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  * @verbose: Show more prints in the log
  48  *
  49  * Returns: 0 if the journal is clean or locked, else an error
  50  */
  51 int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
  52                         bool verbose)
  53 {
  54         int error;
  55         struct gfs2_holder j_gh;
  56         struct gfs2_log_header_host head;
  57         struct gfs2_inode *ip;
  58
  59         ip = GFS2_I(jd->jd_inode);
  60         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
  61                                    GL_EXACT | GL_NOCACHE, &j_gh);
  62         if (error) {
  63                 if (verbose)
  64                         fs_err(sdp, "Error %d locking journal for spectator "
  65                                "mount.\n", error);
  66                 return -EPERM;
  67         }
  68         error = gfs2_jdesc_check(jd);
  69         if (error) {
  70                 if (verbose)
  71                         fs_err(sdp, "Error checking journal for spectator "
  72                                "mount.\n");
  73                 goto out_unlock;
  74         }
  75         error = gfs2_find_jhead(jd, &head, false);
  76         if (error) {
  77                 if (verbose)
  78                         fs_err(sdp, "Error parsing journal for spectator "
  79                                "mount.\n");
  80                 goto out_unlock;
  81         }
  82         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
  83                 error = -EPERM;
  84                 if (verbose)
  85                         fs_err(sdp, "jid=%u: Journal is dirty, so the first "
  86                                "mounter must not be a spectator.\n",
  87                                jd->jd_jid);
  88         }
  89
  90 out_unlock:
  91         gfs2_glock_dq_uninit(&j_gh);
  92         return error;
  93 }
  94
  95 /**
  96  * gfs2_freeze_lock_shared - hold the freeze glock
  97  * @sdp: the superblock
  98  * @freeze_gh: pointer to the requested holder
  99  * @caller_flags: any additional flags needed by the caller
 100  */
 101 int gfs2_freeze_lock_shared(struct gfs2_sbd *sdp, struct gfs2_holder *freeze_gh,
 102                             int caller_flags)
 103 {
 104         int flags = LM_FLAG_NOEXP | GL_EXACT | caller_flags;
 105         int error;
 106
 107         error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, flags,
 108                                    freeze_gh);
 109         if (error && error != GLR_TRYFAILED)
 110                 fs_err(sdp, "can't lock the freeze glock: %d\n", error);
 111         return error;
 112 }
 113
 114 void gfs2_freeze_unlock(struct gfs2_holder *freeze_gh)
 115 {
 116         if (gfs2_holder_initialized(freeze_gh))
 117                 gfs2_glock_dq_uninit(freeze_gh);
 118 }
 119
 120 static void signal_our_withdraw(struct gfs2_sbd *sdp)
 121 {
 122         struct gfs2_glock *live_gl = sdp->sd_live_gh.gh_gl;
 123         struct inode *inode;
 124         struct gfs2_inode *ip;
 125         struct gfs2_glock *i_gl;
 126         u64 no_formal_ino;
 127         int ret = 0;
 128         int tries;
 129
 130         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) || !sdp->sd_jdesc)
 131                 return;
 132
 133         gfs2_ail_drain(sdp); /* frees all transactions */
 134         inode = sdp->sd_jdesc->jd_inode;
 135         ip = GFS2_I(inode);
 136         i_gl = ip->i_gl;
 137         no_formal_ino = ip->i_no_formal_ino;
 138
 139         /* Prevent any glock dq until withdraw recovery is complete */
 140         set_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 141         /*
 142          * Don't tell dlm we're bailing until we have no more buffers in the
 143          * wind. If journal had an IO error, the log code should just purge
 144          * the outstanding buffers rather than submitting new IO. Making the
 145          * file system read-only will flush the journal, etc.
 146          *
 147          * During a normal unmount, gfs2_make_fs_ro calls gfs2_log_shutdown
 148          * which clears SDF_JOURNAL_LIVE. In a withdraw, we must not write
 149          * any UNMOUNT log header, so we can't call gfs2_log_shutdown, and
 150          * therefore we need to clear SDF_JOURNAL_LIVE manually.
 151          */
 152         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 153         if (!sb_rdonly(sdp->sd_vfs)) {
 154                 bool locked = mutex_trylock(&sdp->sd_freeze_mutex);
 155
 156                 gfs2_make_fs_ro(sdp);
 157
 158                 if (locked)
 159                         mutex_unlock(&sdp->sd_freeze_mutex);
 160
 161                 /*
 162                  * Dequeue any pending non-system glock holders that can no
 163                  * longer be granted because the file system is withdrawn.
