Can now mostly assemble DDF arrays
[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 int Assemble(struct supertype *st, char *mddev, int mdfd,
54              mddev_ident_t ident,
55              mddev_dev_t devlist, char *backup_file,
56              int readonly, int runstop,
57              char *update, char *homehost,
58              int verbose, int force)
59 {
60         /*
61          * The task of Assemble is to find a collection of
62          * devices that should (according to their superblocks)
63          * form an array, and to give this collection to the MD driver.
64          * In Linux-2.4 and later, this involves submitting a
65          * SET_ARRAY_INFO ioctl with no arg - to prepare
66          * the array - and then submit a number of
67          * ADD_NEW_DISK ioctls to add disks into
68          * the array.  Finally RUN_ARRAY might
69          * be submitted to start the array.
70          *
71          * Much of the work of Assemble is in finding and/or
72          * checking the disks to make sure they look right.
73          *
74          * If mddev is not set, then scan must be set and we
75          *  read through the config file for dev+uuid mapping
76          *  We recurse, setting mddev, for each device that
77          *    - isn't running
78          *    - has a valid uuid (or any uuid if !uuidset)
79          *
80          * If mddev is set, we try to determine state of md.
81          *   check version - must be at least 0.90.0
82          *   check kernel version.  must be at least 2.4.
83          *    If not, we can possibly fall back on START_ARRAY
84          *   Try to GET_ARRAY_INFO.
85          *     If possible, give up
86          *     If not, try to STOP_ARRAY just to make sure
87          *
88          * If !uuidset and scan, look in conf-file for uuid
89          *       If not found, give up
90          * If !devlist and scan and uuidset, get list of devs from conf-file
91          *
92          * For each device:
93          *   Check superblock - discard if bad
94          *   Check uuid (set if we don't have one) - discard if no match
95          *   Check superblock similarity if we have a superblock - discard if different
96          *   Record events, devicenum
97          * This should give us a list of devices for the array
98          * We should collect the most recent event number
99          *
100          * Count disks with recent enough event count
101          * While force && !enough disks
102          *    Choose newest rejected disks, update event count
103          *     mark clean and rewrite superblock
104          * If recent kernel:
105          *    SET_ARRAY_INFO
106          *    foreach device with recent events : ADD_NEW_DISK
107          *    if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
108          * If old kernel:
109          *    Check the device numbers in superblock are right
110          *    update superblock if any changes
111          *    START_ARRAY
112          *
113          */
114         int clean = 0;
115         int must_close = 0;
116         int old_linux = 0;
117         int vers = 0; /* Keep gcc quite - it really is initialised */
118         struct {
119                 char *devname;
120                 int uptodate; /* set once we decide that this device is as
121                                * recent as everything else in the array.
122                                */
123                 struct mdinfo i;
124         } *devices;
125         int *best = NULL; /* indexed by raid_disk */
126         unsigned int bestcnt = 0;
127         int devcnt = 0;
128         unsigned int okcnt, sparecnt;
129         unsigned int req_cnt;
130         unsigned int i;
131         int most_recent = 0;
132         int chosen_drive;
133         int change = 0;
134         int inargv = 0;
135         int bitmap_done;
136         int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
137         unsigned int num_devs;
138         mddev_dev_t tmpdev;
139         struct mdinfo info;
140         char *avail;
141         int nextspare = 0;
142
143         if (get_linux_version() < 2004000)
144                 old_linux = 1;
145
146         if (mdfd >= 0) {
147                 vers = md_get_version(mdfd);
148                 if (vers <= 0) {
149                         fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
150                         return 1;
151                 }
152                 if (vers < 9000) {
153                         fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
154                                 "    Upgrade your kernel or try --build\n");
155                         return 1;
156                 }
157
158                 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
159                         fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
160                                 mddev);
161                         return 1;
162                 }
163                 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
164         }
165         /*
166          * If any subdevs are listed, then any that don't
167          * match ident are discarded.  Remainder must all match and
168          * become the array.
