fs/gfs2/util.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/crc32.h>
  13 #include <linux/gfs2_ondisk.h>
  14 #include <linux/delay.h>
  15 #include <linux/uaccess.h>
  16
  17 #include "gfs2.h"
  18 #include "incore.h"
  19 #include "glock.h"
  20 #include "glops.h"
  21 #include "log.h"
  22 #include "lops.h"
  23 #include "recovery.h"
  24 #include "rgrp.h"
  25 #include "super.h"
  26 #include "util.h"
  27
  28 struct kmem_cache *gfs2_glock_cachep __read_mostly;
  29 struct kmem_cache *gfs2_glock_aspace_cachep __read_mostly;
  30 struct kmem_cache *gfs2_inode_cachep __read_mostly;
  31 struct kmem_cache *gfs2_bufdata_cachep __read_mostly;
  32 struct kmem_cache *gfs2_rgrpd_cachep __read_mostly;
  33 struct kmem_cache *gfs2_quotad_cachep __read_mostly;
  34 struct kmem_cache *gfs2_qadata_cachep __read_mostly;
  35 struct kmem_cache *gfs2_trans_cachep __read_mostly;
  36 mempool_t *gfs2_page_pool __read_mostly;
  37
  38 void gfs2_assert_i(struct gfs2_sbd *sdp)
  39 {
  40         fs_emerg(sdp, "fatal assertion failed\n");
  41 }
  42
  43 /**
  44  * check_journal_clean - Make sure a journal is clean for a spectator mount
  45  * @sdp: The GFS2 superblock
  46  * @jd: The journal descriptor
  47  *
  48  * Returns: 0 if the journal is clean or locked, else an error
  49  */
  50 int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
  51                         bool verbose)
  52 {
  53         int error;
  54         struct gfs2_holder j_gh;
  55         struct gfs2_log_header_host head;
  56         struct gfs2_inode *ip;
  57
  58         ip = GFS2_I(jd->jd_inode);
  59         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
  60                                    GL_EXACT | GL_NOCACHE, &j_gh);
  61         if (error) {
  62                 if (verbose)
  63                         fs_err(sdp, "Error %d locking journal for spectator "
  64                                "mount.\n", error);
  65                 return -EPERM;
  66         }
  67         error = gfs2_jdesc_check(jd);
  68         if (error) {
  69                 if (verbose)
  70                         fs_err(sdp, "Error checking journal for spectator "
  71                                "mount.\n");
  72                 goto out_unlock;
  73         }
  74         error = gfs2_find_jhead(jd, &head, false);
  75         if (error) {
  76                 if (verbose)
  77                         fs_err(sdp, "Error parsing journal for spectator "
  78                                "mount.\n");
  79                 goto out_unlock;
  80         }
  81         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
  82                 error = -EPERM;
  83                 if (verbose)
  84                         fs_err(sdp, "jid=%u: Journal is dirty, so the first "
  85                                "mounter must not be a spectator.\n",
  86                                jd->jd_jid);
  87         }
  88
  89 out_unlock:
  90         gfs2_glock_dq_uninit(&j_gh);
  91         return error;
  92 }
  93
  94 static void signal_our_withdraw(struct gfs2_sbd *sdp)
  95 {
  96         struct gfs2_glock *gl = sdp->sd_live_gh.gh_gl;
  97         struct inode *inode = sdp->sd_jdesc->jd_inode;
  98         struct gfs2_inode *ip = GFS2_I(inode);
  99         u64 no_formal_ino = ip->i_no_formal_ino;
 100         int ret = 0;
 101         int tries;
 102
 103         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags))
 104                 return;
 105
 106         /* Prevent any glock dq until withdraw recovery is complete */
 107         set_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 108         /*
 109          * Don't tell dlm we're bailing until we have no more buffers in the
 110          * wind. If journal had an IO error, the log code should just purge
 111          * the outstanding buffers rather than submitting new IO. Making the
 112          * file system read-only will flush the journal, etc.
 113          *
 114          * During a normal unmount, gfs2_make_fs_ro calls gfs2_log_shutdown
 115          * which clears SDF_JOURNAL_LIVE. In a withdraw, we must not write
 116          * any UNMOUNT log header, so we can't call gfs2_log_shutdown, and
 117          * therefore we need to clear SDF_JOURNAL_LIVE manually.
 118          */
 119         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 120         if (!sb_rdonly(sdp->sd_vfs))
 121                 ret = gfs2_make_fs_ro(sdp);
 122
 123         if (sdp->sd_lockstruct.ls_ops->lm_lock == NULL) { /* lock_nolock */
 124                 if (!ret)
 125                         ret = -EIO;
 126                 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 127                 goto skip_recovery;
 128         }
 129         /*
 130          * Drop the glock for our journal so another node can recover it.
