fs/gfs2/recovery.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 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 #include <asm/semaphore.h>
  18
  19 #include "gfs2.h"
  20 #include "lm_interface.h"
  21 #include "incore.h"
  22 #include "bmap.h"
  23 #include "glock.h"
  24 #include "glops.h"
  25 #include "lm.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 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
  34                            struct buffer_head **bh)
  35 {
  36         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
  37         struct gfs2_glock *gl = ip->i_gl;
  38         int new = 0;
  39         uint64_t dblock;
  40         uint32_t extlen;
  41         int error;
  42
  43         error = gfs2_extent_map(ip->i_vnode, blk, &new, &dblock, &extlen);
  44         if (error)
  45                 return error;
  46         if (!dblock) {
  47                 gfs2_consist_inode(ip);
  48                 return -EIO;
  49         }
  50
  51         gfs2_meta_ra(gl, dblock, extlen);
  52         error = gfs2_meta_read(gl, dblock, DIO_START | DIO_WAIT, bh);
  53
  54         return error;
  55 }
  56
  57 int gfs2_revoke_add(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
  58 {
  59         struct list_head *head = &sdp->sd_revoke_list;
  60         struct gfs2_revoke_replay *rr;
  61         int found = 0;
  62
  63         list_for_each_entry(rr, head, rr_list) {
  64                 if (rr->rr_blkno == blkno) {
  65                         found = 1;
  66                         break;
  67                 }
  68         }
  69
  70         if (found) {
  71                 rr->rr_where = where;
  72                 return 0;
  73         }
  74
  75         rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_KERNEL);
  76         if (!rr)
  77                 return -ENOMEM;
  78
  79         rr->rr_blkno = blkno;
  80         rr->rr_where = where;
  81         list_add(&rr->rr_list, head);
  82
  83         return 1;
  84 }
  85
  86 int gfs2_revoke_check(struct gfs2_sbd *sdp, uint64_t blkno, unsigned int where)
  87 {
  88         struct gfs2_revoke_replay *rr;
  89         int wrap, a, b, revoke;
  90         int found = 0;
  91
  92         list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
  93                 if (rr->rr_blkno == blkno) {
  94                         found = 1;
  95                         break;
  96                 }
  97         }
  98
  99         if (!found)
 100                 return 0;
 101
 102         wrap = (rr->rr_where < sdp->sd_replay_tail);
 103         a = (sdp->sd_replay_tail < where);
 104         b = (where < rr->rr_where);
 105         revoke = (wrap) ? (a || b) : (a && b);
 106
 107         return revoke;
 108 }
 109
 110 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
 111 {
 112         struct list_head *head = &sdp->sd_revoke_list;
 113         struct gfs2_revoke_replay *rr;
 114
 115         while (!list_empty(head)) {
 116                 rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
 117                 list_del(&rr->rr_list);
 118                 kfree(rr);
 119         }
 120 }
 121
 122 /**
 123  * get_log_header - read the log header for a given segment
 124  * @jd: the journal
 125  * @blk: the block to look at
 126  * @lh: the log header to return
 127  *
 128  * Read the log header for a given segement in a given journal.  Do a few
 129  * sanity checks on it.
 130  *
 131  * Returns: 0 on success,
 132  *          1 if the header was invalid or incomplete,
 133  *          errno on error
 134  */
 135
 136 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
 137                           struct gfs2_log_header *head)
 138 {
 139         struct buffer_head *bh;
 140         struct gfs2_log_header lh;
 141         uint32_t hash;
 142         int error;
 143
 144         error = gfs2_replay_read_block(jd, blk, &bh);
 145         if (error)
 146                 return error;
 147
 148         memcpy(&lh, bh->b_data, sizeof(struct gfs2_log_header));
 149         lh.lh_hash = 0;
 150         hash = gfs2_disk_hash((char *)&lh, sizeof(struct gfs2_log_header));
 151         gfs2_log_header_in(&lh, bh->b_data);
 152
 153         brelse(bh);
 154
 155         if (lh.lh_header.mh_magic != GFS2_MAGIC ||
 156             lh.lh_header.mh_type != GFS2_METATYPE_LH ||
 157             lh.lh_blkno != blk ||
 158             lh.lh_hash != hash)
 159                 return 1;
 160
 161         *head = lh;
 162
 163         return 0;
 164 }
 165
 166 /**
 167  * find_good_lh - find a good log header
 168  * @jd: the journal
 169  * @blk: the segment to start searching from
 170  * @lh: the log header to fill in
 171  * @forward: if true search forward in the log, else search backward
 172  *
 173  * Call get_log_header() to get a log header for a segment, but if the
 174  * segment is bad, either scan forward or backward until we find a good one.
