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 #include <linux/ktime.h>
19 
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "bmap.h"
23 #include "glock.h"
24 #include "glops.h"
25 #include "log.h"
26 #include "lops.h"
27 #include "meta_io.h"
28 #include "recovery.h"
29 #include "super.h"
30 #include "util.h"
31 #include "dir.h"
32 
33 struct workqueue_struct *gfs_recovery_wq;
34 
gfs2_replay_read_block(struct gfs2_jdesc * jd,unsigned int blk,struct buffer_head ** bh)35 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
36 			   struct buffer_head **bh)
37 {
38 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
39 	struct gfs2_glock *gl = ip->i_gl;
40 	int new = 0;
41 	u64 dblock;
42 	u32 extlen;
43 	int error;
44 
45 	error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
46 	if (error)
47 		return error;
48 	if (!dblock) {
49 		gfs2_consist_inode(ip);
50 		return -EIO;
51 	}
52 
53 	*bh = gfs2_meta_ra(gl, dblock, extlen);
54 
55 	return error;
56 }
57 
gfs2_revoke_add(struct gfs2_jdesc * jd,u64 blkno,unsigned int where)58 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
59 {
60 	struct list_head *head = &jd->jd_revoke_list;
61 	struct gfs2_revoke_replay *rr;
62 	int found = 0;
63 
64 	list_for_each_entry(rr, head, rr_list) {
65 		if (rr->rr_blkno == blkno) {
66 			found = 1;
67 			break;
68 		}
69 	}
70 
71 	if (found) {
72 		rr->rr_where = where;
73 		return 0;
74 	}
75 
76 	rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
77 	if (!rr)
78 		return -ENOMEM;
79 
80 	rr->rr_blkno = blkno;
81 	rr->rr_where = where;
82 	list_add(&rr->rr_list, head);
83 
84 	return 1;
85 }
86 
gfs2_revoke_check(struct gfs2_jdesc * jd,u64 blkno,unsigned int where)87 int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
88 {
89 	struct gfs2_revoke_replay *rr;
90 	int wrap, a, b, revoke;
91 	int found = 0;
92 
93 	list_for_each_entry(rr, &jd->jd_revoke_list, rr_list) {
94 		if (rr->rr_blkno == blkno) {
95 			found = 1;
96 			break;
97 		}
98 	}
99 
100 	if (!found)
101 		return 0;
102 
103 	wrap = (rr->rr_where < jd->jd_replay_tail);
104 	a = (jd->jd_replay_tail < where);
105 	b = (where < rr->rr_where);
106 	revoke = (wrap) ? (a || b) : (a && b);
107 
108 	return revoke;
109 }
110 
gfs2_revoke_clean(struct gfs2_jdesc * jd)111 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
112 {
113 	struct list_head *head = &jd->jd_revoke_list;
114 	struct gfs2_revoke_replay *rr;
115 
116 	while (!list_empty(head)) {
117 		rr = list_entry(head->next, struct gfs2_revoke_replay, rr_list);
118 		list_del(&rr->rr_list);
119 		kfree(rr);
120 	}
121 }
122 
123 /**
124  * get_log_header - read the log header for a given segment
125  * @jd: the journal
126  * @blk: the block to look at
127  * @lh: the log header to return
128  *
129  * Read the log header for a given segement in a given journal.  Do a few
130  * sanity checks on it.
