Assemble: allow members of containers to be assembled and auto-assembled.
[thirdparty/mdadm.git] / Assemble.c
1 /*
2  * mdadm - manage Linux "md" devices aka RAID arrays.
3  *
4  * Copyright (C) 2001-2006 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: <neilb@cse.unsw.edu.au>
23  *    Paper: Neil Brown
24  *           School of Computer Science and Engineering
25  *           The University of New South Wales
26  *           Sydney, 2052
27  *           Australia
28  */
29
30 #include        "mdadm.h"
31 #include        <ctype.h>
32
33 static int name_matches(char *found, char *required, char *homehost)
34 {
35         /* See if the name found matches the required name, possibly
36          * prefixed with 'homehost'
37          */
38         char fnd[33];
39
40         strncpy(fnd, found, 32);
41         fnd[32] = 0;
42         if (strcmp(found, required)==0)
43                 return 1;
44         if (homehost) {
45                 int l = strlen(homehost);
46                 if (l < 32 && fnd[l] == ':' &&
47                     strcmp(fnd+l+1, required)==0)
48                         return 1;
49         }
50         return 0;
51 }
52
53 static int is_member_busy(char *metadata_version)
54 {
55         /* check if the given member array is active */
56         struct mdstat_ent *mdstat = mdstat_read(1, 0);
57         struct mdstat_ent *ent;
58         int busy = 0;
59
60         for (ent = mdstat; ent; ent = ent->next) {
61                 if (ent->metadata_version == NULL)
62                         continue;
63                 if (strncmp(ent->metadata_version, "external:", 9) != 0)
64                         continue;
65                 if (!is_subarray(&ent->metadata_version[9]))
66                         continue;
67                 /* Skip first char - it can be '/' or '-' */
68                 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
69                         busy = 1;
70                         break;
71                 }
72         }
73         free_mdstat(mdstat);
74
75         return busy;
76 }
77
78 int Assemble(struct supertype *st, char *mddev,
79              mddev_ident_t ident,
80              mddev_dev_t devlist, char *backup_file,
81              int readonly, int runstop,
82              char *update, char *homehost,
83              int verbose, int force)
84 {
85         /*
86          * The task of Assemble is to find a collection of
87          * devices that should (according to their superblocks)
88          * form an array, and to give this collection to the MD driver.
89          * In Linux-2.4 and later, this involves submitting a
90          * SET_ARRAY_INFO ioctl with no arg - to prepare
91          * the array - and then submit a number of
92          * ADD_NEW_DISK ioctls to add disks into
93          * the array.  Finally RUN_ARRAY might
94          * be submitted to start the array.
95          *
96          * Much of the work of Assemble is in finding and/or
97          * checking the disks to make sure they look right.
98          *
99          * If mddev is not set, then scan must be set and we
100          *  read through the config file for dev+uuid mapping
101          *  We recurse, setting mddev, for each device that
102          *    - isn't running
103          *    - has a valid uuid (or any uuid if !uuidset)
104          *
105          * If mddev is set, we try to determine state of md.
106          *   check version - must be at least 0.90.0
107          *   check kernel version.  must be at least 2.4.
108          *    If not, we can possibly fall back on START_ARRAY
109          *   Try to GET_ARRAY_INFO.
110          *     If possible, give up
111          *     If not, try to STOP_ARRAY just to make sure
112          *
113          * If !uuidset and scan, look in conf-file for uuid
114          *       If not found, give up
115          * If !devlist and scan and uuidset, get list of devs from conf-file
116          *
117          * For each device:
118          *   Check superblock - discard if bad
119          *   Check uuid (set if we don't have one) - discard if no match
120          *   Check superblock similarity if we have a superblock - discard if different
121          *   Record events, devicenum
122          * This should give us a list of devices for the array
123          * We should collect the most recent event number
124          *
125          * Count disks with recent enough event count
126          * While force && !enough disks
127          *    Choose newest rejected disks, update event count
128          *     mark clean and rewrite superblock
129          * If recent kernel:
130          *    SET_ARRAY_INFO
131          *    foreach device with recent events : ADD_NEW_DISK
132          *    if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
133          * If old kernel:
134          *    Check the device numbers in superblock are right
135          *    update superblock if any changes
136          *    START_ARRAY
137          *
138          */
139         int mdfd;
140         int clean;
141         int auto_assem = (mddev == NULL);
142         int old_linux = 0;
143         int vers = vers; /* Keep gcc quite - it really is initialised */
144         struct {
145                 char *devname;
146                 int uptodate; /* set once we decide that this device is as
147                                * recent as everything else in the array.
148                                */
149                 struct mdinfo i;
150         } *devices;
151         int *best = NULL; /* indexed by raid_disk */
152         unsigned int bestcnt = 0;
153         int devcnt = 0;
154         unsigned int okcnt, sparecnt;
155         unsigned int req_cnt;
156         unsigned int i;
157         int most_recent = 0;
158         int chosen_drive;
159         int change = 0;
160         int inargv = 0;
161         int report_missmatch;
162         int bitmap_done;
163         int start_partial_ok = (runstop >= 0) && 
164                 (force || devlist==NULL || auto_assem);
165         unsigned int num_devs;
166         mddev_dev_t tmpdev;
167         struct mdinfo info;
168         struct mdinfo *content = NULL;
169         mdu_array_info_t tmp_inf;
170         char *avail;
171         int nextspare = 0;
172         char *name = NULL;
173         int trustworthy;
174         char chosen_name[1024];
175
176         if (get_linux_version() < 2004000)
177                 old_linux = 1;
178
179         /*
180          * If any subdevs are listed, then any that don't
181          * match ident are discarded.  Remainder must all match and
182          * become the array.