 131          */
 132         if (gfs2_holder_initialized(&sdp->sd_journal_gh)) {
 133                 gfs2_glock_dq_wait(&sdp->sd_journal_gh);
 134                 gfs2_holder_uninit(&sdp->sd_journal_gh);
 135         }
 136         sdp->sd_jinode_gh.gh_flags |= GL_NOCACHE;
 137         gfs2_glock_dq(&sdp->sd_jinode_gh);
 138         if (test_bit(SDF_FS_FROZEN, &sdp->sd_flags)) {
 139                 /* Make sure gfs2_unfreeze works if partially-frozen */
 140                 flush_work(&sdp->sd_freeze_work);
 141                 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 142                 thaw_super(sdp->sd_vfs);
 143         } else {
 144                 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
 145         }
 146
 147         /*
 148          * holder_uninit to force glock_put, to force dlm to let go
 149          */
 150         gfs2_holder_uninit(&sdp->sd_jinode_gh);
 151
 152         /*
 153          * Note: We need to be careful here:
 154          * Our iput of jd_inode will evict it. The evict will dequeue its
 155          * glock, but the glock dq will wait for the withdraw unless we have
 156          * exception code in glock_dq.
 157          */
 158         iput(inode);
 159         /*
 160          * Wait until the journal inode's glock is freed. This allows try locks
 161          * on other nodes to be successful, otherwise we remain the owner of
 162          * the glock as far as dlm is concerned.
 163          */
 164         if (gl->gl_ops->go_free) {
 165                 set_bit(GLF_FREEING, &gl->gl_flags);
 166                 wait_on_bit(&gl->gl_flags, GLF_FREEING, TASK_UNINTERRUPTIBLE);
 167         }
 168
 169         /*
 170          * Dequeue the "live" glock, but keep a reference so it's never freed.
 171          */
 172         gfs2_glock_hold(gl);
 173         gfs2_glock_dq_wait(&sdp->sd_live_gh);
 174         /*
 175          * We enqueue the "live" glock in EX so that all other nodes
 176          * get a demote request and act on it. We don't really want the
 177          * lock in EX, so we send a "try" lock with 1CB to produce a callback.
 178          */
 179         fs_warn(sdp, "Requesting recovery of jid %d.\n",
 180                 sdp->sd_lockstruct.ls_jid);
 181         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | LM_FLAG_NOEXP,
 182                            &sdp->sd_live_gh);
 183         msleep(GL_GLOCK_MAX_HOLD);
 184         /*
 185          * This will likely fail in a cluster, but succeed standalone:
 186          */
 187         ret = gfs2_glock_nq(&sdp->sd_live_gh);
 188
 189         /*
 190          * If we actually got the "live" lock in EX mode, there are no other
 191          * nodes available to replay our journal. So we try to replay it
 192          * ourselves. We hold the "live" glock to prevent other mounters
 193          * during recovery, then just dequeue it and reacquire it in our
 194          * normal SH mode. Just in case the problem that caused us to
 195          * withdraw prevents us from recovering our journal (e.g. io errors
 196          * and such) we still check if the journal is clean before proceeding
 197          * but we may wait forever until another mounter does the recovery.
 198          */
 199         if (ret == 0) {
 200                 fs_warn(sdp, "No other mounters found. Trying to recover our "
 201                         "own journal jid %d.\n", sdp->sd_lockstruct.ls_jid);
 202                 if (gfs2_recover_journal(sdp->sd_jdesc, 1))
 203                         fs_warn(sdp, "Unable to recover our journal jid %d.\n",
 204                                 sdp->sd_lockstruct.ls_jid);
 205                 gfs2_glock_dq_wait(&sdp->sd_live_gh);
 206                 gfs2_holder_reinit(LM_ST_SHARED, LM_FLAG_NOEXP | GL_EXACT,
 207                                    &sdp->sd_live_gh);
 208                 gfs2_glock_nq(&sdp->sd_live_gh);
 209         }
 210
 211         gfs2_glock_queue_put(gl); /* drop the extra reference we acquired */
 212         clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 213
 214         /*
 215          * At this point our journal is evicted, so we need to get a new inode
 216          * for it. Once done, we need to call gfs2_find_jhead which
 217          * calls gfs2_map_journal_extents to map it for us again.
