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 ||
 157             lh.lh_hash != hash)
 158                 return 1;
 159
 160         *head = lh;
 161
 162         return 0;
 163 }
 164
 165 /**
 166  * find_good_lh - find a good log header
 167  * @jd: the journal
 168  * @blk: the segment to start searching from
 169  * @lh: the log header to fill in
 170  * @forward: if true search forward in the log, else search backward
 171  *
 172  * Call get_log_header() to get a log header for a segment, but if the
 173  * segment is bad, either scan forward or backward until we find a good one.
 174  *
 175  * Returns: errno
 176  */
 177
 178 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
 179                         struct gfs2_log_header *head)
 180 {
 181         unsigned int orig_blk = *blk;
 182         int error;
 183
 184         for (;;) {
 185                 error = get_log_header(jd, *blk, head);
 186                 if (error <= 0)
 187                         return error;
 188
 189                 if (++*blk == jd->jd_blocks)
 190                         *blk = 0;
 191
 192                 if (*blk == orig_blk) {
 193                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 194                         return -EIO;
 195                 }
 196         }
 197 }
 198
 199 /**
 200  * jhead_scan - make sure we've found the head of the log
 201  * @jd: the journal
 202  * @head: this is filled in with the log descriptor of the head
 203  *
 204  * At this point, seg and lh should be either the head of the log or just
 205  * before.  Scan forward until we find the head.
 206  *
 207  * Returns: errno
 208  */
 209
 210 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 211 {
 212         unsigned int blk = head->lh_blkno;
 213         struct gfs2_log_header lh;
 214         int error;
 215
 216         for (;;) {
 217                 if (++blk == jd->jd_blocks)
 218                         blk = 0;
 219
 220                 error = get_log_header(jd, blk, &lh);
 221                 if (error < 0)
 222                         return error;
 223                 if (error == 1)
 224                         continue;
 225
 226                 if (lh.lh_sequence == head->lh_sequence) {
 227                         gfs2_consist_inode(GFS2_I(jd->jd_inode));
 228                         return -EIO;
 229                 }
 230                 if (lh.lh_sequence < head->lh_sequence)
 231                         break;
 232
 233                 *head = lh;
 234         }
 235
 236         return 0;
 237 }
 238
 239 /**
 240  * gfs2_find_jhead - find the head of a log
 241  * @jd: the journal
 242  * @head: the log descriptor for the head of the log is returned here
 243  *
 244  * Do a binary search of a journal and find the valid log entry with the
 245  * highest sequence number.  (i.e. the log head)
 246  *
 247  * Returns: errno
 248  */
 249
 250 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 251 {
 252         struct gfs2_log_header lh_1, lh_m;
 253         uint32_t blk_1, blk_2, blk_m;
 254         int error;
 255
 256         blk_1 = 0;
 257         blk_2 = jd->jd_blocks - 1;
 258
 259         for (;;) {
 260                 blk_m = (blk_1 + blk_2) / 2;
 261
 262                 error = find_good_lh(jd, &blk_1, &lh_1);
 263                 if (error)
 264                         return error;
 265
 266                 error = find_good_lh(jd, &blk_m, &lh_m);
 267                 if (error)
 268                         return error;
 269
 270                 if (blk_1 == blk_m || blk_m == blk_2)
 271                         break;
 272
 273                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
 274                         blk_1 = blk_m;
 275                 else
 276                         blk_2 = blk_m;
 277         }
 278
 279         error = jhead_scan(jd, &lh_1);
 280         if (error)
 281                 return error;
 282
 283         *head = lh_1;
 284
 285         return error;
 286 }
 287
 288 /**
 289  * foreach_descriptor - go through the active part of the log
 290  * @jd: the journal
 291  * @start: the first log header in the active region
 292  * @end: the last log header (don't process the contents of this entry))
 293  *
 294  * Call a given function once for every log descriptor in the active
 295  * portion of the log.
