generic DDF ARRAY entry
1618 */
1619 struct mdinfo info;
1620 char nbuf[64];
1621 getinfo_super_ddf(st, &info, NULL);
1622 fname_from_uuid(st, &info, nbuf, ':');
1623
1624 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1625 }
1626
1627 static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1628 {
1629 /* We write a DDF ARRAY member entry for each vd, identifying container
1630 * by uuid and member by unit number and uuid.
1631 */
1632 struct ddf_super *ddf = st->sb;
1633 struct mdinfo info;
1634 unsigned int i;
1635 char nbuf[64];
1636 getinfo_super_ddf(st, &info, NULL);
1637 fname_from_uuid(st, &info, nbuf, ':');
1638
1639 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
1640 struct virtual_entry *ve = &ddf->virt->entries[i];
1641 struct vcl vcl;
1642 char nbuf1[64];
1643 char namebuf[17];
1644 if (all_ff(ve->guid))
1645 continue;
1646 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1647 ddf->currentconf =&vcl;
1648 vcl.vcnum = i;
1649 uuid_from_super_ddf(st, info.uuid);
1650 fname_from_uuid(st, &info, nbuf1, ':');
1651 _ddf_array_name(namebuf, ddf, i);
1652 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1653 namebuf[0] == '\0' ? "" : " /dev/md/", namebuf,
1654 nbuf+5, i, nbuf1+5);
1655 }
1656 }
1657
1658 static void export_examine_super_ddf(struct supertype *st)
1659 {
1660 struct mdinfo info;
1661 char nbuf[64];
1662 getinfo_super_ddf(st, &info, NULL);
1663 fname_from_uuid(st, &info, nbuf, ':');
1664 printf("MD_METADATA=ddf\n");
1665 printf("MD_LEVEL=container\n");
1666 printf("MD_UUID=%s\n", nbuf+5);
1667 printf("MD_DEVICES=%u\n",
1668 be16_to_cpu(((struct ddf_super *)st->sb)->phys->used_pdes));
1669 }
1670
1671 static int copy_metadata_ddf(struct supertype *st, int from, int to)
1672 {
1673 void *buf;
1674 unsigned long long dsize, offset;
1675 int bytes;
1676 struct ddf_header *ddf;
1677 int written = 0;
1678
1679 /* The meta consists of an anchor, a primary, and a secondary.
1680 * This all lives at the end of the device.
1681 * So it is easiest to find the earliest of primary and
1682 * secondary, and copy everything from there.
1683 *
1684 * Anchor is 512 from end. It contains primary_lba and secondary_lba
1685 * we choose one of those
1686 */
1687
1688 if (posix_memalign(&buf, 4096, 4096) != 0)
1689 return 1;
1690
1691 if (!get_dev_size(from, NULL, &dsize))
1692 goto err;
1693
1694 if (lseek64(from, dsize-512, 0) < 0)
1695 goto err;
1696 if (read(from, buf, 512) != 512)
1697 goto err;
1698 ddf = buf;
1699 if (!be32_eq(ddf->magic, DDF_HEADER_MAGIC) ||
1700 !be32_eq(calc_crc(ddf, 512), ddf->crc) ||
1701 (memcmp(ddf->revision, DDF_REVISION_0, 8) != 0 &&
1702 memcmp(ddf->revision, DDF_REVISION_2, 8) != 0))
1703 goto err;
1704
1705 offset = dsize - 512;
1706 if ((be64_to_cpu(ddf->primary_lba) << 9) < offset)
1707 offset = be64_to_cpu(ddf->primary_lba) << 9;
1708 if ((be64_to_cpu(ddf->secondary_lba) << 9) < offset)
1709 offset = be64_to_cpu(ddf->secondary_lba) << 9;
1710
1711 bytes = dsize - offset;
1712
1713 if (lseek64(from, offset, 0) < 0 ||
1714 lseek64(to, offset, 0) < 0)
1715 goto err;
1716 while (written < bytes) {
1717 int n = bytes - written;
1718 if (n > 4096)
1719 n = 4096;
1720 if (read(from, buf, n) != n)
1721 goto err;
1722 if (write(to, buf, n) != n)
1723 goto err;
1724 written += n;
1725 }
1726 free(buf);
1727 return 0;
1728 err:
1729 free(buf);
1730 return 1;
1731 }
1732
1733 static void detail_super_ddf(struct supertype *st, char *homehost,
1734 char *subarray)
1735 {
1736 struct ddf_super *sb = st->sb;
1737 int cnt = be16_to_cpu(sb->virt->populated_vdes);
1738
1739 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1740 printf("\n");
1741 printf(" Seq : %08x\n", be32_to_cpu(sb->active->seq));
1742 printf(" Virtual Disks : %d\n", cnt);
1743 printf("\n");
1744 }
1745
1746 static const char *vendors_with_variable_volume_UUID[] = {
1747 "LSI ",
1748 };
1749
1750 static int volume_id_is_reliable(const struct ddf_super *ddf)
1751 {
1752 int n = ARRAY_SIZE(vendors_with_variable_volume_UUID);
1753 int i;
1754 for (i = 0; i < n; i++)
1755 if (!memcmp(ddf->controller.guid,
1756 vendors_with_variable_volume_UUID[i], 8))
1757 return 0;
1758 return 1;
1759 }
1760
1761 static void uuid_of_ddf_subarray(const struct ddf_super *ddf,
1762 unsigned int vcnum, int uuid[4])
1763 {
1764 char buf[DDF_GUID_LEN+18], sha[20], *p;
1765 struct sha1_ctx ctx;
1766 if (volume_id_is_reliable(ddf)) {
1767 uuid_from_ddf_guid(ddf->virt->entries[vcnum].guid, uuid);
1768 return;
1769 }
1770 /*
1771 * Some fake RAID BIOSes (in particular, LSI ones) change the
1772 * VD GUID at every boot. These GUIDs are not suitable for
1773 * identifying an array. Luckily the header GUID appears to
1774 * remain constant.
1775 * We construct a pseudo-UUID from the header GUID and those
1776 * properties of the subarray that we expect to remain constant.
1777 */
1778 memset(buf, 0, sizeof(buf));
1779 p = buf;
1780 memcpy(p, ddf->anchor.guid, DDF_GUID_LEN);
1781 p += DDF_GUID_LEN;
1782 memcpy(p, ddf->virt->entries[vcnum].name, 16);
1783 p += 16;
1784 *((__u16 *) p) = vcnum;
1785 sha1_init_ctx(&ctx);
1786 sha1_process_bytes(buf, sizeof(buf), &ctx);
1787 sha1_finish_ctx(&ctx, sha);
1788 memcpy(uuid, sha, 4*4);
1789 }
1790
1791 static void brief_detail_super_ddf(struct supertype *st, char *subarray)
1792 {
1793 struct mdinfo info;
1794 char nbuf[64];
1795 struct ddf_super *ddf = st->sb;
1796 unsigned int vcnum = get_vd_num_of_subarray(st);
1797 if (vcnum == DDF_CONTAINER)
1798 uuid_from_super_ddf(st, info.uuid);
1799 else if (vcnum == DDF_NOTFOUND)
1800 return;
1801 else
1802 uuid_of_ddf_subarray(ddf, vcnum, info.uuid);
1803 fname_from_uuid(st, &info, nbuf,':');
1804 printf(" UUID=%s", nbuf + 5);
1805 }
1806
1807 static int match_home_ddf(struct supertype *st, char *homehost)
1808 {
1809 /* It matches 'this' host if the controller is a
1810 * Linux-MD controller with vendor_data matching
1811 * the hostname. It would be nice if we could
1812 * test against controller found in /sys or somewhere...