 218          *
 219          * Note that we don't really want it to look up a FREE block. The
 220          * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
 221          * which would otherwise fail because it requires grabbing an rgrp
 222          * glock, which would fail with -EIO because we're withdrawing.
 223          */
 224         inode = gfs2_inode_lookup(sdp->sd_vfs, DT_UNKNOWN,
 225                                   sdp->sd_jdesc->jd_no_addr, no_formal_ino,
 226                                   GFS2_BLKST_FREE);
 227         if (IS_ERR(inode)) {
 228                 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
 229                         sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
 230                 goto skip_recovery;
 231         }
 232         sdp->sd_jdesc->jd_inode = inode;
 233
 234         /*
 235          * Now wait until recovery is complete.
 236          */
 237         for (tries = 0; tries < 10; tries++) {
 238                 ret = check_journal_clean(sdp, sdp->sd_jdesc, false);
 239                 if (!ret)
 240                         break;
 241                 msleep(HZ);
 242                 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
 243                         sdp->sd_lockstruct.ls_jid);
 244         }
 245 skip_recovery:
 246         if (!ret)
 247                 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
 248                         sdp->sd_lockstruct.ls_jid);
 249         else
 250                 fs_warn(sdp, "Journal recovery skipped for %d until next "
 251                         "mount.\n", sdp->sd_lockstruct.ls_jid);
 252         fs_warn(sdp, "Glock dequeues delayed: %lu\n", sdp->sd_glock_dqs_held);
 253         sdp->sd_glock_dqs_held = 0;
 254         wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY);
 255 }
 256
 257 void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
 258 {
 259         struct va_format vaf;
 260         va_list args;
 261
 262         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
 263             test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
 264                 return;
 265
 266         va_start(args, fmt);
 267         vaf.fmt = fmt;
 268         vaf.va = &args;
 269         fs_err(sdp, "%pV", &vaf);
 270         va_end(args);
 271 }
 272
 273 int gfs2_withdraw(struct gfs2_sbd *sdp)
 274 {
 275         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 276         const struct lm_lockops *lm = ls->ls_ops;
 277
 278         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
 279             test_and_set_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
 280                 if (!test_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags))
 281                         return -1;
 282
 283                 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG,
 284                             TASK_UNINTERRUPTIBLE);
 285                 return -1;
 286         }
 287
 288         set_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
 289
 290         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
 291                 fs_err(sdp, "about to withdraw this file system\n");
 292                 BUG_ON(sdp->sd_args.ar_debug);
 293
 294                 signal_our_withdraw(sdp);
 295
 296                 kobject_uevent(&sdp->sd_kobj, KOBJ_OFFLINE);
 297
 298                 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
 299                         wait_for_completion(&sdp->sd_wdack);
 300
 301                 if (lm->lm_unmount) {
 302                         fs_err(sdp, "telling LM to unmount\n");
 303                         lm->lm_unmount(sdp);
 304                 }
 305                 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
 306                 fs_err(sdp, "File system withdrawn\n");
 307                 dump_stack();
 308                 clear_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
 309                 smp_mb__after_atomic();
 310                 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG);
 311         }
 312
 313         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 314                 panic("GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
 315
 316         return -1;
 317 }
 318
 319 /**
 320  * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
 321  */
 322
 323 void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
 324                             const char *function, char *file, unsigned int line,
 325                             bool delayed)
 326 {
 327         if (gfs2_withdrawn(sdp))
 328                 return;
 329
 330         fs_err(sdp,
 331                "fatal: assertion \"%s\" failed\n"
 332                "   function = %s, file = %s, line = %u\n",
 333                assertion, function, file, line);
 334
 335         /*
 336          * If errors=panic was specified on mount, it won't help to delay the
 337          * withdraw.
