1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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;
38 void gfs2_assert_i(struct gfs2_sbd *sdp)
40 fs_emerg(sdp, "fatal assertion failed\n");
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
49 * Returns: 0 if the journal is clean or locked, else an error
51 int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
55 struct gfs2_holder j_gh;
56 struct gfs2_log_header_host head;
57 struct gfs2_inode *ip;
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);
64 fs_err(sdp, "Error %d locking journal for spectator "
68 error = gfs2_jdesc_check(jd);
71 fs_err(sdp, "Error checking journal for spectator "
75 error = gfs2_find_jhead(jd, &head, false);
78 fs_err(sdp, "Error parsing journal for spectator "
82 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
85 fs_err(sdp, "jid=%u: Journal is dirty, so the first "
86 "mounter must not be a spectator.\n",
91 gfs2_glock_dq_uninit(&j_gh);
96 * gfs2_freeze_lock - 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
101 int gfs2_freeze_lock(struct gfs2_sbd *sdp, struct gfs2_holder *freeze_gh,
104 int flags = LM_FLAG_NOEXP | GL_EXACT | caller_flags;
107 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, flags,
109 if (error && error != GLR_TRYFAILED)
110 fs_err(sdp, "can't lock the freeze lock: %d\n", error);
114 void gfs2_freeze_unlock(struct gfs2_holder *freeze_gh)
116 if (gfs2_holder_initialized(freeze_gh))
117 gfs2_glock_dq_uninit(freeze_gh);
120 static void signal_our_withdraw(struct gfs2_sbd *sdp)
122 struct gfs2_glock *live_gl = sdp->sd_live_gh.gh_gl;
124 struct gfs2_inode *ip;
125 struct gfs2_glock *i_gl;
127 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
131 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) || !sdp->sd_jdesc)
134 gfs2_ail_drain(sdp); /* frees all transactions */
135 inode = sdp->sd_jdesc->jd_inode;
138 no_formal_ino = ip->i_no_formal_ino;
140 /* Prevent any glock dq until withdraw recovery is complete */
141 set_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
143 * Don't tell dlm we're bailing until we have no more buffers in the
144 * wind. If journal had an IO error, the log code should just purge
145 * the outstanding buffers rather than submitting new IO. Making the
146 * file system read-only will flush the journal, etc.
148 * During a normal unmount, gfs2_make_fs_ro calls gfs2_log_shutdown
149 * which clears SDF_JOURNAL_LIVE. In a withdraw, we must not write
150 * any UNMOUNT log header, so we can't call gfs2_log_shutdown, and
151 * therefore we need to clear SDF_JOURNAL_LIVE manually.
153 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
154 if (!sb_rdonly(sdp->sd_vfs)) {
155 struct gfs2_holder freeze_gh;
157 gfs2_holder_mark_uninitialized(&freeze_gh);
158 if (sdp->sd_freeze_gl &&
159 !gfs2_glock_is_locked_by_me(sdp->sd_freeze_gl)) {
160 ret = gfs2_freeze_lock(sdp, &freeze_gh,
161 log_write_allowed ? 0 : LM_FLAG_TRY);
162 if (ret == GLR_TRYFAILED)
166 gfs2_make_fs_ro(sdp);
167 gfs2_freeze_unlock(&freeze_gh);
170 if (sdp->sd_lockstruct.ls_ops->lm_lock == NULL) { /* lock_nolock */
173 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
177 * Drop the glock for our journal so another node can recover it.
179 if (gfs2_holder_initialized(&sdp->sd_journal_gh)) {
180 gfs2_glock_dq_wait(&sdp->sd_journal_gh);
181 gfs2_holder_uninit(&sdp->sd_journal_gh);
183 sdp->sd_jinode_gh.gh_flags |= GL_NOCACHE;
184 gfs2_glock_dq(&sdp->sd_jinode_gh);
185 if (test_bit(SDF_FS_FROZEN, &sdp->sd_flags)) {
186 /* Make sure gfs2_unfreeze works if partially-frozen */
187 flush_work(&sdp->sd_freeze_work);
188 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
189 thaw_super(sdp->sd_vfs);
191 wait_on_bit(&i_gl->gl_flags, GLF_DEMOTE,
192 TASK_UNINTERRUPTIBLE);
196 * holder_uninit to force glock_put, to force dlm to let go
198 gfs2_holder_uninit(&sdp->sd_jinode_gh);
201 * Note: We need to be careful here:
202 * Our iput of jd_inode will evict it. The evict will dequeue its
203 * glock, but the glock dq will wait for the withdraw unless we have
204 * exception code in glock_dq.