131  *
132  * Returns: 0 on success,
133  *          1 if the header was invalid or incomplete,
134  *          errno on error
135  */
136 
get_log_header(struct gfs2_jdesc * jd,unsigned int blk,struct gfs2_log_header_host * head)137 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
138 			  struct gfs2_log_header_host *head)
139 {
140 	struct gfs2_log_header *lh;
141 	struct buffer_head *bh;
142 	u32 hash, crc;
143 	int error;
144 
145 	error = gfs2_replay_read_block(jd, blk, &bh);
146 	if (error)
147 		return error;
148 	lh = (void *)bh->b_data;
149 
150 	hash = crc32(~0, lh, LH_V1_SIZE - 4);
151 	hash = ~crc32_le_shift(hash, 4);  /* assume lh_hash is zero */
152 
153 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
154 		     bh->b_size - LH_V1_SIZE - 4);
155 
156 	error = lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
157 		lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
158 		be32_to_cpu(lh->lh_blkno) != blk ||
159 		be32_to_cpu(lh->lh_hash) != hash ||
160 		(lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc);
161 
162 	brelse(bh);
163 
164 	if (!error) {
165 		head->lh_sequence = be64_to_cpu(lh->lh_sequence);
166 		head->lh_flags = be32_to_cpu(lh->lh_flags);
167 		head->lh_tail = be32_to_cpu(lh->lh_tail);
168 		head->lh_blkno = be32_to_cpu(lh->lh_blkno);
169 	}
170 	return error;
171 }
172 
173 /**
174  * find_good_lh - find a good log header
175  * @jd: the journal
176  * @blk: the segment to start searching from
177  * @lh: the log header to fill in
178  * @forward: if true search forward in the log, else search backward
179  *
180  * Call get_log_header() to get a log header for a segment, but if the
181  * segment is bad, either scan forward or backward until we find a good one.
182  *
183  * Returns: errno
184  */
185 
find_good_lh(struct gfs2_jdesc * jd,unsigned int * blk,struct gfs2_log_header_host * head)186 static int find_good_lh(struct gfs2_jdesc *jd, unsigned int *blk,
187 			struct gfs2_log_header_host *head)
188 {
189 	unsigned int orig_blk = *blk;
190 	int error;
191 
192 	for (;;) {
193 		error = get_log_header(jd, *blk, head);
194 		if (error <= 0)
195 			return error;
196 
197 		if (++*blk == jd->jd_blocks)
198 			*blk = 0;
199 
200 		if (*blk == orig_blk) {
201 			gfs2_consist_inode(GFS2_I(jd->jd_inode));
202 			return -EIO;
203 		}
204 	}
205 }
206 
207 /**
208  * jhead_scan - make sure we've found the head of the log
209  * @jd: the journal
210  * @head: this is filled in with the log descriptor of the head
211  *
212  * At this point, seg and lh should be either the head of the log or just
213  * before.  Scan forward until we find the head.
214  *
215  * Returns: errno
216  */
217 
jhead_scan(struct gfs2_jdesc * jd,struct gfs2_log_header_host * head)218 static int jhead_scan(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
219 {
220 	unsigned int blk = head->lh_blkno;
221 	struct gfs2_log_header_host lh;
222 	int error;
223 
224 	for (;;) {
225 		if (++blk == jd->jd_blocks)
226 			blk = 0;
227 
228 		error = get_log_header(jd, blk, &lh);
229 		if (error < 0)
230 			return error;
231 		if (error == 1)
232 			continue;
233 
234 		if (lh.lh_sequence == head->lh_sequence) {
235 			gfs2_consist_inode(GFS2_I(jd->jd_inode));
236 			return -EIO;
237 		}
238 		if (lh.lh_sequence < head->lh_sequence)
239 			break;
240 
241 		*head = lh;
242 	}
243 
244 	return 0;
245 }
246 
247 /**
248  * gfs2_find_jhead - find the head of a log
249  * @jd: the journal
250  * @head: the log descriptor for the head of the log is returned here
251  *
252  * Do a binary search of a journal and find the valid log entry with the
253  * highest sequence number.  (i.e. the log head)
254  *
255  * Returns: errno
256  */
257 
gfs2_find_jhead(struct gfs2_jdesc * jd,struct gfs2_log_header_host * head)258 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head)
259 {
260 	struct gfs2_log_header_host lh_1, lh_m;
261 	u32 blk_1, blk_2, blk_m;
262 	int error;
263 
264 	blk_1 = 0;
265 	blk_2 = jd->jd_blocks - 1;
266 
267 	for (;;) {
268 		blk_m = (blk_1 + blk_2) / 2;
269 
270 		error = find_good_lh(jd, &blk_1, &lh_1);
271 		if (error)
272 			return error;
273 
274 		error = find_good_lh(jd, &blk_m, &lh_m);
275 		if (error)
276 			return error;
277 
278 		if (blk_1 == blk_m || blk_m == blk_2)
279 			break;
280 
281 		if (lh_1.lh_sequence <= lh_m.lh_sequence)
282 			blk_1 = blk_m;
283 		else
284 			blk_2 = blk_m;
285 	}
286 
287 	error = jhead_scan(jd, &lh_1);
288 	if (error)
289 		return error;
290 
291 	*head = lh_1;
292 
293 	return error;
294 }
295 
296 /**
297  * foreach_descriptor - go through the active part of the log
298  * @jd: the journal
299  * @start: the first log header in the active region
300  * @end: the last log header (don't process the contents of this entry))
301  *
302  * Call a given function once for every log descriptor in the active
303  * portion of the log.
