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