recovery.c

  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
 30struct workqueue_struct *gfs2_recovery_wq;
 31
 32int 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	u64 dblock;
 38	u32 extlen;
 39	int error;
 40
 41	extlen = 32;
 42	error = gfs2_get_extent(&ip->i_inode, blk, &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
 55int 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 = NULL, *iter;
 59
 60	list_for_each_entry(iter, head, rr_list) {
 61		if (iter->rr_blkno == blkno) {
 62			rr = iter;
 63			break;
 64		}
 65	}
 66
 67	if (rr) {
 68		rr->rr_where = where;
 69		return 0;
 70	}
 71
 72	rr = kmalloc(sizeof(struct gfs2_revoke_replay), GFP_NOFS);
 73	if (!rr)
 74		return -ENOMEM;
 75
 76	rr->rr_blkno = blkno;
 77	rr->rr_where = where;
 78	list_add(&rr->rr_list, head);
 79
 80	return 1;
 81}
 82
 83int gfs2_revoke_check(struct gfs2_jdesc *jd, u64 blkno, unsigned int where)
 84{
 85	struct gfs2_revoke_replay *rr = NULL, *iter;
 86	int wrap, a, b, revoke;
 87
 88	list_for_each_entry(iter, &jd->jd_revoke_list, rr_list) {
 89		if (iter->rr_blkno == blkno) {
 90			rr = iter;
 91			break;
 92		}
 93	}
 94
 95	if (!rr)
 96		return 0;
 97
 98	wrap = (rr->rr_where < jd->jd_replay_tail);
 99	a = (jd->jd_replay_tail < where);
100	b = (where < rr->rr_where);
101	revoke = (wrap) ? (a || b) : (a && b);
102
103	return revoke;
104}
105
106void gfs2_revoke_clean(struct gfs2_jdesc *jd)
107{
108	struct list_head *head = &jd->jd_revoke_list;
109	struct gfs2_revoke_replay *rr;
110
111	while (!list_empty(head)) {
112		rr = list_first_entry(head, struct gfs2_revoke_replay, rr_list);
113		list_del(&rr->rr_list);
114		kfree(rr);
115	}
116}
117
118int __get_log_header(struct gfs2_sbd *sdp, const struct gfs2_log_header *lh,
119		     unsigned int blkno, struct gfs2_log_header_host *head)
120{
121	const u32 zero = 0;
122	u32 hash, crc;
123
124	if (lh->lh_header.mh_magic != cpu_to_be32(GFS2_MAGIC) ||
125	    lh->lh_header.mh_type != cpu_to_be32(GFS2_METATYPE_LH) ||
126	    (blkno && be32_to_cpu(lh->lh_blkno) != blkno))
127		return 1;
128
129	hash = crc32(~0, lh, LH_V1_SIZE - 4);
130	hash = ~crc32(hash, &zero, 4); /* assume lh_hash is zero */
131
132	if (be32_to_cpu(lh->lh_hash) != hash)
133		return 1;
134
135	crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
136		     sdp->sd_sb.sb_bsize - LH_V1_SIZE - 4);
137
138	if ((lh->lh_crc != 0 && be32_to_cpu(lh->lh_crc) != crc))
139		return 1;
140
141	head->lh_sequence = be64_to_cpu(lh->lh_sequence);
142	head->lh_flags = be32_to_cpu(lh->lh_flags);
143	head->lh_tail = be32_to_cpu(lh->lh_tail);
144	head->lh_blkno = be32_to_cpu(lh->lh_blkno);
145
146	head->lh_local_total = be64_to_cpu(lh->lh_local_total);
147	head->lh_local_free = be64_to_cpu(lh->lh_local_free);
148	head->lh_local_dinodes = be64_to_cpu(lh->lh_local_dinodes);
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 * @head: 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
166static 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 * @pass: iteration number (foreach_descriptor() is called in a for() loop)
190 *
191 * Call a given function once for every log descriptor in the active
192 * portion of the log.
