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 mempool_t *gfs2_page_pool __read_mostly;
  36
  37 void gfs2_assert_i(struct gfs2_sbd *sdp)
  38 {
  39         fs_emerg(sdp, "fatal assertion failed\n");
  40 }
  41
  42 /**
  43  * check_journal_clean - Make sure a journal is clean for a spectator mount
  44  * @sdp: The GFS2 superblock
  45  * @jd: The journal descriptor
  46  *
  47  * Returns: 0 if the journal is clean or locked, else an error
  48  */
  49 int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
  50                         bool verbose)
  51 {
  52         int error;
  53         struct gfs2_holder j_gh;
  54         struct gfs2_log_header_host head;
  55         struct gfs2_inode *ip;
  56
  57         ip = GFS2_I(jd->jd_inode);
  58         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
  59                                    GL_EXACT | GL_NOCACHE, &j_gh);
  60         if (error) {
  61                 if (verbose)
  62                         fs_err(sdp, "Error %d locking journal for spectator "
  63                                "mount.\n", error);
  64                 return -EPERM;
  65         }
  66         error = gfs2_jdesc_check(jd);
  67         if (error) {
  68                 if (verbose)
  69                         fs_err(sdp, "Error checking journal for spectator "
  70                                "mount.\n");
  71                 goto out_unlock;
  72         }
  73         error = gfs2_find_jhead(jd, &head, false);
  74         if (error) {
  75                 if (verbose)
  76                         fs_err(sdp, "Error parsing journal for spectator "
  77                                "mount.\n");
  78                 goto out_unlock;
  79         }
  80         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
  81                 error = -EPERM;
  82                 if (verbose)
  83                         fs_err(sdp, "jid=%u: Journal is dirty, so the first "
  84                                "mounter must not be a spectator.\n",
  85                                jd->jd_jid);
  86         }
  87
  88 out_unlock:
  89         gfs2_glock_dq_uninit(&j_gh);
  90         return error;
  91 }
  92
  93 static void signal_our_withdraw(struct gfs2_sbd *sdp)
  94 {
  95         struct gfs2_glock *gl = sdp->sd_live_gh.gh_gl;
  96         struct inode *inode = sdp->sd_jdesc->jd_inode;
  97         struct gfs2_inode *ip = GFS2_I(inode);
  98         u64 no_formal_ino = ip->i_no_formal_ino;
  99         int ret = 0;
 100         int tries;
 101
 102         if (test_bit(SDF_NORECOVERY, &sdp->sd_flags))
 103                 return;
 104
 105         /* Prevent any glock dq until withdraw recovery is complete */
 106         set_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 107         /*
 108          * Don't tell dlm we're bailing until we have no more buffers in the
 109          * wind. If journal had an IO error, the log code should just purge
 110          * the outstanding buffers rather than submitting new IO. Making the
 111          * file system read-only will flush the journal, etc.
 112          *
 113          * During a normal unmount, gfs2_make_fs_ro calls gfs2_log_shutdown
 114          * which clears SDF_JOURNAL_LIVE. In a withdraw, we must not write
 115          * any UNMOUNT log header, so we can't call gfs2_log_shutdown, and
 116          * therefore we need to clear SDF_JOURNAL_LIVE manually.
 117          */
 118         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 119         if (!sb_rdonly(sdp->sd_vfs))
 120                 ret = gfs2_make_fs_ro(sdp);
 121
 122         if (sdp->sd_lockstruct.ls_ops->lm_lock == NULL) { /* lock_nolock */
 123                 if (!ret)
 124                         ret = -EIO;
 125                 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 126                 goto skip_recovery;
 127         }
 128         /*
 129          * Drop the glock for our journal so another node can recover it.
 130          */
 131         if (gfs2_holder_initialized(&sdp->sd_journal_gh)) {
 132                 gfs2_glock_dq_wait(&sdp->sd_journal_gh);
 133                 gfs2_holder_uninit(&sdp->sd_journal_gh);
 134         }
 135         sdp->sd_jinode_gh.gh_flags |= GL_NOCACHE;
 136         gfs2_glock_dq(&sdp->sd_jinode_gh);
 137         if (test_bit(SDF_FS_FROZEN, &sdp->sd_flags)) {
 138                 /* Make sure gfs2_unfreeze works if partially-frozen */
 139                 flush_workqueue(gfs2_freeze_wq);
 140                 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
 141                 thaw_super(sdp->sd_vfs);
 142         } else {
 143                 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
 144         }
 145
 146         /*
 147          * holder_uninit to force glock_put, to force dlm to let go
 148          */
 149         gfs2_holder_uninit(&sdp->sd_jinode_gh);
 150
 151         /*
 152          * Note: We need to be careful here:
 153          * Our iput of jd_inode will evict it. The evict will dequeue its
 154          * glock, but the glock dq will wait for the withdraw unless we have
 155          * exception code in glock_dq.
