xref: /openbmc/linux/fs/gfs2/recovery.c (revision 22246614)
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/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/crc32.h>
16 #include <linux/lm_interface.h>
17 
18 #include "gfs2.h"
19 #include "incore.h"
20 #include "bmap.h"
21 #include "glock.h"
22 #include "glops.h"
23 #include "lops.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "super.h"
27 #include "util.h"
28 #include "dir.h"
29 
30 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
31 			   struct buffer_head **bh)
32 {
33 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
34 	struct gfs2_glock *gl = ip->i_gl;
35 	int new = 0;
36 	u64 dblock;
37 	u32 extlen;
38 	int error;
39 
40 	error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
41 	if (error)
42 		return error;
43 	if (!dblock) {
44 		gfs2_consist_inode(ip);
45 		return -EIO;
46 	}
47 
48 	*bh = gfs2_meta_ra(gl, dblock, extlen);
49 
50 	return error;
51 }
52 
53 int gfs2_revoke_add(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
54 {
55 	struct list_head *head = &sdp->sd_revoke_list;
56 	struct gfs2_revoke_replay *rr;
57 	int found = 0;
58 
59 	list_for_each_entry(rr, head, rr_list) {
60 		if (rr->rr_blkno == blkno) {
61 			found = 1;
62 			break;
63 		}
64 	}
65 
66 	if (found) {
67 		rr->rr_where = where;
68 		return 0;
69 	}
70 
71 	rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
72 	if (!rr)
73 		return -ENOMEM;
74 
75 	rr->rr_blkno = blkno;
76 	rr->rr_where = where;
77 	list_add(&rr->rr_list, head);
78 
79 	return 1;
80 }
81 
82 int gfs2_revoke_check(struct gfs2_sbd *sdp, u64 blkno, unsigned int where)
83 {
84 	struct gfs2_revoke_replay *rr;
85 	int wrap, a, b, revoke;
86 	int found = 0;
87 
88 	list_for_each_entry(rr, &sdp->sd_revoke_list, rr_list) {
89 		if (rr->rr_blkno == blkno) {
90 			found = 1;
91 			break;
92 		}
93 	}
94 
95 	if (!found)
96 		return 0;
97 
98 	wrap = (rr->rr_where < sdp->sd_replay_tail);
99 	a = (sdp->sd_replay_tail < where);
100 	b = (where < rr->rr_where);
101 	revoke = (wrap) ? (a || b) : (a && b);
102 
103 	return revoke;
104 }
105 
106 void gfs2_revoke_clean(struct gfs2_sbd *sdp)
107 {
108 	struct list_head *head = &sdp->sd_revoke_list;
109 	struct gfs2_revoke_replay *rr;
110 
111 	while (!list_empty(head)) {
112 		rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
113 		list_del(&rr->rr_list);
114 		kfree(rr);
115 	}
116 }
117 
118 static int gfs2_log_header_in(struct gfs2_log_header_host *lh, const void *buf)
119 {
120 	const struct gfs2_log_header *str = buf;
121 
122 	if (str->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
123 	    str->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH))
124 		return 1;
125 
126 	lh->lh_sequence = be64_to_cpu(str->lh_sequence);
127 	lh->lh_flags = be32_to_cpu(str->lh_flags);
128 	lh->lh_tail = be32_to_cpu(str->lh_tail);
129 	lh->lh_blkno = be32_to_cpu(str->lh_blkno);
130 	lh->lh_hash = be32_to_cpu(str->lh_hash);
131 	return 0;
132 }
133 
134 /**
135  * get_log_header - read the log header for a given segment
136  * @jd: the journal
137  * @blk: the block to look at
138  * @lh: the log header to return
139  *
140  * Read the log header for a given segement in a given journal.  Do a few
141  * sanity checks on it.
