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/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 #include <linux/lm_interface.h>
  18
  19 #include "gfs2.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         u64 dblock;
  39         u32 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         *bh = gfs2_meta_ra(gl, dblock, extlen);
  51
  52         return error;
  53 }
  54
  55 int gfs2_revoke_add(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
  56 {
  57         struct list_head *head = &sdp->sd_revoke_list;
  58         struct gfs2_revoke_replay *rr;
  59         int found = 0;
  60
  61         list_for_each_entry(rr, head, rr_list) {
  62                 if (rr->rr_blkno == blkno) {
  63                         found = 1;
  64                         break;
  65                 }
  66         }
  67
  68         if (found) {
  69                 rr->rr_where = where;
  70                 return 0;
  71         }
  72
  73         rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
  74         if (!rr)
  75                 return -ENOMEM;
  76
  77         rr->rr_blkno = blkno;
  78         rr->rr_where = where;
  79         list_add(&rr->rr_list, head);
  80
  81         return 1;
  82 }
  83
  84 int gfs2_revoke_check(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
  85 {
  86         struct gfs2_revoke_replay *rr;
  87         int wrap, a, b, revoke;
  88         int found = 0;
  89
  90         list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
  91                 if (rr->rr_blkno == blkno) {
  92                         found = 1;
  93                         break;
  94                 }
  95         }
  96
  97         if (!found)
  98                 return 0;
  99
 100         wrap = (rr->rr_where < sdp->sd_replay_tail);
 101         a = (sdp->sd_replay_tail < where);
 102         b = (where < rr->rr_where);
 103         revoke = (wrap) ? (a || b) : (a && b);
 104
 105         return revoke;
 106 }
 107
 108 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
 109 {
 110         struct list_head *head = &sdp->sd_revoke_list;
 111         struct gfs2_revoke_replay *rr;
 112
 113         while (!list_empty(head)) {
 114                 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
 115                 list_del(&rr->rr_list);
 116                 kfree(rr);
 117         }
 118 }
 119
 120 /**
 121  * get_log_header - read the log header for a given segment
 122  * @jd: the journal
 123  * @blk: the block to look at
 124  * @lh: the log header to return
 125  *
 126  * Read the log header for a given segement in a given journal.  Do a few
 127  * sanity checks on it.
 128  *
 129  * Returns: 0 on success,
 130  *          1 if the header was invalid or incomplete,
 131  *          errno on error
 132  */
 133
 134 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
 135                           struct gfs2_log_header *head)
 136 {
 137         struct buffer_head *bh;
 138         struct gfs2_log_header lh;
 139         u32 hash;
 140         int error;
 141
 142         error = gfs2_replay_read_block(jd, blk, &bh);
 143         if (error)
 144                 return error;
 145
 146         memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
 147         lh.lh_hash = 0;
 148         hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
 149         gfs2_log_header_in(&lh, bh->b_data);
 150
 151         brelse(bh);
 152
 153         if (lh.lh_header.mh_magic != GFS2_MAGIC ||
 154             lh.lh_header.mh_type != GFS2_METATYPE_LH ||
 155             lh.lh_blkno != blk || lh.lh_hash != hash)
 156                 return 1;
 157
 158         *head = lh;
 159
 160         return 0;
 161 }
 162
 163 /**
 164  * find_good_lh - find a good log header
 165  * @jd: the journal
 166  * @blk: the segment to start searching from
 167  * @lh: the log header to fill in
 168  * @forward: if true search forward in the log, else search backward
 169  *
 170  * Call get_log_header() to get a log header for a segment, but if the
 171  * segment is bad, either scan forward or backward until we find a good one.
 172  *
 173  * Returns: errno
 174  */
 175
 176 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
 177                         struct gfs2_log_header *head)
 178 {
 179         unsigned int orig_blk = *blk;
 180         int error;
 181
 182         for (;;) {
 183                 error = get_log_header(jd, *blk, head);
 184                 if (error <= 0)
 185                         return error;
 186
 187                 if (++*blk == jd->jd_blocks)
 188                         *blk = 0;
 189
 190                 if (*blk == orig_blk) {
 191                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 192                         return -EIO;
 193                 }
 194         }
 195 }
 196
 197 /**
 198  * jhead_scan - make sure we've found the head of the log
 199  * @jd: the journal
 200  * @head: this is filled in with the log descriptor of the head
 201  *
 202  * At this point, seg and lh should be either the head of the log or just
 203  * before.  Scan forward until we find the head.
