1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/sched/signal.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/statfs.h>
16 #include <linux/seq_file.h>
17 #include <linux/mount.h>
18 #include <linux/kthread.h>
19 #include <linux/delay.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/crc32.h>
22 #include <linux/time.h>
23 #include <linux/wait.h>
24 #include <linux/writeback.h>
25 #include <linux/backing-dev.h>
26 #include <linux/kernel.h>
49 SHOULD_NOT_DELETE_DINODE,
50 SHOULD_DEFER_EVICTION,
54 * gfs2_jindex_free - Clear all the journal index information
55 * @sdp: The GFS2 superblock
59 void gfs2_jindex_free(struct gfs2_sbd *sdp)
61 struct list_head list;
62 struct gfs2_jdesc *jd;
64 spin_lock(&sdp->sd_jindex_spin);
65 list_add(&list, &sdp->sd_jindex_list);
66 list_del_init(&sdp->sd_jindex_list);
68 spin_unlock(&sdp->sd_jindex_spin);
71 while (!list_empty(&list)) {
72 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
73 gfs2_free_journal_extents(jd);
74 list_del(&jd->jd_list);
81 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
83 struct gfs2_jdesc *jd;
85 list_for_each_entry(jd, head, jd_list) {
86 if (jd->jd_jid == jid)
92 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
94 struct gfs2_jdesc *jd;
96 spin_lock(&sdp->sd_jindex_spin);
97 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
98 spin_unlock(&sdp->sd_jindex_spin);
103 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
105 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
106 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
107 u64 size = i_size_read(jd->jd_inode);
109 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
112 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
114 if (gfs2_write_alloc_required(ip, 0, size)) {
115 gfs2_consist_inode(ip);
123 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
124 * @sdp: the filesystem
129 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
131 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
132 struct gfs2_glock *j_gl = ip->i_gl;
133 struct gfs2_log_header_host head;
136 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
137 if (gfs2_withdrawn(sdp))
140 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
141 if (error || gfs2_withdrawn(sdp))
144 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
149 /* Initialize some head of the log stuff */
150 sdp->sd_log_sequence = head.lh_sequence + 1;
151 gfs2_log_pointers_init(sdp, head.lh_blkno);
153 error = gfs2_quota_init(sdp);
154 if (!error && !gfs2_withdrawn(sdp))
155 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
159 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
161 const struct gfs2_statfs_change *str = buf;
163 sc->sc_total = be64_to_cpu(str->sc_total);
164 sc->sc_free = be64_to_cpu(str->sc_free);
165 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
168 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
170 struct gfs2_statfs_change *str = buf;
172 str->sc_total = cpu_to_be64(sc->sc_total);
173 str->sc_free = cpu_to_be64(sc->sc_free);
174 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
177 int gfs2_statfs_init(struct gfs2_sbd *sdp)
179 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
180 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
181 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
182 struct buffer_head *m_bh;
183 struct gfs2_holder gh;
186 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
191 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
195 if (sdp->sd_args.ar_spectator) {
196 spin_lock(&sdp->sd_statfs_spin);
197 gfs2_statfs_change_in(m_sc, m_bh->b_data +
198 sizeof(struct gfs2_dinode));
199 spin_unlock(&sdp->sd_statfs_spin);
201 spin_lock(&sdp->sd_statfs_spin);
202 gfs2_statfs_change_in(m_sc, m_bh->b_data +
203 sizeof(struct gfs2_dinode));
204 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
205 sizeof(struct gfs2_dinode));
206 spin_unlock(&sdp->sd_statfs_spin);
212 gfs2_glock_dq_uninit(&gh);
216 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
219 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
220 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
221 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
225 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
227 spin_lock(&sdp->sd_statfs_spin);
228 l_sc->sc_total += total;
229 l_sc->sc_free += free;
230 l_sc->sc_dinodes += dinodes;
231 gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
232 sizeof(struct gfs2_dinode));
233 if (sdp->sd_args.ar_statfs_percent) {
234 x = 100 * l_sc->sc_free;
235 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
236 if (x >= y || x <= -y)
239 spin_unlock(&sdp->sd_statfs_spin);
242 gfs2_wake_up_statfs(sdp);
245 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
247 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
248 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
249 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
250 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
252 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