183          * If no subdevs, then we scan all devices in the config file, but
184          * there must be something in the identity
185          */
186
187         if (!devlist &&
188             ident->uuid_set == 0 &&
189             ident->super_minor < 0 &&
190             ident->devices == NULL) {
191                 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
192                         mddev ? mddev : "further assembly");
193                 return 1;
194         }
195
196         if (devlist == NULL)
197                 devlist = conf_get_devs();
198         else if (mddev)
199                 inargv = 1;
200
201         report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
202  try_again:
203         /* We come back here when doing auto-assembly and attempting some
204          * set of devices failed.  Those are now marked as ->used==2 and
205          * we ignore them and try again
206          */
207
208         tmpdev = devlist; num_devs = 0;
209         while (tmpdev) {
210                 if (tmpdev->used)
211                         tmpdev->used = 2;
212                 else
213                         num_devs++;
214                 tmpdev = tmpdev->next;
215         }
216         devices = malloc(num_devs * sizeof(*devices));
217
218         if (!st && ident->st) st = ident->st;
219
220         if (verbose>0)
221             fprintf(stderr, Name ": looking for devices for %s\n",
222                     mddev ? mddev : "further assembly");
223
224         /* first walk the list of devices to find a consistent set
225          * that match the criterea, if that is possible.
226          * We flag the ones we like with 'used'.
227          */
228         for (tmpdev = devlist;
229              tmpdev;
230              tmpdev = tmpdev->next) {
231                 char *devname = tmpdev->devname;
232                 int dfd;
233                 struct stat stb;
234                 struct supertype *tst = dup_super(st);
235
236                 if (tmpdev->used > 1) continue;
237
238                 if (ident->devices &&
239                     !match_oneof(ident->devices, devname)) {
240                         if (report_missmatch)
241                                 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
242                         continue;
243                 }
244
245                 dfd = dev_open(devname, O_RDONLY|O_EXCL);
246                 if (dfd < 0) {
247                         if (report_missmatch)
248                                 fprintf(stderr, Name ": cannot open device %s: %s\n",
249                                         devname, strerror(errno));
250                         tmpdev->used = 2;
251                 } else if (fstat(dfd, &stb)< 0) {
252                         /* Impossible! */
253                         fprintf(stderr, Name ": fstat failed for %s: %s\n",
254                                 devname, strerror(errno));
255                         tmpdev->used = 2;
256                 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
257                         fprintf(stderr, Name ": %s is not a block device.\n",
258                                 devname);
259                         tmpdev->used = 2;
260                 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
261                         if (report_missmatch)
262                                 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
263                                         devname);
264                         tmpdev->used = 2;
265                 } else if (tst->ss->load_super(tst,dfd, NULL)) {
266                         if (report_missmatch)
267                                 fprintf( stderr, Name ": no RAID superblock on %s\n",
268                                          devname);
269                 } else {
270                         content = &info;
271                         memset(content, 0, sizeof(*content));
272                         tst->ss->getinfo_super(tst, content);
273                 }
274                 if (dfd >= 0) close(dfd);
275
276                 if (tst && tst->sb && tst->ss->container_content
277                     && tst->loaded_container) {
278                         /* tmpdev is a container.  We need to be either
279                          * looking for a member, or auto-assembling
280                          */
281                         if (st) {
282                                 /* already found some components, this cannot
283                                  * be another one.
284                                  */
285                                 if (report_missmatch)
286                                         fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
287                                                 devname);
288                                 goto loop;
289                         }
290
291                         if (ident->container) {
292                                 if (ident->container[0] == '/' &&
293                                     !same_dev(ident->container, devname)) {
294                                         if (report_missmatch)
295                                                 fprintf(stderr, Name ": %s is not the container required (%s)\n",
296                                                         devname, ident->container);
297                                         goto loop;
298                                 }
299                                 if (ident->container[0] != '/') {
300                                         /* we have a uuid */
301                                         int uuid[4];
302                                         if (!parse_uuid(ident->container, uuid) ||
303                                             !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
304                                                 if (report_missmatch)
305                                                         fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
306                                                                 devname);
307                                                 goto loop;
308                                         }
309                                 }
310                         }
311                         /* It is worth looking inside this container.