 164                  */
 165                 gfs2_gl_dq_holders(sdp);
 166         }
 167
 168         if (sdp->sd_lockstruct.ls_ops->lm_lock == NULL) { /* lock_nolock */
 169                 if (!ret)
 170                         ret = -EIO;
 171                 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 172                 goto skip_recovery;
 173         }
 174         /*
 175          * Drop the glock for our journal so another node can recover it.
 176          */
 177         if (gfs2_holder_initialized(&sdp->sd_journal_gh)) {
 178                 gfs2_glock_dq_wait(&sdp->sd_journal_gh);
 179                 gfs2_holder_uninit(&sdp->sd_journal_gh);
 180         }
 181         sdp->sd_jinode_gh.gh_flags |= GL_NOCACHE;
 182         gfs2_glock_dq(&sdp->sd_jinode_gh);
 183         gfs2_thaw_freeze_initiator(sdp->sd_vfs);
 184         wait_on_bit(&i_gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
 185
 186         /*
 187          * holder_uninit to force glock_put, to force dlm to let go
 188          */
 189         gfs2_holder_uninit(&sdp->sd_jinode_gh);
 190
 191         /*
 192          * Note: We need to be careful here:
 193          * Our iput of jd_inode will evict it. The evict will dequeue its
 194          * glock, but the glock dq will wait for the withdraw unless we have
 195          * exception code in glock_dq.
 196          */
 197         iput(inode);
 198         sdp->sd_jdesc->jd_inode = NULL;
 199         /*
 200          * Wait until the journal inode's glock is freed. This allows try locks
 201          * on other nodes to be successful, otherwise we remain the owner of
 202          * the glock as far as dlm is concerned.
 203          */
 204         if (i_gl->gl_ops->go_free) {
 205                 set_bit(GLF_FREEING, &i_gl->gl_flags);
 206                 wait_on_bit(&i_gl->gl_flags, GLF_FREEING, TASK_UNINTERRUPTIBLE);
 207         }
 208
 209         /*
 210          * Dequeue the "live" glock, but keep a reference so it's never freed.
 211          */
 212         gfs2_glock_hold(live_gl);
 213         gfs2_glock_dq_wait(&sdp->sd_live_gh);
 214         /*
 215          * We enqueue the "live" glock in EX so that all other nodes
 216          * get a demote request and act on it. We don't really want the
 217          * lock in EX, so we send a "try" lock with 1CB to produce a callback.
 218          */
 219         fs_warn(sdp, "Requesting recovery of jid %d.\n",
 220                 sdp->sd_lockstruct.ls_jid);
 221         gfs2_holder_reinit(LM_ST_EXCLUSIVE,
 222                            LM_FLAG_TRY_1CB | LM_FLAG_NOEXP | GL_NOPID,
 223                            &sdp->sd_live_gh);
 224         msleep(GL_GLOCK_MAX_HOLD);
 225         /*
 226          * This will likely fail in a cluster, but succeed standalone:
 227          */
 228         ret = gfs2_glock_nq(&sdp->sd_live_gh);
 229
 230         /*
 231          * If we actually got the "live" lock in EX mode, there are no other
 232          * nodes available to replay our journal. So we try to replay it
 233          * ourselves. We hold the "live" glock to prevent other mounters
 234          * during recovery, then just dequeue it and reacquire it in our
 235          * normal SH mode. Just in case the problem that caused us to
 236          * withdraw prevents us from recovering our journal (e.g. io errors
 237          * and such) we still check if the journal is clean before proceeding
 238          * but we may wait forever until another mounter does the recovery.
 239          */
 240         if (ret == 0) {
 241                 fs_warn(sdp, "No other mounters found. Trying to recover our "
 242                         "own journal jid %d.\n", sdp->sd_lockstruct.ls_jid);
 243                 if (gfs2_recover_journal(sdp->sd_jdesc, 1))
 244                         fs_warn(sdp, "Unable to recover our journal jid %d.\n",
 245                                 sdp->sd_lockstruct.ls_jid);
 246                 gfs2_glock_dq_wait(&sdp->sd_live_gh);
 247                 gfs2_holder_reinit(LM_ST_SHARED,
 248                                    LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
 249                                    &sdp->sd_live_gh);
 250                 gfs2_glock_nq(&sdp->sd_live_gh);
 251         }
 252
 253         gfs2_glock_queue_put(live_gl); /* drop extra reference we acquired */
 254         clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 255
 256         /*
 257          * At this point our journal is evicted, so we need to get a new inode
 258          * for it. Once done, we need to call gfs2_find_jhead which
 259          * calls gfs2_map_journal_extents to map it for us again.
 260          *
 261          * Note that we don't really want it to look up a FREE block. The
 262          * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
 263          * which would otherwise fail because it requires grabbing an rgrp
 264          * glock, which would fail with -EIO because we're withdrawing.
