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/kthread.h>
13 #include <linux/crc32.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/delay.h>
16 #include <linux/uaccess.h>
17
18 #include "gfs2.h"
19 #include "incore.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "log.h"
23 #include "lops.h"
24 #include "recovery.h"
25 #include "rgrp.h"
26 #include "super.h"
27 #include "util.h"
28
29 struct kmem_cache *gfs2_glock_cachep __read_mostly;
30 struct kmem_cache *gfs2_glock_aspace_cachep __read_mostly;
31 struct kmem_cache *gfs2_inode_cachep __read_mostly;
32 struct kmem_cache *gfs2_bufdata_cachep __read_mostly;
33 struct kmem_cache *gfs2_rgrpd_cachep __read_mostly;
34 struct kmem_cache *gfs2_quotad_cachep __read_mostly;
35 struct kmem_cache *gfs2_qadata_cachep __read_mostly;
36 struct kmem_cache *gfs2_trans_cachep __read_mostly;
37 mempool_t *gfs2_page_pool __read_mostly;
38
gfs2_assert_i(struct gfs2_sbd * sdp)39 void gfs2_assert_i(struct gfs2_sbd *sdp)
40 {
41 fs_emerg(sdp, "fatal assertion failed\n");
42 }
43
44 /**
45 * check_journal_clean - Make sure a journal is clean for a spectator mount
46 * @sdp: The GFS2 superblock
47 * @jd: The journal descriptor
48 * @verbose: Show more prints in the log
49 *
50 * Returns: 0 if the journal is clean or locked, else an error
51 */
check_journal_clean(struct gfs2_sbd * sdp,struct gfs2_jdesc * jd,bool verbose)52 int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
53 bool verbose)
54 {
55 int error;
56 struct gfs2_holder j_gh;
57 struct gfs2_log_header_host head;
58 struct gfs2_inode *ip;
59
60 ip = GFS2_I(jd->jd_inode);
61 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
62 GL_EXACT | GL_NOCACHE, &j_gh);
63 if (error) {
64 if (verbose)
65 fs_err(sdp, "Error %d locking journal for spectator "
66 "mount.\n", error);
67 return -EPERM;
68 }
69 error = gfs2_jdesc_check(jd);
70 if (error) {
71 if (verbose)
72 fs_err(sdp, "Error checking journal for spectator "
73 "mount.\n");
74 goto out_unlock;
75 }
76 error = gfs2_find_jhead(jd, &head, false);
77 if (error) {
78 if (verbose)
79 fs_err(sdp, "Error parsing journal for spectator "
80 "mount.\n");
81 goto out_unlock;
82 }
83 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
84 error = -EPERM;
85 if (verbose)
86 fs_err(sdp, "jid=%u: Journal is dirty, so the first "
87 "mounter must not be a spectator.\n",
88 jd->jd_jid);
89 }
90
91 out_unlock:
92 gfs2_glock_dq_uninit(&j_gh);
93 return error;
94 }
95
96 /**
97 * gfs2_freeze_lock_shared - hold the freeze glock
98 * @sdp: the superblock
99 */
gfs2_freeze_lock_shared(struct gfs2_sbd * sdp)100 int gfs2_freeze_lock_shared(struct gfs2_sbd *sdp)
101 {
102 int flags = LM_FLAG_NOEXP | GL_EXACT;
103 int error;
104
105 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, flags,
106 &sdp->sd_freeze_gh);
107 if (error && error != GLR_TRYFAILED)
108 fs_err(sdp, "can't lock the freeze glock: %d\n", error);
109 return error;
110 }
111
gfs2_freeze_unlock(struct gfs2_sbd * sdp)112 void gfs2_freeze_unlock(struct gfs2_sbd *sdp)
113 {
114 if (gfs2_holder_initialized(&sdp->sd_freeze_gh))
115 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
116 }
117
signal_our_withdraw(struct gfs2_sbd * sdp)118 static void signal_our_withdraw(struct gfs2_sbd *sdp)
119 {
120 struct gfs2_glock *live_gl = sdp->sd_live_gh.gh_gl;
121 struct inode *inode;
122 struct gfs2_inode *ip;
123 struct gfs2_glock *i_gl;
124 u64 no_formal_ino;
125 int ret = 0;
126 int tries;
127
128 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) || !sdp->sd_jdesc)
129 return;
130
131 gfs2_ail_drain(sdp); /* frees all transactions */
132 inode = sdp->sd_jdesc->jd_inode;
133 ip = GFS2_I(inode);
134 i_gl = ip->i_gl;
135 no_formal_ino = ip->i_no_formal_ino;
136
137 /* Prevent any glock dq until withdraw recovery is complete */
138 set_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
139 /*
140 * Don't tell dlm we're bailing until we have no more buffers in the
141 * wind. If journal had an IO error, the log code should just purge
142 * the outstanding buffers rather than submitting new IO. Making the
143 * file system read-only will flush the journal, etc.