 156          */
 157         iput(inode);
 158         /*
 159          * Wait until the journal inode's glock is freed. This allows try locks
 160          * on other nodes to be successful, otherwise we remain the owner of
 161          * the glock as far as dlm is concerned.
 162          */
 163         if (gl->gl_ops->go_free) {
 164                 set_bit(GLF_FREEING, &gl->gl_flags);
 165                 wait_on_bit(&gl->gl_flags, GLF_FREEING, TASK_UNINTERRUPTIBLE);
 166         }
 167
 168         /*
 169          * Dequeue the "live" glock, but keep a reference so it's never freed.
 170          */
 171         gfs2_glock_hold(gl);
 172         gfs2_glock_dq_wait(&sdp->sd_live_gh);
 173         /*
 174          * We enqueue the "live" glock in EX so that all other nodes
 175          * get a demote request and act on it. We don't really want the
 176          * lock in EX, so we send a "try" lock with 1CB to produce a callback.
 177          */
 178         fs_warn(sdp, "Requesting recovery of jid %d.\n",
 179                 sdp->sd_lockstruct.ls_jid);
 180         gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | LM_FLAG_NOEXP,
 181                            &sdp->sd_live_gh);
 182         msleep(GL_GLOCK_MAX_HOLD);
 183         /*
 184          * This will likely fail in a cluster, but succeed standalone:
 185          */
 186         ret = gfs2_glock_nq(&sdp->sd_live_gh);
 187
 188         /*
 189          * If we actually got the "live" lock in EX mode, there are no other
 190          * nodes available to replay our journal. So we try to replay it
 191          * ourselves. We hold the "live" glock to prevent other mounters
 192          * during recovery, then just dequeue it and reacquire it in our
 193          * normal SH mode. Just in case the problem that caused us to
 194          * withdraw prevents us from recovering our journal (e.g. io errors
 195          * and such) we still check if the journal is clean before proceeding
 196          * but we may wait forever until another mounter does the recovery.
 197          */
 198         if (ret == 0) {
 199                 fs_warn(sdp, "No other mounters found. Trying to recover our "
 200                         "own journal jid %d.\n", sdp->sd_lockstruct.ls_jid);
 201                 if (gfs2_recover_journal(sdp->sd_jdesc, 1))
 202                         fs_warn(sdp, "Unable to recover our journal jid %d.\n",
 203                                 sdp->sd_lockstruct.ls_jid);
 204                 gfs2_glock_dq_wait(&sdp->sd_live_gh);
 205                 gfs2_holder_reinit(LM_ST_SHARED, LM_FLAG_NOEXP | GL_EXACT,
 206                                    &sdp->sd_live_gh);
 207                 gfs2_glock_nq(&sdp->sd_live_gh);
 208         }
 209
 210         gfs2_glock_queue_put(gl); /* drop the extra reference we acquired */
 211         clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
 212
 213         /*
 214          * At this point our journal is evicted, so we need to get a new inode
 215          * for it. Once done, we need to call gfs2_find_jhead which
 216          * calls gfs2_map_journal_extents to map it for us again.
 217          *
 218          * Note that we don't really want it to look up a FREE block. The
 219          * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
 220          * which would otherwise fail because it requires grabbing an rgrp
 221          * glock, which would fail with -EIO because we're withdrawing.
 222          */
 223         inode = gfs2_inode_lookup(sdp->sd_vfs, DT_UNKNOWN,
 224                                   sdp->sd_jdesc->jd_no_addr, no_formal_ino,
 225                                   GFS2_BLKST_FREE);
 226         if (IS_ERR(inode)) {
 227                 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
 228                         sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
 229                 goto skip_recovery;
 230         }
 231         sdp->sd_jdesc->jd_inode = inode;
 232
 233         /*
 234          * Now wait until recovery is complete.