312                          */
313                 next_member:
314                         if (tmpdev->content)
315                                 content = tmpdev->content;
316                         else
317                                 content = tst->ss->container_content(tst);
318
319                         tmpdev->content = content->next;
320                         if (tmpdev->content == NULL)
321                                 tmpdev->used = 1;
322
323                 } else if (ident->container || ident->member) {
324                         /* No chance of this matching if we don't have
325                          * a container */
326                         if (report_missmatch)
327                                 fprintf(stderr, Name "%s is not a container, and one is required.\n",
328                                         devname);
329                         goto loop;
330                 }
331
332                 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
333                     (!tst || !tst->sb ||
334                      same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) {
335                         if (report_missmatch)
336                                 fprintf(stderr, Name ": %s has wrong uuid.\n",
337                                         devname);
338                         goto loop;
339                 }
340                 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
341                     (!tst || !tst->sb ||
342                      name_matches(content->name, ident->name, homehost)==0)) {
343                         if (report_missmatch)
344                                 fprintf(stderr, Name ": %s has wrong name.\n",
345                                         devname);
346                         goto loop;
347                 }
348                 if (ident->super_minor != UnSet &&
349                     (!tst || !tst->sb ||
350                      ident->super_minor != content->array.md_minor)) {
351                         if (report_missmatch)
352                                 fprintf(stderr, Name ": %s has wrong super-minor.\n",
353                                         devname);
354                         goto loop;
355                 }
356                 if (ident->level != UnSet &&
357                     (!tst || !tst->sb ||
358                      ident->level != content->array.level)) {
359                         if (report_missmatch)
360                                 fprintf(stderr, Name ": %s has wrong raid level.\n",
361                                         devname);
362                         goto loop;
363                 }
364                 if (ident->raid_disks != UnSet &&
365                     (!tst || !tst->sb ||
366                      ident->raid_disks!= content->array.raid_disks)) {
367                         if (report_missmatch)
368                                 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
369                                         devname);
370                         goto loop;
371                 }
372                 if (auto_assem) {
373                         if (tst == NULL || tst->sb == NULL)
374                                 continue;
375                 }
376                 /* If we are this far, then we are nearly commited to this device.
377                  * If the super_block doesn't exist, or doesn't match others,
378                  * then we probably cannot continue
379                  * However if one of the arrays is for the homehost, and
380                  * the other isn't that can disambiguate.
381                  */
382
383                 if (!tst || !tst->sb) {
384                         fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
385                                 devname);
386                         if (st)
387                                 st->ss->free_super(st);
388                         return 1;
389                 }
390
391                 if (tst && tst->sb && tst->ss->container_content
392                     && tst->loaded_container) {
393                         /* we have the one container we need, don't keep
394                          * looking.  If the chosen member is active, skip.
395                          */
396                         if (is_member_busy(content->text_version)) {
397                                 if (auto_assem)
398                                         goto loop;
399                                 fprintf(stderr, Name ": member %s in %s is already assembled\n",
400                                         content->text_version,
401                                         devname);
402                                 tst->ss->free_super(tst);
403                                 return 1;
404                         }
405                         st = tst; tst = NULL;
406                         if (!auto_assem && tmpdev->next != NULL) {
407                                 fprintf(stderr, Name ": %s is a container, but is not "
408                                         "only device given: confused and aborting\n",
409                                         devname);
410                                 st->ss->free_super(st);
411                                 return 1;
412                         }
413                         break;
414                 }
415                 if (st == NULL)
416                         st = dup_super(tst);
417                 if (st->minor_version == -1)
418                         st->minor_version = tst->minor_version;
419                 if (st->ss != tst->ss ||
420                     st->minor_version != tst->minor_version ||
421                     st->ss->compare_super(st, tst) != 0) {
422                         /* Some mismatch. If exactly one array matches this host,
423                          * we can resolve on that one.
424                          * Or, if we are auto assembling, we just ignore the second
425                          * for now.
426                          */
427                         if (auto_assem)
428                                 goto loop;
429                         if (homehost) {
430                                 int first = st->ss->match_home(st, homehost);
431                                 int last = tst->ss->match_home(tst, homehost);
432                                 if (first != last &&
433                                     (first == 1 || last == 1)) {
434                                         /* We can do something */
435                                         if (first) {/* just ignore this one */
436                                                 if (report_missmatch)
437                                                         fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
438                                                                 devname);
439                                                 goto loop;
440                                         } else { /* reject all those sofar */
441                                                 mddev_dev_t td;
442                                                 if (report_missmatch)
443                                                         fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
444                                                                 devname);
445                                                 for (td=devlist; td != tmpdev; td=td->next)
446                                                         if (td->used == 1)
447                                                                 td->used = 0;
448                                                 tmpdev->used = 1;
449                                                 goto loop;
450                                         }
451                                 }
452                         }
453                         fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
454                                 devname);
455                         tst->ss->free_super(tst);
456                         st->ss->free_super(st);
457                         return 1;
458                 }
459
460                 tmpdev->used = 1;
461
462         loop:
463                 if (tmpdev->content)
464                         goto next_member;
465                 if (tst)
466                         tst->ss->free_super(tst);
467         }
468
469         if (!st || !st->sb || !content)
470                 return 2;
471
472         /* Now need to open array the device.  Use create_mddev */
473         if (content == &info)
474                 st->ss->getinfo_super(st, content);
475
476         trustworthy = FOREIGN;
477         switch (st->ss->match_home(st, homehost)) {
478         case 0:
479                 trustworthy = FOREIGN;
480                 name = content->name;
481                 break;
482         case 1:
483                 trustworthy = LOCAL;
484                 name = strchr(content->name, ':');
485                 if (name)
486                         name++;
487                 else
488                         name = content->name;
489                 break;
490         case -1:
491                 trustworthy = FOREIGN;
492                 break;
493         }
