fs/gfs2/recovery.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/gfs2_ondisk.h>
  16 #include <linux/crc32.h>
  17 #include <linux/crc32c.h>
  18 #include <linux/ktime.h>
  19
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "bmap.h"
  23 #include "glock.h"
  24 #include "glops.h"
  25 #include "log.h"
  26 #include "lops.h"
  27 #include "meta_io.h"
  28 #include "recovery.h"
  29 #include "super.h"
  30 #include "util.h"
  31 #include "dir.h"
  32
  33 struct workqueue_struct *gfs_recovery_wq;
  34
  35 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
  36                            struct buffer_head **bh)
  37 {
  38         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
  39         struct gfs2_glock *gl = ip->i_gl;
  40         int new = 0;
  41         u64 dblock;
  42         u32 extlen;
  43         int error;
  44
  45         error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
  46         if (error)
  47                 return error;
  48         if (!dblock) {
  49                 gfs2_consist_inode(ip);
  50                 return -EIO;
  51         }
  52
  53         *bh = gfs2_meta_ra(gl, dblock, extlen);
  54
  55         return error;
  56 }
  57
  58 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
  59 {
  60         struct list_head *head = &jd->jd_revoke_list;
  61         struct gfs2_revoke_replay *rr;
  62         int found = 0;
  63
  64         list_for_each_entry(rr, head, rr_list) {
  65                 if (rr->rr_blkno == blkno) {
  66                         found = 1;
  67                         break;
  68                 }
  69         }
  70
  71         if (found) {
  72                 rr->rr_where = where;
  73                 return 0;
  74         }
  75
  76         rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
  77         if (!rr)
  78                 return -ENOMEM;
  79
  80         rr->rr_blkno = blkno;
  81         rr->rr_where = where;
  82         list_add(&rr->rr_list, head);
  83
  84         return 1;
  85 }
  86
  87 int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
  88 {
  89         struct gfs2_revoke_replay *rr;
  90         int wrap, a, b, revoke;
  91         int found = 0;
  92
  93         list_for_each_entry(rr, &jd->jd_revoke_list, rr_list) {
  94                 if (rr->rr_blkno == blkno) {
  95                         found = 1;
  96                         break;
  97                 }
  98         }
  99
 100         if (!found)
 101                 return 0;
 102
 103         wrap = (rr->rr_where < jd->jd_replay_tail);
 104         a = (jd->jd_replay_tail < where);
 105         b = (where < rr->rr_where);
 106         revoke = (wrap) ? (a || b) : (a && b);
 107
 108         return revoke;
 109 }
 110
 111 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
 112 {
 113         struct list_head *head = &jd->jd_revoke_list;
 114         struct gfs2_revoke_replay *rr;
 115
 116         while (!list_empty(head)) {
 117                 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
 118                 list_del(&rr->rr_list);
 119                 kfree(rr);
 120         }
 121 }
 122
 123 int __get_log_header(struct gfs2_sbd *sdp, const struct gfs2_log_header *lh,
 124                      unsigned int blkno, struct gfs2_log_header_host *head)
 125 {
 126         u32 hash, crc;
 127
 128         if (lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
 129             lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
 130             (blkno && be32_to_cpu(lh->lh_blkno) != blkno))
 131                 return 1;
 132
 133         hash = crc32(~0, lh, LH_V1_SIZE - 4);
 134         hash = ~crc32_le_shift(hash, 4); /* assume lh_hash is zero */
 135
 136         if (be32_to_cpu(lh->lh_hash) != hash)
 137                 return 1;
 138
 139         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 140                      sdp->sd_sb.sb_bsize - LH_V1_SIZE - 4);
 141
 142         if ((lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc))
 143                 return 1;
 144
 145         head->lh_sequence = be64_to_cpu(lh->lh_sequence);
 146         head->lh_flags = be32_to_cpu(lh->lh_flags);
 147         head->lh_tail = be32_to_cpu(lh->lh_tail);
 148         head->lh_blkno = be32_to_cpu(lh->lh_blkno);
 149
 150         return 0;
 151 }
 152 /**
 153  * get_log_header - read the log header for a given segment
 154  * @jd: the journal
 155  * @blk: the block to look at
 156  * @lh: the log header to return
 157  *
 158  * Read the log header for a given segement in a given journal.  Do a few
 159  * sanity checks on it.