253 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
255 spin_lock(&sdp->sd_statfs_spin);
256 m_sc->sc_total += l_sc->sc_total;
257 m_sc->sc_free += l_sc->sc_free;
258 m_sc->sc_dinodes += l_sc->sc_dinodes;
259 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
260 memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
261 0, sizeof(struct gfs2_statfs_change));
262 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
263 spin_unlock(&sdp->sd_statfs_spin);
266 int gfs2_statfs_sync(struct super_block *sb, int type)
268 struct gfs2_sbd *sdp = sb->s_fs_info;
269 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
270 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
271 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
272 struct gfs2_holder gh;
273 struct buffer_head *m_bh;
276 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
281 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
285 spin_lock(&sdp->sd_statfs_spin);
286 gfs2_statfs_change_in(m_sc, m_bh->b_data +
287 sizeof(struct gfs2_dinode));
288 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
289 spin_unlock(&sdp->sd_statfs_spin);
292 spin_unlock(&sdp->sd_statfs_spin);
294 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
298 update_statfs(sdp, m_bh);
299 sdp->sd_statfs_force_sync = 0;
306 gfs2_glock_dq_uninit(&gh);
312 struct list_head list;
313 struct gfs2_holder gh;
317 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
319 * @sdp: the file system
324 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
326 struct gfs2_inode *ip;
327 struct gfs2_jdesc *jd;
330 struct gfs2_log_header_host lh;
333 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
334 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
339 ip = GFS2_I(jd->jd_inode);
340 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
345 list_add(&lfcc->list, &list);
348 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
349 LM_FLAG_NOEXP | GL_NOPID,
354 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
355 error = gfs2_jdesc_check(jd);
358 error = gfs2_find_jhead(jd, &lh, false);
361 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
368 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
371 while (!list_empty(&list)) {
372 lfcc = list_first_entry(&list, struct lfcc, list);
373 list_del(&lfcc->list);
374 gfs2_glock_dq_uninit(&lfcc->gh);
380 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
382 const struct inode *inode = &ip->i_inode;
383 struct gfs2_dinode *str = buf;
385 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
386 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
387 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
388 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
389 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
390 str->di_mode = cpu_to_be32(inode->i_mode);
391 str->di_uid = cpu_to_be32(i_uid_read(inode));
392 str->di_gid = cpu_to_be32(i_gid_read(inode));
393 str->di_nlink = cpu_to_be32(inode->i_nlink);
394 str->di_size = cpu_to_be64(i_size_read(inode));
395 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
396 str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
397 str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
398 str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec);
400 str->di_goal_meta = cpu_to_be64(ip->i_goal);
401 str->di_goal_data = cpu_to_be64(ip->i_goal);
402 str->di_generation = cpu_to_be64(ip->i_generation);
404 str->di_flags = cpu_to_be32(ip->i_diskflags);
405 str->di_height = cpu_to_be16(ip->i_height);
406 str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
407 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
409 str->di_depth = cpu_to_be16(ip->i_depth);
410 str->di_entries = cpu_to_be32(ip->i_entries);
412 str->di_eattr = cpu_to_be64(ip->i_eattr);
413 str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
414 str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
415 str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
419 * gfs2_write_inode - Make sure the inode is stable on the disk
421 * @wbc: The writeback control structure
426 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
428 struct gfs2_inode *ip = GFS2_I(inode);
429 struct gfs2_sbd *sdp = GFS2_SB(inode);
430 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
431 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
433 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
436 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
437 GFS2_LOG_HEAD_FLUSH_NORMAL |
438 GFS2_LFC_WRITE_INODE);
439 if (bdi->wb.dirty_exceeded)
440 gfs2_ail1_flush(sdp, wbc);
442 filemap_fdatawrite(metamapping);
444 ret = filemap_fdatawait(metamapping);
446 mark_inode_dirty_sync(inode);
448 spin_lock(&inode->i_lock);
449 if (!(inode->i_flags & I_DIRTY))
450 gfs2_ordered_del_inode(ip);
451 spin_unlock(&inode->i_lock);
457 * gfs2_dirty_inode - check for atime updates
458 * @inode: The inode in question
459 * @flags: The type of dirty
461 * Unfortunately it can be called under any combination of inode
462 * glock and transaction lock, so we have to check carefully.