494         if (!auto_assem && trustworthy == FOREIGN)
495                 /* If the array is listed in mdadm or on
496                  * command line, then we trust the name
497                  * even if the array doesn't look local
498                  */
499                 trustworthy = LOCAL;
500
501         if (content->name[0] == 0 &&
502             content->array.level == LEVEL_CONTAINER) {
503                 name = content->text_version;
504                 trustworthy = METADATA;
505         }
506         mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
507                             chosen_name);
508         if (mdfd < 0) {
509                 st->ss->free_super(st);
510                 free(devices);
511                 if (auto_assem)
512                         goto try_again;
513                 return 1;
514         }
515         mddev = chosen_name;
516         vers = md_get_version(mdfd);
517         if (vers < 9000) {
518                 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
519                         "    Upgrade your kernel or try --build\n");
520                 close(mdfd);
521                 return 1;
522         }
523         if (ioctl(mdfd, GET_ARRAY_INFO, &tmp_inf)==0) {
524                 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
525                         mddev);
526                 for (tmpdev = devlist ;
527                      tmpdev && tmpdev->used != 1;
528                      tmpdev = tmpdev->next)
529                         ;
530                 if (tmpdev && auto_assem)
531                         fprintf(stderr, Name ":   %s needed for %s...\n",
532                                 mddev, tmpdev->devname);
533                 close(mdfd);
534                 mdfd = -3;
535                 st->ss->free_super(st);
536                 free(devices);
537                 if (auto_assem)
538                         goto try_again;
539                 return 1;
540         }
541         ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
542
543 #ifndef MDASSEMBLE
544         if (content != &info) {
545                 /* This is a member of a container.  Try starting the array. */
546                 return assemble_container_content(st, mdfd, content, runstop,
547                                            chosen_name, verbose);
548         }
549 #endif
550         /* Ok, no bad inconsistancy, we can try updating etc */
551         bitmap_done = 0;
552         for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
553                 char *devname = tmpdev->devname;
554                 struct stat stb;
555                 /* looks like a good enough match to update the super block if needed */
556 #ifndef MDASSEMBLE
557                 if (update) {
558                         int dfd;
559                         /* prepare useful information in info structures */
560                         struct stat stb2;
561                         struct supertype *tst;
562                         fstat(mdfd, &stb2);
563
564                         if (strcmp(update, "uuid")==0 &&
565                             !ident->uuid_set) {
566                                 int rfd;
567                                 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
568                                     read(rfd, ident->uuid, 16) != 16) {
569                                         *(__u32*)(ident->uuid) = random();
570                                         *(__u32*)(ident->uuid+1) = random();
571                                         *(__u32*)(ident->uuid+2) = random();
572                                         *(__u32*)(ident->uuid+3) = random();
573                                 }
574                                 if (rfd >= 0) close(rfd);
575                         }
576                         dfd = dev_open(devname, O_RDWR|O_EXCL);
577
578                         remove_partitions(dfd);
579
580                         tst = dup_super(st);
581                         tst->ss->load_super(tst, dfd, NULL);
582                         tst->ss->getinfo_super(tst, content);
583
584                         memcpy(content->uuid, ident->uuid, 16);
585                         strcpy(content->name, ident->name);
586                         content->array.md_minor = minor(stb2.st_rdev);
587
588                         tst->ss->update_super(tst, content, update,
589                                               devname, verbose,
590                                               ident->uuid_set, homehost);
591                         if (strcmp(update, "uuid")==0 &&
592                             !ident->uuid_set) {
593                                 ident->uuid_set = 1;
594                                 memcpy(ident->uuid, content->uuid, 16);
595                         }
596                         if (dfd < 0)
597                                 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
598                                         devname);
599                         else if (tst->ss->store_super(tst, dfd))
600                                 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
601                                         devname);
602                         if (dfd >= 0)
603                                 close(dfd);
604
605                         if (strcmp(update, "uuid")==0 &&
606                             ident->bitmap_fd >= 0 && !bitmap_done) {
607                                 if (bitmap_update_uuid(ident->bitmap_fd,
608                                                        content->uuid,
609                                                        tst->ss->swapuuid) != 0)
610                                         fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
611                                 else
612                                         bitmap_done = 1;
613                         }
614                         tst->ss->free_super(tst);
615                 } else
616 #endif
617                 {
618                         struct supertype *tst = dup_super(st);
619                         int dfd;
620                         dfd = dev_open(devname, O_RDWR|O_EXCL);
621
622                         remove_partitions(dfd);
623
624                         tst->ss->load_super(tst, dfd, NULL);
625                         tst->ss->getinfo_super(tst, content);
626                         tst->ss->free_super(tst);
627                         close(dfd);
628                 }
629
630                 stat(devname, &stb);
631
632                 if (verbose > 0)
633                         fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
634                                 devname, mddev, content->disk.raid_disk);
635                 devices[devcnt].devname = devname;
636                 devices[devcnt].uptodate = 0;
637                 devices[devcnt].i = *content;
638                 devices[devcnt].i.disk.major = major(stb.st_rdev);
639                 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
640                 if (most_recent < devcnt) {
641                         if (devices[devcnt].i.events
642                             > devices[most_recent].i.events)
643                                 most_recent = devcnt;
644                 }
645                 if (content->array.level == -4)
646                         /* with multipath, the raid_disk from the superblock is meaningless */
647                         i = devcnt;
648                 else
649                         i = devices[devcnt].i.disk.raid_disk;
650                 if (i+1 == 0) {
651                         if (nextspare < content->array.raid_disks)
652                                 nextspare = content->array.raid_disks;
653                         i = nextspare++;
654                 } else {
655                         if (i >= content->array.raid_disks &&
656                             i >= nextspare)
657                                 nextspare = i+1;
658                 }
659                 if (i < 10000) {
660                         if (i >= bestcnt) {
661                                 unsigned int newbestcnt = i+10;
662                                 int *newbest = malloc(sizeof(int)*newbestcnt);
663                                 unsigned int c;
664                                 for (c=0; c < newbestcnt; c++)
665                                         if (c < bestcnt)
666                                                 newbest[c] = best[c];
667                                         else
668                                                 newbest[c] = -1;
669                                 if (best)free(best);
670                                 best = newbest;
671                                 bestcnt = newbestcnt;
672                         }
673                         if (best[i] >=0 &&
674                             devices[best[i]].i.events
675                             == devices[devcnt].i.events
676                             && (devices[best[i]].i.disk.minor
677                                 != devices[devcnt].i.disk.minor)
678                             && st->ss == &super0
679                             && content->array.level != LEVEL_MULTIPATH) {
680                                 /* two different devices with identical superblock.
