xref: /linux/fs/gfs2/recovery.c (revision 9a379e77033f02c4a071891afdf0f0a01eff8ccb)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/module.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/crc32.h>
17 #include <linux/crc32c.h>
18 
19 #include "gfs2.h"
20 #include "incore.h"
21 #include "bmap.h"
22 #include "glock.h"
23 #include "glops.h"
24 #include "log.h"
25 #include "lops.h"
26 #include "meta_io.h"
27 #include "recovery.h"
28 #include "super.h"
29 #include "util.h"
30 #include "dir.h"
31 
32 struct workqueue_struct *gfs_recovery_wq;
33 
34 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
35 			   struct buffer_head **bh)
36 {
37 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
38 	struct gfs2_glock *gl = ip->i_gl;
39 	int new = 0;
40 	u64 dblock;
41 	u32 extlen;
42 	int error;
43 
44 	error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
45 	if (error)
46 		return error;
47 	if (!dblock) {
48 		gfs2_consist_inode(ip);
49 		return -EIO;
50 	}
51 
52 	*bh = gfs2_meta_ra(gl, dblock, extlen);
53 
54 	return error;
55 }
56 
57 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
58 {
59 	struct list_head *head = &jd->jd_revoke_list;
60 	struct gfs2_revoke_replay *rr;
61 	int found = 0;
62 
63 	list_for_each_entry(rr, head, rr_list) {
64 		if (rr->rr_blkno == blkno) {
65 			found = 1;
66 			break;
67 		}
68 	}
69 
70 	if (found) {
71 		rr->rr_where = where;
72 		return 0;
73 	}
74 
75 	rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
76 	if (!rr)
77 		return -ENOMEM;
78 
79 	rr->rr_blkno = blkno;
80 	rr->rr_where = where;
81 	list_add(&rr->rr_list, head);
82 
83 	return 1;
84 }
85 
86 int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
87 {
88 	struct gfs2_revoke_replay *rr;
89 	int wrap, a, b, revoke;
90 	int found = 0;
91 
92 	list_for_each_entry(rr, &jd->jd_revoke_list, rr_list) {
93 		if (rr->rr_blkno == blkno) {
94 			found = 1;
95 			break;
96 		}
97 	}
98 
99 	if (!found)
100 		return 0;
101 
102 	wrap = (rr->rr_where < jd->jd_replay_tail);
103 	a = (jd->jd_replay_tail < where);
104 	b = (where < rr->rr_where);
105 	revoke = (wrap) ? (a || b) : (a && b);
106 
107 	return revoke;
108 }
109 
110 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
111 {
112 	struct list_head *head = &jd->jd_revoke_list;
113 	struct gfs2_revoke_replay *rr;
114 
115 	while (!list_empty(head)) {
116 		rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
117 		list_del(&rr->rr_list);
118 		kfree(rr);
119 	}
120 }
121 
122 /**
123  * get_log_header - read the log header for a given segment
124  * @jd: the journal
125  * @blk: the block to look at
126  * @lh: the log header to return
127  *
128  * Read the log header for a given segement in a given journal.  Do a few
129  * sanity checks on it.
