xref: /linux/fs/gfs2/recovery.c (revision ca64d84e93762f4e587e040a44ad9f6089afc777)
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 #include <linux/module.h>
8 #include <linux/slab.h>
9 #include <linux/spinlock.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/gfs2_ondisk.h>
13 #include <linux/crc32.h>
14 #include <linux/crc32c.h>
15 #include <linux/ktime.h>
16 
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "log.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 struct workqueue_struct *gfs_recovery_wq;
31 
32 int gfs2_replay_read_block(struct gfs2_jdesc *jd, unsigned int blk,
33 			   struct buffer_head **bh)
34 {
35 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
36 	struct gfs2_glock *gl = ip->i_gl;
37 	int new = 0;
38 	u64 dblock;
39 	u32 extlen;
40 	int error;
41 
42 	error = gfs2_extent_map(&ip->i_inode, blk, &new, &dblock, &extlen);
43 	if (error)
44 		return error;
45 	if (!dblock) {
46 		gfs2_consist_inode(ip);
47 		return -EIO;
48 	}
49 
50 	*bh = gfs2_meta_ra(gl, dblock, extlen);
51 
52 	return error;
53 }
54 
55 int gfs2_revoke_add(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
56 {
57 	struct list_head *head = &jd->jd_revoke_list;
58 	struct gfs2_revoke_replay *rr;
59 	int found = 0;
60 
61 	list_for_each_entry(rr, head, rr_list) {
62 		if (rr->rr_blkno == blkno) {
63 			found = 1;
64 			break;
65 		}
66 	}
67 
68 	if (found) {
69 		rr->rr_where = where;
70 		return 0;
71 	}
72 
73 	rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
74 	if (!rr)
75 		return -ENOMEM;
76 
77 	rr->rr_blkno = blkno;
78 	rr->rr_where = where;
79 	list_add(&rr->rr_list, head);
80 
81 	return 1;
82 }
83 
84 int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
85 {
86 	struct gfs2_revoke_replay *rr;
87 	int wrap, a, b, revoke;
88 	int found = 0;
89 
90 	list_for_each_entry(rr, &jd->jd_revoke_list, rr_list) {
91 		if (rr->rr_blkno == blkno) {
92 			found = 1;
93 			break;
94 		}
95 	}
96 
97 	if (!found)
98 		return 0;
99 
100 	wrap = (rr->rr_where < jd->jd_replay_tail);
101 	a = (jd->jd_replay_tail < where);
102 	b = (where < rr->rr_where);
103 	revoke = (wrap) ? (a || b) : (a && b);
104 
105 	return revoke;
106 }
107 
108 void gfs2_revoke_clean(struct gfs2_jdesc *jd)
109 {
110 	struct list_head *head = &jd->jd_revoke_list;
111 	struct gfs2_revoke_replay *rr;
112 
113 	while (!list_empty(head)) {
114 		rr = list_first_entry(head, struct gfs2_revoke_replay, rr_list);
115 		list_del(&rr->rr_list);
116 		kfree(rr);
117 	}
118 }
119 
120 int __get_log_header(struct gfs2_sbd *sdp, const struct gfs2_log_header *lh,
121 		     unsigned int blkno, struct gfs2_log_header_host *head)
122 {
123 	u32 hash, crc;
124 
125 	if (lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
126 	    lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
127 	    (blkno && be32_to_cpu(lh->lh_blkno) != blkno))
128 		return 1;
129 
130 	hash = crc32(~0, lh, LH_V1_SIZE - 4);
131 	hash = ~crc32_le_shift(hash, 4); /* assume lh_hash is zero */
132 
133 	if (be32_to_cpu(lh->lh_hash) != hash)
134 		return 1;
135 
136 	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
137 		     sdp->sd_sb.sb_bsize - LH_V1_SIZE - 4);
138 
139 	if ((lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc))
140 		return 1;
141 
142 	head->lh_sequence = be64_to_cpu(lh->lh_sequence);
143 	head->lh_flags = be32_to_cpu(lh->lh_flags);
144 	head->lh_tail = be32_to_cpu(lh->lh_tail);
145 	head->lh_blkno = be32_to_cpu(lh->lh_blkno);
146 
147 	return 0;
148 }
149 /**
150  * get_log_header - read the log header for a given segment
151  * @jd: the journal
152  * @blk: the block to look at
153  * @lh: the log header to return
154  *
155  * Read the log header for a given segement in a given journal.  Do a few
156  * sanity checks on it.
157  *
158  * Returns: 0 on success,
159  *          1 if the header was invalid or incomplete,
160  *          errno on error
161  */
162 
163 static int get_log_header(struct gfs2_jdesc *jd, unsigned int blk,
164 			  struct gfs2_log_header_host *head)
165 {
166 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
167 	struct buffer_head *bh;
168 	int error;
169 
170 	error = gfs2_replay_read_block(jd, blk, &bh);
171 	if (error)
172 		return error;
173 
174 	error = __get_log_header(sdp, (const struct gfs2_log_header *)bh->b_data,
175 				 blk, head);
176 	brelse(bh);
177 
178 	return error;
179 }
180 
181 /**
182  * foreach_descriptor - go through the active part of the log
183  * @jd: the journal
184  * @start: the first log header in the active region
185  * @end: the last log header (don't process the contents of this entry))
186  *
187  * Call a given function once for every log descriptor in the active
188  * portion of the log.