193 *
194 * Returns: errno
195 */
196
197static int foreach_descriptor(struct gfs2_jdesc *jd, u32 start,
198			      unsigned int end, int pass)
199{
200	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
201	struct buffer_head *bh;
202	struct gfs2_log_descriptor *ld;
203	int error = 0;
204	u32 length;
205	__be64 *ptr;
206	unsigned int offset = sizeof(struct gfs2_log_descriptor);
207	offset += sizeof(__be64) - 1;
208	offset &= ~(sizeof(__be64) - 1);
209
210	while (start != end) {
211		error = gfs2_replay_read_block(jd, start, &bh);
212		if (error)
213			return error;
214		if (gfs2_meta_check(sdp, bh)) {
215			brelse(bh);
216			return -EIO;
217		}
218		ld = (struct gfs2_log_descriptor *)bh->b_data;
219		length = be32_to_cpu(ld->ld_length);
220
221		if (be32_to_cpu(ld->ld_header.mh_type) == GFS2_METATYPE_LH) {
222			struct gfs2_log_header_host lh;
223			error = get_log_header(jd, start, &lh);
224			if (!error) {
225				gfs2_replay_incr_blk(jd, &start);
226				brelse(bh);
227				continue;
228			}
229			if (error == 1) {
230				gfs2_consist_inode(GFS2_I(jd->jd_inode));
231				error = -EIO;
232			}
233			brelse(bh);
234			return error;
235		} else if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LD)) {
236			brelse(bh);
237			return -EIO;
238		}
239		ptr = (__be64 *)(bh->b_data + offset);
240		error = lops_scan_elements(jd, start, ld, ptr, pass);
241		if (error) {
242			brelse(bh);
243			return error;
244		}
245
246		while (length--)
247			gfs2_replay_incr_blk(jd, &start);
248
249		brelse(bh);
250	}
251
252	return 0;
253}
254
255/**
256 * clean_journal - mark a dirty journal as being clean
257 * @jd: the journal
258 * @head: the head journal to start from
259 *
260 * Returns: errno
261 */
262
263static void clean_journal(struct gfs2_jdesc *jd,
264			  struct gfs2_log_header_host *head)
265{
266	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
267
268	gfs2_replay_incr_blk(jd, &head->lh_blkno);
269	head->lh_sequence++;
270	gfs2_write_log_header(sdp, jd, head->lh_sequence, 0, head->lh_blkno,
271			      GFS2_LOG_HEAD_UNMOUNT | GFS2_LOG_HEAD_RECOVERY,
272			      REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC);
273}
274
275
276static 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/**
296 * update_statfs_inode - Update the master statfs inode or zero out the local
297 *			 statfs inode for a given journal.
298 * @jd: The journal
299 * @head: If NULL, @inode is the local statfs inode and we need to zero it out.
300 *	  Otherwise, it @head contains the statfs change info that needs to be
301 *	  synced to the master statfs inode (pointed to by @inode).
302 * @inode: statfs inode to update.
303 */
304static int update_statfs_inode(struct gfs2_jdesc *jd,
305			       struct gfs2_log_header_host *head,
306			       struct inode *inode)
307{
308	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
309	struct gfs2_inode *ip;
310	struct buffer_head *bh;
311	struct gfs2_statfs_change_host sc;
312	int error = 0;
313
314	BUG_ON(!inode);
315	ip = GFS2_I(inode);
316
317	error = gfs2_meta_inode_buffer(ip, &bh);
318	if (error)
319		goto out;
320
321	spin_lock(&sdp->sd_statfs_spin);
322
323	if (head) { /* Update the master statfs inode */
324		gfs2_statfs_change_in(&sc, bh->b_data + sizeof(struct gfs2_dinode));
325		sc.sc_total += head->lh_local_total;
326		sc.sc_free += head->lh_local_free;
327		sc.sc_dinodes += head->lh_local_dinodes;
328		gfs2_statfs_change_out(&sc, bh->b_data + sizeof(struct gfs2_dinode));
329
330		fs_info(sdp, "jid=%u: Updated master statfs Total:%lld, "
331			"Free:%lld, Dinodes:%lld after change "
332			"[%+lld,%+lld,%+lld]\n", jd->jd_jid, sc.sc_total,
333			sc.sc_free, sc.sc_dinodes, head->lh_local_total,
334			head->lh_local_free, head->lh_local_dinodes);
335	} else { /* Zero out the local statfs inode */
336		memset(bh->b_data + sizeof(struct gfs2_dinode), 0,
337		       sizeof(struct gfs2_statfs_change));
338		/* If it's our own journal, reset any in-memory changes too */
339		if (jd->jd_jid == sdp->sd_lockstruct.ls_jid) {
340			memset(&sdp->sd_statfs_local, 0,
341			       sizeof(struct gfs2_statfs_change_host));
342		}
343	}
344	spin_unlock(&sdp->sd_statfs_spin);
345
346	mark_buffer_dirty(bh);
347	brelse(bh);
348	gfs2_inode_metasync(ip->i_gl);
349
350out:
351	return error;
352}
353
354/**
355 * recover_local_statfs - Update the master and local statfs changes for this
356 *			  journal.