1813 */
1814 struct ddf_super *ddf = st->sb;
1815 unsigned int len;
1816
1817 if (!homehost)
1818 return 0;
1819 len = strlen(homehost);
1820
1821 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1822 len < sizeof(ddf->controller.vendor_data) &&
1823 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1824 ddf->controller.vendor_data[len] == 0);
1825 }
1826
1827 static int find_index_in_bvd(const struct ddf_super *ddf,
1828 const struct vd_config *conf, unsigned int n,
1829 unsigned int *n_bvd)
1830 {
1831 /*
1832 * Find the index of the n-th valid physical disk in this BVD.
1833 * Unused entries can be sprinkled in with the used entries,
1834 * but don't count.
1835 */
1836 unsigned int i, j;
1837 for (i = 0, j = 0;
1838 i < ddf->mppe && j < be16_to_cpu(conf->prim_elmnt_count);
1839 i++) {
1840 if (be32_to_cpu(conf->phys_refnum[i]) != 0xffffffff) {
1841 if (n == j) {
1842 *n_bvd = i;
1843 return 1;
1844 }
1845 j++;
1846 }
1847 }
1848 dprintf("couldn't find BVD member %u (total %u)\n",
1849 n, be16_to_cpu(conf->prim_elmnt_count));
1850 return 0;
1851 }
1852
1853 /* Given a member array instance number, and a raid disk within that instance,
1854 * find the vd_config structure. The offset of the given disk in the phys_refnum
1855 * table is returned in n_bvd.
1856 * For two-level members with a secondary raid level the vd_config for
1857 * the appropriate BVD is returned.
1858 * The return value is always &vlc->conf, where vlc is returned in last pointer.
1859 */
1860 static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst,
1861 unsigned int n,
1862 unsigned int *n_bvd, struct vcl **vcl)
1863 {
1864 struct vcl *v;
1865
1866 for (v = ddf->conflist; v; v = v->next) {
1867 unsigned int nsec, ibvd = 0;
1868 struct vd_config *conf;
1869 if (inst != v->vcnum)
1870 continue;
1871 conf = &v->conf;
1872 if (conf->sec_elmnt_count == 1) {
1873 if (find_index_in_bvd(ddf, conf, n, n_bvd)) {
1874 *vcl = v;
1875 return conf;
1876 } else
1877 goto bad;
1878 }
1879 if (v->other_bvds == NULL) {
1880 pr_err("BUG: other_bvds is NULL, nsec=%u\n",
1881 conf->sec_elmnt_count);
1882 goto bad;
1883 }
1884 nsec = n / be16_to_cpu(conf->prim_elmnt_count);
1885 if (conf->sec_elmnt_seq != nsec) {
1886 for (ibvd = 1; ibvd < conf->sec_elmnt_count; ibvd++) {
1887 if (v->other_bvds[ibvd-1]->sec_elmnt_seq ==
1888 nsec)
1889 break;
1890 }
1891 if (ibvd == conf->sec_elmnt_count)
1892 goto bad;
1893 conf = v->other_bvds[ibvd-1];
1894 }
1895 if (!find_index_in_bvd(ddf, conf,
1896 n - nsec*conf->sec_elmnt_count, n_bvd))
1897 goto bad;
1898 dprintf("found disk %u as member %u in bvd %d of array %u\n",
1899 n, *n_bvd, ibvd, inst);
1900 *vcl = v;
1901 return conf;
1902 }
1903 bad:
1904 pr_err("Couldn't find disk %d in array %u\n", n, inst);
1905 return NULL;
1906 }
1907
1908 static int find_phys(const struct ddf_super *ddf, be32 phys_refnum)
1909 {
1910 /* Find the entry in phys_disk which has the given refnum
1911 * and return it's index
1912 */
1913 unsigned int i;
1914 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++)
1915 if (be32_eq(ddf->phys->entries[i].refnum, phys_refnum))
1916 return i;
1917 return -1;
1918 }
1919
1920 static void uuid_from_ddf_guid(const char *guid, int uuid[4])
1921 {
1922 char buf[20];
1923 struct sha1_ctx ctx;
1924 sha1_init_ctx(&ctx);
1925 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1926 sha1_finish_ctx(&ctx, buf);
1927 memcpy(uuid, buf, 4*4);
1928 }
1929
1930 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1931 {
1932 /* The uuid returned here is used for:
1933 * uuid to put into bitmap file (Create, Grow)
1934 * uuid for backup header when saving critical section (Grow)
1935 * comparing uuids when re-adding a device into an array
1936 * In these cases the uuid required is that of the data-array,
1937 * not the device-set.
1938 * uuid to recognise same set when adding a missing device back
1939 * to an array. This is a uuid for the device-set.
1940 *
1941 * For each of these we can make do with a truncated
1942 * or hashed uuid rather than the original, as long as
1943 * everyone agrees.
1944 * In the case of SVD we assume the BVD is of interest,
1945 * though that might be the case if a bitmap were made for
1946 * a mirrored SVD - worry about that later.
1947 * So we need to find the VD configuration record for the
1948 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1949 * The first 16 bytes of the sha1 of these is used.
1950 */
1951 struct ddf_super *ddf = st->sb;
1952 struct vcl *vcl = ddf->currentconf;
1953
1954 if (vcl)
1955 uuid_of_ddf_subarray(ddf, vcl->vcnum, uuid);
1956 else
1957 uuid_from_ddf_guid(ddf->anchor.guid, uuid);
1958 }
1959
1960 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info, char *map)
1961 {
1962 struct ddf_super *ddf = st->sb;
1963 int map_disks = info->array.raid_disks;
1964 __u32 *cptr;
1965
1966 if (ddf->currentconf) {
1967 getinfo_super_ddf_bvd(st, info, map);
1968 return;
1969 }
1970 memset(info, 0, sizeof(*info));
1971
1972 info->array.raid_disks = be16_to_cpu(ddf->phys->used_pdes);
1973 info->array.level = LEVEL_CONTAINER;
1974 info->array.layout = 0;
1975 info->array.md_minor = -1;
1976 cptr = (__u32 *)(ddf->anchor.guid + 16);
1977 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
1978
1979 info->array.chunk_size = 0;
1980 info->container_enough = 1;
1981
1982 info->disk.major = 0;
1983 info->disk.minor = 0;
1984 if (ddf->dlist) {
1985 struct phys_disk_entry *pde = NULL;
1986 info->disk.number = be32_to_cpu(ddf->dlist->disk.refnum);
1987 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1988
1989 info->data_offset = be64_to_cpu(ddf->phys->
1990 entries[info->disk.raid_disk].