169          * If no subdevs, then we scan all devices in the config file, but
170          * there must be something in the identity
171          */
172
173         if (!devlist &&
174             ident->uuid_set == 0 &&
175             ident->super_minor < 0 &&
176             ident->devices == NULL) {
177                 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
178                         mddev ? mddev : "further assembly");
179                 return 1;
180         }
181         if (devlist == NULL)
182                 devlist = conf_get_devs();
183         else if (mdfd >= 0)
184                 inargv = 1;
185
186  try_again:
187
188         tmpdev = devlist; num_devs = 0;
189         while (tmpdev) {
190                 if (tmpdev->used)
191                         tmpdev->used = 2;
192                 else
193                         num_devs++;
194                 tmpdev = tmpdev->next;
195         }
196         devices = malloc(num_devs * sizeof(*devices));
197
198         if (!st && ident->st) st = ident->st;
199
200         if (verbose>0)
201             fprintf(stderr, Name ": looking for devices for %s\n",
202                     mddev ? mddev : "further assembly");
203
204         /* first walk the list of devices to find a consistent set
205          * that match the criterea, if that is possible.
206          * We flag the one we like with 'used'.
207          */
208         for (tmpdev = devlist;
209              tmpdev;
210              tmpdev = tmpdev->next) {
211                 char *devname = tmpdev->devname;
212                 int dfd;
213                 struct stat stb;
214                 struct supertype *tst = dup_super(st);
215
216                 if (tmpdev->used > 1) continue;
217
218                 if (ident->devices &&
219                     !match_oneof(ident->devices, devname)) {
220                         if ((inargv && verbose>=0) || verbose > 0)
221                                 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
222                         continue;
223                 }
224
225                 dfd = dev_open(devname, O_RDONLY|O_EXCL);
226                 if (dfd < 0) {
227                         if ((inargv && verbose >= 0) || verbose > 0)
228                                 fprintf(stderr, Name ": cannot open device %s: %s\n",
229                                         devname, strerror(errno));
230                         tmpdev->used = 2;
231                 } else if (fstat(dfd, &stb)< 0) {
232                         /* Impossible! */
233                         fprintf(stderr, Name ": fstat failed for %s: %s\n",
234                                 devname, strerror(errno));
235                         tmpdev->used = 2;
236                 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
237                         fprintf(stderr, Name ": %s is not a block device.\n",
238                                 devname);
239                         tmpdev->used = 2;
240                 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
241                         if ((inargv && verbose >= 0) || verbose > 0)
242                                 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
243                                         devname);
244                         tmpdev->used = 2;
245                 } else if (tst->ss->load_super(tst,dfd, NULL)) {
246                         if ((inargv && verbose >= 0) || verbose > 0)
247                                 fprintf( stderr, Name ": no RAID superblock on %s\n",
248                                          devname);
249                 } else {
250                         tst->ss->getinfo_super(tst, &info);
251                 }
252                 if (dfd >= 0) close(dfd);
253
254                 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
255                     (!tst || !tst->sb ||
256                      same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
257                         if ((inargv && verbose >= 0) || verbose > 0)
258                                 fprintf(stderr, Name ": %s has wrong uuid.\n",
259                                         devname);
260                         goto loop;
261                 }
262                 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
263                     (!tst || !tst->sb ||
264                      name_matches(info.name, ident->name, homehost)==0)) {
265                         if ((inargv && verbose >= 0) || verbose > 0)
266                                 fprintf(stderr, Name ": %s has wrong name.\n",
267                                         devname);
268                         goto loop;
269                 }
270                 if (ident->super_minor != UnSet &&
271                     (!tst || !tst->sb ||
272                      ident->super_minor != info.array.md_minor)) {
273                         if ((inargv && verbose >= 0) || verbose > 0)
274                                 fprintf(stderr, Name ": %s has wrong super-minor.\n",
275                                         devname);
276                         goto loop;
277                 }
278                 if (ident->level != UnSet &&
279                     (!tst || !tst->sb ||
280                      ident->level != info.array.level)) {
281                         if ((inargv && verbose >= 0) || verbose > 0)
282                                 fprintf(stderr, Name ": %s has wrong raid level.\n",
283                                         devname);
284                         goto loop;
285                 }
286                 if (ident->raid_disks != UnSet &&
287                     (!tst || !tst->sb ||
288                      ident->raid_disks!= info.array.raid_disks)) {
289                         if ((inargv && verbose >= 0) || verbose > 0)
290                                 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
291                                         devname);
292                         goto loop;
293                 }
294                 if (mdfd < 0) {
295                         if (tst == NULL || tst->sb == NULL)
296                                 continue;
297                         if (update == NULL &&
298                             tst->ss->match_home(tst, homehost)==0) {
299                                 if ((inargv && verbose >= 0) || verbose > 0)
300                                         fprintf(stderr, Name ": %s is not built for host %s.\n",
301                                                 devname, homehost);
302                                 /* Auto-assemble, and this is not a usable host */
303                                 /* if update != NULL, we are updating the host
304                                  * name... */
305                                 goto loop;
306                         }
307                 }
308                 /* If we are this far, then we are nearly commited to this device.