681                                  * Could be a mis-detection caused by overlapping
682                                  * partitions.  fail-safe.
683                                  */
684                                 fprintf(stderr, Name ": WARNING %s and %s appear"
685                                         " to have very similar superblocks.\n"
686                                         "      If they are really different, "
687                                         "please --zero the superblock on one\n"
688                                         "      If they are the same or overlap,"
689                                         " please remove one from %s.\n",
690                                         devices[best[i]].devname, devname,
691                                         inargv ? "the list" :
692                                            "the\n      DEVICE list in mdadm.conf"
693                                         );
694                                 close(mdfd);
695                                 return 1;
696                         }
697                         if (best[i] == -1
698                             || (devices[best[i]].i.events
699                                 < devices[devcnt].i.events))
700                                 best[i] = devcnt;
701                 }
702                 devcnt++;
703         }
704
705         if (devcnt == 0) {
706                 fprintf(stderr, Name ": no devices found for %s\n",
707                         mddev);
708                 if (st)
709                         st->ss->free_super(st);
710                 close(mdfd);
711                 return 1;
712         }
713
714         if (update && strcmp(update, "byteorder")==0)
715                 st->minor_version = 90;
716
717         st->ss->getinfo_super(st, content);
718         clean = content->array.state & 1;
719
720         /* now we have some devices that might be suitable.
721          * I wonder how many
722          */
723         avail = malloc(content->array.raid_disks);
724         memset(avail, 0, content->array.raid_disks);
725         okcnt = 0;
726         sparecnt=0;
727         for (i=0; i< bestcnt ;i++) {
728                 int j = best[i];
729                 int event_margin = 1; /* always allow a difference of '1'
730                                        * like the kernel does
731                                        */
732                 if (j < 0) continue;
733                 /* note: we ignore error flags in multipath arrays
734                  * as they don't make sense
735                  */
736                 if (content->array.level != -4)
737                         if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
738                                 if (!(devices[j].i.disk.state
739                                       & (1<<MD_DISK_FAULTY)))
740                                         sparecnt++;
741                                 continue;
742                         }
743                 if (devices[j].i.events+event_margin >=
744                     devices[most_recent].i.events) {
745                         devices[j].uptodate = 1;
746                         if (i < content->array.raid_disks) {
747                                 okcnt++;
748                                 avail[i]=1;
749                         } else
750                                 sparecnt++;
751                 }
752         }
753         while (force && !enough(content->array.level, content->array.raid_disks,
754                                 content->array.layout, 1,
755                                 avail, okcnt)) {
756                 /* Choose the newest best drive which is
757                  * not up-to-date, update the superblock
758                  * and add it.
759                  */
760                 int fd;
761                 struct supertype *tst;
762                 long long current_events;
763                 chosen_drive = -1;
764                 for (i=0; i<content->array.raid_disks && i < bestcnt; i++) {
765                         int j = best[i];
766                         if (j>=0 &&
767                             !devices[j].uptodate &&
768                             devices[j].i.events > 0 &&
769                             (chosen_drive < 0 ||
770                              devices[j].i.events
771                              > devices[chosen_drive].i.events))
772                                 chosen_drive = j;
773                 }
774                 if (chosen_drive < 0)
775                         break;
776                 current_events = devices[chosen_drive].i.events;
777         add_another:
778                 if (verbose >= 0)
779                         fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
780                                 devices[chosen_drive].devname,
781                                 devices[chosen_drive].i.disk.raid_disk,
782                                 (int)(devices[chosen_drive].i.events),
783                                 (int)(devices[most_recent].i.events));
784                 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
785                 if (fd < 0) {
786                         fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
787                                 devices[chosen_drive].devname);
788                         devices[chosen_drive].i.events = 0;
789                         continue;
790                 }
791                 tst = dup_super(st);
792                 if (tst->ss->load_super(tst,fd, NULL)) {
793                         close(fd);
794                         fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
795                                 devices[chosen_drive].devname);
796                         devices[chosen_drive].i.events = 0;
797                         continue;
798                 }
799                 content->events = devices[most_recent].i.events;
800                 tst->ss->update_super(tst, content, "force-one",
801                                      devices[chosen_drive].devname, verbose,
802                                      0, NULL);
803
804                 if (tst->ss->store_super(tst, fd)) {
805                         close(fd);
806                         fprintf(stderr, Name ": Could not re-write superblock on %s\n",
807                                 devices[chosen_drive].devname);
808                         devices[chosen_drive].i.events = 0;
809                         tst->ss->free_super(tst);
810                         continue;
811                 }
812                 close(fd);
813                 devices[chosen_drive].i.events = devices[most_recent].i.events;
814                 devices[chosen_drive].uptodate = 1;
815                 avail[chosen_drive] = 1;
816                 okcnt++;
817                 tst->ss->free_super(tst);
818
819                 /* If there are any other drives of the same vintage,
820                  * add them in as well.  We can't lose and we might gain
821                  */
822                 for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) {
823                         int j = best[i];
824                         if (j >= 0 &&
825                             !devices[j].uptodate &&
826                             devices[j].i.events > 0 &&
827                             devices[j].i.events == current_events) {
828                                 chosen_drive = j;
829                                 goto add_another;
830                         }
831                 }
832         }
833
834         /* Now we want to look at the superblock which the kernel will base things on
835          * and compare the devices that we think are working with the devices that the
836          * superblock thinks are working.