 175  *
 176  * Returns: errno
 177  */
 178
 179 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
 180                         struct gfs2_log_header *head)
 181 {
 182         unsigned int orig_blk = *blk;
 183         int error;
 184
 185         for (;;) {
 186                 error = get_log_header(jd, *blk, head);
 187                 if (error <= 0)
 188                         return error;
 189
 190                 if (++*blk == jd->jd_blocks)
 191                         *blk = 0;
 192
 193                 if (*blk == orig_blk) {
 194                         gfs2_consist_inode(jd->jd_inode->u.generic_ip);
 195                         return -EIO;
 196                 }
 197         }
 198 }
 199
 200 /**
 201  * jhead_scan - make sure we've found the head of the log
 202  * @jd: the journal
 203  * @head: this is filled in with the log descriptor of the head
 204  *
 205  * At this point, seg and lh should be either the head of the log or just
 206  * before.  Scan forward until we find the head.
 207  *
 208  * Returns: errno
 209  */
 210
 211 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 212 {
 213         unsigned int blk = head->lh_blkno;
 214         struct gfs2_log_header lh;
 215         int error;
 216
 217         for (;;) {
 218                 if (++blk == jd->jd_blocks)
 219                         blk = 0;
 220
 221                 error = get_log_header(jd, blk, &lh);
 222                 if (error < 0)
 223                         return error;
 224                 if (error == 1)
 225                         continue;
 226
 227                 if (lh.lh_sequence == head->lh_sequence) {
 228                         gfs2_consist_inode(jd->jd_inode->u.generic_ip);
 229                         return -EIO;
 230                 }
 231                 if (lh.lh_sequence < head->lh_sequence)
 232                         break;
 233
 234                 *head = lh;
 235         }
 236
 237         return 0;
 238 }
 239
 240 /**
 241  * gfs2_find_jhead - find the head of a log
 242  * @jd: the journal
 243  * @head: the log descriptor for the head of the log is returned here
 244  *
 245  * Do a binary search of a journal and find the valid log entry with the
 246  * highest sequence number.  (i.e. the log head)
 247  *
 248  * Returns: errno
 249  */
 250
 251 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 252 {
 253         struct gfs2_log_header lh_1, lh_m;
 254         uint32_t blk_1, blk_2, blk_m;
 255         int error;
 256
 257         blk_1 = 0;
 258         blk_2 = jd->jd_blocks - 1;
 259
 260         for (;;) {
 261                 blk_m = (blk_1 + blk_2) / 2;
 262
 263                 error = find_good_lh(jd, &blk_1, &lh_1);
 264                 if (error)
 265                         return error;
 266
 267                 error = find_good_lh(jd, &blk_m, &lh_m);
 268                 if (error)
 269                         return error;
 270
 271                 if (blk_1 == blk_m || blk_m == blk_2)
 272                         break;
 273
 274                 if (lh_1.lh_sequence <= lh_m.lh_sequence)
 275                         blk_1 = blk_m;
 276                 else
 277                         blk_2 = blk_m;
 278         }
 279
 280         error = jhead_scan(jd, &lh_1);
 281         if (error)
 282                 return error;
 283
 284         *head = lh_1;
 285
 286         return error;
 287 }
 288
 289 /**
 290  * foreach_descriptor - go through the active part of the log
 291  * @jd: the journal
 292  * @start: the first log header in the active region
 293  * @end: the last log header (don't process the contents of this entry))
 294  *
 295  * Call a given function once for every log descriptor in the active
 296  * portion of the log.