309                  * If the super_block doesn't exist, or doesn't match others,
310                  * then we probably cannot continue
311                  * However if one of the arrays is for the homehost, and
312                  * the other isn't that can disambiguate.
313                  */
314
315                 if (!tst || !tst->sb) {
316                         fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
317                                 devname);
318                         if (st)
319                                 st->ss->free_super(st);
320                         return 1;
321                 }
322
323                 if (st == NULL)
324                         st = dup_super(tst);
325                 if (st->minor_version == -1)
326                         st->minor_version = tst->minor_version;
327                 if (st->ss != tst->ss ||
328                     st->minor_version != tst->minor_version ||
329                     st->ss->compare_super(st, tst) != 0) {
330                         /* Some mismatch. If exactly one array matches this host,
331                          * we can resolve on that one.
332                          * Or, if we are auto assembling, we just ignore the second
333                          * for now.
334                          */
335                         if (mdfd < 0)
336                                 goto loop;
337                         if (homehost) {
338                                 int first = st->ss->match_home(st, homehost);
339                                 int last = tst->ss->match_home(tst, homehost);
340                                 if (first+last == 1) {
341                                         /* We can do something */
342                                         if (first) {/* just ignore this one */
343                                                 if ((inargv && verbose >= 0) || verbose > 0)
344                                                         fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
345                                                                 devname);
346                                                 goto loop;
347                                         } else { /* reject all those sofar */
348                                                 mddev_dev_t td;
349                                                 if ((inargv && verbose >= 0) || verbose > 0)
350                                                         fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
351                                                                 devname);
352                                                 for (td=devlist; td != tmpdev; td=td->next)
353                                                         if (td->used == 1)
354                                                                 td->used = 0;
355                                                 tmpdev->used = 1;
356                                                 goto loop;
357                                         }
358                                 }
359                         }
360                         fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
361                                 devname);
362                         tst->ss->free_super(tst);
363                         st->ss->free_super(st);
364                         return 1;
365                 }
366
367                 tmpdev->used = 1;
368
369         loop:
370                 tst->ss->free_super(tst);
371         }
372
373         if (mdfd < 0) {
374                 /* So... it is up to me to open the device.
375                  * We create a name '/dev/md/XXX' based on the info in the
376                  * superblock, and call open_mddev on that
377                  */
378                 mdu_array_info_t inf;
379                 char *c;
380                 if (!st->sb) {
381                         return 2;
382                 }
383                 st->ss->getinfo_super(st, &info);
384                 c = strchr(info.name, ':');
385                 if (c) c++; else c= info.name;
386                 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
387                         /* /dev/md/d0 style for partitionable */
388                         asprintf(&mddev, "/dev/md/d%s", c);
389                 else
390                         asprintf(&mddev, "/dev/md/%s", c);
391                 mdfd = open_mddev(mddev, ident->autof);
392                 if (mdfd < 0) {
393                         st->ss->free_super(st);
394                         free(devices);
395                         goto try_again;
396                 }
397                 vers = md_get_version(mdfd);
398                 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
399                         for (tmpdev = devlist ;
400                              tmpdev && tmpdev->used != 1;
401                              tmpdev = tmpdev->next)
402                                 ;
403                         fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
404                         if (tmpdev)
405                                 fprintf(stderr, Name ":   %s needed for %s...\n",
406                                         mddev, tmpdev->devname);
407                         close(mdfd);
408                         mdfd = -1;
409                         st->ss->free_super(st);
410                         free(devices);
411                         goto try_again;
412                 }
413                 must_close = 1;
414         }
415
416         /* Ok, no bad inconsistancy, we can try updating etc */
417         bitmap_done = 0;
418         for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
419                 char *devname = tmpdev->devname;
420                 struct stat stb;
421                 /* looks like a good enough match to update the super block if needed */
422 #ifndef MDASSEMBLE
423                 if (update) {
424                         int dfd;
425                         /* prepare useful information in info structures */
426                         struct stat stb2;
427                         struct supertype *tst;
428                         fstat(mdfd, &stb2);
429
430                         if (strcmp(update, "uuid")==0 &&