304  *
305  * Returns: errno
306  */
307 
foreach_descriptor(struct gfs2_jdesc * jd,unsigned int start,unsigned int end,int pass)308 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
309 			      unsigned int end, int pass)
310 {
311 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
312 	struct buffer_head *bh;
313 	struct gfs2_log_descriptor *ld;
314 	int error = 0;
315 	u32 length;
316 	__be64 *ptr;
317 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
318 	offset += sizeof(__be64) - 1;
319 	offset &= ~(sizeof(__be64) - 1);
320 
321 	while (start != end) {
322 		error = gfs2_replay_read_block(jd, start, &bh);
323 		if (error)
324 			return error;
325 		if (gfs2_meta_check(sdp, bh)) {
326 			brelse(bh);
327 			return -EIO;
328 		}
329 		ld = (struct gfs2_log_descriptor *)bh->b_data;
330 		length = be32_to_cpu(ld->ld_length);
331 
332 		if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
333 			struct gfs2_log_header_host lh;
334 			error = get_log_header(jd, start, &lh);
335 			if (!error) {
336 				gfs2_replay_incr_blk(jd, &start);
337 				brelse(bh);
338 				continue;
339 			}
340 			if (error == 1) {
341 				gfs2_consist_inode(GFS2_I(jd->jd_inode));
342 				error = -EIO;
343 			}
344 			brelse(bh);
345 			return error;
346 		} else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
347 			brelse(bh);
348 			return -EIO;
349 		}
350 		ptr = (__be64 *)(bh->b_data + offset);
351 		error = lops_scan_elements(jd, start, ld, ptr, pass);
352 		if (error) {
353 			brelse(bh);
354 			return error;
355 		}
356 
357 		while (length--)
358 			gfs2_replay_incr_blk(jd, &start);
359 
360 		brelse(bh);
361 	}
362 
363 	return 0;
364 }
365 
366 /**
367  * clean_journal - mark a dirty journal as being clean
368  * @jd: the journal
369  * @head: the head journal to start from
370  *
371  * Returns: errno
372  */
373 
clean_journal(struct gfs2_jdesc * jd,struct gfs2_log_header_host * head)374 static void clean_journal(struct gfs2_jdesc *jd,
375 			  struct gfs2_log_header_host *head)
376 {
377 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
378 
379 	sdp->sd_log_flush_head = head->lh_blkno;
380 	gfs2_replay_incr_blk(jd, &sdp->sd_log_flush_head);
381 	gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0,
382 			      GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
383 			      REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
384 }
385 
386 
gfs2_recovery_done(struct gfs2_sbd * sdp,unsigned int jid,unsigned int message)387 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
388                                unsigned int message)
389 {
390 	char env_jid[20];
391 	char env_status[20];
392 	char *envp[] = { env_jid, env_status, NULL };
393 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
394 
395         ls->ls_recover_jid_done = jid;
396         ls->ls_recover_jid_status = message;
397 	sprintf(env_jid, "JID=%u", jid);
398 	sprintf(env_status, "RECOVERY=%s",
399 		message == LM_RD_SUCCESS ? "Done" : "Failed");
400         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
401 
402 	if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
403 		sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
404 }
405 
gfs2_recover_func(struct work_struct * work)406 void gfs2_recover_func(struct work_struct *work)
407 {
408 	struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
409 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
410 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
411 	struct gfs2_log_header_host head;
412 	struct gfs2_holder j_gh, ji_gh, thaw_gh;
413 	ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
414 	int ro = 0;
415 	unsigned int pass;
416 	int error = 0;
417 	int jlocked = 0;
418 
419 	t_start = ktime_get();
420 	if (sdp->sd_args.ar_spectator)
421 		goto fail;
422 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
423 		fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
424 			jd->jd_jid);
425 		jlocked = 1;
426 		/* Acquire the journal lock so we can do recovery */
427 
428 		error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
429 					  LM_ST_EXCLUSIVE,
