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 v.2.
   8  */
   9
  10 #include <linux/sched.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
  18 #include "gfs2.h"
  19 #include "lm_interface.h"
  20 #include "incore.h"
  21 #include "bmap.h"
  22 #include "glock.h"
  23 #include "glops.h"
  24 #include "lm.h"
  25 #include "lops.h"
  26 #include "meta_io.h"
  27 #include "recovery.h"
  28 #include "super.h"
  29 #include "util.h"
  30 #include "dir.h"
  31
  32 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
  33                            struct buffer_head **bh)
  34 {
  35         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
  36         struct gfs2_glock *gl = ip->i_gl;
  37         int new = 0;
  38         uint64_t dblock;
  39         uint32_t extlen;
  40         int error;
  41
  42         error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
  43         if (error)
  44                 return error;
  45         if (!dblock) {
  46                 gfs2_consist_inode(ip);
  47                 return -EIO;
  48         }
  49
  50         gfs2_meta_ra(gl, dblock, extlen);
  51         error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
  52
  53         return error;
  54 }
  55
  56 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
  57 {
  58         struct list_head *head = &sdp->sd_revoke_list;
  59         struct gfs2_revoke_replay *rr;
  60         int found = 0;
  61
  62         list_for_each_entry(rr, head, rr_list) {
  63                 if (rr->rr_blkno == blkno) {
  64                         found = 1;
  65                         break;
  66                 }
  67         }
  68
  69         if (found) {
  70                 rr->rr_where = where;
  71                 return 0;
  72         }
  73
  74         rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
  75         if (!rr)
  76                 return -ENOMEM;
  77
  78         rr->rr_blkno = blkno;
  79         rr->rr_where = where;
  80         list_add(&rr->rr_list, head);
  81
  82         return 1;
  83 }
  84
  85 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
  86 {
  87         struct gfs2_revoke_replay *rr;
  88         int wrap, a, b, revoke;
  89         int found = 0;
  90
  91         list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
  92                 if (rr->rr_blkno == blkno) {
  93                         found = 1;
  94                         break;
  95                 }
  96         }
  97
  98         if (!found)
  99                 return 0;
 100
 101         wrap = (rr->rr_where < sdp->sd_replay_tail);
 102         a = (sdp->sd_replay_tail < where);
 103         b = (where < rr->rr_where);
 104         revoke = (wrap) ? (a || b) : (a && b);
 105
 106         return revoke;
 107 }
 108
 109 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
 110 {
 111         struct list_head *head = &sdp->sd_revoke_list;
 112         struct gfs2_revoke_replay *rr;
 113
 114         while (!list_empty(head)) {
 115                 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
 116                 list_del(&rr->rr_list);
 117                 kfree(rr);
 118         }
 119 }
 120
 121 /**
 122  * get_log_header - read the log header for a given segment
 123  * @jd: the journal
 124  * @blk: the block to look at
 125  * @lh: the log header to return
 126  *
 127  * Read the log header for a given segement in a given journal.  Do a few
 128  * sanity checks on it.
 129  *
 130  * Returns: 0 on success,
 131  *          1 if the header was invalid or incomplete,
 132  *          errno on error
 133  */
 134
 135 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
 136                           struct gfs2_log_header *head)
 137 {
 138         struct buffer_head *bh;
 139         struct gfs2_log_header lh;
 140         uint32_t hash;
 141         int error;
 142
 143         error = gfs2_replay_read_block(jd, blk, &bh);
 144         if (error)
 145                 return error;
 146
 147         memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
 148         lh.lh_hash = 0;
 149         hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
 150         gfs2_log_header_in(&lh, bh->b_data);
 151
 152         brelse(bh);
 153
 154         if (lh.lh_header.mh_magic != GFS2_MAGIC ||
 155             lh.lh_header.mh_type != GFS2_METATYPE_LH ||
 156             lh.lh_blkno != blk || lh.lh_hash != hash)
 157                 return 1;
 158
 159         *head = lh;
 160
 161         return 0;
 162 }
 163
 164 /**
 165  * find_good_lh - find a good log header
 166  * @jd: the journal
 167  * @blk: the segment to start searching from
 168  * @lh: the log header to fill in
 169  * @forward: if true search forward in the log, else search backward
 170  *
 171  * Call get_log_header() to get a log header for a segment, but if the
 172  * segment is bad, either scan forward or backward until we find a good one.
 173  *
 174  * Returns: errno
 175  */
 176
 177 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
 178                         struct gfs2_log_header *head)
 179 {
 180         unsigned int orig_blk = *blk;
 181         int error;
 182
 183         for (;;) {
 184                 error = get_log_header(jd, *blk, head);
 185                 if (error <= 0)
 186                         return error;
 187
 188                 if (++*blk == jd->jd_blocks)
 189                         *blk = 0;
 190
 191                 if (*blk == orig_blk) {
 192                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 193                         return -EIO;
 194                 }
 195         }
 196 }
 197
 198 /**
 199  * jhead_scan - make sure we've found the head of the log
 200  * @jd: the journal
 201  * @head: this is filled in with the log descriptor of the head
 202  *
 203  * At this point, seg and lh should be either the head of the log or just
 204  * before.  Scan forward until we find the head.
