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
;
281 d_mark_dontcache(inode
);
284 * Now wait until recovery is complete.
286 for (tries
= 0; tries
< 10; tries
++) {
287 ret
= check_journal_clean(sdp
, sdp
->sd_jdesc
, false);
291 fs_warn(sdp
, "Waiting for journal recovery jid %d.\n",
292 sdp
->sd_lockstruct
.ls_jid
);
296 fs_warn(sdp
, "Journal recovery complete for jid %d.\n",
297 sdp
->sd_lockstruct
.ls_jid
);
299 fs_warn(sdp
, "Journal recovery skipped for jid %d until next "
300 "mount.\n", sdp
->sd_lockstruct
.ls_jid
);
301 fs_warn(sdp
, "Glock dequeues delayed: %lu\n", sdp
->sd_glock_dqs_held
);
302 sdp
->sd_glock_dqs_held
= 0;
303 wake_up_bit(&sdp
->sd_flags
, SDF_WITHDRAW_RECOVERY
);
306 void gfs2_lm(struct gfs2_sbd
*sdp
, const char *fmt
, ...)
308 struct va_format vaf
;
311 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_WITHDRAW
&&
312 test_bit(SDF_WITHDRAWN
, &sdp
->sd_flags
))
318 fs_err(sdp
, "%pV", &vaf
);
322 int gfs2_withdraw(struct gfs2_sbd
*sdp
)
324 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
325 const struct lm_lockops
*lm
= ls
->ls_ops
;
327 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_WITHDRAW
&&
328 test_and_set_bit(SDF_WITHDRAWN
, &sdp
->sd_flags
)) {
329 if (!test_bit(SDF_WITHDRAW_IN_PROG
, &sdp
->sd_flags
))
332 wait_on_bit(&sdp
->sd_flags
, SDF_WITHDRAW_IN_PROG
,
333 TASK_UNINTERRUPTIBLE
);
337 set_bit(SDF_WITHDRAW_IN_PROG
, &sdp
->sd_flags
);
339 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_WITHDRAW
) {
340 fs_err(sdp
, "about to withdraw this file system\n");
341 BUG_ON(sdp
->sd_args
.ar_debug
);
343 signal_our_withdraw(sdp
);
345 kobject_uevent(&sdp
->sd_kobj
, KOBJ_OFFLINE
);
347 if (!strcmp(sdp
->sd_lockstruct
.ls_ops
->lm_proto_name
, "lock_dlm"))
348 wait_for_completion(&sdp
->sd_wdack
);
350 if (lm
->lm_unmount
) {
351 fs_err(sdp
, "telling LM to unmount\n");
354 set_bit(SDF_SKIP_DLM_UNLOCK
, &sdp
->sd_flags
);
355 fs_err(sdp
, "File system withdrawn\n");
357 clear_bit(SDF_WITHDRAW_IN_PROG
, &sdp
->sd_flags
);
358 smp_mb__after_atomic();
359 wake_up_bit(&sdp
->sd_flags
, SDF_WITHDRAW_IN_PROG
);
362 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_PANIC
)
363 panic("GFS2: fsid=%s: panic requested\n", sdp
->sd_fsname
);
369 * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
372 void gfs2_assert_withdraw_i(struct gfs2_sbd
*sdp
, char *assertion
,
373 const char *function
, char *file
, unsigned int line
,
376 if (gfs2_withdrawn(sdp
))
380 "fatal: assertion \"%s\" failed\n"
381 " function = %s, file = %s, line = %u\n",
382 assertion
, function
, file
, line
);
385 * If errors=panic was specified on mount, it won't help to delay the
388 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_PANIC
)
392 gfs2_withdraw_delayed(sdp
);
399 * gfs2_assert_warn_i - Print a message to the console if @assertion is false
402 void gfs2_assert_warn_i(struct gfs2_sbd
*sdp
, char *assertion
,
403 const char *function
, char *file
, unsigned int line
)
405 if (time_before(jiffies
,
406 sdp
->sd_last_warning
+
407 gfs2_tune_get(sdp
, gt_complain_secs
) * HZ
))
410 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_WITHDRAW
)
411 fs_warn(sdp
, "warning: assertion \"%s\" failed at function = %s, file = %s, line = %u\n",
412 assertion
, function
, file
, line
);
414 if (sdp
->sd_args
.ar_debug
)
419 if (sdp
->sd_args
.ar_errors
== GFS2_ERRORS_PANIC
)
420 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed\n"