 297  *
 298  * Returns: errno
 299  */
 300
 301 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
 302                               unsigned int end, int pass)
 303 {
 304         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 305         struct gfs2_sbd *sdp = ip->i_sbd;
 306         struct buffer_head *bh;
 307         struct gfs2_log_descriptor *ld;
 308         int error = 0;
 309         u32 length;
 310         __be64 *ptr;
 311         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 312         offset += (sizeof(__be64)-1);
 313         offset &= ~(sizeof(__be64)-1);
 314
 315         while (start != end) {
 316                 error = gfs2_replay_read_block(jd, start, &bh);
 317                 if (error)
 318                         return error;
 319                 if (gfs2_meta_check(sdp, bh)) {
 320                         brelse(bh);
 321                         return -EIO;
 322                 }
 323                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 324                 length = be32_to_cpu(ld->ld_length);
 325
 326                 if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
 327                         struct gfs2_log_header lh;
 328                         error = get_log_header(jd, start, &lh);
 329                         if (!error) {
 330                                 gfs2_replay_incr_blk(sdp, &start);
 331                                 continue;
 332                         }
 333                         if (error == 1) {
 334                                 gfs2_consist_inode(jd->jd_inode->u.generic_ip);
 335                                 error = -EIO;
 336                         }
 337                         brelse(bh);
 338                         return error;
 339                 } else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
 340                         brelse(bh);
 341                         return -EIO;
 342                 }
 343                 ptr = (__be64 *)(bh->b_data + offset);
 344                 error = lops_scan_elements(jd, start, ld, ptr, pass);
 345                 if (error) {
 346                         brelse(bh);
 347                         return error;
 348                 }
 349
 350                 while (length--)
 351                         gfs2_replay_incr_blk(sdp, &start);
 352
 353                 brelse(bh);
 354         }
 355
 356         return 0;
 357 }
 358
 359 /**
 360  * clean_journal - mark a dirty journal as being clean
 361  * @sdp: the filesystem
 362  * @jd: the journal
 363  * @gl: the journal's glock
 364  * @head: the head journal to start from
 365  *
 366  * Returns: errno
 367  */
 368
 369 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header *head)
 370 {
 371         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 372         struct gfs2_sbd *sdp = ip->i_sbd;
 373         unsigned int lblock;
 374         int new = 0;
 375         uint64_t dblock;
 376         struct gfs2_log_header *lh;
 377         uint32_t hash;
 378         struct buffer_head *bh;
 379         int error;
 380         int boundary;
 381
 382         lblock = head->lh_blkno;
 383         gfs2_replay_incr_blk(sdp, &lblock);
 384         error = gfs2_block_map(ip->i_vnode, lblock, &new, &dblock, &boundary);
 385         if (error)
 386                 return error;
 387         if (!dblock) {
 388                 gfs2_consist_inode(ip);
 389                 return -EIO;
 390         }
 391
 392         bh = sb_getblk(sdp->sd_vfs, dblock);
 393         lock_buffer(bh);
 394         memset(bh->b_data, 0, bh->b_size);
 395         set_buffer_uptodate(bh);
 396         clear_buffer_dirty(bh);
 397         unlock_buffer(bh);
 398
 399         lh = (struct gfs2_log_header *)bh->b_data;
 400         memset(lh, 0, sizeof(struct gfs2_log_header));
 401         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 402         lh->lh_header.mh_type = cpu_to_be16(GFS2_METATYPE_LH);
 403         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 404         lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
 405         lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
 406         lh->lh_blkno = cpu_to_be32(lblock);
 407         hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
 408         lh->lh_hash = cpu_to_be32(hash);
 409
 410         set_buffer_dirty(bh);
 411         if (sync_dirty_buffer(bh))
 412                 gfs2_io_error_bh(sdp, bh);
 413         brelse(bh);
 414
 415         return error;
 416 }
 417
 418 /**
 419  * gfs2_recover_journal - recovery a given journal
 420  * @jd: the struct gfs2_jdesc describing the journal
 421  *
 422  * Acquire the journal's lock, check to see if the journal is clean, and
 423  * do recovery if necessary.