144 *
145 * During a normal unmount, gfs2_make_fs_ro calls gfs2_log_shutdown
146 * which clears SDF_JOURNAL_LIVE. In a withdraw, we must not write
147 * any UNMOUNT log header, so we can't call gfs2_log_shutdown, and
148 * therefore we need to clear SDF_JOURNAL_LIVE manually.
149 */
150 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
151 if (!sb_rdonly(sdp->sd_vfs)) {
152 bool locked = mutex_trylock(&sdp->sd_freeze_mutex);
153
154 wake_up(&sdp->sd_logd_waitq);
155 wake_up(&sdp->sd_quota_wait);
156
157 wait_event_timeout(sdp->sd_log_waitq,
158 gfs2_log_is_empty(sdp),
159 HZ * 5);
160
161 sdp->sd_vfs->s_flags |= SB_RDONLY;
162
163 if (locked)
164 mutex_unlock(&sdp->sd_freeze_mutex);
165
166 /*
167 * Dequeue any pending non-system glock holders that can no
168 * longer be granted because the file system is withdrawn.
169 */
170 gfs2_gl_dq_holders(sdp);
171 }
172
173 if (sdp->sd_lockstruct.ls_ops->lm_lock == NULL) { /* lock_nolock */
174 if (!ret)
175 ret = -EIO;
176 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
177 goto skip_recovery;
178 }
179 /*
180 * Drop the glock for our journal so another node can recover it.
181 */
182 if (gfs2_holder_initialized(&sdp->sd_journal_gh)) {
183 gfs2_glock_dq_wait(&sdp->sd_journal_gh);
184 gfs2_holder_uninit(&sdp->sd_journal_gh);
185 }
186 sdp->sd_jinode_gh.gh_flags |= GL_NOCACHE;
187 gfs2_glock_dq(&sdp->sd_jinode_gh);
188 gfs2_thaw_freeze_initiator(sdp->sd_vfs);
189 wait_on_bit(&i_gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
190
191 /*
192 * holder_uninit to force glock_put, to force dlm to let go
193 */
194 gfs2_holder_uninit(&sdp->sd_jinode_gh);
195
196 /*
197 * Note: We need to be careful here:
198 * Our iput of jd_inode will evict it. The evict will dequeue its
199 * glock, but the glock dq will wait for the withdraw unless we have
200 * exception code in glock_dq.
201 */
202 iput(inode);
203 sdp->sd_jdesc->jd_inode = NULL;
204 /*
205 * Wait until the journal inode's glock is freed. This allows try locks
206 * on other nodes to be successful, otherwise we remain the owner of
207 * the glock as far as dlm is concerned.
208 */
209 if (i_gl->gl_ops->go_unlocked) {
210 set_bit(GLF_UNLOCKED, &i_gl->gl_flags);
211 wait_on_bit(&i_gl->gl_flags, GLF_UNLOCKED, TASK_UNINTERRUPTIBLE);
212 }
213
214 /*
215 * Dequeue the "live" glock, but keep a reference so it's never freed.
216 */
217 gfs2_glock_hold(live_gl);
218 gfs2_glock_dq_wait(&sdp->sd_live_gh);
219 /*
220 * We enqueue the "live" glock in EX so that all other nodes
221 * get a demote request and act on it. We don't really want the
222 * lock in EX, so we send a "try" lock with 1CB to produce a callback.
223 */
224 fs_warn(sdp, "Requesting recovery of jid %d.\n",
225 sdp->sd_lockstruct.ls_jid);
226 gfs2_holder_reinit(LM_ST_EXCLUSIVE,
227 LM_FLAG_TRY_1CB | LM_FLAG_NOEXP | GL_NOPID,
228 &sdp->sd_live_gh);
229 msleep(GL_GLOCK_MAX_HOLD);
230 /*
231 * This will likely fail in a cluster, but succeed standalone:
232 */
233 ret = gfs2_glock_nq(&sdp->sd_live_gh);
234
235 /*
236 * If we actually got the "live" lock in EX mode, there are no other
237 * nodes available to replay our journal. So we try to replay it
238 * ourselves. We hold the "live" glock to prevent other mounters
239 * during recovery, then just dequeue it and reacquire it in our
240 * normal SH mode. Just in case the problem that caused us to
241 * withdraw prevents us from recovering our journal (e.g. io errors
242 * and such) we still check if the journal is clean before proceeding
243 * but we may wait forever until another mounter does the recovery.