 160  *
 161  * Returns: 0 on success,
 162  *          1 if the header was invalid or incomplete,
 163  *          errno on error
 164  */
 165
 166 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
 167                           struct gfs2_log_header_host *head)
 168 {
 169         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 170         struct buffer_head *bh;
 171         int error;
 172
 173         error = gfs2_replay_read_block(jd, blk, &bh);
 174         if (error)
 175                 return error;
 176
 177         error = __get_log_header(sdp, (const struct gfs2_log_header *)bh->b_data,
 178                                  blk, head);
 179         brelse(bh);
 180
 181         return error;
 182 }
 183
 184 /**
 185  * find_good_lh - find a good log header
 186  * @jd: the journal
 187  * @blk: the segment to start searching from
 188  * @lh: the log header to fill in
 189  * @forward: if true search forward in the log, else search backward
 190  *
 191  * Call get_log_header() to get a log header for a segment, but if the
 192  * segment is bad, either scan forward or backward until we find a good one.
 193  *
 194  * Returns: errno
 195  */
 196
 197 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
 198                         struct gfs2_log_header_host *head)
 199 {
 200         unsigned int orig_blk = *blk;
 201         int error;
 202
 203         for (;;) {
 204                 error = get_log_header(jd, *blk, head);
 205                 if (error <= 0)
 206                         return error;
 207
 208                 if (++*blk == jd->jd_blocks)
 209                         *blk = 0;
 210
 211                 if (*blk == orig_blk) {
 212                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 213                         return -EIO;
 214                 }
 215         }
 216 }
 217
 218 /**
 219  * jhead_scan - make sure we've found the head of the log
 220  * @jd: the journal
 221  * @head: this is filled in with the log descriptor of the head
 222  *
 223  * At this point, seg and lh should be either the head of the log or just
 224  * before.  Scan forward until we find the head.
 225  *
 226  * Returns: errno
 227  */
 228
 229 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
 230 {
 231         unsigned int blk = head->lh_blkno;
 232         struct gfs2_log_header_host lh;
 233         int error;
 234
 235         for (;;) {
 236                 if (++blk == jd->jd_blocks)
 237                         blk = 0;
 238
 239                 error = get_log_header(jd, blk, &lh);
 240                 if (error < 0)
 241                         return error;
 242                 if (error == 1)
 243                         continue;
 244
 245                 if (lh.lh_sequence == head->lh_sequence) {
 246                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 247                         return -EIO;
 248                 }
 249                 if (lh.lh_sequence < head->lh_sequence)
 250                         break;
 251
 252                 *head = lh;
 253         }
 254
 255         return 0;
 256 }
 257
 258 /**
 259  * gfs2_find_jhead - find the head of a log
 260  * @jd: the journal
 261  * @head: the log descriptor for the head of the log is returned here
 262  *
 263  * Do a binary search of a journal and find the valid log entry with the
 264  * highest sequence number.  (i.e. the log head)
 265  *
 266  * Returns: errno
 267  */
 268
 269 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
 270 {
 271         struct gfs2_log_header_host lh_1, lh_m;
 272         u32 blk_1, blk_2, blk_m;
 273         int error;
 274
 275         blk_1 = 0;
 276         blk_2 = jd->jd_blocks - 1;
 277
 278         for (;;) {
 279                 blk_m = (blk_1 + blk_2) / 2;
 280
 281                 error = find_good_lh(jd, &blk_1, &lh_1);
 282                 if (error)
 283                         return error;
 284
 285                 error = find_good_lh(jd, &blk_m, &lh_m);
 286                 if (error)
 287                         return error;
 288
 289                 if (blk_1 == blk_m || blk_m == blk_2)
 290                         break;
 291
 292                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
 293                         blk_1 = blk_m;
 294                 else
 295                         blk_2 = blk_m;
 296         }
 297
 298         error = jhead_scan(jd, &lh_1);
 299         if (error)
 300                 return error;
 301
 302         *head = lh_1;
 303
 304         return error;
 305 }
 306
 307 /**
 308  * foreach_descriptor - go through the active part of the log
 309  * @jd: the journal
 310  * @start: the first log header in the active region
 311  * @end: the last log header (don't process the contents of this entry))
 312  *
 313  * Call a given function once for every log descriptor in the active
 314  * portion of the log.