357 *
358 * Previously, statfs updates would be read in from the local statfs inode and
359 * synced to the master statfs inode during recovery.
360 *
361 * We now use the statfs updates in the journal head to update the master statfs
362 * inode instead of reading in from the local statfs inode. To preserve backward
363 * compatibility with kernels that can't do this, we still need to keep the
364 * local statfs inode up to date by writing changes to it. At some point in the
365 * future, we can do away with the local statfs inodes altogether and keep the
366 * statfs changes solely in the journal.
367 *
368 * @jd: the journal
369 * @head: the journal head
370 *
371 * Returns: errno
372 */
373static void recover_local_statfs(struct gfs2_jdesc *jd,
374				 struct gfs2_log_header_host *head)
375{
376	int error;
377	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
378
379	if (!head->lh_local_total && !head->lh_local_free
380	    && !head->lh_local_dinodes) /* No change */
381		goto zero_local;
382
383	 /* First update the master statfs inode with the changes we
384	  * found in the journal. */
385	error = update_statfs_inode(jd, head, sdp->sd_statfs_inode);
386	if (error)
387		goto out;
388
389zero_local:
390	/* Zero out the local statfs inode so any changes in there
391	 * are not re-recovered. */
392	error = update_statfs_inode(jd, NULL,
393				    find_local_statfs_inode(sdp, jd->jd_jid));
394out:
395	return;
396}
397
398void gfs2_recover_func(struct work_struct *work)
399{
400	struct gfs2_jdesc *jd = container_of(work, struct gfs2_jdesc, jd_work);
401	struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
402	struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
403	struct gfs2_log_header_host head;
404	struct gfs2_holder j_gh, ji_gh;
405	ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
406	int ro = 0;
407	unsigned int pass;
408	int error = 0;
409	int jlocked = 0;
410
411	if (gfs2_withdrawn(sdp)) {
412		fs_err(sdp, "jid=%u: Recovery not attempted due to withdraw.\n",
413		       jd->jd_jid);
414		goto fail;
415	}
416	t_start = ktime_get();
417	if (sdp->sd_args.ar_spectator)
418		goto fail;
419	if (jd->jd_jid != sdp->sd_lockstruct.ls_jid) {
420		fs_info(sdp, "jid=%u: Trying to acquire journal glock...\n",
421			jd->jd_jid);
422		jlocked = 1;
423		/* Acquire the journal glock 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_RECOVER | LM_FLAG_TRY |
428					  GL_NOCACHE,
429					  &j_gh);
430		switch (error) {
431		case 0:
432			break;
433
434		case GLR_TRYFAILED:
435			fs_info(sdp, "jid=%u: Busy\n", jd->jd_jid);
436			error = 0;
437			goto fail;
438
439		default:
440			goto fail;
441		}
442
443		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
444					   LM_FLAG_RECOVER | GL_NOCACHE,
445					   &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	fs_info(sdp, "jid=%u: Journal head lookup took %lldms\n", jd->jd_jid,
464		ktime_ms_delta(t_jhd, t_jlck));
465
466	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
467		mutex_lock(&sdp->sd_freeze_mutex);
468
469		if (test_bit(SDF_FROZEN, &sdp->sd_flags)) {