464 * At the moment this deals only with atime - it should be possible
465 * to expand that role in future, once a review of the locking has
469 static void gfs2_dirty_inode(struct inode *inode, int flags)
471 struct gfs2_inode *ip = GFS2_I(inode);
472 struct gfs2_sbd *sdp = GFS2_SB(inode);
473 struct buffer_head *bh;
474 struct gfs2_holder gh;
476 int need_endtrans = 0;
479 if (unlikely(!ip->i_gl)) {
480 /* This can only happen during incomplete inode creation. */
481 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
485 if (unlikely(gfs2_withdrawn(sdp)))
487 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
488 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
490 fs_err(sdp, "dirty_inode: glock %d\n", ret);
491 gfs2_dump_glock(NULL, ip->i_gl, true);
495 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
498 if (current->journal_info == NULL) {
499 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
501 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
507 ret = gfs2_meta_inode_buffer(ip, &bh);
509 gfs2_trans_add_meta(ip->i_gl, bh);
510 gfs2_dinode_out(ip, bh->b_data);
518 gfs2_glock_dq_uninit(&gh);
522 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
523 * @sdp: the filesystem
528 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
530 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
532 gfs2_flush_delete_work(sdp);
533 if (!log_write_allowed && current == sdp->sd_quotad_process)
534 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
535 else if (sdp->sd_quotad_process)
536 kthread_stop(sdp->sd_quotad_process);
537 sdp->sd_quotad_process = NULL;
539 if (!log_write_allowed && current == sdp->sd_logd_process)
540 fs_warn(sdp, "The logd daemon is withdrawing.\n");
541 else if (sdp->sd_logd_process)
542 kthread_stop(sdp->sd_logd_process);
543 sdp->sd_logd_process = NULL;
545 if (log_write_allowed) {
546 gfs2_quota_sync(sdp->sd_vfs, 0);
547 gfs2_statfs_sync(sdp->sd_vfs, 0);
549 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
550 GFS2_LFC_MAKE_FS_RO);
551 wait_event_timeout(sdp->sd_log_waitq,
552 gfs2_log_is_empty(sdp),
554 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
556 wait_event_timeout(sdp->sd_log_waitq,
557 gfs2_log_is_empty(sdp),
560 gfs2_quota_cleanup(sdp);
562 if (!log_write_allowed)
563 sdp->sd_vfs->s_flags |= SB_RDONLY;
567 * gfs2_put_super - Unmount the filesystem
568 * @sb: The VFS superblock
572 static void gfs2_put_super(struct super_block *sb)
574 struct gfs2_sbd *sdp = sb->s_fs_info;
575 struct gfs2_jdesc *jd;
577 /* No more recovery requests */
578 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
581 /* Wait on outstanding recovery */
583 spin_lock(&sdp->sd_jindex_spin);
584 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
585 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
587 spin_unlock(&sdp->sd_jindex_spin);
588 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
589 TASK_UNINTERRUPTIBLE);
592 spin_unlock(&sdp->sd_jindex_spin);
594 if (!sb_rdonly(sb)) {
595 gfs2_make_fs_ro(sdp);
597 WARN_ON(gfs2_withdrawing(sdp));
599 /* At this point, we're through modifying the disk */
603 iput(sdp->sd_jindex);
604 iput(sdp->sd_statfs_inode);
605 iput(sdp->sd_rindex);
606 iput(sdp->sd_quota_inode);
608 gfs2_glock_put(sdp->sd_rename_gl);
609 gfs2_glock_put(sdp->sd_freeze_gl);
611 if (!sdp->sd_args.ar_spectator) {
612 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
613 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
614 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
615 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
616 brelse(sdp->sd_sc_bh);
617 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
618 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
619 free_local_statfs_inodes(sdp);
620 iput(sdp->sd_qc_inode);
623 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
624 gfs2_clear_rgrpd(sdp);
625 gfs2_jindex_free(sdp);
626 /* Take apart glock structures and buffer lists */
627 gfs2_gl_hash_clear(sdp);
628 truncate_inode_pages_final(&sdp->sd_aspace);
629 gfs2_delete_debugfs_file(sdp);
630 /* Unmount the locking protocol */
631 gfs2_lm_unmount(sdp);
633 /* At this point, we're through participating in the lockspace */
634 gfs2_sys_fs_del(sdp);
639 * gfs2_sync_fs - sync the filesystem
640 * @sb: the superblock
641 * @wait: true to wait for completion
643 * Flushes the log to disk.
