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);
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 gfs2_glock_put(live_gl); /* drop extra reference we acquired */
236 clear_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags);
237
238 /*
239 * If we actually got the "live" lock in EX mode, there are no other
240 * nodes available to replay our journal.
241 */
242 if (ret == 0) {
243 fs_warn(sdp, "No other mounters found.\n");
244 /*
245 * We are about to release the lockspace. By keeping live_gl
246 * locked here, we ensure that the next mounter coming along
247 * will be a "first" mounter which will perform recovery.
248 */
249 goto skip_recovery;
250 }
251
252 /*
253 * At this point our journal is evicted, so we need to get a new inode
254 * for it. Once done, we need to call gfs2_find_jhead which
255 * calls gfs2_map_journal_extents to map it for us again.
256 *
257 * Note that we don't really want it to look up a FREE block. The
258 * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
259 * which would otherwise fail because it requires grabbing an rgrp
260 * glock, which would fail with -EIO because we're withdrawing.
261 */
262 inode = gfs2_inode_lookup(sdp->sd_vfs, DT_UNKNOWN,
263 sdp->sd_jdesc->jd_no_addr, no_formal_ino,
264 GFS2_BLKST_FREE);
265 if (IS_ERR(inode)) {
266 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
267 sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
268 goto skip_recovery;
269 }
270 sdp->sd_jdesc->jd_inode = inode;
271 d_mark_dontcache(inode);
272
273 /*
274 * Now wait until recovery is complete.
275 */
276 for (tries = 0; tries < 10; tries++) {
277 ret = check_journal_clean(sdp, sdp->sd_jdesc, false);
278 if (!ret)
279 break;
280 msleep(HZ);
281 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
282 sdp->sd_lockstruct.ls_jid);
283 }
284 skip_recovery:
285 if (!ret)
286 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
287 sdp->sd_lockstruct.ls_jid);
288 else
289 fs_warn(sdp, "Journal recovery skipped for jid %d until next "
290 "mount.\n", sdp->sd_lockstruct.ls_jid);
291 fs_warn(sdp, "Glock dequeues delayed: %lu\n", sdp->sd_glock_dqs_held);
292 sdp->sd_glock_dqs_held = 0;
293 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY);
294 }
295
gfs2_lm(struct gfs2_sbd * sdp,const char * fmt,...)296 void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
297 {
298 struct va_format vaf;
299 va_list args;
300
301 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
302 test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
303 return;
304
305 va_start(args, fmt);
306 vaf.fmt = fmt;
307 vaf.va = &args;
308 fs_err(sdp, "%pV", &vaf);
309 va_end(args);
310 }
311
gfs2_withdraw(struct gfs2_sbd * sdp)312 int gfs2_withdraw(struct gfs2_sbd *sdp)
313 {
314 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
315 const struct lm_lockops *lm = ls->ls_ops;
316
317 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
318 unsigned long old = READ_ONCE(sdp->sd_flags), new;
319
320 do {
321 if (old & BIT(SDF_WITHDRAWN)) {
322 wait_on_bit(&sdp->sd_flags,
323 SDF_WITHDRAW_IN_PROG,
324 TASK_UNINTERRUPTIBLE);
325 return -1;
326 }
327 new = old | BIT(SDF_WITHDRAWN) | BIT(SDF_WITHDRAW_IN_PROG);
328 } while (unlikely(!try_cmpxchg(&sdp->sd_flags, &old, new)));
329
330 fs_err(sdp, "about to withdraw this file system\n");
331 BUG_ON(sdp->sd_args.ar_debug);
332
333 signal_our_withdraw(sdp);
334
335 kobject_uevent(&sdp->sd_kobj, KOBJ_OFFLINE);
336
337 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
338 wait_for_completion(&sdp->sd_wdack);
339
340 if (lm->lm_unmount) {
341 fs_err(sdp, "telling LM to unmount\n");
342 lm->lm_unmount(sdp);
343 }
344 fs_err(sdp, "File system withdrawn\n");
345 dump_stack();
346 clear_bit(SDF_WITHDRAW_IN_PROG, &sdp->sd_flags);
347 smp_mb__after_atomic();
348 wake_up_bit(&sdp->sd_flags, SDF_WITHDRAW_IN_PROG);
349 }
350
351 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
352 panic("GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
353
354 return -1;
355 }
356
357 /*
358 * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
359 */
360
gfs2_assert_withdraw_i(struct gfs2_sbd * sdp,char * assertion,const char * function,char * file,unsigned int line,bool delayed)361 void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
362 const char *function, char *file, unsigned int line,
363 bool delayed)
364 {
365 if (gfs2_withdrawing_or_withdrawn(sdp))
366 return;
367
368 fs_err(sdp,
369 "fatal: assertion \"%s\" failed - "
370 "function = %s, file = %s, line = %u\n",
371 assertion, function, file, line);
372
373 /*
374 * If errors=panic was specified on mount, it won't help to delay the
375 * withdraw.