431                             !ident->uuid_set) {
432                                 int rfd;
433                                 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
434                                     read(rfd, ident->uuid, 16) != 16) {
435                                         *(__u32*)(ident->uuid) = random();
436                                         *(__u32*)(ident->uuid+1) = random();
437                                         *(__u32*)(ident->uuid+2) = random();
438                                         *(__u32*)(ident->uuid+3) = random();
439                                 }
440                                 if (rfd >= 0) close(rfd);
441                         }
442                         dfd = dev_open(devname, O_RDWR|O_EXCL);
443
444                         remove_partitions(dfd);
445
446                         tst = dup_super(st);
447                         tst->ss->load_super(tst, dfd, NULL);
448                         tst->ss->getinfo_super(tst, &info);
449
450                         memcpy(info.uuid, ident->uuid, 16);
451                         strcpy(info.name, ident->name);
452                         info.array.md_minor = minor(stb2.st_rdev);
453
454                         tst->ss->update_super(tst, &info, update,
455                                               devname, verbose,
456                                               ident->uuid_set, homehost);
457                         if (strcmp(update, "uuid")==0 &&
458                             !ident->uuid_set) {
459                                 ident->uuid_set = 1;
460                                 memcpy(ident->uuid, info.uuid, 16);
461                         }
462                         if (dfd < 0)
463                                 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
464                                         devname);
465                         else if (tst->ss->store_super(tst, dfd))
466                                 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
467                                         devname);
468                         if (dfd >= 0)
469                                 close(dfd);
470
471                         if (strcmp(update, "uuid")==0 &&
472                             ident->bitmap_fd >= 0 && !bitmap_done) {
473                                 if (bitmap_update_uuid(ident->bitmap_fd,
474                                                        info.uuid,
475                                                        tst->ss->swapuuid) != 0)
476                                         fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
477                                 else
478                                         bitmap_done = 1;
479                         }
480                         tst->ss->free_super(tst);
481                 } else
482 #endif
483                 {
484                         struct supertype *tst = dup_super(st);;
485                         int dfd;
486                         dfd = dev_open(devname, O_RDWR|O_EXCL);
487
488                         remove_partitions(dfd);
489
490                         tst->ss->load_super(tst, dfd, NULL);
491                         tst->ss->getinfo_super(tst, &info);
492                         tst->ss->free_super(tst);
493                         close(dfd);
494                 }
495
496                 stat(devname, &stb);
497
498                 if (verbose > 0)
499                         fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
500                                 devname, mddev, info.disk.raid_disk);
501                 devices[devcnt].devname = devname;
502                 devices[devcnt].uptodate = 0;
503                 devices[devcnt].i = info;
504                 devices[devcnt].i.disk.major = major(stb.st_rdev);
505                 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
506                 if (most_recent < devcnt) {
507                         if (devices[devcnt].i.events
508                             > devices[most_recent].i.events)
509                                 most_recent = devcnt;
510                 }
511                 if (info.array.level == -4)
512                         /* with multipath, the raid_disk from the superblock is meaningless */
513                         i = devcnt;
514                 else
515                         i = devices[devcnt].i.disk.raid_disk;
516                 if (i+1 == 0) {
517                         if (nextspare < info.array.raid_disks)
518                                 nextspare = info.array.raid_disks;
519                         i = nextspare++;
520                 } else {
521                         if (i >= info.array.raid_disks &&
522                             i >= nextspare)
523                                 nextspare = i+1;
524                 }
525                 if (i < 10000) {
526                         if (i >= bestcnt) {
527                                 unsigned int newbestcnt = i+10;
528                                 int *newbest = malloc(sizeof(int)*newbestcnt);
529                                 unsigned int c;
530                                 for (c=0; c < newbestcnt; c++)
531                                         if (c < bestcnt)
532                                                 newbest[c] = best[c];
533                                         else
534                                                 newbest[c] = -1;
535                                 if (best)free(best);
536                                 best = newbest;
537                                 bestcnt = newbestcnt;
538                         }
539                         if (best[i] >=0 &&
540                             devices[best[i]].i.events
541                             == devices[devcnt].i.events
542                             && (devices[best[i]].i.disk.minor
543                                 != devices[devcnt].i.disk.minor)
544                             && st->ss->major == 0
545                             && info.array.level != -4) {
546                                 /* two different devices with identical superblock.
547                                  * Could be a mis-detection caused by overlapping
548                                  * partitions.  fail-safe.