1991 config_size);
1992 info->component_size = ddf->dlist->size - info->data_offset;
1993 if (info->disk.raid_disk >= 0)
1994 pde = ddf->phys->entries + info->disk.raid_disk;
1995 if (pde &&
1996 !(be16_to_cpu(pde->state) & DDF_Failed) &&
1997 !(be16_to_cpu(pde->state) & DDF_Missing))
1998 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
1999 else
2000 info->disk.state = 1 << MD_DISK_FAULTY;
2001
2002 } else {
2003 /* There should always be a dlist, but just in case...*/
2004 info->disk.number = -1;
2005 info->disk.raid_disk = -1;
2006 info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
2007 }
2008 info->events = be32_to_cpu(ddf->active->seq);
2009 info->array.utime = DECADE + be32_to_cpu(ddf->active->timestamp);
2010
2011 info->recovery_start = MaxSector;
2012 info->reshape_active = 0;
2013 info->recovery_blocked = 0;
2014 info->name[0] = 0;
2015
2016 info->array.major_version = -1;
2017 info->array.minor_version = -2;
2018 strcpy(info->text_version, "ddf");
2019 info->safe_mode_delay = 0;
2020
2021 uuid_from_super_ddf(st, info->uuid);
2022
2023 if (map) {
2024 int i, e = 0;
2025 int max = be16_to_cpu(ddf->phys->max_pdes);
2026 for (i = e = 0 ; i < map_disks ; i++, e++) {
2027 while (e < max &&
2028 be32_to_cpu(ddf->phys->entries[e].refnum) == 0xffffffff)
2029 e++;
2030 if (i < info->array.raid_disks && e < max &&
2031 !(be16_to_cpu(ddf->phys->entries[e].state) &
2032 DDF_Failed))
2033 map[i] = 1;
2034 else
2035 map[i] = 0;
2036 }
2037 }
2038 }
2039
2040 /* size of name must be at least 17 bytes! */
2041 static void _ddf_array_name(char *name, const struct ddf_super *ddf, int i)
2042 {
2043 int j;
2044 memcpy(name, ddf->virt->entries[i].name, 16);
2045 name[16] = 0;
2046 for(j = 0; j < 16; j++)
2047 if (name[j] == ' ')
2048 name[j] = 0;
2049 }
2050
2051 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info, char *map)
2052 {
2053 struct ddf_super *ddf = st->sb;
2054 struct vcl *vc = ddf->currentconf;
2055 int cd = ddf->currentdev;
2056 int n_prim;
2057 int j;
2058 struct dl *dl = NULL;
2059 int map_disks = info->array.raid_disks;
2060 __u32 *cptr;
2061 struct vd_config *conf;
2062
2063 memset(info, 0, sizeof(*info));
2064 if (layout_ddf2md(&vc->conf, &info->array) == -1)
2065 return;
2066 info->array.md_minor = -1;
2067 cptr = (__u32 *)(vc->conf.guid + 16);
2068 info->array.ctime = DECADE + __be32_to_cpu(*cptr);
2069 info->array.utime = DECADE + be32_to_cpu(vc->conf.timestamp);
2070 info->array.chunk_size = 512 << vc->conf.chunk_shift;
2071 info->custom_array_size = be64_to_cpu(vc->conf.array_blocks);
2072
2073 conf = &vc->conf;
2074 n_prim = be16_to_cpu(conf->prim_elmnt_count);
2075 if (conf->sec_elmnt_count > 1 && cd >= n_prim) {
2076 int ibvd = cd / n_prim - 1;
2077 cd %= n_prim;
2078 conf = vc->other_bvds[ibvd];
2079 }
2080
2081 if (cd >= 0 && (unsigned)cd < ddf->mppe) {
2082 info->data_offset =
2083 be64_to_cpu(LBA_OFFSET(ddf, conf)[cd]);
2084 if (vc->block_sizes)
2085 info->component_size = vc->block_sizes[cd];
2086 else
2087 info->component_size = be64_to_cpu(conf->blocks);
2088
2089 for (dl = ddf->dlist; dl ; dl = dl->next)
2090 if (be32_eq(dl->disk.refnum, conf->phys_refnum[cd]))
2091 break;
2092 }
2093
2094 info->disk.major = 0;
2095 info->disk.minor = 0;
2096 info->disk.state = 0;
2097 if (dl && dl->pdnum >= 0) {
2098 info->disk.major = dl->major;
2099 info->disk.minor = dl->minor;
2100 info->disk.raid_disk = cd + conf->sec_elmnt_seq
2101 * be16_to_cpu(conf->prim_elmnt_count);
2102 info->disk.number = dl->pdnum;
2103 info->disk.state = 0;
2104 if (info->disk.number >= 0 &&
2105 (be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Online) &&
2106 !(be16_to_cpu(ddf->phys->entries[info->disk.number].state) & DDF_Failed))
2107 info->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2108 info->events = be32_to_cpu(ddf->active->seq);
2109 }
2110
2111 info->container_member = ddf->currentconf->vcnum;
2112
2113 info->recovery_start = MaxSector;
2114 info->resync_start = 0;
2115 info->reshape_active = 0;
2116 info->recovery_blocked = 0;
2117 if (!(ddf->virt->entries[info->container_member].state &
2118 DDF_state_inconsistent) &&
2119 (ddf->virt->entries[info->container_member].init_state &
2120 DDF_initstate_mask) == DDF_init_full)
2121 info->resync_start = MaxSector;
2122
2123 uuid_from_super_ddf(st, info->uuid);
2124
2125 info->array.major_version = -1;
2126 info->array.minor_version = -2;
2127 sprintf(info->text_version, "/%s/%d",
2128 st->container_devnm,
2129 info->container_member);
2130 info->safe_mode_delay = DDF_SAFE_MODE_DELAY;
2131
2132 _ddf_array_name(info->name, ddf, info->container_member);
2133
2134 if (map)
2135 for (j = 0; j < map_disks; j++) {
2136 map[j] = 0;
2137 if (j < info->array.raid_disks) {
2138 int i = find_phys(ddf, vc->conf.phys_refnum[j]);
2139 if (i >= 0 &&
2140 (be16_to_cpu(ddf->phys->entries[i].state)
2141 & DDF_Online) &&
2142 !(be16_to_cpu(ddf->phys->entries[i].state)
2143 & DDF_Failed))
2144 map[i] = 1;
2145 }
2146 }
2147 }
2148
2149 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
2150 char *update,
2151 char *devname, int verbose,
2152 int uuid_set, char *homehost)
2153 {
2154 /* For 'assemble' and 'force' we need to return non-zero if any
2155 * change was made. For others, the return value is ignored.
2156 * Update options are:
2157 * force-one : This device looks a bit old but needs to be included,
2158 * update age info appropriately.
2159 * assemble: clear any 'faulty' flag to allow this device to
2160 * be assembled.
2161 * force-array: Array is degraded but being forced, mark it clean
2162 * if that will be needed to assemble it.
2163 *
2164 * newdev: not used ????
2165 * grow: Array has gained a new device - this is currently for
2166 * linear only
2167 * resync: mark as dirty so a resync will happen.