837          * If there are differences and --force is given, then update this chosen
838          * superblock.
839          */
840         chosen_drive = -1;
841         st->ss->free_super(st);
842         for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
843                 int j = best[i];
844                 int fd;
845
846                 if (j<0)
847                         continue;
848                 if (!devices[j].uptodate)
849                         continue;
850                 chosen_drive = j;
851                 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
852                         fprintf(stderr, Name ": Cannot open %s: %s\n",
853                                 devices[j].devname, strerror(errno));
854                         close(mdfd);
855                         return 1;
856                 }
857                 if (st->ss->load_super(st,fd, NULL)) {
858                         close(fd);
859                         fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
860                                 devices[j].devname);
861                         close(mdfd);
862                         return 1;
863                 }
864                 close(fd);
865         }
866         if (st->sb == NULL) {
867                 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
868                 close(mdfd);
869                 return 1;
870         }
871         st->ss->getinfo_super(st, content);
872 #ifndef MDASSEMBLE
873         sysfs_init(content, mdfd, 0);
874 #endif
875         for (i=0; i<bestcnt; i++) {
876                 int j = best[i];
877                 unsigned int desired_state;
878
879                 if (i < content->array.raid_disks)
880                         desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
881                 else
882                         desired_state = 0;
883
884                 if (j<0)
885                         continue;
886                 if (!devices[j].uptodate)
887                         continue;
888
889                 devices[j].i.disk.state = desired_state;
890
891                 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
892                                          verbose, 0, NULL)) {
893                         if (force) {
894                                 if (verbose >= 0)
895                                         fprintf(stderr, Name ": "
896                                                 "clearing FAULTY flag for device %d in %s for %s\n",
897                                                 j, mddev, devices[j].devname);
898                                 change = 1;
899                         } else {
900                                 if (verbose >= -1)
901                                         fprintf(stderr, Name ": "
902                                                 "device %d in %s has wrong state in superblock, but %s seems ok\n",
903                                                 i, mddev, devices[j].devname);
904                         }
905                 }
906 #if 0
907                 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
908                         fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
909                                 i, mddev);
910                 }
911 #endif
912         }
913         if (force && !clean &&
914             !enough(content->array.level, content->array.raid_disks,
915                     content->array.layout, clean,
916                     avail, okcnt)) {
917                 change += st->ss->update_super(st, content, "force-array",
918                                         devices[chosen_drive].devname, verbose,
919                                                0, NULL);
920                 clean = 1;
921         }
922
923         if (change) {
924                 int fd;
925                 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
926                 if (fd < 0) {
927                         fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
928                                 devices[chosen_drive].devname);
929                         close(mdfd);
930                         return 1;
931                 }
932                 if (st->ss->store_super(st, fd)) {
933                         close(fd);
934                         fprintf(stderr, Name ": Could not re-write superblock on %s\n",
935                                 devices[chosen_drive].devname);
936                         close(mdfd);
937                         return 1;
938                 }
939                 close(fd);
940         }
941
942         /* If we are in the middle of a reshape we may need to restore saved data
943          * that was moved aside due to the reshape overwriting live data
944          * The code of doing this lives in Grow.c
945          */
946 #ifndef MDASSEMBLE
947         if (content->reshape_active) {
948                 int err = 0;
949                 int *fdlist = malloc(sizeof(int)* bestcnt);
950                 for (i=0; i<bestcnt; i++) {
951                         int j = best[i];
952                         if (j >= 0) {
953                                 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
954                                 if (fdlist[i] < 0) {
955                                         fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
956                                                 devices[j].devname);
957                                         err = 1;
958                                         break;
959                                 }
960                         } else
961                                 fdlist[i] = -1;
962                 }
963                 if (!err)
964                         err = Grow_restart(st, content, fdlist, bestcnt, backup_file);
965                 while (i>0) {
966                         i--;
967                         if (fdlist[i]>=0) close(fdlist[i]);
968                 }
969                 if (err) {
970                         fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
971                         close(mdfd);
972                         return err;
973                 }
974         }
975 #endif
976         /* count number of in-sync devices according to the superblock.