421 "GFS2: fsid=%s: function = %s, file = %s, line = %u\n",
422 sdp
->sd_fsname
, assertion
,
423 sdp
->sd_fsname
, function
, file
, line
);
425 sdp
->sd_last_warning
= jiffies
;
429 * gfs2_consist_i - Flag a filesystem consistency error and withdraw
432 void gfs2_consist_i(struct gfs2_sbd
*sdp
, const char *function
,
433 char *file
, unsigned int line
)
436 "fatal: filesystem consistency error - function = %s, file = %s, line = %u\n",
437 function
, file
, line
);
442 * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
445 void gfs2_consist_inode_i(struct gfs2_inode
*ip
,
446 const char *function
, char *file
, unsigned int line
)
448 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
451 "fatal: filesystem consistency error\n"
452 " inode = %llu %llu\n"
453 " function = %s, file = %s, line = %u\n",
454 (unsigned long long)ip
->i_no_formal_ino
,
455 (unsigned long long)ip
->i_no_addr
,
456 function
, file
, line
);
457 gfs2_dump_glock(NULL
, ip
->i_gl
, 1);
462 * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
465 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd
*rgd
,
466 const char *function
, char *file
, unsigned int line
)
468 struct gfs2_sbd
*sdp
= rgd
->rd_sbd
;
469 char fs_id_buf
[sizeof(sdp
->sd_fsname
) + 7];
471 sprintf(fs_id_buf
, "fsid=%s: ", sdp
->sd_fsname
);
472 gfs2_rgrp_dump(NULL
, rgd
, fs_id_buf
);
474 "fatal: filesystem consistency error\n"
476 " function = %s, file = %s, line = %u\n",
477 (unsigned long long)rgd
->rd_addr
,
478 function
, file
, line
);
479 gfs2_dump_glock(NULL
, rgd
->rd_gl
, 1);
484 * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
485 * Returns: -1 if this call withdrew the machine,
486 * -2 if it was already withdrawn
489 int gfs2_meta_check_ii(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
,
490 const char *type
, const char *function
, char *file
,
496 "fatal: invalid metadata block\n"
498 " function = %s, file = %s, line = %u\n",
499 (unsigned long long)bh
->b_blocknr
, type
,
500 function
, file
, line
);
501 me
= gfs2_withdraw(sdp
);
502 return (me
) ? -1 : -2;
506 * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
507 * Returns: -1 if this call withdrew the machine,
508 * -2 if it was already withdrawn
511 int gfs2_metatype_check_ii(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
,
512 u16 type
, u16 t
, const char *function
,
513 char *file
, unsigned int line
)
518 "fatal: invalid metadata block\n"
519 " bh = %llu (type: exp=%u, found=%u)\n"
520 " function = %s, file = %s, line = %u\n",
521 (unsigned long long)bh
->b_blocknr
, type
, t
,
522 function
, file
, line
);
523 me
= gfs2_withdraw(sdp
);
524 return (me
) ? -1 : -2;
528 * gfs2_io_error_i - Flag an I/O error and withdraw
529 * Returns: -1 if this call withdrew the machine,
530 * 0 if it was already withdrawn
533 int gfs2_io_error_i(struct gfs2_sbd
*sdp
, const char *function
, char *file
,
538 " function = %s, file = %s, line = %u\n",
539 function
, file
, line
);
540 return gfs2_withdraw(sdp
);
544 * gfs2_io_error_bh_i - Flag a buffer I/O error
545 * @withdraw: withdraw the filesystem
548 void gfs2_io_error_bh_i(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
,
549 const char *function
, char *file
, unsigned int line
,
552 if (gfs2_withdrawn(sdp
))
555 fs_err(sdp
, "fatal: I/O error\n"
557 " function = %s, file = %s, line = %u\n",
558 (unsigned long long)bh
->b_blocknr
, function
, file
, line
);