2168 * uuid: Change the uuid of the array to match what is given
2169 * homehost: update the recorded homehost
2170 * name: update the name - preserving the homehost
2171 * _reshape_progress: record new reshape_progress position.
2172 *
2173 * Following are not relevant for this version:
2174 * sparc2.2 : update from old dodgey metadata
2175 * super-minor: change the preferred_minor number
2176 * summaries: update redundant counters.
2177 */
2178 int rv = 0;
2179 // struct ddf_super *ddf = st->sb;
2180 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2181 // struct virtual_entry *ve = find_ve(ddf);
2182
2183 /* we don't need to handle "force-*" or "assemble" as
2184 * there is no need to 'trick' the kernel. When the metadata is
2185 * first updated to activate the array, all the implied modifications
2186 * will just happen.
2187 */
2188
2189 if (strcmp(update, "grow") == 0) {
2190 /* FIXME */
2191 } else if (strcmp(update, "resync") == 0) {
2192 // info->resync_checkpoint = 0;
2193 } else if (strcmp(update, "homehost") == 0) {
2194 /* homehost is stored in controller->vendor_data,
2195 * or it is when we are the vendor
2196 */
2197 // if (info->vendor_is_local)
2198 // strcpy(ddf->controller.vendor_data, homehost);
2199 rv = -1;
2200 } else if (strcmp(update, "name") == 0) {
2201 /* name is stored in virtual_entry->name */
2202 // memset(ve->name, ' ', 16);
2203 // strncpy(ve->name, info->name, 16);
2204 rv = -1;
2205 } else if (strcmp(update, "_reshape_progress") == 0) {
2206 /* We don't support reshape yet */
2207 } else if (strcmp(update, "assemble") == 0 ) {
2208 /* Do nothing, just succeed */
2209 rv = 0;
2210 } else
2211 rv = -1;
2212
2213 // update_all_csum(ddf);
2214
2215 return rv;
2216 }
2217
2218 static void make_header_guid(char *guid)
2219 {
2220 be32 stamp;
2221 /* Create a DDF Header of Virtual Disk GUID */
2222
2223 /* 24 bytes of fiction required.
2224 * first 8 are a 'vendor-id' - "Linux-MD"
2225 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2226 * Remaining 8 random number plus timestamp
2227 */
2228 memcpy(guid, T10, sizeof(T10));
2229 stamp = cpu_to_be32(0xdeadbeef);
2230 memcpy(guid+8, &stamp, 4);
2231 stamp = cpu_to_be32(0);
2232 memcpy(guid+12, &stamp, 4);
2233 stamp = cpu_to_be32(time(0) - DECADE);
2234 memcpy(guid+16, &stamp, 4);
2235 stamp._v32 = random32();
2236 memcpy(guid+20, &stamp, 4);
2237 }
2238
2239 static unsigned int find_unused_vde(const struct ddf_super *ddf)
2240 {
2241 unsigned int i;
2242 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2243 if (all_ff(ddf->virt->entries[i].guid))
2244 return i;
2245 }
2246 return DDF_NOTFOUND;
2247 }
2248
2249 static unsigned int find_vde_by_name(const struct ddf_super *ddf,
2250 const char *name)
2251 {
2252 unsigned int i;
2253 if (name == NULL)
2254 return DDF_NOTFOUND;
2255 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++) {
2256 if (all_ff(ddf->virt->entries[i].guid))
2257 continue;
2258 if (!strncmp(name, ddf->virt->entries[i].name,
2259 sizeof(ddf->virt->entries[i].name)))
2260 return i;
2261 }
2262 return DDF_NOTFOUND;
2263 }
2264
2265 static unsigned int find_vde_by_guid(const struct ddf_super *ddf,
2266 const char *guid)
2267 {
2268 unsigned int i;
2269 if (guid == NULL || all_ff(guid))
2270 return DDF_NOTFOUND;
2271 for (i = 0; i < be16_to_cpu(ddf->virt->max_vdes); i++)
2272 if (!memcmp(ddf->virt->entries[i].guid, guid, DDF_GUID_LEN))
2273 return i;
2274 return DDF_NOTFOUND;
2275 }
2276
2277 static int init_super_ddf(struct supertype *st,
2278 mdu_array_info_t *info,
2279 struct shape *s, char *name, char *homehost,
2280 int *uuid, unsigned long long data_offset)
2281 {
2282 /* This is primarily called by Create when creating a new array.
2283 * We will then get add_to_super called for each component, and then
2284 * write_init_super called to write it out to each device.
2285 * For DDF, Create can create on fresh devices or on a pre-existing
2286 * array.
2287 * To create on a pre-existing array a different method will be called.
2288 * This one is just for fresh drives.
2289 *
2290 * We need to create the entire 'ddf' structure which includes:
2291 * DDF headers - these are easy.
2292 * Controller data - a Sector describing this controller .. not that
2293 * this is a controller exactly.
2294 * Physical Disk Record - one entry per device, so
2295 * leave plenty of space.
2296 * Virtual Disk Records - again, just leave plenty of space.
2297 * This just lists VDs, doesn't give details.
2298 * Config records - describe the VDs that use this disk
2299 * DiskData - describes 'this' device.
2300 * BadBlockManagement - empty
2301 * Diag Space - empty
2302 * Vendor Logs - Could we put bitmaps here?
2303 *
2304 */
2305 struct ddf_super *ddf;
2306 char hostname[17];
2307 int hostlen;
2308 int max_phys_disks, max_virt_disks;
2309 unsigned long long sector;
2310 int clen;
2311 int i;
2312 int pdsize, vdsize;
2313 struct phys_disk *pd;
2314 struct virtual_disk *vd;
2315
2316 if (st->sb)
2317 return init_super_ddf_bvd(st, info, s->size, name, homehost, uuid,
2318 data_offset);
2319
2320 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
2321 pr_err("could not allocate superblock\n");
2322 return 0;
2323 }
2324 memset(ddf, 0, sizeof(*ddf));
2325 st->sb = ddf;
2326
2327 if (info == NULL) {
2328 /* zeroing superblock */
2329 return 0;
2330 }
2331
2332 /* At least 32MB *must* be reserved for the ddf. So let's just
2333 * start 32MB from the end, and put the primary header there.
2334 * Don't do secondary for now.
2335 * We don't know exactly where that will be yet as it could be
2336 * different on each device. So just set up the lengths.