 235          */
 236         for (tries = 0; tries < 10; tries++) {
 237                 ret = check_journal_clean(sdp, sdp->sd_jdesc, false);
 238                 if (!ret)
 239                         break;
 240                 msleep(HZ);
 241                 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
 242                         sdp->sd_lockstruct.ls_jid);
 243         }
 244 skip_recovery:
 245         if (!ret)
 246                 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
 247                         sdp->sd_lockstruct.ls_jid);
 248         else
 249                 fs_warn(sdp, "Journal recovery skipped for %d until next "
 250                         "mount.\n", sdp->sd_lockstruct.ls_jid);
 251         fs_warn(sdp, "Glock dequeues delayed: %lu\n", sdp->sd_glock_dqs_held);
 252         sdp->sd_glock_dqs_held = 0;
 253         wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY);
 254 }
 255
 256 void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
 257 {
 258         struct va_format vaf;
 259         va_list args;
 260
 261         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
 262             test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
 263                 return;
 264
 265         va_start(args, fmt);
 266         vaf.fmt = fmt;
 267         vaf.va = &args;
 268         fs_err(sdp, "%pV", &vaf);
 269         va_end(args);
 270 }
 271
 272 int gfs2_withdraw(struct gfs2_sbd *sdp)
 273 {
 274         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 275         const struct lm_lockops *lm = ls->ls_ops;
 276
 277         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
 278             test_and_set_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
 279                 if (!test_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags))
 280                         return -1;
 281
 282                 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG,
 283                             TASK_UNINTERRUPTIBLE);
 284                 return -1;
 285         }
 286
 287         set_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
 288
 289         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
 290                 fs_err(sdp, "about to withdraw this file system\n");
 291                 BUG_ON(sdp->sd_args.ar_debug);
 292
 293                 signal_our_withdraw(sdp);
 294
 295                 kobject_uevent(&sdp->sd_kobj, KOBJ_OFFLINE);
 296
 297                 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
 298                         wait_for_completion(&sdp->sd_wdack);
 299
 300                 if (lm->lm_unmount) {
 301                         fs_err(sdp, "telling LM to unmount\n");
 302                         lm->lm_unmount(sdp);
 303                 }
 304                 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
 305                 fs_err(sdp, "File system withdrawn\n");
 306                 dump_stack();
 307                 clear_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
 308                 smp_mb__after_atomic();
 309                 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG);
 310         }
 311
 312         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 313                 panic("GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
 314
 315         return -1;
 316 }
 317
 318 /**
 319  * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
 320  */
 321
 322 void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
 323                             const char *function, char *file, unsigned int line,
 324                             bool delayed)
 325 {
 326         if (gfs2_withdrawn(sdp))
 327                 return;
 328
 329         fs_err(sdp,
 330                "fatal: assertion \"%s\" failed\n"
 331                "   function = %s, file = %s, line = %u\n",
 332                assertion, function, file, line);
 333
 334         /*
 335          * If errors=panic was specified on mount, it won't help to delay the
 336          * withdraw.