646 static int gfs2_sync_fs(struct super_block *sb, int wait)
648 struct gfs2_sbd *sdp = sb->s_fs_info;
650 gfs2_quota_sync(sb, -1);
652 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
654 return sdp->sd_log_error;
657 void gfs2_freeze_func(struct work_struct *work)
660 struct gfs2_holder freeze_gh;
661 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
662 struct super_block *sb = sdp->sd_vfs;
664 atomic_inc(&sb->s_active);
665 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
667 gfs2_assert_withdraw(sdp, 0);
669 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
670 error = thaw_super(sb);
672 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
674 gfs2_assert_withdraw(sdp, 0);
676 gfs2_freeze_unlock(&freeze_gh);
678 deactivate_super(sb);
679 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
680 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
685 * gfs2_freeze - prevent further writes to the filesystem
686 * @sb: the VFS structure for the filesystem
690 static int gfs2_freeze(struct super_block *sb)
692 struct gfs2_sbd *sdp = sb->s_fs_info;
695 mutex_lock(&sdp->sd_freeze_mutex);
696 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
702 if (gfs2_withdrawn(sdp)) {
707 error = gfs2_lock_fs_check_clean(sdp);
712 fs_err(sdp, "waiting for recovery before freeze\n");
713 else if (error == -EIO) {
714 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
715 "to recovery error.\n");
718 fs_err(sdp, "error freezing FS: %d\n", error);
720 fs_err(sdp, "retrying...\n");
723 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
725 mutex_unlock(&sdp->sd_freeze_mutex);
730 * gfs2_unfreeze - reallow writes to the filesystem
731 * @sb: the VFS structure for the filesystem
735 static int gfs2_unfreeze(struct super_block *sb)
737 struct gfs2_sbd *sdp = sb->s_fs_info;
739 mutex_lock(&sdp->sd_freeze_mutex);
740 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
741 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
742 mutex_unlock(&sdp->sd_freeze_mutex);
746 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
747 mutex_unlock(&sdp->sd_freeze_mutex);
748 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
752 * statfs_slow_fill - fill in the sg for a given RG
754 * @sc: the sc structure
756 * Returns: 0 on success, -ESTALE if the LVB is invalid
759 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
760 struct gfs2_statfs_change_host *sc)
762 gfs2_rgrp_verify(rgd);
763 sc->sc_total += rgd->rd_data;
764 sc->sc_free += rgd->rd_free;
765 sc->sc_dinodes += rgd->rd_dinodes;
770 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
771 * @sdp: the filesystem
772 * @sc: the sc info that will be returned
774 * Any error (other than a signal) will cause this routine to fall back
775 * to the synchronous version.
777 * FIXME: This really shouldn't busy wait like this.
782 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
784 struct gfs2_rgrpd *rgd_next;
785 struct gfs2_holder *gha, *gh;
786 unsigned int slots = 64;
791 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
792 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
795 for (x = 0; x < slots; x++)
796 gfs2_holder_mark_uninitialized(gha + x);
798 rgd_next = gfs2_rgrpd_get_first(sdp);
803 for (x = 0; x < slots; x++) {
806 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
807 err = gfs2_glock_wait(gh);
809 gfs2_holder_uninit(gh);
813 struct gfs2_rgrpd *rgd =
814 gfs2_glock2rgrp(gh->gh_gl);
816 error = statfs_slow_fill(rgd, sc);
818 gfs2_glock_dq_uninit(gh);
822 if (gfs2_holder_initialized(gh))
824 else if (rgd_next && !error) {
825 error = gfs2_glock_nq_init(rgd_next->rd_gl,
829 rgd_next = gfs2_rgrpd_get_next(rgd_next);
833 if (signal_pending(current))
834 error = -ERESTARTSYS;
848 * gfs2_statfs_i - Do a statfs
849 * @sdp: the filesystem
850 * @sc: the sc structure
855 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
857 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
858 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
860 spin_lock(&sdp->sd_statfs_spin);
863 sc->sc_total += l_sc->sc_total;
864 sc->sc_free += l_sc->sc_free;
865 sc->sc_dinodes += l_sc->sc_dinodes;
867 spin_unlock(&sdp->sd_statfs_spin);
871 if (sc->sc_free > sc->sc_total)
872 sc->sc_free = sc->sc_total;
873 if (sc->sc_dinodes < 0)
880 * gfs2_statfs - Gather and return stats about the filesystem