2337 */
2338
2339 ddf->anchor.magic = DDF_HEADER_MAGIC;
2340 make_header_guid(ddf->anchor.guid);
2341
2342 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
2343 ddf->anchor.seq = cpu_to_be32(1);
2344 ddf->anchor.timestamp = cpu_to_be32(time(0) - DECADE);
2345 ddf->anchor.openflag = 0xFF;
2346 ddf->anchor.foreignflag = 0;
2347 ddf->anchor.enforcegroups = 0; /* Is this best?? */
2348 ddf->anchor.pad0 = 0xff;
2349 memset(ddf->anchor.pad1, 0xff, 12);
2350 memset(ddf->anchor.header_ext, 0xff, 32);
2351 ddf->anchor.primary_lba = cpu_to_be64(~(__u64)0);
2352 ddf->anchor.secondary_lba = cpu_to_be64(~(__u64)0);
2353 ddf->anchor.type = DDF_HEADER_ANCHOR;
2354 memset(ddf->anchor.pad2, 0xff, 3);
2355 ddf->anchor.workspace_len = cpu_to_be32(32768); /* Must be reserved */
2356 /* Put this at bottom of 32M reserved.. */
2357 ddf->anchor.workspace_lba = cpu_to_be64(~(__u64)0);
2358 max_phys_disks = 1023; /* Should be enough, 4095 is also allowed */
2359 ddf->anchor.max_pd_entries = cpu_to_be16(max_phys_disks);
2360 max_virt_disks = 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2361 ddf->anchor.max_vd_entries = cpu_to_be16(max_virt_disks);
2362 ddf->max_part = 64;
2363 ddf->anchor.max_partitions = cpu_to_be16(ddf->max_part);
2364 ddf->mppe = 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2365 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
2366 ddf->anchor.config_record_len = cpu_to_be16(ddf->conf_rec_len);
2367 ddf->anchor.max_primary_element_entries = cpu_to_be16(ddf->mppe);
2368 memset(ddf->anchor.pad3, 0xff, 54);
2369 /* Controller section is one sector long immediately
2370 * after the ddf header */
2371 sector = 1;
2372 ddf->anchor.controller_section_offset = cpu_to_be32(sector);
2373 ddf->anchor.controller_section_length = cpu_to_be32(1);
2374 sector += 1;
2375
2376 /* phys is 8 sectors after that */
2377 pdsize = ROUND_UP(sizeof(struct phys_disk) +
2378 sizeof(struct phys_disk_entry)*max_phys_disks,
2379 512);
2380 switch(pdsize/512) {
2381 case 2: case 8: case 32: case 128: case 512: break;
2382 default: abort();
2383 }
2384 ddf->anchor.phys_section_offset = cpu_to_be32(sector);
2385 ddf->anchor.phys_section_length =
2386 cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
2387 sector += pdsize/512;
2388
2389 /* virt is another 32 sectors */
2390 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
2391 sizeof(struct virtual_entry) * max_virt_disks,
2392 512);
2393 switch(vdsize/512) {
2394 case 2: case 8: case 32: case 128: case 512: break;
2395 default: abort();
2396 }
2397 ddf->anchor.virt_section_offset = cpu_to_be32(sector);
2398 ddf->anchor.virt_section_length =
2399 cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
2400 sector += vdsize/512;
2401
2402 clen = ddf->conf_rec_len * (ddf->max_part+1);
2403 ddf->anchor.config_section_offset = cpu_to_be32(sector);
2404 ddf->anchor.config_section_length = cpu_to_be32(clen);
2405 sector += clen;
2406
2407 ddf->anchor.data_section_offset = cpu_to_be32(sector);
2408 ddf->anchor.data_section_length = cpu_to_be32(1);
2409 sector += 1;
2410
2411 ddf->anchor.bbm_section_length = cpu_to_be32(0);
2412 ddf->anchor.bbm_section_offset = cpu_to_be32(0xFFFFFFFF);
2413 ddf->anchor.diag_space_length = cpu_to_be32(0);
2414 ddf->anchor.diag_space_offset = cpu_to_be32(0xFFFFFFFF);
2415 ddf->anchor.vendor_length = cpu_to_be32(0);
2416 ddf->anchor.vendor_offset = cpu_to_be32(0xFFFFFFFF);
2417
2418 memset(ddf->anchor.pad4, 0xff, 256);
2419
2420 memcpy(&ddf->primary, &ddf->anchor, 512);
2421 memcpy(&ddf->secondary, &ddf->anchor, 512);
2422
2423 ddf->primary.openflag = 1; /* I guess.. */
2424 ddf->primary.type = DDF_HEADER_PRIMARY;
2425
2426 ddf->secondary.openflag = 1; /* I guess.. */
2427 ddf->secondary.type = DDF_HEADER_SECONDARY;
2428
2429 ddf->active = &ddf->primary;
2430
2431 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
2432
2433 /* 24 more bytes of fiction required.
2434 * first 8 are a 'vendor-id' - "Linux-MD"
2435 * Remaining 16 are serial number.... maybe a hostname would do?
2436 */
2437 memcpy(ddf->controller.guid, T10, sizeof(T10));
2438 gethostname(hostname, sizeof(hostname));
2439 hostname[sizeof(hostname) - 1] = 0;
2440 hostlen = strlen(hostname);
2441 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
2442 for (i = strlen(T10) ; i+hostlen < 24; i++)
2443 ddf->controller.guid[i] = ' ';
2444
2445 ddf->controller.type.vendor_id = cpu_to_be16(0xDEAD);
2446 ddf->controller.type.device_id = cpu_to_be16(0xBEEF);
2447 ddf->controller.type.sub_vendor_id = cpu_to_be16(0);
2448 ddf->controller.type.sub_device_id = cpu_to_be16(0);
2449 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
2450 memset(ddf->controller.pad, 0xff, 8);
2451 memset(ddf->controller.vendor_data, 0xff, 448);
2452 if (homehost && strlen(homehost) < 440)
2453 strcpy((char*)ddf->controller.vendor_data, homehost);
2454
2455 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
2456 pr_err("could not allocate pd\n");
2457 return 0;
2458 }
2459 ddf->phys = pd;
2460 ddf->pdsize = pdsize;
2461
2462 memset(pd, 0xff, pdsize);
2463 memset(pd, 0, sizeof(*pd));
2464 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2465 pd->used_pdes = cpu_to_be16(0);
2466 pd->max_pdes = cpu_to_be16(max_phys_disks);
2467 memset(pd->pad, 0xff, 52);
2468 for (i = 0; i < max_phys_disks; i++)
2469 memset(pd->entries[i].guid, 0xff, DDF_GUID_LEN);
2470
2471 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
2472 pr_err("could not allocate vd\n");
2473 return 0;
2474 }
2475 ddf->virt = vd;
2476 ddf->vdsize = vdsize;
2477 memset(vd, 0, vdsize);
2478 vd->magic = DDF_VIRT_RECORDS_MAGIC;
2479 vd->populated_vdes = cpu_to_be16(0);
2480 vd->max_vdes = cpu_to_be16(max_virt_disks);
2481 memset(vd->pad, 0xff, 52);
2482
2483 for (i=0; i<max_virt_disks; i++)
2484 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
2485
2486 st->sb = ddf;
2487 ddf_set_updates_pending(ddf, NULL);
2488 return 1;
2489 }
2490
2491 static int chunk_to_shift(int chunksize)
2492 {
2493 return ffs(chunksize/512)-1;
2494 }
2495
2496 struct extent {
2497 unsigned long long start, size;
2498 };
2499 static int cmp_extent(const void *av, const void *bv)
2500 {
2501 const struct extent *a = av;
2502 const struct extent *b = bv;
2503 if (a->start < b->start)
2504 return -1;
2505 if (a->start > b->start)
2506 return 1;
2507 return 0;
2508 }
2509
2510 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
2511 {
2512 /* Find a list of used extents on the given physical device
2513 * (dnum) of the given ddf.