142  *
143  * Returns: 0 on success,
144  *          1 if the header was invalid or incomplete,
145  *          errno on error
146  */
147 
148 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
149 			  struct gfs2_log_header_host *head)
150 {
151 	struct buffer_head *bh;
152 	struct gfs2_log_header_host uninitialized_var(lh);
153 	const u32 nothing = 0;
154 	u32 hash;
155 	int error;
156 
157 	error = gfs2_replay_read_block(jd, blk, &bh);
158 	if (error)
159 		return error;
160 
161 	hash = crc32_le((u32)~0, bh->b_data, sizeof(struct gfs2_log_header) -
162 					     sizeof(u32));
163 	hash = crc32_le(hash, (unsigned char const *)&nothing, sizeof(nothing));
164 	hash ^= (u32)~0;
165 	error = gfs2_log_header_in(&lh, bh->b_data);
166 	brelse(bh);
167 
168 	if (error || lh.lh_blkno != blk || lh.lh_hash != hash)
169 		return 1;
170 
171 	*head = lh;
172 
173 	return 0;
174 }
175 
176 /**
177  * find_good_lh - find a good log header
178  * @jd: the journal
179  * @blk: the segment to start searching from
180  * @lh: the log header to fill in
181  * @forward: if true search forward in the log, else search backward
182  *
183  * Call get_log_header() to get a log header for a segment, but if the
184  * segment is bad, either scan forward or backward until we find a good one.
185  *
186  * Returns: errno
187  */
188 
189 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
190 			struct gfs2_log_header_host *head)
191 {
192 	unsigned int orig_blk = *blk;
193 	int error;
194 
195 	for (;;) {
196 		error = get_log_header(jd, *blk, head);
197 		if (error <= 0)
198 			return error;
199 
200 		if (++*blk == jd->jd_blocks)
201 			*blk = 0;
202 
203 		if (*blk == orig_blk) {
204 			gfs2_consist_inode(GFS2_I(jd->jd_inode));
205 			return -EIO;
206 		}
207 	}
208 }
209 
210 /**
211  * jhead_scan - make sure we've found the head of the log
212  * @jd: the journal
213  * @head: this is filled in with the log descriptor of the head
214  *
215  * At this point, seg and lh should be either the head of the log or just
216  * before.  Scan forward until we find the head.
217  *
218  * Returns: errno
219  */
220 
221 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
222 {
223 	unsigned int blk = head->lh_blkno;
224 	struct gfs2_log_header_host lh;
225 	int error;
226 
227 	for (;;) {
228 		if (++blk == jd->jd_blocks)
229 			blk = 0;
230 
231 		error = get_log_header(jd, blk, &lh);
232 		if (error < 0)
233 			return error;
234 		if (error == 1)
235 			continue;
236 
237 		if (lh.lh_sequence == head->lh_sequence) {
238 			gfs2_consist_inode(GFS2_I(jd->jd_inode));
239 			return -EIO;
240 		}
241 		if (lh.lh_sequence < head->lh_sequence)
242 			break;
243 
244 		*head = lh;
245 	}
246 
247 	return 0;
248 }
249 
250 /**
251  * gfs2_find_jhead - find the head of a log
252  * @jd: the journal
253  * @head: the log descriptor for the head of the log is returned here
254  *
255  * Do a binary search of a journal and find the valid log entry with the
256  * highest sequence number.  (i.e. the log head)
257  *
258  * Returns: errno
259  */
260 
261 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
262 {
263 	struct gfs2_log_header_host lh_1, lh_m;
264 	u32 blk_1, blk_2, blk_m;
265 	int error;
266 
267 	blk_1 = 0;
268 	blk_2 = jd->jd_blocks - 1;
269 
270 	for (;;) {
271 		blk_m = (blk_1 + blk_2) / 2;
272 
273 		error = find_good_lh(jd, &blk_1, &lh_1);
274 		if (error)
275 			return error;
276 
277 		error = find_good_lh(jd, &blk_m, &lh_m);
278 		if (error)
279 			return error;
280 
281 		if (blk_1 == blk_m || blk_m == blk_2)
282 			break;
283 
284 		if (lh_1.lh_sequence <= lh_m.lh_sequence)
285 			blk_1 = blk_m;
286 		else
287 			blk_2 = blk_m;
288 	}
289 
290 	error = jhead_scan(jd, &lh_1);
291 	if (error)
292 		return error;
293 
294 	*head = lh_1;
295 
296 	return error;
297 }
298 
299 /**
300  * foreach_descriptor - go through the active part of the log
301  * @jd: the journal
302  * @start: the first log header in the active region
303  * @end: the last log header (don't process the contents of this entry))
304  *
305  * Call a given function once for every log descriptor in the active
306  * portion of the log.