977          * We must have this number to start the array without -s or -R
978          */
979         req_cnt = content->array.working_disks;
980
981         /* Almost ready to actually *do* something */
982         if (!old_linux) {
983                 int rv;
984
985                 /* First, fill in the map, so that udev can find our name
986                  * as soon as we become active.
987                  */
988                 map_update(NULL, fd2devnum(mdfd), content->text_version,
989                            content->uuid, chosen_name);
990
991                 rv = set_array_info(mdfd, st, content);
992                 if (rv) {
993                         fprintf(stderr, Name ": failed to set array info for %s: %s\n",
994                                 mddev, strerror(errno));
995                         close(mdfd);
996                         return 1;
997                 }
998                 if (ident->bitmap_fd >= 0) {
999                         if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1000                                 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1001                                 close(mdfd);
1002                                 return 1;
1003                         }
1004                 } else if (ident->bitmap_file) {
1005                         /* From config file */
1006                         int bmfd = open(ident->bitmap_file, O_RDWR);
1007                         if (bmfd < 0) {
1008                                 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1009                                         ident->bitmap_file);
1010                                 close(mdfd);
1011                                 return 1;
1012                         }
1013                         if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1014                                 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1015                                 close(bmfd);
1016                                 close(mdfd);
1017                                 return 1;
1018                         }
1019                         close(bmfd);
1020                 }
1021
1022                 /* First, add the raid disks, but add the chosen one last */
1023                 for (i=0; i<= bestcnt; i++) {
1024                         int j;
1025                         if (i < bestcnt) {
1026                                 j = best[i];
1027                                 if (j == chosen_drive)
1028                                         continue;
1029                         } else
1030                                 j = chosen_drive;
1031
1032                         if (j >= 0 /* && devices[j].uptodate */) {
1033                                 rv = add_disk(mdfd, st, content, &devices[j].i);
1034
1035                                 if (rv) {
1036                                         fprintf(stderr, Name ": failed to add "
1037                                                         "%s to %s: %s\n",
1038                                                 devices[j].devname,
1039                                                 mddev,
1040                                                 strerror(errno));
1041                                         if (i < content->array.raid_disks
1042                                             || i == bestcnt)
1043                                                 okcnt--;
1044                                         else
1045                                                 sparecnt--;
1046                                 } else if (verbose > 0)
1047                                         fprintf(stderr, Name ": added %s "
1048                                                         "to %s as %d\n",
1049                                                 devices[j].devname, mddev,
1050                                                 devices[j].i.disk.raid_disk);
1051                         } else if (verbose > 0 && i < content->array.raid_disks)
1052                                 fprintf(stderr, Name ": no uptodate device for "
1053                                                 "slot %d of %s\n",
1054                                         i, mddev);
1055                 }
1056
1057                 if (content->array.level == LEVEL_CONTAINER) {
1058                         if (verbose >= 0) {
1059                                 fprintf(stderr, Name ": Container %s has been "
1060                                         "assembled with %d drive%s",
1061                                         mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1062                                 if (okcnt < content->array.raid_disks)
1063                                         fprintf(stderr, " (out of %d)",
1064                                                 content->array.raid_disks);
1065                                 fprintf(stderr, "\n");
1066                         }
1067                         sysfs_uevent(content, "change");
1068                         close(mdfd);
1069                         return 0;
1070                 }
1071
1072                 if (runstop == 1 ||
1073                     (runstop <= 0 &&
1074                      ( enough(content->array.level, content->array.raid_disks,
1075                               content->array.layout, clean, avail, okcnt) &&
1076                        (okcnt >= req_cnt || start_partial_ok)
1077                              ))) {
1078                         if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
1079                                 if (verbose >= 0) {
1080                                         fprintf(stderr, Name ": %s has been started with %d drive%s",
1081                                                 mddev, okcnt, okcnt==1?"":"s");
1082                                         if (okcnt < content->array.raid_disks)
1083                                                 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1084                                         if (sparecnt)
1085                                                 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1086                                         fprintf(stderr, ".\n");
1087                                 }
1088                                 close(mdfd);
1089                                 if (auto_assem) {
1090                                         int usecs = 1;
1091                                         /* There is a nasty race with 'mdadm --monitor'.
1092                                          * If it opens this device before we close it,
1093                                          * it gets an incomplete open on which IO
1094                                          * doesn't work and the capacity is
1095                                          * wrong.
1096                                          * If we reopen (to check for layered devices)
1097                                          * before --monitor closes, we loose.
1098                                          *
1099                                          * So: wait upto 1 second for there to be
1100                                          * a non-zero capacity.