 265          */
 266         inode = gfs2_inode_lookup(sdp->sd_vfs, DT_UNKNOWN,
 267                                   sdp->sd_jdesc->jd_no_addr, no_formal_ino,
 268                                   GFS2_BLKST_FREE);
 269         if (IS_ERR(inode)) {
 270                 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
 271                         sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
 272                 goto skip_recovery;
 273         }
 274         sdp->sd_jdesc->jd_inode = inode;
 275         d_mark_dontcache(inode);
 276
 277         /*
 278          * Now wait until recovery is complete.
 279          */
 280         for (tries = 0; tries < 10; tries++) {
 281                 ret = check_journal_clean(sdp, sdp->sd_jdesc, false);
 282                 if (!ret)
 283                         break;
 284                 msleep(HZ);
 285                 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
 286                         sdp->sd_lockstruct.ls_jid);
 287         }
 288 skip_recovery:
 289         if (!ret)
 290                 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
 291                         sdp->sd_lockstruct.ls_jid);
 292         else
 293                 fs_warn(sdp, "Journal recovery skipped for jid %d until next "
 294                         "mount.\n", sdp->sd_lockstruct.ls_jid);
 295         fs_warn(sdp, "Glock dequeues delayed: %lu\n", sdp->sd_glock_dqs_held);
 296         sdp->sd_glock_dqs_held = 0;
 297         wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY);
 298 }
 299
 300 void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
 301 {
 302         struct va_format vaf;
 303         va_list args;
 304
 305         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
 306             test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
 307                 return;
 308
 309         va_start(args, fmt);
 310         vaf.fmt = fmt;
 311         vaf.va = &args;
 312         fs_err(sdp, "%pV", &vaf);
 313         va_end(args);
 314 }
 315
 316 int gfs2_withdraw(struct gfs2_sbd *sdp)
 317 {
 318         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 319         const struct lm_lockops *lm = ls->ls_ops;
 320
 321         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
 322             test_and_set_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
 323                 if (!test_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags))
 324                         return -1;
 325
 326                 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG,
 327                             TASK_UNINTERRUPTIBLE);
 328                 return -1;
 329         }
 330
 331         set_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
 332
 333         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
 334                 fs_err(sdp, "about to withdraw this file system\n");
 335                 BUG_ON(sdp->sd_args.ar_debug);
 336
 337                 signal_our_withdraw(sdp);
 338
 339                 kobject_uevent(&sdp->sd_kobj, KOBJ_OFFLINE);
 340
 341                 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
 342                         wait_for_completion(&sdp->sd_wdack);
 343
 344                 if (lm->lm_unmount) {
 345                         fs_err(sdp, "telling LM to unmount\n");
 346                         lm->lm_unmount(sdp);
 347                 }
 348                 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
 349                 fs_err(sdp, "File system withdrawn\n");
 350                 dump_stack();
 351                 clear_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
 352                 smp_mb__after_atomic();
 353                 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG);
 354         }
 355
 356         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 357                 panic("GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
 358
 359         return -1;
 360 }
 361
 362 /*
 363  * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
 364  */
 365
 366 void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
 367                             const char *function, char *file, unsigned int line,
 368                             bool delayed)
 369 {
 370         if (gfs2_withdrawn(sdp))
 371                 return;
 372
 373         fs_err(sdp,
 374                "fatal: assertion \"%s\" failed\n"
 375                "   function = %s, file = %s, line = %u\n",
 376                assertion, function, file, line);
 377
 378         /*
 379          * If errors=panic was specified on mount, it won't help to delay the
 380          * withdraw.
