xref: /openbmc/linux/fs/gfs2/recovery.c (revision 5b4cb650)
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 
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 
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 
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 
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 int __get_log_header(struct gfs2_sbd *sdp, const struct gfs2_log_header *lh,
124 		     unsigned int blkno, struct gfs2_log_header_host *head)
125 {
126 	u32 hash, crc;
127 
128 	if (lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
129 	    lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
130 	    (blkno && be32_to_cpu(lh->lh_blkno) != blkno))
131 		return 1;
132 
133 	hash = crc32(~0, lh, LH_V1_SIZE - 4);
134 	hash = ~crc32_le_shift(hash, 4); /* assume lh_hash is zero */
135 
136 	if (be32_to_cpu(lh->lh_hash) != hash)
137 		return 1;
138 
139 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
140 		     sdp->sd_sb.sb_bsize - LH_V1_SIZE - 4);
141 
142 	if ((lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc))
143 		return 1;
144 
145 	head->lh_sequence = be64_to_cpu(lh->lh_sequence);
146 	head->lh_flags = be32_to_cpu(lh->lh_flags);
147 	head->lh_tail = be32_to_cpu(lh->lh_tail);
148 	head->lh_blkno = be32_to_cpu(lh->lh_blkno);
149 
150 	return 0;
151 }
152 /**
153  * get_log_header - read the log header for a given segment
154  * @jd: the journal
155  * @blk: the block to look at
156  * @lh: the log header to return
157  *
158  * Read the log header for a given segement in a given journal.  Do a few
159  * sanity checks on it.
160  *
161  * Returns: 0 on success,
162  *          1 if the header was invalid or incomplete,
163  *          errno on error
164  */
165 
166 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
167 			  struct gfs2_log_header_host *head)
168 {
169 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
170 	struct buffer_head *bh;
171 	int error;
172 
173 	error = gfs2_replay_read_block(jd, blk, &bh);
174 	if (error)
175 		return error;
176 
177 	error = __get_log_header(sdp, (const struct gfs2_log_header *)bh->b_data,
178 				 blk, head);
179 	brelse(bh);
180 
181 	return error;
182 }
183 
184 /**
185  * foreach_descriptor - go through the active part of the log
186  * @jd: the journal
187  * @start: the first log header in the active region
188  * @end: the last log header (don't process the contents of this entry))
189  *
190  * Call a given function once for every log descriptor in the active
191  * portion of the log.
192  *
193  * Returns: errno
194  */
195 
196 static int foreach_descriptor(struct gfs2_jdesc *jd, unsigned int start,
197 			      unsigned int end, int pass)
198 {
199 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
200 	struct buffer_head *bh;
201 	struct gfs2_log_descriptor *ld;
202 	int error = 0;
203 	u32 length;
204 	__be64 *ptr;
205 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
206 	offset += sizeof(__be64) - 1;
207 	offset &= ~(sizeof(__be64) - 1);
208 
209 	while (start != end) {
210 		error = gfs2_replay_read_block(jd, start, &bh);
211 		if (error)
212 			return error;
213 		if (gfs2_meta_check(sdp, bh)) {
214 			brelse(bh);
215 			return -EIO;
216 		}
217 		ld = (struct gfs2_log_descriptor *)bh->b_data;
218 		length = be32_to_cpu(ld->ld_length);
219 
220 		if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
221 			struct gfs2_log_header_host lh;
222 			error = get_log_header(jd, start, &lh);
223 			if (!error) {
224 				gfs2_replay_incr_blk(jd, &start);
225 				brelse(bh);
226 				continue;
227 			}
228 			if (error == 1) {
229 				gfs2_consist_inode(GFS2_I(jd->jd_inode));
230 				error = -EIO;
231 			}
232 			brelse(bh);
233 			return error;
234 		} else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
235 			brelse(bh);
236 			return -EIO;
237 		}
238 		ptr = (__be64 *)(bh->b_data + offset);
239 		error = lops_scan_elements(jd, start, ld, ptr, pass);
240 		if (error) {
241 			brelse(bh);
242 			return error;
243 		}
244 
245 		while (length--)
246 			gfs2_replay_incr_blk(jd, &start);
247 
248 		brelse(bh);
249 	}
250 
251 	return 0;
252 }
253 
254 /**
255  * clean_journal - mark a dirty journal as being clean
256  * @jd: the journal
257  * @head: the head journal to start from
258  *
259  * Returns: errno
260  */
261 
262 static void clean_journal(struct gfs2_jdesc *jd,
263 			  struct gfs2_log_header_host *head)
264 {
265 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
266 
267 	sdp->sd_log_flush_head = head->lh_blkno;
268 	gfs2_replay_incr_blk(jd, &sdp->sd_log_flush_head);
269 	gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0,
270 			      GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
271 			      REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
272 }
273 
274 
275 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
276                                unsigned int message)
277 {
278 	char env_jid[20];
279 	char env_status[20];
280 	char *envp[] = { env_jid, env_status, NULL };
281 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
282 
283         ls->ls_recover_jid_done = jid;
284         ls->ls_recover_jid_status = message;
285 	sprintf(env_jid, "JID=%u", jid);
286 	sprintf(env_status, "RECOVERY=%s",
287 		message == LM_RD_SUCCESS ? "Done" : "Failed");