 204  *
 205  * Returns: errno
 206  */
 207
 208 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 209 {
 210         unsigned int blk = head->lh_blkno;
 211         struct gfs2_log_header lh;
 212         int error;
 213
 214         for (;;) {
 215                 if (++blk == jd->jd_blocks)
 216                         blk = 0;
 217
 218                 error = get_log_header(jd, blk, &lh);
 219                 if (error < 0)
 220                         return error;
 221                 if (error == 1)
 222                         continue;
 223
 224                 if (lh.lh_sequence == head->lh_sequence) {
 225                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 226                         return -EIO;
 227                 }
 228                 if (lh.lh_sequence < head->lh_sequence)
 229                         break;
 230
 231                 *head = lh;
 232         }
 233
 234         return 0;
 235 }
 236
 237 /**
 238  * gfs2_find_jhead - find the head of a log
 239  * @jd: the journal
 240  * @head: the log descriptor for the head of the log is returned here
 241  *
 242  * Do a binary search of a journal and find the valid log entry with the
 243  * highest sequence number.  (i.e. the log head)
 244  *
 245  * Returns: errno
 246  */
 247
 248 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 249 {
 250         struct gfs2_log_header lh_1, lh_m;
 251         u32 blk_1, blk_2, blk_m;
 252         int error;
 253
 254         blk_1 = 0;
 255         blk_2 = jd->jd_blocks - 1;
 256
 257         for (;;) {
 258                 blk_m = (blk_1 + blk_2) / 2;
 259
 260                 error = find_good_lh(jd, &blk_1, &lh_1);
 261                 if (error)
 262                         return error;
 263
 264                 error = find_good_lh(jd, &blk_m, &lh_m);
 265                 if (error)
 266                         return error;
 267
 268                 if (blk_1 == blk_m || blk_m == blk_2)
 269                         break;
 270
 271                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
 272                         blk_1 = blk_m;
 273                 else
 274                         blk_2 = blk_m;
 275         }
 276
 277         error = jhead_scan(jd, &lh_1);
 278         if (error)
 279                 return error;
 280
 281         *head = lh_1;
 282
 283         return error;
 284 }
 285
 286 /**
 287  * foreach_descriptor - go through the active part of the log
 288  * @jd: the journal
 289  * @start: the first log header in the active region
 290  * @end: the last log header (don't process the contents of this entry))
 291  *
 292  * Call a given function once for every log descriptor in the active
 293  * portion of the log.
 294  *
 295  * Returns: errno
 296  */
 297
 298 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
 299                               unsigned int end, int pass)
 300 {
 301         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 302         struct buffer_head *bh;
 303         struct gfs2_log_descriptor *ld;
 304         int error = 0;
 305         u32 length;
 306         __be64 *ptr;
 307         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 308         offset += sizeof(__be64) - 1;
 309         offset &= ~(sizeof(__be64) - 1);
 310
 311         while (start != end) {
 312                 error = gfs2_replay_read_block(jd, start, &bh);
 313                 if (error)
 314                         return error;
 315                 if (gfs2_meta_check(sdp, bh)) {
 316                         brelse(bh);
 317                         return -EIO;
 318                 }
 319                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 320                 length = be32_to_cpu(ld->ld_length);
 321
 322                 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
 323                         struct gfs2_log_header lh;
 324                         error = get_log_header(jd, start, &lh);
 325                         if (!error) {
 326                                 gfs2_replay_incr_blk(sdp, &start);
 327                                 brelse(bh);
 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         struct gfs2_log_header *lh;
 372         u32 hash;
 373         struct buffer_head *bh;
 374         int error;
 375         struct buffer_head bh_map;
 376
 377         lblock = head->lh_blkno;
 378         gfs2_replay_incr_blk(sdp, &lblock);
 379         error = gfs2_block_map(&ip->i_inode, lblock, 0, &bh_map, 1);
 380         if (error)
 381                 return error;
 382         if (!bh_map.b_blocknr) {
 383                 gfs2_consist_inode(ip);
 384                 return -EIO;
 385         }
 386
 387         bh = sb_getblk(sdp->sd_vfs, bh_map.b_blocknr);
 388         lock_buffer(bh);
 389         memset(bh->b_data, 0, bh->b_size);
 390         set_buffer_uptodate(bh);
 391         clear_buffer_dirty(bh);
 392         unlock_buffer(bh);
 393
 394         lh = (struct gfs2_log_header *)bh->b_data;
 395         memset(lh, 0, sizeof(struct gfs2_log_header));
 396         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 397         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 398         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 399         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
 400         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
 401         lh->lh_blkno = cpu_to_be32(lblock);
 402         hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
 403         lh->lh_hash = cpu_to_be32(hash);
 404
 405         set_buffer_dirty(bh);
 406         if (sync_dirty_buffer(bh))
 407                 gfs2_io_error_bh(sdp, bh);
 408         brelse(bh);
 409
 410         return error;
 411 }
 412
 413 /**
 414  * gfs2_recover_journal - recovery a given journal
 415  * @jd: the struct gfs2_jdesc describing the journal
 416  *
 417  * Acquire the journal's lock, check to see if the journal is clean, and
 418  * do recovery if necessary.