244 */
245 if (ret == 0) {
246 fs_warn(sdp, "No other mounters found. Trying to recover our "
247 "own journal jid %d.\n", sdp->sd_lockstruct.ls_jid);
248 if (gfs2_recover_journal(sdp->sd_jdesc, 1))
249 fs_warn(sdp, "Unable to recover our journal jid %d.\n",
250 sdp->sd_lockstruct.ls_jid);
251 gfs2_glock_dq_wait(&sdp->sd_live_gh);
252 gfs2_holder_reinit(LM_ST_SHARED,
253 LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
254 &sdp->sd_live_gh);
255 gfs2_glock_nq(&sdp->sd_live_gh);
256 }
257
258 gfs2_glock_put(live_gl); /* drop extra reference we acquired */
259 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
260
261 /*
262 * At this point our journal is evicted, so we need to get a new inode
263 * for it. Once done, we need to call gfs2_find_jhead which
264 * calls gfs2_map_journal_extents to map it for us again.
265 *
266 * Note that we don't really want it to look up a FREE block. The
267 * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
268 * which would otherwise fail because it requires grabbing an rgrp
269 * glock, which would fail with -EIO because we're withdrawing.
270 */
271 inode = gfs2_inode_lookup(sdp->sd_vfs, DT_UNKNOWN,
272 sdp->sd_jdesc->jd_no_addr, no_formal_ino,
273 GFS2_BLKST_FREE);
274 if (IS_ERR(inode)) {
275 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
276 sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
277 goto skip_recovery;
278 }
279 sdp->sd_jdesc->jd_inode = inode;
280 d_mark_dontcache(inode);
281
282 /*
283 * Now wait until recovery is complete.
284 */
285 for (tries = 0; tries < 10; tries++) {
286 ret = check_journal_clean(sdp, sdp->sd_jdesc, false);
287 if (!ret)
288 break;
289 msleep(HZ);
290 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
291 sdp->sd_lockstruct.ls_jid);
292 }
293 skip_recovery:
294 if (!ret)
295 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
296 sdp->sd_lockstruct.ls_jid);
297 else
298 fs_warn(sdp, "Journal recovery skipped for jid %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);
303 }
304
gfs2_lm(struct gfs2_sbd * sdp,const char * fmt,...)305 void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
306 {
307 struct va_format vaf;
308 va_list args;
309
310 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
311 test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
312 return;
313
314 va_start(args, fmt);
315 vaf.fmt = fmt;
316 vaf.va = &args;
317 fs_err(sdp, "%pV", &vaf);
318 va_end(args);
319 }
320
gfs2_withdraw(struct gfs2_sbd * sdp)321 int gfs2_withdraw(struct gfs2_sbd *sdp)
322 {
323 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
324 const struct lm_lockops *lm = ls->ls_ops;
325
326 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
327 unsigned long old = READ_ONCE(sdp->sd_flags), new;
328
329 do {
330 if (old & BIT(SDF_WITHDRAWN)) {
331 wait_on_bit(&sdp->sd_flags,
332 SDF_WITHDRAW_IN_PROG,
333 TASK_UNINTERRUPTIBLE);
334 return -1;
335 }
336 new = old | BIT(SDF_WITHDRAWN) | BIT(SDF_WITHDRAW_IN_PROG);
337 } while (unlikely(!try_cmpxchg(&sdp->sd_flags, &old, new)));
338
339 fs_err(sdp, "about to withdraw this file system\n");
340 BUG_ON(sdp->sd_args.ar_debug);
341
342 signal_our_withdraw(sdp);
343
344 kobject_uevent(&sdp->sd_kobj, KOBJ_OFFLINE);
345
346 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
347 wait_for_completion(&sdp->sd_wdack);
348
349 if (lm->lm_unmount) {
350 fs_err(sdp, "telling LM to unmount\n");
351 lm->lm_unmount(sdp);
352 }
353 fs_err(sdp, "File system withdrawn\n");
354 dump_stack();
355 clear_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
356 smp_mb__after_atomic();
357 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG);
358 }
359
360 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
361 panic("GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
362
363 return -1;
364 }
365
366 /*
367 * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
368 */
369
gfs2_assert_withdraw_i(struct gfs2_sbd * sdp,char * assertion,const char * function,char * file,unsigned int line,bool delayed)370 void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
371 const char *function, char *file, unsigned int line,
372 bool delayed)
373 {
374 if (gfs2_withdrawing_or_withdrawn(sdp))
375 return;
376
377 fs_err(sdp,
378 "fatal: assertion \"%s\" failed - "
379 "function = %s, file = %s, line = %u\n",
380 assertion, function, file, line);
381
382 /*
383 * If errors=panic was specified on mount, it won't help to delay the
384 * withdraw.