2514 * Return a malloced array of 'struct extent'
2515 */
2516 struct extent *rv;
2517 int n = 0;
2518 unsigned int i;
2519 __u16 state;
2520
2521 if (dl->pdnum < 0)
2522 return NULL;
2523 state = be16_to_cpu(ddf->phys->entries[dl->pdnum].state);
2524
2525 if ((state & (DDF_Online|DDF_Failed|DDF_Missing)) != DDF_Online)
2526 return NULL;
2527
2528 rv = xmalloc(sizeof(struct extent) * (ddf->max_part + 2));
2529
2530 for (i = 0; i < ddf->max_part; i++) {
2531 const struct vd_config *bvd;
2532 unsigned int ibvd;
2533 struct vcl *v = dl->vlist[i];
2534 if (v == NULL ||
2535 get_pd_index_from_refnum(v, dl->disk.refnum, ddf->mppe,
2536 &bvd, &ibvd) == DDF_NOTFOUND)
2537 continue;
2538 rv[n].start = be64_to_cpu(LBA_OFFSET(ddf, bvd)[ibvd]);
2539 rv[n].size = be64_to_cpu(bvd->blocks);
2540 n++;
2541 }
2542 qsort(rv, n, sizeof(*rv), cmp_extent);
2543
2544 rv[n].start = be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
2545 rv[n].size = 0;
2546 return rv;
2547 }
2548
2549 static unsigned long long find_space(
2550 struct ddf_super *ddf, struct dl *dl,
2551 unsigned long long data_offset,
2552 unsigned long long *size)
2553 {
2554 /* Find if the requested amount of space is available.
2555 * If it is, return start.
2556 * If not, set *size to largest space.
2557 * If data_offset != INVALID_SECTORS, then the space must start
2558 * at this location.
2559 */
2560 struct extent *e = get_extents(ddf, dl);
2561 int i = 0;
2562 unsigned long long pos = 0;
2563 unsigned long long max_size = 0;
2564
2565 if (!e) {
2566 *size = 0;
2567 return INVALID_SECTORS;
2568 }
2569 do {
2570 unsigned long long esize = e[i].start - pos;
2571 if (data_offset != INVALID_SECTORS &&
2572 pos <= data_offset &&
2573 e[i].start > data_offset) {
2574 pos = data_offset;
2575 esize = e[i].start - pos;
2576 }
2577 if (data_offset != INVALID_SECTORS &&
2578 pos != data_offset) {
2579 i++;
2580 continue;
2581 }
2582 if (esize >= *size) {
2583 /* Found! */
2584 free(e);
2585 return pos;
2586 }
2587 if (esize > max_size)
2588 max_size = esize;
2589 pos = e[i].start + e[i].size;
2590 i++;
2591 } while (e[i-1].size);
2592 *size = max_size;
2593 free(e);
2594 return INVALID_SECTORS;
2595 }
2596
2597 static int init_super_ddf_bvd(struct supertype *st,
2598 mdu_array_info_t *info,
2599 unsigned long long size,
2600 char *name, char *homehost,
2601 int *uuid, unsigned long long data_offset)
2602 {
2603 /* We are creating a BVD inside a pre-existing container.
2604 * so st->sb is already set.
2605 * We need to create a new vd_config and a new virtual_entry
2606 */
2607 struct ddf_super *ddf = st->sb;
2608 unsigned int venum, i;
2609 struct virtual_entry *ve;
2610 struct vcl *vcl;
2611 struct vd_config *vc;
2612
2613 if (find_vde_by_name(ddf, name) != DDF_NOTFOUND) {
2614 pr_err("This ddf already has an array called %s\n", name);
2615 return 0;
2616 }
2617 venum = find_unused_vde(ddf);
2618 if (venum == DDF_NOTFOUND) {
2619 pr_err("Cannot find spare slot for virtual disk\n");
2620 return 0;
2621 }
2622 ve = &ddf->virt->entries[venum];
2623
2624 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2625 * timestamp, random number
2626 */
2627 make_header_guid(ve->guid);
2628 ve->unit = cpu_to_be16(info->md_minor);
2629 ve->pad0 = 0xFFFF;
2630 ve->guid_crc._v16 = crc32(0, (unsigned char *)ddf->anchor.guid,
2631 DDF_GUID_LEN);
2632 ve->type = cpu_to_be16(0);
2633 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
2634 if (info->state & 1) /* clean */
2635 ve->init_state = DDF_init_full;
2636 else
2637 ve->init_state = DDF_init_not;
2638
2639 memset(ve->pad1, 0xff, 14);
2640 memset(ve->name, ' ', 16);
2641 if (name)
2642 strncpy(ve->name, name, 16);
2643 ddf->virt->populated_vdes =
2644 cpu_to_be16(be16_to_cpu(ddf->virt->populated_vdes)+1);
2645
2646 /* Now create a new vd_config */
2647 if (posix_memalign((void**)&vcl, 512,
2648 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
2649 pr_err("could not allocate vd_config\n");
2650 return 0;
2651 }
2652 vcl->vcnum = venum;
2653 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2654 vc = &vcl->conf;
2655
2656 vc->magic = DDF_VD_CONF_MAGIC;
2657 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2658 vc->timestamp = cpu_to_be32(time(0)-DECADE);
2659 vc->seqnum = cpu_to_be32(1);
2660 memset(vc->pad0, 0xff, 24);
2661 vc->chunk_shift = chunk_to_shift(info->chunk_size);
2662 if (layout_md2ddf(info, vc) == -1 ||
2663 be16_to_cpu(vc->prim_elmnt_count) > ddf->mppe) {
2664 pr_err("unsupported RAID level/layout %d/%d with %d disks\n",
2665 info->level, info->layout, info->raid_disks);
2666 free(vcl);
2667 return 0;
2668 }
2669 vc->sec_elmnt_seq = 0;
2670 if (alloc_other_bvds(ddf, vcl) != 0) {
2671 pr_err("could not allocate other bvds\n");
2672 free(vcl);
2673 return 0;
2674 }
2675 vc->blocks = cpu_to_be64(size * 2);
2676 vc->array_blocks = cpu_to_be64(
2677 calc_array_size(info->level, info->raid_disks, info->layout,
2678 info->chunk_size, size * 2));
2679 memset(vc->pad1, 0xff, 8);
2680 vc->spare_refs[0] = cpu_to_be32(0xffffffff);
2681 vc->spare_refs[1] = cpu_to_be32(0xffffffff);
2682 vc->spare_refs[2] = cpu_to_be32(0xffffffff);
2683 vc->spare_refs[3] = cpu_to_be32(0xffffffff);
2684 vc->spare_refs[4] = cpu_to_be32(0xffffffff);
2685 vc->spare_refs[5] = cpu_to_be32(0xffffffff);
2686 vc->spare_refs[6] = cpu_to_be32(0xffffffff);
2687 vc->spare_refs[7] = cpu_to_be32(0xffffffff);
2688 memset(vc->cache_pol, 0, 8);
2689 vc->bg_rate = 0x80;
2690 memset(vc->pad2, 0xff, 3);
2691 memset(vc->pad3, 0xff, 52);
2692 memset(vc->pad4, 0xff, 192);
2693 memset(vc->v0, 0xff, 32);
2694 memset(vc->v1, 0xff, 32);
2695 memset(vc->v2, 0xff, 16);
2696 memset(vc->v3, 0xff, 16);
2697 memset(vc->vendor, 0xff, 32);
2698
2699 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2700 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2701
2702 for (i = 1; i < vc->sec_elmnt_count; i++) {
2703 memcpy(vcl->other_bvds[i-1], vc, ddf->conf_rec_len * 512);
2704 vcl->other_bvds[i-1]->sec_elmnt_seq = i;
2705 }
2706
2707 vcl->next = ddf->conflist;
2708 ddf->conflist = vcl;
2709 ddf->currentconf = vcl;
2710 ddf_set_updates_pending(ddf, NULL);
2711 return 1;
2712 }
2713
2714 static void add_to_super_ddf_bvd(struct supertype *st,
2715 mdu_disk_info_t *dk, int fd, char *devname,
2716 unsigned long long data_offset)
2717 {
2718 /* fd and devname identify a device within the ddf container (st).