549                                  */
550                                 fprintf(stderr, Name ": WARNING %s and %s appear"
551                                         " to have very similar superblocks.\n"
552                                         "      If they are really different, "
553                                         "please --zero the superblock on one\n"
554                                         "      If they are the same or overlap,"
555                                         " please remove one from %s.\n",
556                                         devices[best[i]].devname, devname,
557                                         inargv ? "the list" :
558                                            "the\n      DEVICE list in mdadm.conf"
559                                         );
560                                 if (must_close) close(mdfd);
561                                 return 1;
562                         }
563                         if (best[i] == -1
564                             || (devices[best[i]].i.events
565                                 < devices[devcnt].i.events))
566                                 best[i] = devcnt;
567                 }
568                 devcnt++;
569         }
570
571         if (update && strcmp(update, "byteorder")==0)
572                 st->minor_version = 90;
573
574         if (devcnt == 0) {
575                 fprintf(stderr, Name ": no devices found for %s\n",
576                         mddev);
577                 st->ss->free_super(st);
578                 if (must_close) close(mdfd);
579                 return 1;
580         }
581
582         st->ss->getinfo_super(st, &info);
583         clean = info.array.state & 1;
584
585         /* now we have some devices that might be suitable.
586          * I wonder how many
587          */
588         avail = malloc(info.array.raid_disks);
589         memset(avail, 0, info.array.raid_disks);
590         okcnt = 0;
591         sparecnt=0;
592         for (i=0; i< bestcnt ;i++) {
593                 int j = best[i];
594                 int event_margin = 1; /* always allow a difference of '1'
595                                        * like the kernel does
596                                        */
597                 if (j < 0) continue;
598                 /* note: we ignore error flags in multipath arrays
599                  * as they don't make sense
600                  */
601                 if (info.array.level != -4)
602                         if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
603                                 if (!(devices[j].i.disk.state
604                                       & (1<<MD_DISK_FAULTY)))
605                                         sparecnt++;
606                                 continue;
607                         }
608                 if (devices[j].i.events+event_margin >=
609                     devices[most_recent].i.events) {
610                         devices[j].uptodate = 1;
611                         if (i < info.array.raid_disks) {
612                                 okcnt++;
613                                 avail[i]=1;
614                         } else
615                                 sparecnt++;
616                 }
617         }
618         while (force && !enough(info.array.level, info.array.raid_disks,
619                                 info.array.layout, 1,
620                                 avail, okcnt)) {
621                 /* Choose the newest best drive which is
622                  * not up-to-date, update the superblock
623                  * and add it.
624                  */
625                 int fd;
626                 struct supertype *tst;
627                 long long current_events;
628                 chosen_drive = -1;
629                 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
630                         int j = best[i];
631                         if (j>=0 &&
632                             !devices[j].uptodate &&
633                             devices[j].i.events > 0 &&
634                             (chosen_drive < 0 ||
635                              devices[j].i.events
636                              > devices[chosen_drive].i.events))
637                                 chosen_drive = j;
638                 }
639                 if (chosen_drive < 0)
640                         break;
641                 current_events = devices[chosen_drive].i.events;
642         add_another:
643                 if (verbose >= 0)
644                         fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
645                                 devices[chosen_drive].devname,
646                                 devices[chosen_drive].i.disk.raid_disk,
647                                 (int)(devices[chosen_drive].i.events),
648                                 (int)(devices[most_recent].i.events));
649                 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
650                 if (fd < 0) {
651                         fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
652                                 devices[chosen_drive].devname);
653                         devices[chosen_drive].i.events = 0;
654                         continue;
655                 }
656                 tst = dup_super(st);
657                 if (tst->ss->load_super(st,fd, NULL)) {
658                         close(fd);
659                         fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
660                                 devices[chosen_drive].devname);
661                         devices[chosen_drive].i.events = 0;
662                         continue;
663                 }
664                 info.events = devices[most_recent].i.events;
665                 tst->ss->update_super(tst, &info, "force-one",
666                                      devices[chosen_drive].devname, verbose,
667                                      0, NULL);
668
669                 if (tst->ss->store_super(tst, fd)) {
670                         close(fd);
671                         fprintf(stderr, Name ": Could not re-write superblock on %s\n",
672                                 devices[chosen_drive].devname);
673                         devices[chosen_drive].i.events = 0;
674                         tst->ss->free_super(tst);
675                         continue;
676                 }
677                 close(fd);
678                 devices[chosen_drive].i.events = devices[most_recent].i.events;
679                 devices[chosen_drive].uptodate = 1;
680                 avail[chosen_drive] = 1;
681                 okcnt++;
682                 tst->ss->free_super(tst);
683
684                 /* If there are any other drives of the same vintage,
685                  * add them in as well.  We can't lose and we might gain
686                  */
687                 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
688                         int j = best[i];
689                         if (j >= 0 &&
690                             !devices[j].uptodate &&
691                             devices[j].i.events > 0 &&
692                             devices[j].i.events == current_events) {
693                                 chosen_drive = j;
694                                 goto add_another;
695                         }
696                 }
697         }
698
699         /* Now we want to look at the superblock which the kernel will base things on
700          * and compare the devices that we think are working with the devices that the
701          * superblock thinks are working.