 205  *
 206  * Returns: errno
 207  */
 208
 209 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 210 {
 211         unsigned int blk = head->lh_blkno;
 212         struct gfs2_log_header lh;
 213         int error;
 214
 215         for (;;) {
 216                 if (++blk == jd->jd_blocks)
 217                         blk = 0;
 218
 219                 error = get_log_header(jd, blk, &lh);
 220                 if (error < 0)
 221                         return error;
 222                 if (error == 1)
 223                         continue;
 224
 225                 if (lh.lh_sequence == head->lh_sequence) {
 226                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 227                         return -EIO;
 228                 }
 229                 if (lh.lh_sequence < head->lh_sequence)
 230                         break;
 231
 232                 *head = lh;
 233         }
 234
 235         return 0;
 236 }
 237
 238 /**
 239  * gfs2_find_jhead - find the head of a log
 240  * @jd: the journal
 241  * @head: the log descriptor for the head of the log is returned here
 242  *
 243  * Do a binary search of a journal and find the valid log entry with the
 244  * highest sequence number.  (i.e. the log head)
 245  *
 246  * Returns: errno
 247  */
 248
 249 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 250 {
 251         struct gfs2_log_header lh_1, lh_m;
 252         uint32_t blk_1, blk_2, blk_m;
 253         int error;
 254
 255         blk_1 = 0;
 256         blk_2 = jd->jd_blocks - 1;
 257
 258         for (;;) {
 259                 blk_m = (blk_1 + blk_2) / 2;
 260
 261                 error = find_good_lh(jd, &blk_1, &lh_1);
 262                 if (error)
 263                         return error;
 264
 265                 error = find_good_lh(jd, &blk_m, &lh_m);
 266                 if (error)
 267                         return error;
 268
 269                 if (blk_1 == blk_m || blk_m == blk_2)
 270                         break;
 271
 272                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
 273                         blk_1 = blk_m;
 274                 else
 275                         blk_2 = blk_m;
 276         }
 277
 278         error = jhead_scan(jd, &lh_1);
 279         if (error)
 280                 return error;
 281
 282         *head = lh_1;
 283
 284         return error;
 285 }
 286
 287 /**
 288  * foreach_descriptor - go through the active part of the log
 289  * @jd: the journal
 290  * @start: the first log header in the active region
 291  * @end: the last log header (don't process the contents of this entry))
 292  *
 293  * Call a given function once for every log descriptor in the active
 294  * portion of the log.
 295  *
 296  * Returns: errno
 297  */
 298
 299 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
 300                               unsigned int end, int pass)
 301 {
 302         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 303         struct buffer_head *bh;
 304         struct gfs2_log_descriptor *ld;
 305         int error = 0;
 306         u32 length;
 307         __be64 *ptr;
 308         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 309         offset += (sizeof(__be64)-1);
 310         offset &= ~(sizeof(__be64)-1);
 311
 312         while (start != end) {
 313                 error = gfs2_replay_read_block(jd, start, &bh);
 314                 if (error)
 315                         return error;
 316                 if (gfs2_meta_check(sdp, bh)) {
 317                         brelse(bh);
 318                         return -EIO;
 319                 }
 320                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 321                 length = be32_to_cpu(ld->ld_length);
 322
 323                 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
 324                         struct gfs2_log_header lh;
 325                         error = get_log_header(jd, start, &lh);
 326                         if (!error) {
 327                                 gfs2_replay_incr_blk(sdp, &start);
 328                                 continue;
 329                         }
 330                         if (error == 1) {
 331                                 gfs2_consist_inode(GFS2_I(jd->jd_inode));
 332                                 error = -EIO;
 333                         }
 334                         brelse(bh);
 335                         return error;
 336                 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
 337                         brelse(bh);
 338                         return -EIO;
 339                 }
 340                 ptr = (__be64 *)(bh->b_data + offset);
 341                 error = lops_scan_elements(jd, start, ld, ptr, pass);
 342                 if (error) {
 343                         brelse(bh);
 344                         return error;
 345                 }
 346
 347                 while (length--)
 348                         gfs2_replay_incr_blk(sdp, &start);
 349
 350                 brelse(bh);
 351         }
 352
 353         return 0;
 354 }
 355
 356 /**
 357  * clean_journal - mark a dirty journal as being clean
 358  * @sdp: the filesystem
 359  * @jd: the journal
 360  * @gl: the journal's glock
 361  * @head: the head journal to start from
 362  *
 363  * Returns: errno
 364  */
 365
 366 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 367 {
 368         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 369         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 370         unsigned int lblock;
 371         int new = 0;
 372         uint64_t dblock;
 373         struct gfs2_log_header *lh;
 374         uint32_t hash;
 375         struct buffer_head *bh;
 376         int error;
 377         int boundary;
 378
 379         lblock = head->lh_blkno;
 380         gfs2_replay_incr_blk(sdp, &lblock);
 381         error = gfs2_block_map(&ip->i_inode, lblock, &new, &dblock, &boundary);
 382         if (error)
 383                 return error;
 384         if (!dblock) {
 385                 gfs2_consist_inode(ip);
 386                 return -EIO;
 387         }
 388
 389         bh = sb_getblk(sdp->sd_vfs, dblock);
 390         lock_buffer(bh);
 391         memset(bh->b_data, 0, bh->b_size);
 392         set_buffer_uptodate(bh);
 393         clear_buffer_dirty(bh);
 394         unlock_buffer(bh);
 395
 396         lh = (struct gfs2_log_header *)bh->b_data;
 397         memset(lh, 0, sizeof(struct gfs2_log_header));
 398         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 399         lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
 400         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 401         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
 402         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
 403         lh->lh_blkno = cpu_to_be32(lblock);
 404         hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
 405         lh->lh_hash = cpu_to_be32(hash);
 406
 407         set_buffer_dirty(bh);
 408         if (sync_dirty_buffer(bh))
 409                 gfs2_io_error_bh(sdp, bh);
 410         brelse(bh);
 411
 412         return error;
 413 }
 414
 415 /**
 416  * gfs2_recover_journal - recovery a given journal
 417  * @jd: the struct gfs2_jdesc describing the journal
 418  *
 419  * Acquire the journal's lock, check to see if the journal is clean, and
 420  * do recovery if necessary.