130  *
131  * Returns: 0 on success,
132  *          1 if the header was invalid or incomplete,
133  *          errno on error
134  */
135 
136 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
137 			  struct gfs2_log_header_host *head)
138 {
139 	struct gfs2_log_header *lh;
140 	struct buffer_head *bh;
141 	u32 hash, crc;
142 	int error;
143 
144 	error = gfs2_replay_read_block(jd, blk, &bh);
145 	if (error)
146 		return error;
147 	lh = (void *)bh->b_data;
148 
149 	hash = crc32(~0, lh, LH_V1_SIZE - 4);
150 	hash = ~crc32_le_shift(hash, 4);  /* assume lh_hash is zero */
151 
152 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
153 		     bh->b_size - LH_V1_SIZE - 4);
154 
155 	error = lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
156 		lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
157 		be32_to_cpu(lh->lh_blkno) != blk ||
158 		be32_to_cpu(lh->lh_hash) != hash ||
159 		(lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc);
160 
161 	brelse(bh);
162 
163 	if (!error) {
164 		head->lh_sequence = be64_to_cpu(lh->lh_sequence);
165 		head->lh_flags = be32_to_cpu(lh->lh_flags);
166 		head->lh_tail = be32_to_cpu(lh->lh_tail);
167 		head->lh_blkno = be32_to_cpu(lh->lh_blkno);
168 	}
169 	return error;
170 }
171 
172 /**
173  * find_good_lh - find a good log header
174  * @jd: the journal
175  * @blk: the segment to start searching from
176  * @lh: the log header to fill in
177  * @forward: if true search forward in the log, else search backward
178  *
179  * Call get_log_header() to get a log header for a segment, but if the
180  * segment is bad, either scan forward or backward until we find a good one.
181  *
182  * Returns: errno
183  */
184 
185 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
186 			struct gfs2_log_header_host *head)
187 {
188 	unsigned int orig_blk = *blk;
189 	int error;
190 
191 	for (;;) {
192 		error = get_log_header(jd, *blk, head);
193 		if (error <= 0)
194 			return error;
195 
196 		if (++*blk == jd->jd_blocks)
197 			*blk = 0;
198 
199 		if (*blk == orig_blk) {
200 			gfs2_consist_inode(GFS2_I(jd->jd_inode));
201 			return -EIO;
202 		}
203 	}
204 }
205 
206 /**
207  * jhead_scan - make sure we've found the head of the log
208  * @jd: the journal
209  * @head: this is filled in with the log descriptor of the head
210  *
211  * At this point, seg and lh should be either the head of the log or just
212  * before.  Scan forward until we find the head.
213  *
214  * Returns: errno
215  */
216 
217 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
218 {
219 	unsigned int blk = head->lh_blkno;
220 	struct gfs2_log_header_host lh;
221 	int error;
222 
223 	for (;;) {
224 		if (++blk == jd->jd_blocks)
225 			blk = 0;
226 
227 		error = get_log_header(jd, blk, &lh);
228 		if (error < 0)
229 			return error;
230 		if (error == 1)
231 			continue;
232 
233 		if (lh.lh_sequence == head->lh_sequence) {
234 			gfs2_consist_inode(GFS2_I(jd->jd_inode));
235 			return -EIO;
236 		}
237 		if (lh.lh_sequence < head->lh_sequence)
238 			break;
239 
240 		*head = lh;
241 	}
242 
243 	return 0;
244 }
245 
246 /**
247  * gfs2_find_jhead - find the head of a log
248  * @jd: the journal
249  * @head: the log descriptor for the head of the log is returned here
250  *
251  * Do a binary search of a journal and find the valid log entry with the
252  * highest sequence number.  (i.e. the log head)
253  *
254  * Returns: errno
255  */
256 
257 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
258 {
259 	struct gfs2_log_header_host lh_1, lh_m;
260 	u32 blk_1, blk_2, blk_m;
261 	int error;
262 
263 	blk_1 = 0;
264 	blk_2 = jd->jd_blocks - 1;
265 
266 	for (;;) {
267 		blk_m = (blk_1 + blk_2) / 2;
268 
269 		error = find_good_lh(jd, &blk_1, &lh_1);
270 		if (error)
271 			return error;
272 
273 		error = find_good_lh(jd, &blk_m, &lh_m);
274 		if (error)
275 			return error;
276 
277 		if (blk_1 == blk_m || blk_m == blk_2)
278 			break;
279 
280 		if (lh_1.lh_sequence <= lh_m.lh_sequence)
281 			blk_1 = blk_m;
282 		else
283 			blk_2 = blk_m;
284 	}
285 
286 	error = jhead_scan(jd, &lh_1);
287 	if (error)
288 		return error;
289 
290 	*head = lh_1;
291 
292 	return error;
293 }
294 
295 /**
296  * foreach_descriptor - go through the active part of the log
297  * @jd: the journal
298  * @start: the first log header in the active region
299  * @end: the last log header (don't process the contents of this entry))
300  *
301  * Call a given function once for every log descriptor in the active
302  * portion of the log.