702          * If there are differences and --force is given, then update this chosen
703          * superblock.
704          */
705         chosen_drive = -1;
706         st->ss->free_super(st);
707         for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
708                 int j = best[i];
709                 int fd;
710
711                 if (j<0)
712                         continue;
713                 if (!devices[j].uptodate)
714                         continue;
715                 chosen_drive = j;
716                 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
717                         fprintf(stderr, Name ": Cannot open %s: %s\n",
718                                 devices[j].devname, strerror(errno));
719                         if (must_close) close(mdfd);
720                         return 1;
721                 }
722                 if (st->ss->load_super(st,fd, NULL)) {
723                         close(fd);
724                         fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
725                                 devices[j].devname);
726                         if (must_close) close(mdfd);
727                         return 1;
728                 }
729                 close(fd);
730         }
731         if (st->sb == NULL) {
732                 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
733                 if (must_close) close(mdfd);
734                 return 1;
735         }
736         st->ss->getinfo_super(st, &info);
737         for (i=0; i<bestcnt; i++) {
738                 int j = best[i];
739                 unsigned int desired_state;
740
741                 if (i < info.array.raid_disks)
742                         desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
743                 else
744                         desired_state = 0;
745
746                 if (j<0)
747                         continue;
748                 if (!devices[j].uptodate)
749                         continue;
750
751                 devices[j].i.disk.state = desired_state;
752
753                 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
754                                          verbose, 0, NULL)) {
755                         if (force) {
756                                 if (verbose >= 0)
757                                         fprintf(stderr, Name ": "
758                                                 "clearing FAULTY flag for device %d in %s for %s\n",
759                                                 j, mddev, devices[j].devname);
760                                 change = 1;
761                         } else {
762                                 if (verbose >= -1)
763                                         fprintf(stderr, Name ": "
764                                                 "device %d in %s has wrong state in superblock, but %s seems ok\n",
765                                                 i, mddev, devices[j].devname);
766                         }
767                 }
768 #if 0
769                 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
770                         fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
771                                 i, mddev);
772                 }
773 #endif
774         }
775         if (force && !clean &&
776             !enough(info.array.level, info.array.raid_disks,
777                     info.array.layout, clean,
778                     avail, okcnt)) {
779                 change += st->ss->update_super(st, &info, "force-array",
780                                         devices[chosen_drive].devname, verbose,
781                                                0, NULL);
782                 clean = 1;
783         }
784
785         if (change) {
786                 int fd;
787                 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
788                 if (fd < 0) {
789                         fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
790                                 devices[chosen_drive].devname);
791                         if (must_close) close(mdfd);
792                         return 1;
793                 }
794                 if (st->ss->store_super(st, fd)) {
795                         close(fd);
796                         fprintf(stderr, Name ": Could not re-write superblock on %s\n",
797                                 devices[chosen_drive].devname);
798                         if (must_close) close(mdfd);
799                         return 1;
800                 }
801                 close(fd);
802         }
803
804         /* If we are in the middle of a reshape we may need to restore saved data
805          * that was moved aside due to the reshape overwriting live data
806          * The code of doing this lives in Grow.c
807          */
808 #ifndef MDASSEMBLE
809         if (info.reshape_active) {
810                 int err = 0;
811                 int *fdlist = malloc(sizeof(int)* bestcnt);
812                 for (i=0; i<bestcnt; i++) {
813                         int j = best[i];
814                         if (j >= 0) {
815                                 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
816                                 if (fdlist[i] < 0) {
817                                         fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
818                                                 devices[j].devname);
819                                         err = 1;
820                                         break;
821                                 }
822                         } else
823                                 fdlist[i] = -1;
824                 }
825                 if (!err)
826                         err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
827                 while (i>0) {
828                         i--;
829                         if (fdlist[i]>=0) close(fdlist[i]);
830                 }
831                 if (err) {
832                         fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
833                         if (must_close) close(mdfd);
834                         return err;
835                 }
836         }
837 #endif
838         /* count number of in-sync devices according to the superblock.