2719 * dk identifies a location in the new BVD.
2720 * We need to find suitable free space in that device and update
2721 * the phys_refnum and lba_offset for the newly created vd_config.
2722 * We might also want to update the type in the phys_disk
2723 * section.
2724 *
2725 * Alternately: fd == -1 and we have already chosen which device to
2726 * use and recorded in dlist->raid_disk;
2727 */
2728 struct dl *dl;
2729 struct ddf_super *ddf = st->sb;
2730 struct vd_config *vc;
2731 unsigned int i;
2732 unsigned long long blocks, pos;
2733 unsigned int raid_disk = dk->raid_disk;
2734
2735 if (fd == -1) {
2736 for (dl = ddf->dlist; dl ; dl = dl->next)
2737 if (dl->raiddisk == dk->raid_disk)
2738 break;
2739 } else {
2740 for (dl = ddf->dlist; dl ; dl = dl->next)
2741 if (dl->major == dk->major &&
2742 dl->minor == dk->minor)
2743 break;
2744 }
2745 if (!dl || dl->pdnum < 0 || ! (dk->state & (1<<MD_DISK_SYNC)))
2746 return;
2747
2748 vc = &ddf->currentconf->conf;
2749 if (vc->sec_elmnt_count > 1) {
2750 unsigned int n = be16_to_cpu(vc->prim_elmnt_count);
2751 if (raid_disk >= n)
2752 vc = ddf->currentconf->other_bvds[raid_disk / n - 1];
2753 raid_disk %= n;
2754 }
2755
2756 blocks = be64_to_cpu(vc->blocks);
2757 if (ddf->currentconf->block_sizes)
2758 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2759
2760 pos = find_space(ddf, dl, data_offset, &blocks);
2761 if (pos == INVALID_SECTORS)
2762 return;
2763
2764 ddf->currentdev = dk->raid_disk;
2765 vc->phys_refnum[raid_disk] = dl->disk.refnum;
2766 LBA_OFFSET(ddf, vc)[raid_disk] = cpu_to_be64(pos);
2767
2768 for (i = 0; i < ddf->max_part ; i++)
2769 if (dl->vlist[i] == NULL)
2770 break;
2771 if (i == ddf->max_part)
2772 return;
2773 dl->vlist[i] = ddf->currentconf;
2774
2775 if (fd >= 0)
2776 dl->fd = fd;
2777 if (devname)
2778 dl->devname = devname;
2779
2780 /* Check if we can mark array as optimal yet */
2781 i = ddf->currentconf->vcnum;
2782 ddf->virt->entries[i].state =
2783 (ddf->virt->entries[i].state & ~DDF_state_mask)
2784 | get_svd_state(ddf, ddf->currentconf);
2785 be16_clear(ddf->phys->entries[dl->pdnum].type,
2786 cpu_to_be16(DDF_Global_Spare));
2787 be16_set(ddf->phys->entries[dl->pdnum].type,
2788 cpu_to_be16(DDF_Active_in_VD));
2789 dprintf("added disk %d/%08x to VD %d/%s as disk %d\n",
2790 dl->pdnum, be32_to_cpu(dl->disk.refnum),
2791 ddf->currentconf->vcnum, guid_str(vc->guid),
2792 dk->raid_disk);
2793 ddf_set_updates_pending(ddf, vc);
2794 }
2795
2796 static unsigned int find_unused_pde(const struct ddf_super *ddf)
2797 {
2798 unsigned int i;
2799 for (i = 0; i < be16_to_cpu(ddf->phys->max_pdes); i++) {
2800 if (all_ff(ddf->phys->entries[i].guid))
2801 return i;
2802 }
2803 return DDF_NOTFOUND;
2804 }
2805
2806 static void _set_config_size(struct phys_disk_entry *pde, const struct dl *dl)
2807 {
2808 __u64 cfs, t;
2809 cfs = min(dl->size - 32*1024*2ULL, be64_to_cpu(dl->primary_lba));
2810 t = be64_to_cpu(dl->secondary_lba);
2811 if (t != ~(__u64)0)
2812 cfs = min(cfs, t);
2813 /*
2814 * Some vendor DDF structures interpret workspace_lba
2815 * very differently than we do: Make a sanity check on the value.
2816 */
2817 t = be64_to_cpu(dl->workspace_lba);
2818 if (t < cfs) {
2819 __u64 wsp = cfs - t;
2820 if (wsp > 1024*1024*2ULL && wsp > dl->size / 16) {
2821 pr_err("%x:%x: workspace size 0x%llx too big, ignoring\n",
2822 dl->major, dl->minor, (unsigned long long)wsp);
2823 } else
2824 cfs = t;
2825 }
2826 pde->config_size = cpu_to_be64(cfs);
2827 dprintf("%x:%x config_size %llx, DDF structure is %llx blocks\n",
2828 dl->major, dl->minor,
2829 (unsigned long long)cfs, (unsigned long long)(dl->size-cfs));
2830 }
2831
2832 /* Add a device to a container, either while creating it or while
2833 * expanding a pre-existing container
2834 */
2835 static int add_to_super_ddf(struct supertype *st,
2836 mdu_disk_info_t *dk, int fd, char *devname,
2837 unsigned long long data_offset)
2838 {
2839 struct ddf_super *ddf = st->sb;
2840 struct dl *dd;
2841 time_t now;
2842 struct tm *tm;
2843 unsigned long long size;
2844 struct phys_disk_entry *pde;
2845 unsigned int n, i;
2846 struct stat stb;
2847 __u32 *tptr;
2848
2849 if (ddf->currentconf) {
2850 add_to_super_ddf_bvd(st, dk, fd, devname, data_offset);
2851 return 0;
2852 }
2853
2854 /* This is device numbered dk->number. We need to create
2855 * a phys_disk entry and a more detailed disk_data entry.