430 					  LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
431 					  &j_gh);
432 		switch (error) {
433 		case 0:
434 			break;
435 
436 		case GLR_TRYFAILED:
437 			fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
438 			error = 0;
439 
440 		default:
441 			goto fail;
442 		};
443 
444 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
445 					   LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
446 		if (error)
447 			goto fail_gunlock_j;
448 	} else {
449 		fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
450 	}
451 
452 	t_jlck = ktime_get();
453 	fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
454 
455 	error = gfs2_jdesc_check(jd);
456 	if (error)
457 		goto fail_gunlock_ji;
458 
459 	error = gfs2_find_jhead(jd, &head);
460 	if (error)
461 		goto fail_gunlock_ji;
462 	t_jhd = ktime_get();
463 
464 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
465 		fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
466 			jd->jd_jid);
467 
468 		/* Acquire a shared hold on the freeze lock */
469 
470 		error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
471 					   LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
472 					   &thaw_gh);
473 		if (error)
474 			goto fail_gunlock_ji;
475 
476 		if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
477 			ro = 1;
478 		} else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
479 			if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
480 				ro = 1;
481 		} else {
482 			if (sb_rdonly(sdp->sd_vfs)) {
483 				/* check if device itself is read-only */
484 				ro = bdev_read_only(sdp->sd_vfs->s_bdev);
485 				if (!ro) {
486 					fs_info(sdp, "recovery required on "
487 						"read-only filesystem.\n");
488 					fs_info(sdp, "write access will be "
489 						"enabled during recovery.\n");
490 				}
491 			}
492 		}
493 
494 		if (ro) {
495 			fs_warn(sdp, "jid=%u: Can't replay: read-only block "
496 				"device\n", jd->jd_jid);
497 			error = -EROFS;
498 			goto fail_gunlock_thaw;
499 		}
500 
501 		t_tlck = ktime_get();
502 		fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
503 
504 		for (pass = 0; pass < 2; pass++) {
505 			lops_before_scan(jd, &head, pass);
506 			error = foreach_descriptor(jd, head.lh_tail,
507 						   head.lh_blkno, pass);
508 			lops_after_scan(jd, error, pass);
509 			if (error)
510 				goto fail_gunlock_thaw;
511 		}
512 
513 		clean_journal(jd, &head);
514 
515 		gfs2_glock_dq_uninit(&thaw_gh);
516 		t_rep = ktime_get();
517 		fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
518 			"jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
519 			jd->jd_jid, ktime_ms_delta(t_rep, t_start),
520 			ktime_ms_delta(t_jlck, t_start),
521 			ktime_ms_delta(t_jhd, t_jlck),
522 			ktime_ms_delta(t_tlck, t_jhd),
523 			ktime_ms_delta(t_rep, t_tlck));
524 	}
525 
526 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
527 
528 	if (jlocked) {
529 		gfs2_glock_dq_uninit(&ji_gh);
530 		gfs2_glock_dq_uninit(&j_gh);
531 	}
532 
533 	fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
534 	goto done;
535 
536 fail_gunlock_thaw:
537 	gfs2_glock_dq_uninit(&thaw_gh);
538 fail_gunlock_ji:
539 	if (jlocked) {
540 		gfs2_glock_dq_uninit(&ji_gh);
541 fail_gunlock_j:
542 		gfs2_glock_dq_uninit(&j_gh);
543 	}
544 
545 	fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
546 fail:
547 	jd->jd_recover_error = error;
548 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
549 done:
550 	clear_bit(JDF_RECOVERY, &jd->jd_flags);
551 	smp_mb__after_atomic();
552 	wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
553 }
554 
gfs2_recover_journal(struct gfs2_jdesc * jd,bool wait)555 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
556 {
557 	int rv;
558 
559 	if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
560 		return -EBUSY;
561 
562 	/* we have JDF_RECOVERY, queue should always succeed */
563 	rv = queue_work(gfs_recovery_wq, &jd->jd_work);
564 	BUG_ON(!rv);
565 
566 	if (wait)
567 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
568 			    TASK_UNINTERRUPTIBLE);
569 
570 	return wait ? jd->jd_recover_error : 0;
571 }
572 
573