 296  *
 297  * Returns: errno
 298  */
 299
 300 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
 301                               unsigned int end, int pass)
 302 {
 303         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 304         struct buffer_head *bh;
 305         struct gfs2_log_descriptor *ld;
 306         int error = 0;
 307         u32 length;
 308         __be64 *ptr;
 309         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 310         offset += (sizeof(__be64)-1);
 311         offset &= ~(sizeof(__be64)-1);
 312
 313         while (start != end) {
 314                 error = gfs2_replay_read_block(jd, start, &bh);
 315                 if (error)
 316                         return error;
 317                 if (gfs2_meta_check(sdp, bh)) {
 318                         brelse(bh);
 319                         return -EIO;
 320                 }
 321                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 322                 length = be32_to_cpu(ld->ld_length);
 323
 324                 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
 325                         struct gfs2_log_header lh;
 326                         error = get_log_header(jd, start, &lh);
 327                         if (!error) {
 328                                 gfs2_replay_incr_blk(sdp, &start);
 329                                 continue;
 330                         }
 331                         if (error == 1) {
 332                                 gfs2_consist_inode(GFS2_I(jd->jd_inode));
 333                                 error = -EIO;
 334                         }
 335                         brelse(bh);
 336                         return error;
 337                 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
 338                         brelse(bh);
 339                         return -EIO;
 340                 }
 341                 ptr = (__be64 *)(bh->b_data + offset);
 342                 error = lops_scan_elements(jd, start, ld, ptr, pass);
 343                 if (error) {
 344                         brelse(bh);
 345                         return error;
 346                 }
 347
 348                 while (length--)
 349                         gfs2_replay_incr_blk(sdp, &start);
 350
 351                 brelse(bh);
 352         }
 353
 354         return 0;
 355 }
 356
 357 /**
 358  * clean_journal - mark a dirty journal as being clean
 359  * @sdp: the filesystem
 360  * @jd: the journal
 361  * @gl: the journal's glock
 362  * @head: the head journal to start from
 363  *
 364  * Returns: errno
 365  */
 366
 367 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 368 {
 369         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 370         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 371         unsigned int lblock;
 372         int new = 0;
 373         uint64_t dblock;
 374         struct gfs2_log_header *lh;
 375         uint32_t hash;
 376         struct buffer_head *bh;
 377         int error;
 378         int boundary;
 379
 380         lblock = head->lh_blkno;
 381         gfs2_replay_incr_blk(sdp, &lblock);
 382         error = gfs2_block_map(&ip->i_inode, lblock, &new, &dblock, &boundary);
 383         if (error)
 384                 return error;
 385         if (!dblock) {
 386                 gfs2_consist_inode(ip);
 387                 return -EIO;
 388         }
 389
 390         bh = sb_getblk(sdp->sd_vfs, dblock);
 391         lock_buffer(bh);
 392         memset(bh->b_data, 0, bh->b_size);
 393         set_buffer_uptodate(bh);
 394         clear_buffer_dirty(bh);
 395         unlock_buffer(bh);
 396
 397         lh = (struct gfs2_log_header *)bh->b_data;
 398         memset(lh, 0, sizeof(struct gfs2_log_header));
 399         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 400         lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
 401         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 402         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
 403         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
 404         lh->lh_blkno = cpu_to_be32(lblock);
 405         hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
 406         lh->lh_hash = cpu_to_be32(hash);
 407
 408         set_buffer_dirty(bh);
 409         if (sync_dirty_buffer(bh))
 410                 gfs2_io_error_bh(sdp, bh);
 411         brelse(bh);
 412
 413         return error;
 414 }
 415
 416 /**
 417  * gfs2_recover_journal - recovery a given journal
 418  * @jd: the struct gfs2_jdesc describing the journal
 419  *
 420  * Acquire the journal's lock, check to see if the journal is clean, and
 421  * do recovery if necessary.