 419  *
 420  * Returns: errno
 421  */
 422
 423 int gfs2_recover_journal(struct gfs2_jdesc *jd)
 424 {
 425         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 426         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 427         struct gfs2_log_header head;
 428         struct gfs2_holder j_gh, ji_gh, t_gh;
 429         unsigned long t;
 430         int ro = 0;
 431         unsigned int pass;
 432         int error;
 433
 434         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 435                 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
 436                         jd->jd_jid);
 437
 438                 /* Aquire the journal lock so we can do recovery */
 439
 440                 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
 441                                           LM_ST_EXCLUSIVE,
 442                                           LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
 443                                           &j_gh);
 444                 switch (error) {
 445                 case 0:
 446                         break;
 447
 448                 case GLR_TRYFAILED:
 449                         fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
 450                         error = 0;
 451
 452                 default:
 453                         goto fail;
 454                 };
 455
 456                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 457                                            LM_FLAG_NOEXP, &ji_gh);
 458                 if (error)
 459                         goto fail_gunlock_j;
 460         } else {
 461                 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
 462         }
 463
 464         fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
 465
 466         error = gfs2_jdesc_check(jd);
 467         if (error)
 468                 goto fail_gunlock_ji;
 469
 470         error = gfs2_find_jhead(jd, &head);
 471         if (error)
 472                 goto fail_gunlock_ji;
 473
 474         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 475                 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
 476                         jd->jd_jid);
 477
 478                 t = jiffies;
 479
 480                 /* Acquire a shared hold on the transaction lock */
 481
 482                 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 483                                            LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
 484                                            GL_NOCANCEL | GL_NOCACHE, &t_gh);
 485                 if (error)
 486                         goto fail_gunlock_ji;
 487
 488                 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
 489                         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 490                                 ro = 1;
 491                 } else {
 492                         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 493                                 ro = 1;
 494                 }
 495
 496                 if (ro) {
 497                         fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
 498                                 jd->jd_jid);
 499                         error = -EROFS;
 500                         goto fail_gunlock_tr;
 501                 }
 502
 503                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
 504
 505                 for (pass = 0; pass < 2; pass++) {
 506                         lops_before_scan(jd, &head, pass);
 507                         error = foreach_descriptor(jd, head.lh_tail,
 508                                                    head.lh_blkno, pass);
 509                         lops_after_scan(jd, error, pass);
 510                         if (error)
 511                                 goto fail_gunlock_tr;
 512                 }
 513
 514                 error = clean_journal(jd, &head);
 515                 if (error)
 516                         goto fail_gunlock_tr;
 517
 518                 gfs2_glock_dq_uninit(&t_gh);
 519                 t = DIV_ROUND_UP(jiffies - t, HZ);
 520                 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
 521                         jd->jd_jid, t);
 522         }
 523
 524         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 525                 gfs2_glock_dq_uninit(&ji_gh);
 526
 527         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
 528
 529         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 530                 gfs2_glock_dq_uninit(&j_gh);
 531
 532         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
 533         return 0;
 534
 535 fail_gunlock_tr:
 536         gfs2_glock_dq_uninit(&t_gh);
 537 fail_gunlock_ji:
 538         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 539                 gfs2_glock_dq_uninit(&ji_gh);
 540 fail_gunlock_j:
 541                 gfs2_glock_dq_uninit(&j_gh);
 542         }
 543
 544         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
 545
 546 fail:
 547         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
 548         return error;
 549 }
 550
 551 /**
 552  * gfs2_check_journals - Recover any dirty journals
 553  * @sdp: the filesystem
 554  *
 555  */
 556
 557 void gfs2_check_journals(struct gfs2_sbd *sdp)
 558 {
 559         struct gfs2_jdesc *jd;
 560
 561         for (;;) {
 562                 jd = gfs2_jdesc_find_dirty(sdp);
 563                 if (!jd)
 564                         break;
 565
 566                 if (jd != sdp->sd_jdesc)
 567                         gfs2_recover_journal(jd);
 568         }
 569 }
 570