385 */
386 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
387 delayed = false;
388
389 if (delayed)
390 gfs2_withdraw_delayed(sdp);
391 else
392 gfs2_withdraw(sdp);
393 dump_stack();
394 }
395
396 /*
397 * gfs2_assert_warn_i - Print a message to the console if @assertion is false
398 */
399
gfs2_assert_warn_i(struct gfs2_sbd * sdp,char * assertion,const char * function,char * file,unsigned int line)400 void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
401 const char *function, char *file, unsigned int line)
402 {
403 if (time_before(jiffies,
404 sdp->sd_last_warning +
405 gfs2_tune_get(sdp, gt_complain_secs) * HZ))
406 return;
407
408 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
409 fs_warn(sdp, "warning: assertion \"%s\" failed - "
410 "function = %s, file = %s, line = %u\n",
411 assertion, function, file, line);
412
413 if (sdp->sd_args.ar_debug)
414 BUG();
415 else
416 dump_stack();
417
418 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
419 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed - "
420 "function = %s, file = %s, line = %u\n",
421 sdp->sd_fsname, assertion,
422 function, file, line);
423
424 sdp->sd_last_warning = jiffies;
425 }
426
427 /*
428 * gfs2_consist_i - Flag a filesystem consistency error and withdraw
429 */
430
gfs2_consist_i(struct gfs2_sbd * sdp,const char * function,char * file,unsigned int line)431 void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
432 char *file, unsigned int line)
433 {
434 gfs2_lm(sdp,
435 "fatal: filesystem consistency error - "
436 "function = %s, file = %s, line = %u\n",
437 function, file, line);
438 gfs2_withdraw(sdp);
439 }
440
441 /*
442 * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
443 */
444
gfs2_consist_inode_i(struct gfs2_inode * ip,const char * function,char * file,unsigned int line)445 void gfs2_consist_inode_i(struct gfs2_inode *ip,
446 const char *function, char *file, unsigned int line)
447 {
448 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
449
450 gfs2_lm(sdp,
451 "fatal: filesystem consistency error - "
452 "inode = %llu %llu, "
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);
458 gfs2_withdraw(sdp);
459 }
460
461 /*
462 * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
463 */
464
gfs2_consist_rgrpd_i(struct gfs2_rgrpd * rgd,const char * function,char * file,unsigned int line)465 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
466 const char *function, char *file, unsigned int line)
467 {
468 struct gfs2_sbd *sdp = rgd->rd_sbd;
469 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
470
471 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
472 gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
473 gfs2_lm(sdp,
474 "fatal: filesystem consistency error - "
475 "RG = %llu, "
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);
480 gfs2_withdraw(sdp);
481 }
482
483 /*
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
487 */
488
gfs2_meta_check_ii(struct gfs2_sbd * sdp,struct buffer_head * bh,const char * function,char * file,unsigned int line)489 int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
490 const char *function, char *file,
491 unsigned int line)
492 {
493 int me;
494
495 gfs2_lm(sdp,
496 "fatal: invalid metadata block - "
497 "bh = %llu (bad magic number), "
498 "function = %s, file = %s, line = %u\n",
499 (unsigned long long)bh->b_blocknr,
500 function, file, line);
501 me = gfs2_withdraw(sdp);
502 return (me) ? -1 : -2;
503 }
504
505 /*
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
509 */
510
gfs2_metatype_check_ii(struct gfs2_sbd * sdp,struct buffer_head * bh,u16 type,u16 t,const char * function,char * file,unsigned int line)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)
514 {
515 int me;
516
517 gfs2_lm(sdp,
518 "fatal: invalid metadata block - "
519 "bh = %llu (type: exp=%u, found=%u), "
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;
525 }
526
527 /*
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
531 */
532
gfs2_io_error_i(struct gfs2_sbd * sdp,const char * function,char * file,unsigned int line)533 int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
534 unsigned int line)
535 {
536 gfs2_lm(sdp,
537 "fatal: I/O error - "
538 "function = %s, file = %s, line = %u\n",
539 function, file, line);
540 return gfs2_withdraw(sdp);
541 }
542
543 /*
544 * gfs2_io_error_bh_i - Flag a buffer I/O error
545 * @withdraw: withdraw the filesystem
546 */
547
gfs2_io_error_bh_i(struct gfs2_sbd * sdp,struct buffer_head * bh,const char * function,char * file,unsigned int line,bool withdraw)548 void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
549 const char *function, char *file, unsigned int line,
550 bool withdraw)
551 {
552 if (gfs2_withdrawing_or_withdrawn(sdp))
553 return;
554
555 fs_err(sdp, "fatal: I/O error - "
556 "block = %llu, "
557 "function = %s, file = %s, line = %u\n",
558 (unsigned long long)bh->b_blocknr, function, file, line);
559 if (withdraw)
560 gfs2_withdraw(sdp);
561 }
562
563