1101                                          */
1102                                         while (usecs < 1000) {
1103                                                 mdfd = open(mddev, O_RDONLY);
1104                                                 if (mdfd >= 0) {
1105                                                         unsigned long long size;
1106                                                         if (get_dev_size(mdfd, NULL, &size) &&
1107                                                             size > 0)
1108                                                                 break;
1109                                                         close(mdfd);
1110                                                 }
1111                                                 usleep(usecs);
1112                                                 usecs <<= 1;
1113                                         }
1114                                 }
1115                                 return 0;
1116                         }
1117                         fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1118                                 mddev, strerror(errno));
1119
1120                         if (!enough(content->array.level, content->array.raid_disks,
1121                                     content->array.layout, 1, avail, okcnt))
1122                                 fprintf(stderr, Name ": Not enough devices to "
1123                                         "start the array.\n");
1124                         else if (!enough(content->array.level,
1125                                          content->array.raid_disks,
1126                                          content->array.layout, clean,
1127                                          avail, okcnt))
1128                                 fprintf(stderr, Name ": Not enough devices to "
1129                                         "start the array while not clean "
1130                                         "- consider --force.\n");
1131
1132                         if (auto_assem)
1133                                 ioctl(mdfd, STOP_ARRAY, NULL);
1134                         close(mdfd);
1135                         return 1;
1136                 }
1137                 if (runstop == -1) {
1138                         fprintf(stderr, Name ": %s assembled from %d drive%s",
1139                                 mddev, okcnt, okcnt==1?"":"s");
1140                         if (okcnt != content->array.raid_disks)
1141                                 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1142                         fprintf(stderr, ", but not started.\n");
1143                         close(mdfd);
1144                         return 0;
1145                 }
1146                 if (verbose >= -1) {
1147                         fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1148                         if (sparecnt)
1149                                 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1150                         if (!enough(content->array.level, content->array.raid_disks,
1151                                     content->array.layout, 1, avail, okcnt))
1152                                 fprintf(stderr, " - not enough to start the array.\n");
1153                         else if (!enough(content->array.level,
1154                                          content->array.raid_disks,
1155                                          content->array.layout, clean,
1156                                          avail, okcnt))
1157                                 fprintf(stderr, " - not enough to start the "
1158                                         "array while not clean - consider "
1159                                         "--force.\n");
1160                         else {
1161                                 if (req_cnt == content->array.raid_disks)
1162                                         fprintf(stderr, " - need all %d to start it", req_cnt);
1163                                 else
1164                                         fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1165                                 fprintf(stderr, " (use --run to insist).\n");
1166                         }
1167                 }
1168                 if (auto_assem)
1169                         ioctl(mdfd, STOP_ARRAY, NULL);
1170                 return 1;
1171         } else {
1172                 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1173                  * been updated to point to the current locations of devices.
1174                  * so we can just start the array
1175                  */
1176                 unsigned long dev;
1177                 dev = makedev(devices[chosen_drive].i.disk.major,
1178                             devices[chosen_drive].i.disk.minor);
1179                 if (ioctl(mdfd, START_ARRAY, dev)) {
1180                     fprintf(stderr, Name ": Cannot start array: %s\n",
1181                             strerror(errno));
1182                 }
1183
1184         }
1185         close(mdfd);
1186         return 0;
1187 }
1188
1189 #ifndef MDASSEMBLE
1190 int assemble_container_content(struct supertype *st, int mdfd,
1191                                struct mdinfo *content, int runstop,
1192                                char *chosen_name, int verbose)
1193 {
1194         struct mdinfo *dev, *sra;
1195         int working = 0, preexist = 0;
1196         struct map_ent *map = NULL;
1197
1198         sysfs_init(content, mdfd, 0);
1199
1200         sra = sysfs_read(mdfd, 0, GET_VERSION);
1201         if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1202                 if (sysfs_set_array(content, md_get_version(mdfd)) != 0)
1203                         return 1;
1204         if (sra)
1205                 sysfs_free(sra);
1206
1207         for (dev = content->devs; dev; dev = dev->next)
1208                 if (sysfs_add_disk(content, dev) == 0)
1209                         working++;
1210                 else if (errno == EEXIST)
1211                         preexist++;
1212         if (working == 0)
1213                 /* Nothing new, don't try to start */ ;
1214         else if (runstop > 0 ||
1215                  (working + preexist) >= content->array.working_disks) {
1216                 switch(content->array.level) {
1217                 case LEVEL_LINEAR:
1218                 case LEVEL_MULTIPATH:
1219                 case 0:
1220                         sysfs_set_str(content, NULL, "array_state",
1221                                       "active");
1222                         break;
1223                 default:
1224                         sysfs_set_str(content, NULL, "array_state",
1225                                       "readonly");
1226                         /* start mdmon if needed. */
1227                         if (!mdmon_running(st->container_dev))
1228                                 start_mdmon(st->container_dev);
1229                         ping_monitor(devnum2devname(st->container_dev));
1230                         break;
1231                 }
1232                 sysfs_set_safemode(content, content->safe_mode_delay);
1233                 if (verbose >= 0) {
1234                         fprintf(stderr, Name
1235                                 ": Started %s with %d devices",
1236                                 chosen_name, working + preexist);
1237                         if (preexist)
1238                                 fprintf(stderr, " (%d new)", working);
1239                         fprintf(stderr, "\n");
1240                 }
1241                 /* FIXME should have an O_EXCL and wait for read-auto */
1242         } else
1243                 if (verbose >= 0)
1244                         fprintf(stderr, Name
1245                                 ": %s assembled with %d devices but "
1246                                 "not started\n",
1247                                 chosen_name, working);
1248         map_update(&map, fd2devnum(mdfd),
1249                    content->text_version,
1250                    content->uuid, chosen_name);
1251
1252         return 0;
1253 }
1254 #endif
1255