470			mutex_unlock(&sdp->sd_freeze_mutex);
471			fs_warn(sdp, "jid=%u: Can't replay: filesystem "
472				"is frozen\n", jd->jd_jid);
473			goto fail_gunlock_ji;
474		}
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_nofreeze;
499		}
500
501		t_tlck = ktime_get();
502		fs_info(sdp, "jid=%u: Replaying journal...0x%x to 0x%x\n",
503			jd->jd_jid, head.lh_tail, head.lh_blkno);
504
505		/* We take the sd_log_flush_lock here primarily to prevent log
506		 * flushes and simultaneous journal replays from stomping on
507		 * each other wrt jd_log_bio. */
508		down_read(&sdp->sd_log_flush_lock);
509		for (pass = 0; pass < 2; pass++) {
510			lops_before_scan(jd, &head, pass);
511			error = foreach_descriptor(jd, head.lh_tail,
512						   head.lh_blkno, pass);
513			lops_after_scan(jd, error, pass);
514			if (error) {
515				up_read(&sdp->sd_log_flush_lock);
516				goto fail_gunlock_nofreeze;
517			}
518		}
519
520		recover_local_statfs(jd, &head);
521		clean_journal(jd, &head);
522		up_read(&sdp->sd_log_flush_lock);
523
524		mutex_unlock(&sdp->sd_freeze_mutex);
525		t_rep = ktime_get();
526		fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
527			"jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
528			jd->jd_jid, ktime_ms_delta(t_rep, t_start),
529			ktime_ms_delta(t_jlck, t_start),
530			ktime_ms_delta(t_jhd, t_jlck),
531			ktime_ms_delta(t_tlck, t_jhd),
532			ktime_ms_delta(t_rep, t_tlck));
533	}
534
535	if (jd->jd_jid == sdp->sd_lockstruct.ls_jid)
536		gfs2_log_pointers_init(sdp, &head);
537
538	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);
539
540	if (jlocked) {
541		gfs2_glock_dq_uninit(&ji_gh);
542		gfs2_glock_dq_uninit(&j_gh);
543	}
544
545	fs_info(sdp, "jid=%u: Done\n", jd->jd_jid);
546	goto done;
547
548fail_gunlock_nofreeze:
549	mutex_unlock(&sdp->sd_freeze_mutex);
550fail_gunlock_ji:
551	if (jlocked) {
552		gfs2_glock_dq_uninit(&ji_gh);
553fail_gunlock_j:
554		gfs2_glock_dq_uninit(&j_gh);
555	}
556
557	fs_info(sdp, "jid=%u: %s\n", jd->jd_jid, (error) ? "Failed" : "Done");
558fail:
559	jd->jd_recover_error = error;
560	gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_GAVEUP);
561done:
562	clear_bit(JDF_RECOVERY, &jd->jd_flags);
563	smp_mb__after_atomic();
564	wake_up_bit(&jd->jd_flags, JDF_RECOVERY);
565}
566
567int gfs2_recover_journal(struct gfs2_jdesc *jd, bool wait)
568{
569	int rv;
570
571	if (test_and_set_bit(JDF_RECOVERY, &jd->jd_flags))
572		return -EBUSY;
573
574	/* we have JDF_RECOVERY, queue should always succeed */
575	rv = queue_work(gfs2_recovery_wq, &jd->jd_work);
576	BUG_ON(!rv);
577
578	if (wait)
579		wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
580			    TASK_UNINTERRUPTIBLE);
581
582	return wait ? jd->jd_recover_error : 0;
583}
584
585void gfs2_log_pointers_init(struct gfs2_sbd *sdp,
586			    struct gfs2_log_header_host *head)
587{
588	sdp->sd_log_sequence = head->lh_sequence + 1;
589	gfs2_replay_incr_blk(sdp->sd_jdesc, &head->lh_blkno);
590	sdp->sd_log_tail = head->lh_blkno;
591	sdp->sd_log_flush_head = head->lh_blkno;
592	sdp->sd_log_flush_tail = head->lh_blkno;
593	sdp->sd_log_head = head->lh_blkno;
594}