 337          */
 338         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 339                 delayed = false;
 340
 341         if (delayed)
 342                 gfs2_withdraw_delayed(sdp);
 343         else
 344                 gfs2_withdraw(sdp);
 345         dump_stack();
 346 }
 347
 348 /**
 349  * gfs2_assert_warn_i - Print a message to the console if @assertion is false
 350  */
 351
 352 void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
 353                         const char *function, char *file, unsigned int line)
 354 {
 355         if (time_before(jiffies,
 356                         sdp->sd_last_warning +
 357                         gfs2_tune_get(sdp, gt_complain_secs) * HZ))
 358                 return;
 359
 360         if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
 361                 fs_warn(sdp, "warning: assertion \"%s\" failed at function = %s, file = %s, line = %u\n",
 362                         assertion, function, file, line);
 363
 364         if (sdp->sd_args.ar_debug)
 365                 BUG();
 366         else
 367                 dump_stack();
 368
 369         if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
 370                 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed\n"
 371                       "GFS2: fsid=%s:   function = %s, file = %s, line = %u\n",
 372                       sdp->sd_fsname, assertion,
 373                       sdp->sd_fsname, function, file, line);
 374
 375         sdp->sd_last_warning = jiffies;
 376 }
 377
 378 /**
 379  * gfs2_consist_i - Flag a filesystem consistency error and withdraw
 380  */
 381
 382 void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
 383                     char *file, unsigned int line)
 384 {
 385         gfs2_lm(sdp,
 386                 "fatal: filesystem consistency error - function = %s, file = %s, line = %u\n",
 387                 function, file, line);
 388         gfs2_withdraw(sdp);
 389 }
 390
 391 /**
 392  * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
 393  */
 394
 395 void gfs2_consist_inode_i(struct gfs2_inode *ip,
 396                           const char *function, char *file, unsigned int line)
 397 {
 398         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 399
 400         gfs2_lm(sdp,
 401                 "fatal: filesystem consistency error\n"
 402                 "  inode = %llu %llu\n"
 403                 "  function = %s, file = %s, line = %u\n",
 404                 (unsigned long long)ip->i_no_formal_ino,
 405                 (unsigned long long)ip->i_no_addr,
 406                 function, file, line);
 407         gfs2_withdraw(sdp);
 408 }
 409
 410 /**
 411  * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
 412  */
 413
 414 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
 415                           const char *function, char *file, unsigned int line)
 416 {
 417         struct gfs2_sbd *sdp = rgd->rd_sbd;
 418         char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
 419
 420         sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
 421         gfs2_rgrp_dump(NULL, rgd->rd_gl, fs_id_buf);
 422         gfs2_lm(sdp,
 423                 "fatal: filesystem consistency error\n"
 424                 "  RG = %llu\n"
 425                 "  function = %s, file = %s, line = %u\n",
 426                 (unsigned long long)rgd->rd_addr,
 427                 function, file, line);
 428         gfs2_withdraw(sdp);
 429 }
 430
 431 /**
 432  * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
 433  * Returns: -1 if this call withdrew the machine,
 434  *          -2 if it was already withdrawn
 435  */
 436
 437 int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
 438                        const char *type, const char *function, char *file,
 439                        unsigned int line)
 440 {
 441         int me;
 442
 443         gfs2_lm(sdp,
 444                 "fatal: invalid metadata block\n"
 445                 "  bh = %llu (%s)\n"
 446                 "  function = %s, file = %s, line = %u\n",
 447                 (unsigned long long)bh->b_blocknr, type,
 448                 function, file, line);
 449         me = gfs2_withdraw(sdp);
 450         return (me) ? -1 : -2;
 451 }
 452
 453 /**
 454  * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
 455  * Returns: -1 if this call withdrew the machine,
 456  *          -2 if it was already withdrawn
 457  */
 458
 459 int gfs2_metatype_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
 460                            u16 type, u16 t, const char *function,
 461                            char *file, unsigned int line)
 462 {
 463         int me;
 464
 465         gfs2_lm(sdp,
 466                 "fatal: invalid metadata block\n"
 467                 "  bh = %llu (type: exp=%u, found=%u)\n"
 468                 "  function = %s, file = %s, line = %u\n",
 469                 (unsigned long long)bh->b_blocknr, type, t,
 470                 function, file, line);
 471         me = gfs2_withdraw(sdp);
 472         return (me) ? -1 : -2;
 473 }
 474
 475 /**
 476  * gfs2_io_error_i - Flag an I/O error and withdraw
 477  * Returns: -1 if this call withdrew the machine,
 478  *          0 if it was already withdrawn
 479  */
 480
 481 int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
 482                     unsigned int line)
 483 {
 484         gfs2_lm(sdp,
 485                 "fatal: I/O error\n"
 486                 "  function = %s, file = %s, line = %u\n",
 487                 function, file, line);
 488         return gfs2_withdraw(sdp);
 489 }
 490
 491 /**
 492  * gfs2_io_error_bh_i - Flag a buffer I/O error
 493  * @withdraw: withdraw the filesystem
 494  */
 495
 496 void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
 497                         const char *function, char *file, unsigned int line,
 498                         bool withdraw)
 499 {
 500         if (gfs2_withdrawn(sdp))
 501                 return;
 502
 503         fs_err(sdp, "fatal: I/O error\n"
 504                "  block = %llu\n"
 505                "  function = %s, file = %s, line = %u\n",
 506                (unsigned long long)bh->b_blocknr, function, file, line);
 507         if (withdraw)
 508                 gfs2_withdraw(sdp);
 509 }
 510