 424  *
 425  * Returns: errno
 426  */
 427
 428 int gfs2_recover_journal(struct gfs2_jdesc *jd)
 429 {
 430         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 431         struct gfs2_sbd *sdp = ip->i_sbd;
 432         struct gfs2_log_header head;
 433         struct gfs2_holder j_gh, ji_gh, t_gh;
 434         unsigned long t;
 435         int ro = 0;
 436         unsigned int pass;
 437         int error;
 438
 439         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 440                 fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
 441                         jd->jd_jid);
 442
 443                 /* Aquire the journal lock so we can do recovery */
 444
 445                 error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
 446                                           LM_ST_EXCLUSIVE,
 447                                           LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
 448                                           &j_gh);
 449                 switch (error) {
 450                 case 0:
 451                         break;
 452
 453                 case GLR_TRYFAILED:
 454                         fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
 455                         error = 0;
 456
 457                 default:
 458                         goto fail;
 459                 };
 460
 461                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 462                                            LM_FLAG_NOEXP, &ji_gh);
 463                 if (error)
 464                         goto fail_gunlock_j;
 465         } else {
 466                 fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
 467         }
 468
 469         fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
 470
 471         error = gfs2_jdesc_check(jd);
 472         if (error)
 473                 goto fail_gunlock_ji;
 474
 475         error = gfs2_find_jhead(jd, &head);
 476         if (error)
 477                 goto fail_gunlock_ji;
 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                 t = jiffies;
 484
 485                 /* Acquire a shared hold on the transaction lock */
 486
 487                 error = gfs2_glock_nq_init(sdp->sd_trans_gl,
 488                                            LM_ST_SHARED,
 489                                            LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
 490                                            GL_NOCANCEL | GL_NOCACHE,
 491                                            &t_gh);
 492                 if (error)
 493                         goto fail_gunlock_ji;
 494
 495                 if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
 496                         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
 497                                 ro = 1;
 498                 } else {
 499                         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 500                                 ro = 1;
 501                 }
 502
 503                 if (ro) {
 504                         fs_warn(sdp, "jid=%u: Can't replay: read-only FS\n",
 505                                 jd->jd_jid);
 506                         error = -EROFS;
 507                         goto fail_gunlock_tr;
 508                 }
 509
 510                 fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
 511
 512                 for (pass = 0; pass < 2; pass++) {
 513                         lops_before_scan(jd, &head, pass);
 514                         error = foreach_descriptor(jd, head.lh_tail,
 515                                                    head.lh_blkno, pass);
 516                         lops_after_scan(jd, error, pass);
 517                         if (error)
 518                                 goto fail_gunlock_tr;
 519                 }
 520
 521                 error = clean_journal(jd, &head);
 522                 if (error)
 523                         goto fail_gunlock_tr;
 524
 525                 gfs2_glock_dq_uninit(&t_gh);
 526                 t = DIV_ROUND_UP(jiffies - t, HZ);
 527                 fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
 528                         jd->jd_jid, t);
 529         }
 530
 531         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 532                 gfs2_glock_dq_uninit(&ji_gh);
 533
 534         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
 535
 536         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
 537                 gfs2_glock_dq_uninit(&j_gh);
 538
 539         fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
 540         return 0;
 541
 542 fail_gunlock_tr:
 543         gfs2_glock_dq_uninit(&t_gh);
 544 fail_gunlock_ji:
 545         if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
 546                 gfs2_glock_dq_uninit(&ji_gh);
 547 fail_gunlock_j:
 548                 gfs2_glock_dq_uninit(&j_gh);
 549         }
 550
 551         fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
 552
 553 fail:
 554         gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
 555         return error;
 556 }
 557
 558 /**
 559  * gfs2_check_journals - Recover any dirty journals
 560  * @sdp: the filesystem
 561  *
 562  */
 563
 564 void gfs2_check_journals(struct gfs2_sbd *sdp)
 565 {
 566         struct gfs2_jdesc *jd;
 567
 568         for (;;) {
 569                 jd = gfs2_jdesc_find_dirty(sdp);
 570                 if (!jd)
 571                         break;
 572
 573                 if (jd != sdp->sd_jdesc)
 574                         gfs2_recover_journal(jd);
 575         }
 576 }
 577