839          * We must have this number to start the array without -s or -R
840          */
841         req_cnt = info.array.working_disks;
842
843         /* Almost ready to actually *do* something */
844         if (!old_linux) {
845                 int rv;
846
847 #ifndef MDASSEMBLE
848                 struct mdinfo *sra;
849                 if (st->ss->external) {
850                         char ver[100];
851                         strcat(strcpy(ver, "external:"), st->ss->text_version);
852                         sra = sysfs_read(mdfd, 0, 0);
853                         if ((vers % 100) < 2 ||
854                             sra == NULL ||
855                             sysfs_set_str(sra, NULL, "metadata_version",
856                                           ver) < 0) {
857                                 fprintf(stderr, Name ": This kernel does not "
858                                         "support external metadata.\n");
859                                 return 1;
860                         }
861                         rv = sysfs_set_array(sra, &info);
862                 } else
863 #endif
864                 if ((vers % 100) >= 1) { /* can use different versions */
865                         mdu_array_info_t inf;
866                         memset(&inf, 0, sizeof(inf));
867                         inf.major_version = st->ss->major;
868                         inf.minor_version = st->minor_version;
869                         rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
870                 } else
871                         rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
872
873                 if (rv) {
874                         fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
875                                 mddev, strerror(errno));
876                         if (must_close) close(mdfd);
877                         return 1;
878                 }
879                 if (ident->bitmap_fd >= 0) {
880                         if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
881                                 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
882                                 if (must_close) close(mdfd);
883                                 return 1;
884                         }
885                 } else if (ident->bitmap_file) {
886                         /* From config file */
887                         int bmfd = open(ident->bitmap_file, O_RDWR);
888                         if (bmfd < 0) {
889                                 fprintf(stderr, Name ": Could not open bitmap file %s\n",
890                                         ident->bitmap_file);
891                                 if (must_close) close(mdfd);
892                                 return 1;
893                         }
894                         if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
895                                 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
896                                 close(bmfd);
897                                 if (must_close) close(mdfd);
898                                 return 1;
899                         }
900                         close(bmfd);
901                 }
902
903                 /* First, add the raid disks, but add the chosen one last */
904                 for (i=0; i<= bestcnt; i++) {
905                         int j;
906                         if (i < bestcnt) {
907                                 j = best[i];
908                                 if (j == chosen_drive)
909                                         continue;
910                         } else
911                                 j = chosen_drive;
912
913                         if (j >= 0 /* && devices[j].uptodate */) {
914 #ifndef MDASSEMBLE
915                                 if (st->ss->external) {
916                                         int fd = dev_open(devices[j].devname,
917                                                           O_RDONLY);
918                                         if (fd < 0)
919                                                 rv = 1;
920                                         else {
921                                                 devices[j].i.disk.number =
922                                                         devices[j].i.disk.raid_disk;
923                                                 st->ss->getinfo_super_n(st,
924                                                                &devices[j].i);
925                                                 rv = sysfs_add_disk(sra, fd,
926                                                               &devices[j].i);
927                                                 close(fd);
928                                         }
929                                 } else
930 #endif
931                                         rv = ioctl(mdfd, ADD_NEW_DISK,
932                                           &devices[j].i.disk);
933                                 if (rv) {
934                                         fprintf(stderr, Name ": failed to add "
935                                                         "%s to %s: %s\n",
936                                                 devices[j].devname,
937                                                 mddev,
938                                                 strerror(errno));
939                                         if (i < info.array.raid_disks
940                                             || i == bestcnt)
941                                                 okcnt--;
942                                         else
943                                                 sparecnt--;
944                                 } else if (verbose > 0)
945                                         fprintf(stderr, Name ": added %s "
946                                                         "to %s as %d\n",
947                                                 devices[j].devname, mddev,
948                                                 devices[j].i.disk.raid_disk);
949                         } else if (verbose > 0 && i < info.array.raid_disks)
950                                 fprintf(stderr, Name ": no uptodate device for "
951                                                 "slot %d of %s\n",
952                                         i, mddev);
953                 }
954
955                 if (info.array.level == LEVEL_CONTAINER) {
956                         if (verbose >= 0) {
957                                 fprintf(stderr, Name ": Container %s has been "
958                                         "assembled with %d drive%s",
959                                         mddev, okcnt, okcnt==1?"":"s");
960                                 if (okcnt < info.array.raid_disks)
961                                         fprintf(stderr, " (out of %d)",
962                                                 info.array.raid_disks);
963                                 fprintf(stderr, "\n");
964                         }
965                         if (must_close)
966                                 close(mdfd);
967                         return 0;
968                 }
969
970                 if (runstop == 1 ||
971                     (runstop <= 0 &&
972                      ( enough(info.array.level, info.array.raid_disks,
973                               info.array.layout, clean, avail, okcnt) &&
974                        (okcnt >= req_cnt || start_partial_ok)
975                              ))) {
976                         if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
977                                 if (verbose >= 0) {
978                                         fprintf(stderr, Name ": %s has been started with %d drive%s",
979                                                 mddev, okcnt, okcnt==1?"":"s");
980                                         if (okcnt < info.array.raid_disks)
981                                                 fprintf(stderr, " (out of %d)", info.array.raid_disks);
982                                         if (sparecnt)
983                                                 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
984                                         fprintf(stderr, ".\n");
985                                 }
986                                 if (must_close) {
987                                         int usecs = 1;
988                                         close(mdfd);
989                                         /* There is a nasty race with 'mdadm --monitor'.