881 * @dentry: The name of the link
884 * Returns: 0 on success or error code
887 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
889 struct super_block *sb = dentry->d_sb;
890 struct gfs2_sbd *sdp = sb->s_fs_info;
891 struct gfs2_statfs_change_host sc;
894 error = gfs2_rindex_update(sdp);
898 if (gfs2_tune_get(sdp, gt_statfs_slow))
899 error = gfs2_statfs_slow(sdp, &sc);
901 error = gfs2_statfs_i(sdp, &sc);
906 buf->f_type = GFS2_MAGIC;
907 buf->f_bsize = sdp->sd_sb.sb_bsize;
908 buf->f_blocks = sc.sc_total;
909 buf->f_bfree = sc.sc_free;
910 buf->f_bavail = sc.sc_free;
911 buf->f_files = sc.sc_dinodes + sc.sc_free;
912 buf->f_ffree = sc.sc_free;
913 buf->f_namelen = GFS2_FNAMESIZE;
919 * gfs2_drop_inode - Drop an inode (test for remote unlink)
920 * @inode: The inode to drop
922 * If we've received a callback on an iopen lock then it's because a
923 * remote node tried to deallocate the inode but failed due to this node
924 * still having the inode open. Here we mark the link count zero
925 * since we know that it must have reached zero if the GLF_DEMOTE flag
926 * is set on the iopen glock. If we didn't do a disk read since the
927 * remote node removed the final link then we might otherwise miss
928 * this event. This check ensures that this node will deallocate the
929 * inode's blocks, or alternatively pass the baton on to another
930 * node for later deallocation.
933 static int gfs2_drop_inode(struct inode *inode)
935 struct gfs2_inode *ip = GFS2_I(inode);
937 if (inode->i_nlink &&
938 gfs2_holder_initialized(&ip->i_iopen_gh)) {
939 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
940 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
945 * When under memory pressure when an inode's link count has dropped to
946 * zero, defer deleting the inode to the delete workqueue. This avoids
947 * calling into DLM under memory pressure, which can deadlock.
949 if (!inode->i_nlink &&
950 unlikely(current->flags & PF_MEMALLOC) &&
951 gfs2_holder_initialized(&ip->i_iopen_gh)) {
952 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
955 if (!gfs2_queue_delete_work(gl, 0))
956 gfs2_glock_queue_put(gl);
960 return generic_drop_inode(inode);
963 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
969 } while (!IS_ROOT(d1));
974 * gfs2_show_options - Show mount options for /proc/mounts
975 * @s: seq_file structure
976 * @root: root of this (sub)tree
978 * Returns: 0 on success or error code
981 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
983 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
984 struct gfs2_args *args = &sdp->sd_args;
987 if (is_ancestor(root, sdp->sd_master_dir))
988 seq_puts(s, ",meta");
989 if (args->ar_lockproto[0])
990 seq_show_option(s, "lockproto", args->ar_lockproto);
991 if (args->ar_locktable[0])
992 seq_show_option(s, "locktable", args->ar_locktable);
993 if (args->ar_hostdata[0])
994 seq_show_option(s, "hostdata", args->ar_hostdata);
995 if (args->ar_spectator)
996 seq_puts(s, ",spectator");
997 if (args->ar_localflocks)
998 seq_puts(s, ",localflocks");
1000 seq_puts(s, ",debug");
1001 if (args->ar_posix_acl)
1002 seq_puts(s, ",acl");
1003 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1005 switch (args->ar_quota) {
1006 case GFS2_QUOTA_OFF:
1009 case GFS2_QUOTA_ACCOUNT:
1019 seq_printf(s, ",quota=%s", state);
1021 if (args->ar_suiddir)
1022 seq_puts(s, ",suiddir");
1023 if (args->ar_data != GFS2_DATA_DEFAULT) {
1025 switch (args->ar_data) {
1026 case GFS2_DATA_WRITEBACK:
1027 state = "writeback";
1029 case GFS2_DATA_ORDERED:
1036 seq_printf(s, ",data=%s", state);
1038 if (args->ar_discard)
1039 seq_puts(s, ",discard");
1040 val = sdp->sd_tune.gt_logd_secs;
1042 seq_printf(s, ",commit=%d", val);
1043 val = sdp->sd_tune.gt_statfs_quantum;
1045 seq_printf(s, ",statfs_quantum=%d", val);
1046 else if (sdp->sd_tune.gt_statfs_slow)
1047 seq_puts(s, ",statfs_quantum=0");
1048 val = sdp->sd_tune.gt_quota_quantum;
1050 seq_printf(s, ",quota_quantum=%d", val);
1051 if (args->ar_statfs_percent)
1052 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1053 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1056 switch (args->ar_errors) {
1057 case GFS2_ERRORS_WITHDRAW:
1060 case GFS2_ERRORS_PANIC:
1067 seq_printf(s, ",errors=%s", state);
1069 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1070 seq_puts(s, ",nobarrier");