303  *
304  * Returns: errno
305  */
306 
307 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
308 			      unsigned int end, int pass)
309 {
310 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
311 	struct buffer_head *bh;
312 	struct gfs2_log_descriptor *ld;
313 	int error = 0;
314 	u32 length;
315 	__be64 *ptr;
316 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
317 	offset += sizeof(__be64) - 1;
318 	offset &= ~(sizeof(__be64) - 1);
319 
320 	while (start != end) {
321 		error = gfs2_replay_read_block(jd, start, &bh);
322 		if (error)
323 			return error;
324 		if (gfs2_meta_check(sdp, bh)) {
325 			brelse(bh);
326 			return -EIO;
327 		}
328 		ld = (struct gfs2_log_descriptor *)bh->b_data;
329 		length = be32_to_cpu(ld->ld_length);
330 
331 		if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
332 			struct gfs2_log_header_host lh;
333 			error = get_log_header(jd, start, &lh);
334 			if (!error) {
335 				gfs2_replay_incr_blk(jd, &start);
336 				brelse(bh);
337 				continue;
338 			}
339 			if (error == 1) {
340 				gfs2_consist_inode(GFS2_I(jd->jd_inode));
341 				error = -EIO;
342 			}
343 			brelse(bh);
344 			return error;
345 		} else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
346 			brelse(bh);
347 			return -EIO;
348 		}
349 		ptr = (__be64 *)(bh->b_data + offset);
350 		error = lops_scan_elements(jd, start, ld, ptr, pass);
351 		if (error) {
352 			brelse(bh);
353 			return error;
354 		}
355 
356 		while (length--)
357 			gfs2_replay_incr_blk(jd, &start);
358 
359 		brelse(bh);
360 	}
361 
362 	return 0;
363 }
364 
365 /**
366  * clean_journal - mark a dirty journal as being clean
367  * @jd: the journal
368  * @head: the head journal to start from
369  *
370  * Returns: errno
371  */
372 
373 static void clean_journal(struct gfs2_jdesc *jd,
374 			  struct gfs2_log_header_host *head)
375 {
376 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
377 
378 	sdp->sd_log_flush_head = head->lh_blkno;
379 	gfs2_replay_incr_blk(jd, &sdp->sd_log_flush_head);
380 	gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0,
381 			      GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
382 			      REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
383 }
384 
385 
386 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
387                                unsigned int message)
388 {
389 	char env_jid[20];
390 	char env_status[20];
391 	char *envp[] = { env_jid, env_status, NULL };
392 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
393 
394         ls->ls_recover_jid_done = jid;
395         ls->ls_recover_jid_status = message;
396 	sprintf(env_jid, "JID=%u", jid);
397 	sprintf(env_status, "RECOVERY=%s",
398 		message == LM_RD_SUCCESS ? "Done" : "Failed");
399         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
400 
401 	if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
402 		sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
403 }
404 
405 void gfs2_recover_func(struct work_struct *work)
406 {
407 	struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
408 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
409 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
410 	struct gfs2_log_header_host head;
411 	struct gfs2_holder j_gh, ji_gh, thaw_gh;
412 	unsigned long t;
413 	int ro = 0;
414 	unsigned int pass;
415 	int error;
416 	int jlocked = 0;
417 
418 	if (sdp->sd_args.ar_spectator ||
419 	    (jd->jd_jid != sdp->sd_lockstruct.ls_jid)) {
420 		fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
421 			jd->jd_jid);
422 		jlocked = 1;
423 		/* Acquire the journal lock so we can do recovery */
424 
425 		error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
426 					  LM_ST_EXCLUSIVE,