990                                          * If it opens this device before we close it,
991                                          * it gets an incomplete open on which IO
992                                          * doesn't work and the capacity is
993                                          * wrong.
994                                          * If we reopen (to check for layered devices)
995                                          * before --monitor closes, we loose.
996                                          *
997                                          * So: wait upto 1 second for there to be
998                                          * a non-zero capacity.
999                                          */
1000                                         while (usecs < 1000) {
1001                                                 mdfd = open(mddev, O_RDONLY);
1002                                                 if (mdfd >= 0) {
1003                                                         unsigned long long size;
1004                                                         if (get_dev_size(mdfd, NULL, &size) &&
1005                                                             size > 0)
1006                                                                 break;
1007                                                         close(mdfd);
1008                                                 }
1009                                                 usleep(usecs);
1010                                                 usecs <<= 1;
1011                                         }
1012                                 }
1013                                 return 0;
1014                         }
1015                         fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1016                                 mddev, strerror(errno));
1017
1018                         if (!enough(info.array.level, info.array.raid_disks,
1019                                     info.array.layout, 1, avail, okcnt))
1020                                 fprintf(stderr, Name ": Not enough devices to "
1021                                         "start the array.\n");
1022                         else if (!enough(info.array.level,
1023                                          info.array.raid_disks,
1024                                          info.array.layout, clean,
1025                                          avail, okcnt))
1026                                 fprintf(stderr, Name ": Not enough devices to "
1027                                         "start the array while not clean "
1028                                         "- consider --force.\n");
1029
1030                         if (must_close) close(mdfd);
1031                         return 1;
1032                 }
1033                 if (runstop == -1) {
1034                         fprintf(stderr, Name ": %s assembled from %d drive%s",
1035                                 mddev, okcnt, okcnt==1?"":"s");
1036                         if (okcnt != info.array.raid_disks)
1037                                 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1038                         fprintf(stderr, ", but not started.\n");
1039                         if (must_close) close(mdfd);
1040                         return 0;
1041                 }
1042                 if (verbose >= -1) {
1043                         fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1044                         if (sparecnt)
1045                                 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1046                         if (!enough(info.array.level, info.array.raid_disks,
1047                                     info.array.layout, 1, avail, okcnt))
1048                                 fprintf(stderr, " - not enough to start the array.\n");
1049                         else if (!enough(info.array.level,
1050                                          info.array.raid_disks,
1051                                          info.array.layout, clean,
1052                                          avail, okcnt))
1053                                 fprintf(stderr, " - not enough to start the "
1054                                         "array while not clean - consider "
1055                                         "--force.\n");
1056                         else {
1057                                 if (req_cnt == info.array.raid_disks)
1058                                         fprintf(stderr, " - need all %d to start it", req_cnt);
1059                                 else
1060                                         fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1061                                 fprintf(stderr, " (use --run to insist).\n");
1062                         }
1063                 }
1064                 if (must_close) close(mdfd);
1065                 return 1;
1066         } else {
1067                 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1068                  * been updated to point to the current locations of devices.
1069                  * so we can just start the array
1070                  */
1071                 unsigned long dev;
1072                 dev = makedev(devices[chosen_drive].i.disk.major,
1073                             devices[chosen_drive].i.disk.minor);
1074                 if (ioctl(mdfd, START_ARRAY, dev)) {
1075                     fprintf(stderr, Name ": Cannot start array: %s\n",
1076                             strerror(errno));
1077                 }
1078
1079         }
1080         if (must_close) close(mdfd);
1081         return 0;
1082 }