189  *
190  * Returns: errno
191  */
192 
193 static int foreach_descriptor(struct gfs2_jdesc *jd, u32 start,
194 			      unsigned int end, int pass)
195 {
196 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
197 	struct buffer_head *bh;
198 	struct gfs2_log_descriptor *ld;
199 	int error = 0;
200 	u32 length;
201 	__be64 *ptr;
202 	unsigned int offset = sizeof(struct gfs2_log_descriptor);
203 	offset += sizeof(__be64) - 1;
204 	offset &= ~(sizeof(__be64) - 1);
205 
206 	while (start != end) {
207 		error = gfs2_replay_read_block(jd, start, &bh);
208 		if (error)
209 			return error;
210 		if (gfs2_meta_check(sdp, bh)) {
211 			brelse(bh);
212 			return -EIO;
213 		}
214 		ld = (struct gfs2_log_descriptor *)bh->b_data;
215 		length = be32_to_cpu(ld->ld_length);
216 
217 		if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
218 			struct gfs2_log_header_host lh;
219 			error = get_log_header(jd, start, &lh);
220 			if (!error) {
221 				gfs2_replay_incr_blk(jd, &start);
222 				brelse(bh);
223 				continue;
224 			}
225 			if (error == 1) {
226 				gfs2_consist_inode(GFS2_I(jd->jd_inode));
227 				error = -EIO;
228 			}
229 			brelse(bh);
230 			return error;
231 		} else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
232 			brelse(bh);
233 			return -EIO;
234 		}
235 		ptr = (__be64 *)(bh->b_data + offset);
236 		error = lops_scan_elements(jd, start, ld, ptr, pass);
237 		if (error) {
238 			brelse(bh);
239 			return error;
240 		}
241 
242 		while (length--)
243 			gfs2_replay_incr_blk(jd, &start);
244 
245 		brelse(bh);
246 	}
247 
248 	return 0;
249 }
250 
251 /**
252  * clean_journal - mark a dirty journal as being clean
253  * @jd: the journal
254  * @head: the head journal to start from
255  *
256  * Returns: errno
257  */
258 
259 static void clean_journal(struct gfs2_jdesc *jd,
260 			  struct gfs2_log_header_host *head)
261 {
262 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
263 	u32 lblock = head->lh_blkno;
264 
265 	gfs2_replay_incr_blk(jd, &lblock);
266 	gfs2_write_log_header(sdp, jd, head->lh_sequence + 1, 0, lblock,
267 			      GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
268 			      REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
269 	if (jd->jd_jid == sdp->sd_lockstruct.ls_jid) {
270 		sdp->sd_log_flush_head = lblock;
271 		gfs2_log_incr_head(sdp);
272 	}
273 }
274 
275 
276 static void gfs2_recovery_done(struct gfs2_sbd *sdp, unsigned int jid,
277                                unsigned int message)
278 {
279 	char env_jid[20];
280 	char env_status[20];
281 	char *envp[] = { env_jid, env_status, NULL };
282 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
283 
284         ls->ls_recover_jid_done = jid;
285         ls->ls_recover_jid_status = message;
286 	sprintf(env_jid, "JID=%u", jid);
287 	sprintf(env_status, "RECOVERY=%s",
288 		message == LM_RD_SUCCESS ? "Done" : "Failed");
289         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
290 
291 	if (sdp->sd_lockstruct.ls_ops->lm_recovery_result)
292 		sdp->sd_lockstruct.ls_ops->lm_recovery_result(sdp, jid, message);
293 }
294 
295 void gfs2_recover_func(struct work_struct *work)
296 {
297 	struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
298 	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
299 	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
300 	struct gfs2_log_header_host head;
301 	struct gfs2_holder j_gh, ji_gh, thaw_gh;
302 	ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
303 	int ro = 0;
304 	unsigned int pass;
305 	int error = 0;
306 	int jlocked = 0;
307 
308 	if (gfs2_withdrawn(sdp)) {
309 		fs_err(sdp, "jid=%u: Recovery not attempted due to withdraw.\n",
310 		       jd->jd_jid);
311 		goto fail;
312 	}
313 	t_start = ktime_get();
314 	if (sdp->sd_args.ar_spectator)
315 		goto fail;
316 	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
317 		fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
318 			jd->jd_jid);
319 		jlocked = 1;
320 		/* Acquire the journal lock so we can do recovery */
321 
322 		error = gfs2_glock_nq_num(sdp, jd->jd_jid, &gfs2_journal_glops,
323 					  LM_ST_EXCLUSIVE,
324 					  LM_FLAG_NOEXP | LM_FLAG_TRY | GL_NOCACHE,
325 					  &j_gh);
326 		switch (error) {
327 		case 0:
328 			break;
329 
330 		case GLR_TRYFAILED:
331 			fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
332 			error = 0;
333 
334 		default:
335 			goto fail;
336 		}
337 
338 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
339 					   LM_FLAG_NOEXP | GL_NOCACHE, &ji_gh);
340 		if (error)
341 			goto fail_gunlock_j;
342 	} else {
343 		fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
344 	}
345 
346 	t_jlck = ktime_get();
347 	fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
348 
349 	error = gfs2_jdesc_check(jd);
350 	if (error)
351 		goto fail_gunlock_ji;
352 
353 	error = gfs2_find_jhead(jd, &head, true);
354 	if (error)
355 		goto fail_gunlock_ji;
356 	t_jhd = ktime_get();
357 	fs_info(sdp, "jid=%u: Journal head lookup took %lldms\n", jd->jd_jid,
358 		ktime_ms_delta(t_jhd, t_jlck));
359 
360 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
361 		fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
362 			jd->jd_jid);
363 
364 		/* Acquire a shared hold on the freeze lock */
365 
366 		error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
367 					   LM_FLAG_NOEXP | LM_FLAG_PRIORITY,
368 					   &thaw_gh);
369 		if (error)
370 			goto fail_gunlock_ji;
371 
372 		if (test_bit(SDF_RORECOVERY, &sdp->sd_flags)) {
373 			ro = 1;
374 		} else if (test_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags)) {
375 			if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
376 				ro = 1;
377 		} else {
378 			if (sb_rdonly(sdp->sd_vfs)) {
379 				/* check if device itself is read-only */
380 				ro = bdev_read_only(sdp->sd_vfs->s_bdev);
381 				if (!ro) {
382 					fs_info(sdp, "recovery required on "
383 						"read-only filesystem.\n");
384 					fs_info(sdp, "write access will be "
385 						"enabled during recovery.\n");
386 				}
387 			}
388 		}
389 
390 		if (ro) {
391 			fs_warn(sdp, "jid=%u: Can't replay: read-only block "
392 				"device\n", jd->jd_jid);
393 			error = -EROFS;
394 			goto fail_gunlock_thaw;
395 		}
396 
397 		t_tlck = ktime_get();
398 		fs_info(sdp, "jid=%u: Replaying journal...0x%x to 0x%x\n",
399 			jd->jd_jid, head.lh_tail, head.lh_blkno);
400 
401 		/* We take the sd_log_flush_lock here primarily to prevent log
402 		 * flushes and simultaneous journal replays from stomping on
403 		 * each other wrt sd_log_bio. */
404 		down_read(&sdp->sd_log_flush_lock);
405 		for (pass = 0; pass < 2; pass++) {
406 			lops_before_scan(jd, &head, pass);
407 			error = foreach_descriptor(jd, head.lh_tail,
408 						   head.lh_blkno, pass);
409 			lops_after_scan(jd, error, pass);
410 			if (error)
411 				goto fail_gunlock_thaw;
412 		}
413 
414 		clean_journal(jd, &head);
415 		up_read(&sdp->sd_log_flush_lock);
416 
417 		gfs2_glock_dq_uninit(&thaw_gh);
418 		t_rep = ktime_get();
419 		fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
420 			"jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
421 			jd->jd_jid, ktime_ms_delta(t_rep, t_start),
422 			ktime_ms_delta(t_jlck, t_start),
423 			ktime_ms_delta(t_jhd, t_jlck),
424 			ktime_ms_delta(t_tlck, t_jhd),
425 			ktime_ms_delta(t_rep, t_tlck));
426 	}
427 
428 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
429 
430 	if (jlocked) {
431 		gfs2_glock_dq_uninit(&ji_gh);
432 		gfs2_glock_dq_uninit(&j_gh);
433 	}
434 
435 	fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
436 	goto done;
437 
438 fail_gunlock_thaw:
439 	gfs2_glock_dq_uninit(&thaw_gh);
440 fail_gunlock_ji:
441 	if (jlocked) {
442 		gfs2_glock_dq_uninit(&ji_gh);
443 fail_gunlock_j:
444 		gfs2_glock_dq_uninit(&j_gh);
445 	}
446 
447 	fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
448 fail:
449 	jd->jd_recover_error = error;
450 	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
451 done:
452 	clear_bit(JDF_RECOVERY, &jd->jd_flags);
453 	smp_mb__after_atomic();
454 	wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
455 }
456 
457 int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
458 {
459 	int rv;
460 
461 	if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
462 		return -EBUSY;
463 
464 	/* we have JDF_RECOVERY, queue should always succeed */
465 	rv = queue_work(gfs_recovery_wq, &jd->jd_work);
466 	BUG_ON(!rv);
467 
468 	if (wait)
469 		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
470 			    TASK_UNINTERRUPTIBLE);
471 
472 	return wait ? jd->jd_recover_error : 0;
473 }
474 
475