2856 */
2857 fstat(fd, &stb);
2858 n = find_unused_pde(ddf);
2859 if (n == DDF_NOTFOUND) {
2860 pr_err("No free slot in array, cannot add disk\n");
2861 return 1;
2862 }
2863 pde = &ddf->phys->entries[n];
2864 get_dev_size(fd, NULL, &size);
2865 if (size <= 32*1024*1024) {
2866 pr_err("device size must be at least 32MB\n");
2867 return 1;
2868 }
2869 size >>= 9;
2870
2871 if (posix_memalign((void**)&dd, 512,
2872 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2873 pr_err("could allocate buffer for new disk, aborting\n");
2874 return 1;
2875 }
2876 dd->major = major(stb.st_rdev);
2877 dd->minor = minor(stb.st_rdev);
2878 dd->devname = devname;
2879 dd->fd = fd;
2880 dd->spare = NULL;
2881
2882 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2883 now = time(0);
2884 tm = localtime(&now);
2885 sprintf(dd->disk.guid, "%8s%04d%02d%02d", T10,
2886 (__u16)tm->tm_year+1900,
2887 (__u8)tm->tm_mon+1, (__u8)tm->tm_mday);
2888 tptr = (__u32 *)(dd->disk.guid + 16);
2889 *tptr++ = random32();
2890 *tptr = random32();
2891
2892 do {
2893 /* Cannot be bothered finding a CRC of some irrelevant details*/
2894 dd->disk.refnum._v32 = random32();
2895 for (i = be16_to_cpu(ddf->active->max_pd_entries);
2896 i > 0; i--)
2897 if (be32_eq(ddf->phys->entries[i-1].refnum,
2898 dd->disk.refnum))
2899 break;
2900 } while (i > 0);
2901
2902 dd->disk.forced_ref = 1;
2903 dd->disk.forced_guid = 1;
2904 memset(dd->disk.vendor, ' ', 32);
2905 memcpy(dd->disk.vendor, "Linux", 5);
2906 memset(dd->disk.pad, 0xff, 442);
2907 for (i = 0; i < ddf->max_part ; i++)
2908 dd->vlist[i] = NULL;
2909
2910 dd->pdnum = n;
2911
2912 if (st->update_tail) {
2913 int len = (sizeof(struct phys_disk) +
2914 sizeof(struct phys_disk_entry));
2915 struct phys_disk *pd;
2916
2917 pd = xmalloc(len);
2918 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2919 pd->used_pdes = cpu_to_be16(n);
2920 pde = &pd->entries[0];
2921 dd->mdupdate = pd;
2922 } else
2923 ddf->phys->used_pdes = cpu_to_be16(
2924 1 + be16_to_cpu(ddf->phys->used_pdes));
2925
2926 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2927 pde->refnum = dd->disk.refnum;
2928 pde->type = cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2929 pde->state = cpu_to_be16(DDF_Online);
2930 dd->size = size;
2931 /*
2932 * If there is already a device in dlist, try to reserve the same
2933 * amount of workspace. Otherwise, use 32MB.
2934 * We checked disk size above already.
2935 */
2936 #define __calc_lba(new, old, lba, mb) do { \
2937 unsigned long long dif; \
2938 if ((old) != NULL) \
2939 dif = (old)->size - be64_to_cpu((old)->lba); \
2940 else \
2941 dif = (new)->size; \
2942 if ((new)->size > dif) \
2943 (new)->lba = cpu_to_be64((new)->size - dif); \
2944 else \
2945 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2946 } while (0)
2947 __calc_lba(dd, ddf->dlist, workspace_lba, 32);
2948 __calc_lba(dd, ddf->dlist, primary_lba, 16);
2949 if (ddf->dlist == NULL ||
2950 be64_to_cpu(ddf->dlist->secondary_lba) != ~(__u64)0)
2951 __calc_lba(dd, ddf->dlist, secondary_lba, 32);
2952 _set_config_size(pde, dd);
2953
2954 sprintf(pde->path, "%17.17s","Information: nil") ;
2955 memset(pde->pad, 0xff, 6);
2956
2957 if (st->update_tail) {
2958 dd->next = ddf->add_list;
2959 ddf->add_list = dd;
2960 } else {
2961 dd->next = ddf->dlist;
2962 ddf->dlist = dd;
2963 ddf_set_updates_pending(ddf, NULL);
2964 }
2965
2966 return 0;
2967 }
2968
2969 static int remove_from_super_ddf(struct supertype *st, mdu_disk_info_t *dk)
2970 {
2971 struct ddf_super *ddf = st->sb;
2972 struct dl *dl;
2973
2974 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2975 * disappeared from the container.
2976 * We need to arrange that it disappears from the metadata and
2977 * internal data structures too.
2978 * Most of the work is done by ddf_process_update which edits
2979 * the metadata and closes the file handle and attaches the memory
2980 * where free_updates will free it.
2981 */
2982 for (dl = ddf->dlist; dl ; dl = dl->next)
2983 if (dl->major == dk->major &&
2984 dl->minor == dk->minor)
2985 break;
2986 if (!dl || dl->pdnum < 0)
2987 return -1;
2988
2989 if (st->update_tail) {
2990 int len = (sizeof(struct phys_disk) +
2991 sizeof(struct phys_disk_entry));
2992 struct phys_disk *pd;
2993
2994 pd = xmalloc(len);
2995 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2996 pd->used_pdes = cpu_to_be16(dl->pdnum);
2997 pd->entries[0].state = cpu_to_be16(DDF_Missing);
2998 append_metadata_update(st, pd, len);
2999 }
3000 return 0;
3001 }
3002
3003 /*
3004 * This is the write_init_super method for a ddf container. It is
3005 * called when creating a container or adding another device to a
3006 * container.
3007 */
3008
3009 static int __write_ddf_structure(struct dl *d, struct ddf_super *ddf, __u8 type)
3010 {
3011 unsigned long long sector;
3012 struct ddf_header *header;
3013 int fd, i, n_config, conf_size, buf_size;
3014 int ret = 0;
3015 char *conf;
3016
3017 fd = d->fd;
3018
3019 switch (type) {
3020 case DDF_HEADER_PRIMARY:
3021 header = &ddf->primary;
3022 sector = be64_to_cpu(header->primary_lba);
3023 break;
3024 case DDF_HEADER_SECONDARY:
3025 header = &ddf->secondary;
3026 sector = be64_to_cpu(header->secondary_lba);
3027 break;
3028 default:
3029 return 0;
3030 }
3031 if (sector == ~(__u64)0)
3032 return 0;
3033
3034 header->type = type;
3035 header->openflag = 1;
3036 header->crc = calc_crc(header, 512);
3037
3038 lseek64(fd, sector<<9, 0);
3039 if (write(fd, header, 512) < 0)
3040 goto out;
3041
3042 ddf->controller.crc = calc_crc(&ddf->controller, 512);
3043 if (write(fd, &ddf->controller, 512) < 0)
3044 goto out;
3045
3046 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
3047 if (write(fd, ddf->phys, ddf->pdsize) < 0)
3048 goto out;
3049 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
3050 if (write(fd, ddf->virt, ddf->vdsize) < 0)
3051 goto out;