307  *
308  * Returns: errno
309  */
310 
311 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
312 			      unsigned int end, int pass)
313 {
314 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
315 	struct buffer_head *bh;
316 	struct gfs2_log_descriptor *ld;
317 	int error = 0;
318 	u32 length;
319 	__be64 *ptr;
320 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
321 	offset += sizeof(__be64) - 1;
322 	offset &= ~(sizeof(__be64) - 1);
323 
324 	while (start != end) {
325 		error = gfs2_replay_read_block(jd, start, &bh);
326 		if (error)
327 			return error;
328 		if (gfs2_meta_check(sdp, bh)) {
329 			brelse(bh);
330 			return -EIO;
331 		}
332 		ld = (struct gfs2_log_descriptor *)bh->b_data;
333 		length = be32_to_cpu(ld->ld_length);
334 
335 		if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
336 			struct gfs2_log_header_host lh;
337 			error = get_log_header(jd, start, &lh);
338 			if (!error) {
339 				gfs2_replay_incr_blk(sdp, &start);
340 				brelse(bh);
341 				continue;
342 			}
343 			if (error == 1) {
344 				gfs2_consist_inode(GFS2_I(jd->jd_inode));
345 				error = -EIO;
346 			}
347 			brelse(bh);
348 			return error;
349 		} else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
350 			brelse(bh);
351 			return -EIO;
352 		}
353 		ptr = (__be64 *)(bh->b_data + offset);
354 		error = lops_scan_elements(jd, start, ld, ptr, pass);
355 		if (error) {
356 			brelse(bh);
357 			return error;
358 		}
359 
360 		while (length--)
361 			gfs2_replay_incr_blk(sdp, &start);
362 
363 		brelse(bh);
364 	}
365 
366 	return 0;
367 }
368 
369 /**
370  * clean_journal - mark a dirty journal as being clean
371  * @sdp: the filesystem
372  * @jd: the journal
373  * @gl: the journal's glock
374  * @head: the head journal to start from
375  *
376  * Returns: errno
377  */
378 
379 static int clean_journal(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
380 {
381 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
382 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
383 	unsigned int lblock;
384 	struct gfs2_log_header *lh;
385 	u32 hash;
386 	struct buffer_head *bh;
387 	int error;
388 	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
389 
390 	lblock = head->lh_blkno;
391 	gfs2_replay_incr_blk(sdp, &lblock);
392 	bh_map.b_size = 1 << ip->i_inode.i_blkbits;
393 	error = gfs2_block_map(&ip->i_inode, lblock, &bh_map, 0);
394 	if (error)
395 		return error;
396 	if (!bh_map.b_blocknr) {
397 		gfs2_consist_inode(ip);
398 		return -EIO;
399 	}
400 
401 	bh = sb_getblk(sdp->sd_vfs, bh_map.b_blocknr);
402 	lock_buffer(bh);
403 	memset(bh->b_data, 0, bh->b_size);
404 	set_buffer_uptodate(bh);
405 	clear_buffer_dirty(bh);
406 	unlock_buffer(bh);
407 
408 	lh = (struct gfs2_log_header *)bh->b_data;
409 	memset(lh, 0, sizeof(struct gfs2_log_header));
410 	lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
411 	lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
412 	lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
413 	lh->lh_sequence = cpu_to_be64(head->lh_sequence + 1);
414 	lh->lh_flags = cpu_to_be32(GFS2_LOG_HEAD_UNMOUNT);
415 	lh->lh_blkno = cpu_to_be32(lblock);
416 	hash = gfs2_disk_hash((const char *)lh, sizeof(struct gfs2_log_header));
417 	lh->lh_hash = cpu_to_be32(hash);
418 
419 	set_buffer_dirty(bh);
420 	if (sync_dirty_buffer(bh))
421 		gfs2_io_error_bh(sdp, bh);
422 	brelse(bh);
423 
424 	return error;
425 }
426 
427 
428 static void gfs2_lm_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
429 				  unsigned int message)
430 {
431 	if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
432 		sdp->sd_lockstruct.ls_ops->lm_recovery_done(
433 			sdp->sd_lockstruct.ls_lockspace, jid, message);
434 }
435 
436 
437 /**
438  * gfs2_recover_journal - recovery a given journal
439  * @jd: the struct gfs2_jdesc describing the journal
440  *
441  * Acquire the journal's lock, check to see if the journal is clean, and
442  * do recovery if necessary.