288         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
289 
290 	if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
291 		sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
292 }
293 
294 void gfs2_recover_func(struct work_struct *work)
295 {
296 	struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
297 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
298 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
299 	struct gfs2_log_header_host head;
300 	struct gfs2_holder j_gh, ji_gh, thaw_gh;
301 	ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
302 	int ro = 0;
303 	unsigned int pass;
304 	int error = 0;
305 	int jlocked = 0;
306 
307 	t_start = ktime_get();
308 	if (sdp->sd_args.ar_spectator)
309 		goto fail;
310 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
311 		fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
312 			jd->jd_jid);
313 		jlocked = 1;
314 		/* Acquire the journal lock so we can do recovery */
315 
316 		error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
317 					  LM_ST_EXCLUSIVE,
318 					  LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
319 					  &j_gh);
320 		switch (error) {
321 		case 0:
322 			break;
323 
324 		case GLR_TRYFAILED:
325 			fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
326 			error = 0;
327 
328 		default:
329 			goto fail;
330 		};
331 
332 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
333 					   LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
334 		if (error)
335 			goto fail_gunlock_j;
336 	} else {
337 		fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
338 	}
339 
340 	t_jlck = ktime_get();
341 	fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
342 
343 	error = gfs2_jdesc_check(jd);
344 	if (error)
345 		goto fail_gunlock_ji;
346 
347 	error = gfs2_find_jhead(jd, &head);
348 	if (error)
349 		goto fail_gunlock_ji;
350 	t_jhd = ktime_get();
351 	fs_info(sdp, "jid=%u: Journal head lookup took %lldms\n", jd->jd_jid,
352 		ktime_ms_delta(t_jhd, t_jlck));
353 
354 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
355 		fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
356 			jd->jd_jid);
357 
358 		/* Acquire a shared hold on the freeze lock */
359 
360 		error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
361 					   LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
362 					   &thaw_gh);
363 		if (error)
364 			goto fail_gunlock_ji;
365 
366 		if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
367 			ro = 1;
368 		} else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
369 			if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
370 				ro = 1;
371 		} else {
372 			if (sb_rdonly(sdp->sd_vfs)) {
373 				/* check if device itself is read-only */
374 				ro = bdev_read_only(sdp->sd_vfs->s_bdev);
375 				if (!ro) {
376 					fs_info(sdp, "recovery required on "
377 						"read-only filesystem.\n");
378 					fs_info(sdp, "write access will be "
379 						"enabled during recovery.\n");
380 				}
381 			}
382 		}
383 
384 		if (ro) {
385 			fs_warn(sdp, "jid=%u: Can't replay: read-only block "
386 				"device\n", jd->jd_jid);
387 			error = -EROFS;
388 			goto fail_gunlock_thaw;
389 		}
390 
391 		t_tlck = ktime_get();
392 		fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
393 
394 		for (pass = 0; pass < 2; pass++) {
395 			lops_before_scan(jd, &head, pass);
396 			error = foreach_descriptor(jd, head.lh_tail,
397 						   head.lh_blkno, pass);
398 			lops_after_scan(jd, error, pass);
399 			if (error)
400 				goto fail_gunlock_thaw;
401 		}
402 
403 		clean_journal(jd, &head);
404 
405 		gfs2_glock_dq_uninit(&thaw_gh);
406 		t_rep = ktime_get();
407 		fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
408 			"jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
409 			jd->jd_jid, ktime_ms_delta(t_rep, t_start),
410 			ktime_ms_delta(t_jlck, t_start),
411 			ktime_ms_delta(t_jhd, t_jlck),
412 			ktime_ms_delta(t_tlck, t_jhd),
413 			ktime_ms_delta(t_rep, t_tlck));
414 	}
415 
416 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
417 
418 	if (jlocked) {
419 		gfs2_glock_dq_uninit(&ji_gh);
420 		gfs2_glock_dq_uninit(&j_gh);
421 	}
422 
423 	fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
424 	goto done;
425 
426 fail_gunlock_thaw:
427 	gfs2_glock_dq_uninit(&thaw_gh);
428 fail_gunlock_ji:
429 	if (jlocked) {
430 		gfs2_glock_dq_uninit(&ji_gh);
431 fail_gunlock_j:
432 		gfs2_glock_dq_uninit(&j_gh);
433 	}
434 
435 	fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
436 fail:
437 	jd->jd_recover_error = error;
438 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
439 done:
440 	clear_bit(JDF_RECOVERY, &jd->jd_flags);
441 	smp_mb__after_atomic();
442 	wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
443 }
444 
445 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
446 {
447 	int rv;
448 
449 	if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
450 		return -EBUSY;
451 
452 	/* we have JDF_RECOVERY, queue should always succeed */
453 	rv = queue_work(gfs_recovery_wq, &jd->jd_work);
454 	BUG_ON(!rv);
455 
456 	if (wait)
457 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
458 			    TASK_UNINTERRUPTIBLE);
459 
460 	return wait ? jd->jd_recover_error : 0;
461 }
462 
463