 421  *
 422  * Returns: errno
 423  */
 424
 425 int gfs2_recover_journal(struct gfs2_jdesc *jd)
 426 {
 427         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 428         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 429         struct gfs2_log_header head;
 430         struct gfs2_holder j_gh, ji_gh, t_gh;
 431         unsigned long t;
 432         int ro = 0;
 433         unsigned int pass;
 434         int error;
 435
 436         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 437                 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
 438                         jd->jd_jid);
 439
 440                 /* Aquire the journal lock so we can do recovery */
 441
 442                 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
 443                                           LM_ST_EXCLUSIVE,
 444                                           LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
 445                                           &j_gh);
 446                 switch (error) {
 447                 case 0:
 448                         break;
 449
 450                 case GLR_TRYFAILED:
 451                         fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
 452                         error = 0;
 453
 454                 default:
 455                         goto fail;
 456                 };
 457
 458                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 459                                            LM_FLAG_NOEXP, &ji_gh);
 460                 if (error)
 461                         goto fail_gunlock_j;
 462         } else {
 463                 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
 464         }
 465
 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
 476         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 477                 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
 478                         jd->jd_jid);
 479
 480                 t = jiffies;
 481
 482                 /* Acquire a shared hold on the transaction lock */
 483
 484                 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 485                                            LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
 486                                            GL_NOCANCEL | GL_NOCACHE, &t_gh);
 487                 if (error)
 488                         goto fail_gunlock_ji;
 489
 490                 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
 491                         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 492                                 ro = 1;
 493                 } else {
 494                         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 495                                 ro = 1;
 496                 }
 497
 498                 if (ro) {
 499                         fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
 500                                 jd->jd_jid);
 501                         error = -EROFS;
 502                         goto fail_gunlock_tr;
 503                 }
 504
 505                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
 506
 507                 for (pass = 0; pass < 2; pass++) {
 508                         lops_before_scan(jd, &head, pass);
 509                         error = foreach_descriptor(jd, head.lh_tail,
 510                                                    head.lh_blkno, pass);
 511                         lops_after_scan(jd, error, pass);
 512                         if (error)
 513                                 goto fail_gunlock_tr;
 514                 }
 515
 516                 error = clean_journal(jd, &head);
 517                 if (error)
 518                         goto fail_gunlock_tr;
 519
 520                 gfs2_glock_dq_uninit(&t_gh);
 521                 t = DIV_ROUND_UP(jiffies - t, HZ);
 522                 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
 523                         jd->jd_jid, t);
 524         }
 525
 526         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 527                 gfs2_glock_dq_uninit(&ji_gh);
 528
 529         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
 530
 531         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 532                 gfs2_glock_dq_uninit(&j_gh);
 533
 534         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
 535         return 0;
 536
 537 fail_gunlock_tr:
 538         gfs2_glock_dq_uninit(&t_gh);
 539 fail_gunlock_ji:
 540         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 541                 gfs2_glock_dq_uninit(&ji_gh);
 542 fail_gunlock_j:
 543                 gfs2_glock_dq_uninit(&j_gh);
 544         }
 545
 546         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
 547
 548 fail:
 549         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
 550         return error;
 551 }
 552
 553 /**
 554  * gfs2_check_journals - Recover any dirty journals
 555  * @sdp: the filesystem
 556  *
 557  */
 558
 559 void gfs2_check_journals(struct gfs2_sbd *sdp)
 560 {
 561         struct gfs2_jdesc *jd;
 562
 563         for (;;) {
 564                 jd = gfs2_jdesc_find_dirty(sdp);
 565                 if (!jd)
 566                         break;
 567
 568                 if (jd != sdp->sd_jdesc)
 569                         gfs2_recover_journal(jd);
 570         }
 571 }
 572