208 * Wait until the journal inode's glock is freed. This allows try locks
209 * on other nodes to be successful, otherwise we remain the owner of
210 * the glock as far as dlm is concerned.
212 if (i_gl->gl_ops->go_free) {
213 set_bit(GLF_FREEING, &i_gl->gl_flags);
214 wait_on_bit(&i_gl->gl_flags, GLF_FREEING, TASK_UNINTERRUPTIBLE);
218 * Dequeue the "live" glock, but keep a reference so it's never freed.
220 gfs2_glock_hold(live_gl);
221 gfs2_glock_dq_wait(&sdp->sd_live_gh);
223 * We enqueue the "live" glock in EX so that all other nodes
224 * get a demote request and act on it. We don't really want the
225 * lock in EX, so we send a "try" lock with 1CB to produce a callback.
227 fs_warn(sdp, "Requesting recovery of jid %d.\n",
228 sdp->sd_lockstruct.ls_jid);
229 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | LM_FLAG_NOEXP,
231 msleep(GL_GLOCK_MAX_HOLD);
233 * This will likely fail in a cluster, but succeed standalone:
235 ret = gfs2_glock_nq(&sdp->sd_live_gh);
238 * If we actually got the "live" lock in EX mode, there are no other
239 * nodes available to replay our journal. So we try to replay it
240 * ourselves. We hold the "live" glock to prevent other mounters
241 * during recovery, then just dequeue it and reacquire it in our
242 * normal SH mode. Just in case the problem that caused us to
243 * withdraw prevents us from recovering our journal (e.g. io errors
244 * and such) we still check if the journal is clean before proceeding
245 * but we may wait forever until another mounter does the recovery.
248 fs_warn(sdp, "No other mounters found. Trying to recover our "
249 "own journal jid %d.\n", sdp->sd_lockstruct.ls_jid);
250 if (gfs2_recover_journal(sdp->sd_jdesc, 1))
251 fs_warn(sdp, "Unable to recover our journal jid %d.\n",
252 sdp->sd_lockstruct.ls_jid);
253 gfs2_glock_dq_wait(&sdp->sd_live_gh);
254 gfs2_holder_reinit(LM_ST_SHARED, LM_FLAG_NOEXP | GL_EXACT,
256 gfs2_glock_nq(&sdp->sd_live_gh);
259 gfs2_glock_queue_put(live_gl); /* drop extra reference we acquired */
260 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
263 * At this point our journal is evicted, so we need to get a new inode
264 * for it. Once done, we need to call gfs2_find_jhead which
265 * calls gfs2_map_journal_extents to map it for us again.
267 * Note that we don't really want it to look up a FREE block. The
268 * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
269 * which would otherwise fail because it requires grabbing an rgrp
270 * glock, which would fail with -EIO because we're withdrawing.
272 inode = gfs2_inode_lookup(sdp->sd_vfs, DT_UNKNOWN,
273 sdp->sd_jdesc->jd_no_addr, no_formal_ino,
276 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
277 sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
280 sdp->sd_jdesc->jd_inode = inode;
283 * Now wait until recovery is complete.