 338          */
 339         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 340                 delayed = false;
 341
 342         if (delayed)
 343                 gfs2_withdraw_delayed(sdp);
 344         else
 345                 gfs2_withdraw(sdp);
 346         dump_stack();
 347 }
 348
 349 /**
 350  * gfs2_assert_warn_i - Print a message to the console if @assertion is false
 351  */
 352
 353 void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
 354                         const char *function, char *file, unsigned int line)
 355 {
 356         if (time_before(jiffies,
 357                         sdp->sd_last_warning +
 358                         gfs2_tune_get(sdp, gt_complain_secs) * HZ))
 359                 return;
 360
 361         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
 362                 fs_warn(sdp, "warning: assertion \"%s\" failed at function = %s, file = %s, line = %u\n",
 363                         assertion, function, file, line);
 364
 365         if (sdp->sd_args.ar_debug)
 366                 BUG();
 367         else
 368                 dump_stack();
 369
 370         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 371                 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed\n"
 372                       "GFS2: fsid=%s:   function = %s, file = %s, line = %u\n",
 373                       sdp->sd_fsname, assertion,
 374                       sdp->sd_fsname, function, file, line);
 375
 376         sdp->sd_last_warning = jiffies;
 377 }
 378
 379 /**
 380  * gfs2_consist_i - Flag a filesystem consistency error and withdraw
 381  */
 382
 383 void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
 384                     char *file, unsigned int line)
 385 {
 386         gfs2_lm(sdp,
 387                 "fatal: filesystem consistency error - function = %s, file = %s, line = %u\n",
 388                 function, file, line);
 389         gfs2_withdraw(sdp);
 390 }
 391
 392 /**
 393  * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
 394  */
 395
 396 void gfs2_consist_inode_i(struct gfs2_inode *ip,
 397                           const char *function, char *file, unsigned int line)
 398 {
 399         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 400
 401         gfs2_lm(sdp,
 402                 "fatal: filesystem consistency error\n"
 403                 "  inode = %llu %llu\n"
 404                 "  function = %s, file = %s, line = %u\n",
 405                 (unsigned long long)ip->i_no_formal_ino,
 406                 (unsigned long long)ip->i_no_addr,
 407                 function, file, line);
 408         gfs2_withdraw(sdp);
 409 }
 410
 411 /**
 412  * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
 413  */
 414
 415 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
 416                           const char *function, char *file, unsigned int line)
 417 {
 418         struct gfs2_sbd *sdp = rgd->rd_sbd;
 419         char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
 420
 421         sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
 422         gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
 423         gfs2_lm(sdp,
 424                 "fatal: filesystem consistency error\n"
 425                 "  RG = %llu\n"
 426                 "  function = %s, file = %s, line = %u\n",
 427                 (unsigned long long)rgd->rd_addr,
 428                 function, file, line);
 429         gfs2_withdraw(sdp);
 430 }
 431
 432 /**
 433  * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
 434  * Returns: -1 if this call withdrew the machine,
 435  *          -2 if it was already withdrawn
 436  */
 437
 438 int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
 439                        const char *type, const char *function, char *file,
 440                        unsigned int line)
 441 {
 442         int me;
 443
 444         gfs2_lm(sdp,
 445                 "fatal: invalid metadata block\n"
 446                 "  bh = %llu (%s)\n"
 447                 "  function = %s, file = %s, line = %u\n",
 448                 (unsigned long long)bh->b_blocknr, type,
 449                 function, file, line);
 450         me = gfs2_withdraw(sdp);
 451         return (me) ? -1 : -2;
 452 }
 453
 454 /**
 455  * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
 456  * Returns: -1 if this call withdrew the machine,
 457  *          -2 if it was already withdrawn
 458  */
 459
 460 int gfs2_metatype_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
 461                            u16 type, u16 t, const char *function,
 462                            char *file, unsigned int line)
 463 {
 464         int me;
 465
 466         gfs2_lm(sdp,
 467                 "fatal: invalid metadata block\n"
 468                 "  bh = %llu (type: exp=%u, found=%u)\n"
 469                 "  function = %s, file = %s, line = %u\n",
 470                 (unsigned long long)bh->b_blocknr, type, t,
 471                 function, file, line);
 472         me = gfs2_withdraw(sdp);
 473         return (me) ? -1 : -2;
 474 }
 475
 476 /**
 477  * gfs2_io_error_i - Flag an I/O error and withdraw
 478  * Returns: -1 if this call withdrew the machine,
 479  *          0 if it was already withdrawn
 480  */
 481
 482 int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
 483                     unsigned int line)
 484 {
 485         gfs2_lm(sdp,
 486                 "fatal: I/O error\n"
 487                 "  function = %s, file = %s, line = %u\n",
 488                 function, file, line);
 489         return gfs2_withdraw(sdp);
 490 }
 491
 492 /**
 493  * gfs2_io_error_bh_i - Flag a buffer I/O error
 494  * @withdraw: withdraw the filesystem
 495  */
 496
 497 void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
 498                         const char *function, char *file, unsigned int line,
 499                         bool withdraw)
 500 {
 501         if (gfs2_withdrawn(sdp))
 502                 return;
 503
 504         fs_err(sdp, "fatal: I/O error\n"
 505                "  block = %llu\n"
 506                "  function = %s, file = %s, line = %u\n",
 507                (unsigned long long)bh->b_blocknr, function, file, line);
 508         if (withdraw)
 509                 gfs2_withdraw(sdp);
 510 }
 511