443  *
444  * Returns: errno
445  */
446 
447 int gfs2_recover_journal(struct gfs2_jdesc *jd)
448 {
449 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
450 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
451 	struct gfs2_log_header_host head;
452 	struct gfs2_holder j_gh, ji_gh, t_gh;
453 	unsigned long t;
454 	int ro = 0;
455 	unsigned int pass;
456 	int error;
457 
458 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
459 		fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
460 			jd->jd_jid);
461 
462 		/* Acquire the journal lock so we can do recovery */
463 
464 		error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
465 					  LM_ST_EXCLUSIVE,
466 					  LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
467 					  &j_gh);
468 		switch (error) {
469 		case 0:
470 			break;
471 
472 		case GLR_TRYFAILED:
473 			fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
474 			error = 0;
475 
476 		default:
477 			goto fail;
478 		};
479 
480 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
481 					   LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
482 		if (error)
483 			goto fail_gunlock_j;
484 	} else {
485 		fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
486 	}
487 
488 	fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
489 
490 	error = gfs2_jdesc_check(jd);
491 	if (error)
492 		goto fail_gunlock_ji;
493 
494 	error = gfs2_find_jhead(jd, &head);
495 	if (error)
496 		goto fail_gunlock_ji;
497 
498 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
499 		fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
500 			jd->jd_jid);
501 
502 		t = jiffies;
503 
504 		/* Acquire a shared hold on the transaction lock */
505 
506 		error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
507 					   LM_FLAG_NOEXP | LM_FLAG_PRIORITY |
508 					   GL_NOCANCEL | GL_NOCACHE, &t_gh);
509 		if (error)
510 			goto fail_gunlock_ji;
511 
512 		if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
513 			if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
514 				ro = 1;
515 		} else {
516 			if (sdp->sd_vfs->s_flags & MS_RDONLY) {
517 				/* check if device itself is read-only */
518 				ro = bdev_read_only(sdp->sd_vfs->s_bdev);
519 				if (!ro) {
520 					fs_info(sdp, "recovery required on "
521 						"read-only filesystem.\n");
522 					fs_info(sdp, "write access will be "
523 						"enabled during recovery.\n");
524 				}
525 			}
526 		}
527 
528 		if (ro) {
529 			fs_warn(sdp, "jid=%u: Can't replay: read-only block "
530 				"device\n", jd->jd_jid);
531 			error = -EROFS;
532 			goto fail_gunlock_tr;
533 		}
534 
535 		fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
536 
537 		for (pass = 0; pass < 2; pass++) {
538 			lops_before_scan(jd, &head, pass);
539 			error = foreach_descriptor(jd, head.lh_tail,
540 						   head.lh_blkno, pass);
541 			lops_after_scan(jd, error, pass);
542 			if (error)
543 				goto fail_gunlock_tr;
544 		}
545 
546 		error = clean_journal(jd, &head);
547 		if (error)
548 			goto fail_gunlock_tr;
549 
550 		gfs2_glock_dq_uninit(&t_gh);
551 		t = DIV_ROUND_UP(jiffies - t, HZ);
552 		fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
553 			jd->jd_jid, t);
554 	}
555 
556 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
557 		gfs2_glock_dq_uninit(&ji_gh);
558 
559 	gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
560 
561 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid)
562 		gfs2_glock_dq_uninit(&j_gh);
563 
564 	fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
565 	return 0;
566 
567 fail_gunlock_tr:
568 	gfs2_glock_dq_uninit(&t_gh);
569 fail_gunlock_ji:
570 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
571 		gfs2_glock_dq_uninit(&ji_gh);
572 fail_gunlock_j:
573 		gfs2_glock_dq_uninit(&j_gh);
574 	}
575 
576 	fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
577 
578 fail:
579 	gfs2_lm_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
580 	return error;
581 }
582 
583 /**
584  * gfs2_check_journals - Recover any dirty journals
585  * @sdp: the filesystem
586  *
587  */
588 
589 void gfs2_check_journals(struct gfs2_sbd *sdp)
590 {
591 	struct gfs2_jdesc *jd;
592 
593 	for (;;) {
594 		jd = gfs2_jdesc_find_dirty(sdp);
595 		if (!jd)
596 			break;
597 
598 		if (jd != sdp->sd_jdesc)
599 			gfs2_recover_journal(jd);
600 	}
601 }
602 
603