 381          */
 382         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 383                 delayed = false;
 384
 385         if (delayed)
 386                 gfs2_withdraw_delayed(sdp);
 387         else
 388                 gfs2_withdraw(sdp);
 389         dump_stack();
 390 }
 391
 392 /*
 393  * gfs2_assert_warn_i - Print a message to the console if @assertion is false
 394  */
 395
 396 void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
 397                         const char *function, char *file, unsigned int line)
 398 {
 399         if (time_before(jiffies,
 400                         sdp->sd_last_warning +
 401                         gfs2_tune_get(sdp, gt_complain_secs) * HZ))
 402                 return;
 403
 404         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
 405                 fs_warn(sdp, "warning: assertion \"%s\" failed at function = %s, file = %s, line = %u\n",
 406                         assertion, function, file, line);
 407
 408         if (sdp->sd_args.ar_debug)
 409                 BUG();
 410         else
 411                 dump_stack();
 412
 413         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 414                 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed\n"
 415                       "GFS2: fsid=%s:   function = %s, file = %s, line = %u\n",
 416                       sdp->sd_fsname, assertion,
 417                       sdp->sd_fsname, function, file, line);
 418
 419         sdp->sd_last_warning = jiffies;
 420 }
 421
 422 /*
 423  * gfs2_consist_i - Flag a filesystem consistency error and withdraw
 424  */
 425
 426 void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
 427                     char *file, unsigned int line)
 428 {
 429         gfs2_lm(sdp,
 430                 "fatal: filesystem consistency error - function = %s, file = %s, line = %u\n",
 431                 function, file, line);
 432         gfs2_withdraw(sdp);
 433 }
 434
 435 /*
 436  * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
 437  */
 438
 439 void gfs2_consist_inode_i(struct gfs2_inode *ip,
 440                           const char *function, char *file, unsigned int line)
 441 {
 442         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 443
 444         gfs2_lm(sdp,
 445                 "fatal: filesystem consistency error\n"
 446                 "  inode = %llu %llu\n"
 447                 "  function = %s, file = %s, line = %u\n",
 448                 (unsigned long long)ip->i_no_formal_ino,
 449                 (unsigned long long)ip->i_no_addr,
 450                 function, file, line);
 451         gfs2_dump_glock(NULL, ip->i_gl, 1);
 452         gfs2_withdraw(sdp);
 453 }
 454
 455 /*
 456  * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
 457  */
 458
 459 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
 460                           const char *function, char *file, unsigned int line)
 461 {
 462         struct gfs2_sbd *sdp = rgd->rd_sbd;
 463         char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
 464
 465         sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
 466         gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
 467         gfs2_lm(sdp,
 468                 "fatal: filesystem consistency error\n"
 469                 "  RG = %llu\n"
 470                 "  function = %s, file = %s, line = %u\n",
 471                 (unsigned long long)rgd->rd_addr,
 472                 function, file, line);
 473         gfs2_dump_glock(NULL, rgd->rd_gl, 1);
 474         gfs2_withdraw(sdp);
 475 }
 476
 477 /*
 478  * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
 479  * Returns: -1 if this call withdrew the machine,
 480  *          -2 if it was already withdrawn
 481  */
 482
 483 int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
 484                        const char *type, const char *function, char *file,
 485                        unsigned int line)
 486 {
 487         int me;
 488
 489         gfs2_lm(sdp,
 490                 "fatal: invalid metadata block\n"
 491                 "  bh = %llu (%s)\n"
 492                 "  function = %s, file = %s, line = %u\n",
 493                 (unsigned long long)bh->b_blocknr, type,
 494                 function, file, line);
 495         me = gfs2_withdraw(sdp);
 496         return (me) ? -1 : -2;
 497 }
 498
 499 /*
 500  * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
 501  * Returns: -1 if this call withdrew the machine,
 502  *          -2 if it was already withdrawn
 503  */
 504
 505 int gfs2_metatype_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
 506                            u16 type, u16 t, const char *function,
 507                            char *file, unsigned int line)
 508 {
 509         int me;
 510
 511         gfs2_lm(sdp,
 512                 "fatal: invalid metadata block\n"
 513                 "  bh = %llu (type: exp=%u, found=%u)\n"
 514                 "  function = %s, file = %s, line = %u\n",
 515                 (unsigned long long)bh->b_blocknr, type, t,
 516                 function, file, line);
 517         me = gfs2_withdraw(sdp);
 518         return (me) ? -1 : -2;
 519 }
 520
 521 /*
 522  * gfs2_io_error_i - Flag an I/O error and withdraw
 523  * Returns: -1 if this call withdrew the machine,
 524  *          0 if it was already withdrawn
 525  */
 526
 527 int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
 528                     unsigned int line)
 529 {
 530         gfs2_lm(sdp,
 531                 "fatal: I/O error\n"
 532                 "  function = %s, file = %s, line = %u\n",
 533                 function, file, line);
 534         return gfs2_withdraw(sdp);
 535 }
 536
 537 /*
 538  * gfs2_io_error_bh_i - Flag a buffer I/O error
 539  * @withdraw: withdraw the filesystem
 540  */
 541
 542 void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
 543                         const char *function, char *file, unsigned int line,
 544                         bool withdraw)
 545 {
 546         if (gfs2_withdrawn(sdp))
 547                 return;
 548
 549         fs_err(sdp, "fatal: I/O error\n"
 550                "  block = %llu\n"
 551                "  function = %s, file = %s, line = %u\n",
 552                (unsigned long long)bh->b_blocknr, function, file, line);
 553         if (withdraw)
 554                 gfs2_withdraw(sdp);
 555 }
 556