285 for (tries = 0; tries < 10; tries++) {
286 ret = check_journal_clean(sdp, sdp->sd_jdesc, false);
290 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
291 sdp->sd_lockstruct.ls_jid);
295 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
296 sdp->sd_lockstruct.ls_jid);
298 fs_warn(sdp, "Journal recovery skipped for %d until next "
299 "mount.\n", sdp->sd_lockstruct.ls_jid);
300 fs_warn(sdp, "Glock dequeues delayed: %lu\n", sdp->sd_glock_dqs_held);
301 sdp->sd_glock_dqs_held = 0;
302 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY);
305 void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
307 struct va_format vaf;
310 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
311 test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
317 fs_err(sdp, "%pV", &vaf);
321 int gfs2_withdraw(struct gfs2_sbd *sdp)
323 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
324 const struct lm_lockops *lm = ls->ls_ops;
326 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
327 test_and_set_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
328 if (!test_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags))
331 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG,
332 TASK_UNINTERRUPTIBLE);
336 set_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
338 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
339 fs_err(sdp, "about to withdraw this file system\n");
340 BUG_ON(sdp->sd_args.ar_debug);
342 signal_our_withdraw(sdp);
344 kobject_uevent(&sdp->sd_kobj, KOBJ_OFFLINE);
346 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
347 wait_for_completion(&sdp->sd_wdack);
349 if (lm->lm_unmount) {
350 fs_err(sdp, "telling LM to unmount\n");
353 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
354 fs_err(sdp, "File system withdrawn\n");
356 clear_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
357 smp_mb__after_atomic();
358 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG);
361 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
362 panic("GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
368 * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
371 void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
372 const char *function, char *file, unsigned int line,
375 if (gfs2_withdrawn(sdp))
379 "fatal: assertion \"%s\" failed\n"
380 " function = %s, file = %s, line = %u\n",
381 assertion, function, file, line);
384 * If errors=panic was specified on mount, it won't help to delay the
387 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
391 gfs2_withdraw_delayed(sdp);
398 * gfs2_assert_warn_i - Print a message to the console if @assertion is false
401 void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
402 const char *function, char *file, unsigned int line)
404 if (time_before(jiffies,
405 sdp->sd_last_warning +
406 gfs2_tune_get(sdp, gt_complain_secs) * HZ))
409 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
410 fs_warn(sdp, "warning: assertion \"%s\" failed at function = %s, file = %s, line = %u\n",
411 assertion, function, file, line);
413 if (sdp->sd_args.ar_debug)
418 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
419 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed\n"
420 "GFS2: fsid=%s: function = %s, file = %s, line = %u\n",
421 sdp->sd_fsname, assertion,
422 sdp->sd_fsname, function, file, line);
424 sdp->sd_last_warning = jiffies;
428 * gfs2_consist_i - Flag a filesystem consistency error and withdraw
431 void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
432 char *file, unsigned int line)
435 "fatal: filesystem consistency error - function = %s, file = %s, line = %u\n",
436 function, file, line);
441 * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
444 void gfs2_consist_inode_i(struct gfs2_inode *ip,
445 const char *function, char *file, unsigned int line)
447 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
450 "fatal: filesystem consistency error\n"
451 " inode = %llu %llu\n"
452 " function = %s, file = %s, line = %u\n",
453 (unsigned long long)ip->i_no_formal_ino,
454 (unsigned long long)ip->i_no_addr,
455 function, file, line);
460 * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
463 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
464 const char *function, char *file, unsigned int line)
466 struct gfs2_sbd *sdp = rgd->rd_sbd;
467 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
469 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
470 gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
472 "fatal: filesystem consistency error\n"
474 " function = %s, file = %s, line = %u\n",
475 (unsigned long long)rgd->rd_addr,
476 function, file, line);
481 * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
482 * Returns: -1 if this call withdrew the machine,
483 * -2 if it was already withdrawn
486 int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
487 const char *type, const char *function, char *file,
493 "fatal: invalid metadata block\n"
495 " function = %s, file = %s, line = %u\n",
496 (unsigned long long)bh->b_blocknr, type,
497 function, file, line);
498 me = gfs2_withdraw(sdp);
499 return (me) ? -1 : -2;
503 * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
504 * Returns: -1 if this call withdrew the machine,
505 * -2 if it was already withdrawn
508 int gfs2_metatype_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
509 u16 type, u16 t, const char *function,
510 char *file, unsigned int line)
515 "fatal: invalid metadata block\n"
516 " bh = %llu (type: exp=%u, found=%u)\n"
517 " function = %s, file = %s, line = %u\n",
518 (unsigned long long)bh->b_blocknr, type, t,
519 function, file, line);
520 me = gfs2_withdraw(sdp);
521 return (me) ? -1 : -2;
525 * gfs2_io_error_i - Flag an I/O error and withdraw
526 * Returns: -1 if this call withdrew the machine,
527 * 0 if it was already withdrawn
530 int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
535 " function = %s, file = %s, line = %u\n",
536 function, file, line);
537 return gfs2_withdraw(sdp);
541 * gfs2_io_error_bh_i - Flag a buffer I/O error
542 * @withdraw: withdraw the filesystem
545 void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
546 const char *function, char *file, unsigned int line,
549 if (gfs2_withdrawn(sdp))
552 fs_err(sdp, "fatal: I/O error\n"
554 " function = %s, file = %s, line = %u\n",
555 (unsigned long long)bh->b_blocknr, function, file, line);