427 					  LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
428 					  &j_gh);
429 		switch (error) {
430 		case 0:
431 			break;
432 
433 		case GLR_TRYFAILED:
434 			fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
435 			error = 0;
436 
437 		default:
438 			goto fail;
439 		};
440 
441 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
442 					   LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
443 		if (error)
444 			goto fail_gunlock_j;
445 	} else {
446 		fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
447 	}
448 
449 	fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
450 
451 	error = gfs2_jdesc_check(jd);
452 	if (error)
453 		goto fail_gunlock_ji;
454 
455 	error = gfs2_find_jhead(jd, &head);
456 	if (error)
457 		goto fail_gunlock_ji;
458 
459 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
460 		fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
461 			jd->jd_jid);
462 
463 		t = jiffies;
464 
465 		/* Acquire a shared hold on the freeze lock */
466 
467 		error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
468 					   LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
469 					   &thaw_gh);
470 		if (error)
471 			goto fail_gunlock_ji;
472 
473 		if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
474 			ro = 1;
475 		} else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
476 			if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
477 				ro = 1;
478 		} else {
479 			if (sb_rdonly(sdp->sd_vfs)) {
480 				/* check if device itself is read-only */
481 				ro = bdev_read_only(sdp->sd_vfs->s_bdev);
482 				if (!ro) {
483 					fs_info(sdp, "recovery required on "
484 						"read-only filesystem.\n");
485 					fs_info(sdp, "write access will be "
486 						"enabled during recovery.\n");
487 				}
488 			}
489 		}
490 
491 		if (ro) {
492 			fs_warn(sdp, "jid=%u: Can't replay: read-only block "
493 				"device\n", jd->jd_jid);
494 			error = -EROFS;
495 			goto fail_gunlock_thaw;
496 		}
497 
498 		fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
499 
500 		for (pass = 0; pass < 2; pass++) {
501 			lops_before_scan(jd, &head, pass);
502 			error = foreach_descriptor(jd, head.lh_tail,
503 						   head.lh_blkno, pass);
504 			lops_after_scan(jd, error, pass);
505 			if (error)
506 				goto fail_gunlock_thaw;
507 		}
508 
509 		clean_journal(jd, &head);
510 
511 		gfs2_glock_dq_uninit(&thaw_gh);
512 		t = DIV_ROUND_UP(jiffies - t, HZ);
513 		fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
514 			jd->jd_jid, t);
515 	}
516 
517 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
518 
519 	if (jlocked) {
520 		gfs2_glock_dq_uninit(&ji_gh);
521 		gfs2_glock_dq_uninit(&j_gh);
522 	}
523 
524 	fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
525 	goto done;
526 
527 fail_gunlock_thaw:
528 	gfs2_glock_dq_uninit(&thaw_gh);
529 fail_gunlock_ji:
530 	if (jlocked) {
531 		gfs2_glock_dq_uninit(&ji_gh);
532 fail_gunlock_j:
533 		gfs2_glock_dq_uninit(&j_gh);
534 	}
535 
536 	fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
537 fail:
538 	jd->jd_recover_error = error;
539 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
540 done:
541 	clear_bit(JDF_RECOVERY, &jd->jd_flags);
542 	smp_mb__after_atomic();
543 	wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
544 }
545 
546 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
547 {
548 	int rv;
549 
550 	if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
551 		return -EBUSY;
552 
553 	/* we have JDF_RECOVERY, queue should always succeed */
554 	rv = queue_work(gfs_recovery_wq, &jd->jd_work);
555 	BUG_ON(!rv);
556 
557 	if (wait)
558 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
559 			    TASK_UNINTERRUPTIBLE);
560 
561 	return wait ? jd->jd_recover_error : 0;
562 }
563 
564