376 */
377 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
378 delayed = false;
379
380 if (delayed)
381 gfs2_withdraw_delayed(sdp);
382 else
383 gfs2_withdraw(sdp);
384 dump_stack();
385 }
386
387 /*
388 * gfs2_assert_warn_i - Print a message to the console if @assertion is false
389 */
390
gfs2_assert_warn_i(struct gfs2_sbd * sdp,char * assertion,const char * function,char * file,unsigned int line)391 void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
392 const char *function, char *file, unsigned int line)
393 {
394 if (time_before(jiffies,
395 sdp->sd_last_warning +
396 gfs2_tune_get(sdp, gt_complain_secs) * HZ))
397 return;
398
399 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
400 fs_warn(sdp, "warning: assertion \"%s\" failed - "
401 "function = %s, file = %s, line = %u\n",
402 assertion, function, file, line);
403
404 if (sdp->sd_args.ar_debug)
405 BUG();
406 else
407 dump_stack();
408
409 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
410 panic("GFS2: fsid=%s: warning: assertion \"%s\" failed - "
411 "function = %s, file = %s, line = %u\n",
412 sdp->sd_fsname, assertion,
413 function, file, line);
414
415 sdp->sd_last_warning = jiffies;
416 }
417
418 /*
419 * gfs2_consist_i - Flag a filesystem consistency error and withdraw
420 */
421
gfs2_consist_i(struct gfs2_sbd * sdp,const char * function,char * file,unsigned int line)422 void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
423 char *file, unsigned int line)
424 {
425 gfs2_lm(sdp,
426 "fatal: filesystem consistency error - "
427 "function = %s, file = %s, line = %u\n",
428 function, file, line);
429 gfs2_withdraw(sdp);
430 }
431
432 /*
433 * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
434 */
435
gfs2_consist_inode_i(struct gfs2_inode * ip,const char * function,char * file,unsigned int line)436 void gfs2_consist_inode_i(struct gfs2_inode *ip,
437 const char *function, char *file, unsigned int line)
438 {
439 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
440
441 gfs2_lm(sdp,
442 "fatal: filesystem consistency error - "
443 "inode = %llu %llu, "
444 "function = %s, file = %s, line = %u\n",
445 (unsigned long long)ip->i_no_formal_ino,
446 (unsigned long long)ip->i_no_addr,
447 function, file, line);
448 gfs2_dump_glock(NULL, ip->i_gl, 1);
449 gfs2_withdraw(sdp);
450 }
451
452 /*
453 * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
454 */
455
gfs2_consist_rgrpd_i(struct gfs2_rgrpd * rgd,const char * function,char * file,unsigned int line)456 void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
457 const char *function, char *file, unsigned int line)
458 {
459 struct gfs2_sbd *sdp = rgd->rd_sbd;
460 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
461
462 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
463 gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
464 gfs2_lm(sdp,
465 "fatal: filesystem consistency error - "
466 "RG = %llu, "
467 "function = %s, file = %s, line = %u\n",
468 (unsigned long long)rgd->rd_addr,
469 function, file, line);
470 gfs2_dump_glock(NULL, rgd->rd_gl, 1);
471 gfs2_withdraw(sdp);
472 }
473
474 /*
475 * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
476 * Returns: -1 if this call withdrew the machine,
477 * -2 if it was already withdrawn
478 */
479
gfs2_meta_check_ii(struct gfs2_sbd * sdp,struct buffer_head * bh,const char * function,char * file,unsigned int line)480 int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
481 const char *function, char *file,
482 unsigned int line)
483 {
484 int me;
485
486 gfs2_lm(sdp,
487 "fatal: invalid metadata block - "
488 "bh = %llu (bad magic number), "
489 "function = %s, file = %s, line = %u\n",
490 (unsigned long long)bh->b_blocknr,
491 function, file, line);
492 me = gfs2_withdraw(sdp);
493 return (me) ? -1 : -2;
494 }
495
496 /*
497 * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
498 * Returns: -1 if this call withdrew the machine,
499 * -2 if it was already withdrawn
500 */
501
gfs2_metatype_check_ii(struct gfs2_sbd * sdp,struct buffer_head * bh,u16 type,u16 t,const char * function,char * file,unsigned int line)502 int gfs2_metatype_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
503 u16 type, u16 t, const char *function,
504 char *file, unsigned int line)
505 {
506 int me;
507
508 gfs2_lm(sdp,
509 "fatal: invalid metadata block - "
510 "bh = %llu (type: exp=%u, found=%u), "
511 "function = %s, file = %s, line = %u\n",
512 (unsigned long long)bh->b_blocknr, type, t,
513 function, file, line);
514 me = gfs2_withdraw(sdp);
515 return (me) ? -1 : -2;
516 }
517
518 /*
519 * gfs2_io_error_i - Flag an I/O error and withdraw
520 * Returns: -1 if this call withdrew the machine,
521 * 0 if it was already withdrawn
522 */
523
gfs2_io_error_i(struct gfs2_sbd * sdp,const char * function,char * file,unsigned int line)524 int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
525 unsigned int line)
526 {
527 gfs2_lm(sdp,
528 "fatal: I/O error - "
529 "function = %s, file = %s, line = %u\n",
530 function, file, line);
531 return gfs2_withdraw(sdp);
532 }
533
534 /*
535 * gfs2_io_error_bh_i - Flag a buffer I/O error
536 * @withdraw: withdraw the filesystem
537 */
538
gfs2_io_error_bh_i(struct gfs2_sbd * sdp,struct buffer_head * bh,const char * function,char * file,unsigned int line,bool withdraw)539 void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
540 const char *function, char *file, unsigned int line,
541 bool withdraw)
542 {
543 if (gfs2_withdrawing_or_withdrawn(sdp))
544 return;
545
546 fs_err(sdp, "fatal: I/O error - "
547 "block = %llu, "
548 "function = %s, file = %s, line = %u\n",
549 (unsigned long long)bh->b_blocknr, function, file, line);
550 if (withdraw)
551 gfs2_withdraw(sdp);
552 }
553
554