 422  *
 423  * Returns: errno
 424  */
 425
 426 int gfs2_recover_journal(struct gfs2_jdesc *jd)
 427 {
 428         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 429         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 430         struct gfs2_log_header head;
 431         struct gfs2_holder j_gh, ji_gh, t_gh;
 432         unsigned long t;
 433         int ro = 0;
 434         unsigned int pass;
 435         int error;
 436
 437         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 438                 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
 439                         jd->jd_jid);
 440
 441                 /* Aquire the journal lock so we can do recovery */
 442
 443                 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
 444                                           LM_ST_EXCLUSIVE,
 445                                           LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
 446                                           &j_gh);
 447                 switch (error) {
 448                 case 0:
 449                         break;
 450
 451                 case GLR_TRYFAILED:
 452                         fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
 453                         error = 0;
 454
 455                 default:
 456                         goto fail;
 457                 };
 458
 459                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 460                                            LM_FLAG_NOEXP, &ji_gh);
 461                 if (error)
 462                         goto fail_gunlock_j;
 463         } else {
 464                 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
 465         }
 466
 467         fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
 468
 469         error = gfs2_jdesc_check(jd);
 470         if (error)
 471                 goto fail_gunlock_ji;
 472
 473         error = gfs2_find_jhead(jd, &head);
 474         if (error)
 475                 goto fail_gunlock_ji;
 476
 477         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 478                 fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
 479                         jd->jd_jid);
 480
 481                 t = jiffies;
 482
 483                 /* Acquire a shared hold on the transaction lock */
 484
 485                 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
 486                                            LM_ST_SHARED,
 487                                            LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
 488                                            GL_NOCANCEL | GL_NOCACHE,
 489                                            &t_gh);
 490                 if (error)
 491                         goto fail_gunlock_ji;
 492
 493                 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 (sdp->sd_vfs->s_flags & MS_RDONLY)
 498                                 ro = 1;
 499                 }
 500
 501                 if (ro) {
 502                         fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
 503                                 jd->jd_jid);
 504                         error = -EROFS;
 505                         goto fail_gunlock_tr;
 506                 }
 507
 508                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
 509
 510                 for (pass = 0; pass < 2; pass++) {
 511                         lops_before_scan(jd, &head, pass);
 512                         error = foreach_descriptor(jd, head.lh_tail,
 513                                                    head.lh_blkno, pass);
 514                         lops_after_scan(jd, error, pass);
 515                         if (error)
 516                                 goto fail_gunlock_tr;
 517                 }
 518
 519                 error = clean_journal(jd, &head);
 520                 if (error)
 521                         goto fail_gunlock_tr;
 522
 523                 gfs2_glock_dq_uninit(&t_gh);
 524                 t = DIV_ROUND_UP(jiffies - t, HZ);
 525                 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
 526                         jd->jd_jid, t);
 527         }
 528
 529         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 530                 gfs2_glock_dq_uninit(&ji_gh);
 531
 532         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
 533
 534         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 535                 gfs2_glock_dq_uninit(&j_gh);
 536
 537         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
 538         return 0;
 539
 540 fail_gunlock_tr:
 541         gfs2_glock_dq_uninit(&t_gh);
 542 fail_gunlock_ji:
 543         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 544                 gfs2_glock_dq_uninit(&ji_gh);
 545 fail_gunlock_j:
 546                 gfs2_glock_dq_uninit(&j_gh);
 547         }
 548
 549         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
 550
 551 fail:
 552         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
 553         return error;
 554 }
 555
 556 /**
 557  * gfs2_check_journals - Recover any dirty journals
 558  * @sdp: the filesystem
 559  *
 560  */
 561
 562 void gfs2_check_journals(struct gfs2_sbd *sdp)
 563 {
 564         struct gfs2_jdesc *jd;
 565
 566         for (;;) {
 567                 jd = gfs2_jdesc_find_dirty(sdp);
 568                 if (!jd)
 569                         break;
 570
 571                 if (jd != sdp->sd_jdesc)
 572                         gfs2_recover_journal(jd);
 573         }
 574 }
 575