 315  *
 316  * Returns: errno
 317  */
 318
 319 static int foreach_descriptor(struct gfs2_jdesc *jd, u32 start,
 320                               unsigned int end, int pass)
 321 {
 322         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 323         struct buffer_head *bh;
 324         struct gfs2_log_descriptor *ld;
 325         int error = 0;
 326         u32 length;
 327         __be64 *ptr;
 328         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 329         offset += sizeof(__be64) - 1;
 330         offset &= ~(sizeof(__be64) - 1);
 331
 332         while (start != end) {
 333                 error = gfs2_replay_read_block(jd, start, &bh);
 334                 if (error)
 335                         return error;
 336                 if (gfs2_meta_check(sdp, bh)) {
 337                         brelse(bh);
 338                         return -EIO;
 339                 }
 340                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 341                 length = be32_to_cpu(ld->ld_length);
 342
 343                 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
 344                         struct gfs2_log_header_host lh;
 345                         error = get_log_header(jd, start, &lh);
 346                         if (!error) {
 347                                 gfs2_replay_incr_blk(jd, &start);
 348                                 brelse(bh);
 349                                 continue;
 350                         }
 351                         if (error == 1) {
 352                                 gfs2_consist_inode(GFS2_I(jd->jd_inode));
 353                                 error = -EIO;
 354                         }
 355                         brelse(bh);
 356                         return error;
 357                 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
 358                         brelse(bh);
 359                         return -EIO;
 360                 }
 361                 ptr = (__be64 *)(bh->b_data + offset);
 362                 error = lops_scan_elements(jd, start, ld, ptr, pass);
 363                 if (error) {
 364                         brelse(bh);
 365                         return error;
 366                 }
 367
 368                 while (length--)
 369                         gfs2_replay_incr_blk(jd, &start);
 370
 371                 brelse(bh);
 372         }
 373
 374         return 0;
 375 }
 376
 377 /**
 378  * clean_journal - mark a dirty journal as being clean
 379  * @jd: the journal
 380  * @head: the head journal to start from
 381  *
 382  * Returns: errno
 383  */
 384
 385 static void clean_journal(struct gfs2_jdesc *jd,
 386                           struct gfs2_log_header_host *head)
 387 {
 388         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 389         u32 lblock = head->lh_blkno;
 390
 391         gfs2_replay_incr_blk(jd, &lblock);
 392         if (jd->jd_jid == sdp->sd_lockstruct.ls_jid)
 393                 sdp->sd_log_flush_head = lblock;
 394         gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0, lblock,
 395                               GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
 396                               REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
 397 }
 398
 399
 400 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
 401                                unsigned int message)
 402 {
 403         char env_jid[20];
 404         char env_status[20];
 405         char *envp[] = { env_jid, env_status, NULL };
 406         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 407
 408         ls->ls_recover_jid_done = jid;
 409         ls->ls_recover_jid_status = message;
 410         sprintf(env_jid, "JID=%u", jid);
 411         sprintf(env_status, "RECOVERY=%s",
 412                 message == LM_RD_SUCCESS ? "Done" : "Failed");
 413         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 414
 415         if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
 416                 sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
 417 }
 418
 419 void gfs2_recover_func(struct work_struct *work)
 420 {
 421         struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
 422         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 423         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 424         struct gfs2_log_header_host head;
 425         struct gfs2_holder j_gh, ji_gh, thaw_gh;
 426         ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
 427         int ro = 0;
 428         unsigned int pass;
 429         int error = 0;
 430         int jlocked = 0;
 431
 432         t_start = ktime_get();
 433         if (sdp->sd_args.ar_spectator)
 434                 goto fail;
 435         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 436                 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
 437                         jd->jd_jid);
 438                 jlocked = 1;
 439                 /* Acquire the journal lock so we can do recovery */
 440
 441                 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
 442                                           LM_ST_EXCLUSIVE,
 443                                           LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
 444                                           &j_gh);
 445                 switch (error) {
 446                 case 0:
 447                         break;
 448
 449                 case GLR_TRYFAILED:
 450                         fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
 451                         error = 0;
 452
 453                 default:
 454                         goto fail;
 455                 };
 456
 457                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 458                                            LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
 459                 if (error)
 460                         goto fail_gunlock_j;
 461         } else {
 462                 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
 463         }
 464
 465         t_jlck = ktime_get();
 466         fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
 467
 468         error = gfs2_jdesc_check(jd);
 469         if (error)
 470                 goto fail_gunlock_ji;
 471
 472         error = gfs2_find_jhead(jd, &head);
 473         if (error)
 474                 goto fail_gunlock_ji;
 475         t_jhd = ktime_get();
 476         fs_info(sdp, "jid=%u: Journal head lookup took %lldms\n", jd->jd_jid,
 477                 ktime_ms_delta(t_jhd, t_jlck));
 478
 479         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 480                 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
 481                         jd->jd_jid);
 482
 483                 /* Acquire a shared hold on the freeze lock */
 484
 485                 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
 486                                            LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
 487                                            &thaw_gh);
 488                 if (error)
 489                         goto fail_gunlock_ji;
 490
 491                 if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
 492                         ro = 1;
 493                 } else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
 494                         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 495                                 ro = 1;
 496                 } else {
 497                         if (sb_rdonly(sdp->sd_vfs)) {
 498                                 /* check if device itself is read-only */
 499                                 ro = bdev_read_only(sdp->sd_vfs->s_bdev);
 500                                 if (!ro) {
 501                                         fs_info(sdp, "recovery required on "
 502                                                 "read-only filesystem.\n");
 503                                         fs_info(sdp, "write access will be "
 504                                                 "enabled during recovery.\n");
 505                                 }
 506                         }
 507                 }
 508
 509                 if (ro) {
 510                         fs_warn(sdp, "jid=%u: Can't replay: read-only block "
 511                                 "device\n", jd->jd_jid);
 512                         error = -EROFS;
 513                         goto fail_gunlock_thaw;
 514                 }
 515
 516                 t_tlck = ktime_get();
 517                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
 518
 519                 for (pass = 0; pass < 2; pass++) {
 520                         lops_before_scan(jd, &head, pass);
 521                         error = foreach_descriptor(jd, head.lh_tail,
 522                                                    head.lh_blkno, pass);
 523                         lops_after_scan(jd, error, pass);
 524                         if (error)
 525                                 goto fail_gunlock_thaw;
 526                 }
 527
 528                 clean_journal(jd, &head);
 529
 530                 gfs2_glock_dq_uninit(&thaw_gh);
 531                 t_rep = ktime_get();
 532                 fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
 533                         "jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
 534                         jd->jd_jid, ktime_ms_delta(t_rep, t_start),
 535                         ktime_ms_delta(t_jlck, t_start),
 536                         ktime_ms_delta(t_jhd, t_jlck),
 537                         ktime_ms_delta(t_tlck, t_jhd),
 538                         ktime_ms_delta(t_rep, t_tlck));
 539         }
 540
 541         gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
 542
 543         if (jlocked) {
 544                 gfs2_glock_dq_uninit(&ji_gh);
 545                 gfs2_glock_dq_uninit(&j_gh);
 546         }
 547
 548         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
 549         goto done;
 550
 551 fail_gunlock_thaw:
 552         gfs2_glock_dq_uninit(&thaw_gh);
 553 fail_gunlock_ji:
 554         if (jlocked) {
 555                 gfs2_glock_dq_uninit(&ji_gh);
 556 fail_gunlock_j:
 557                 gfs2_glock_dq_uninit(&j_gh);
 558         }
 559
 560         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
 561 fail:
 562         jd->jd_recover_error = error;
 563         gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
 564 done:
 565         clear_bit(JDF_RECOVERY, &jd->jd_flags);
 566         smp_mb__after_atomic();
 567         wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
 568 }
 569
 570 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
 571 {
 572         int rv;
 573
 574         if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
 575                 return -EBUSY;
 576
 577         /* we have JDF_RECOVERY, queue should always succeed */
 578         rv = queue_work(gfs_recovery_wq, &jd->jd_work);
 579         BUG_ON(!rv);
 580
 581         if (wait)
 582                 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
 583                             TASK_UNINTERRUPTIBLE);
 584
 585         return wait ? jd->jd_recover_error : 0;
 586 }
 587