1071 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1072 seq_puts(s, ",demote_interface_used");
1073 if (args->ar_rgrplvb)
1074 seq_puts(s, ",rgrplvb");
1075 if (args->ar_loccookie)
1076 seq_puts(s, ",loccookie");
1080 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1082 struct inode *inode = &ip->i_inode;
1083 struct gfs2_glock *gl = ip->i_gl;
1085 if (unlikely(!gl)) {
1086 /* This can only happen during incomplete inode creation. */
1087 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1091 truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1092 truncate_inode_pages(&inode->i_data, 0);
1094 if (atomic_read(&gl->gl_revokes) == 0) {
1095 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1096 clear_bit(GLF_DIRTY, &gl->gl_flags);
1100 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1102 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1103 struct gfs2_rgrpd *rgd;
1104 struct gfs2_holder gh;
1107 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1108 gfs2_consist_inode(ip);
1112 error = gfs2_rindex_update(sdp);
1116 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1120 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1122 gfs2_consist_inode(ip);
1127 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1128 LM_FLAG_NODE_SCOPE, &gh);
1132 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1133 sdp->sd_jdesc->jd_blocks);
1135 goto out_rg_gunlock;
1137 gfs2_free_di(rgd, ip);
1139 gfs2_final_release_pages(ip);
1141 gfs2_trans_end(sdp);
1144 gfs2_glock_dq_uninit(&gh);
1146 gfs2_quota_unhold(ip);
1151 * gfs2_glock_put_eventually
1152 * @gl: The glock to put
1154 * When under memory pressure, trigger a deferred glock put to make sure we
1155 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1158 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1160 if (current->flags & PF_MEMALLOC)
1161 gfs2_glock_queue_put(gl);
1166 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1168 struct gfs2_inode *ip = GFS2_I(inode);
1169 struct gfs2_sbd *sdp = GFS2_SB(inode);
1170 struct gfs2_holder *gh = &ip->i_iopen_gh;
1171 long timeout = 5 * HZ;
1174 gh->gh_flags |= GL_NOCACHE;
1175 gfs2_glock_dq_wait(gh);
1178 * If there are no other lock holders, we'll get the lock immediately.
1179 * Otherwise, the other nodes holding the lock will be notified about
1180 * our locking request. If they don't have the inode open, they'll
1181 * evict the cached inode and release the lock. Otherwise, if they
1182 * poke the inode glock, we'll take this as an indication that they
1183 * still need the iopen glock and that they'll take care of deleting
1184 * the inode when they're done. As a last resort, if another node
1185 * keeps holding the iopen glock without showing any activity on the
1186 * inode glock, we'll eventually time out.
1188 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1189 * locking request as an optimization to notify lock holders as soon as
1190 * possible. Without that flag, they'd be notified implicitly by the
1191 * second locking request.
1194 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1195 error = gfs2_glock_nq(gh);
1196 if (error != GLR_TRYFAILED)
1199 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1200 error = gfs2_glock_nq(gh);
1204 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1205 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1206 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1208 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1212 return gfs2_glock_holder_ready(gh) == 0;
1216 * evict_should_delete - determine whether the inode is eligible for deletion
1217 * @inode: The inode to evict
1218 * @gh: The glock holder structure
1220 * This function determines whether the evicted inode is eligible to be deleted
1221 * and locks the inode glock.
1223 * Returns: the fate of the dinode
1225 static enum dinode_demise evict_should_delete(struct inode *inode,
1226 struct gfs2_holder *gh)
1228 struct gfs2_inode *ip = GFS2_I(inode);
1229 struct super_block *sb = inode->i_sb;
1230 struct gfs2_sbd *sdp = sb->s_fs_info;
1233 if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1236 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1237 return SHOULD_DEFER_EVICTION;
1239 /* Deletes should never happen under memory pressure anymore. */
1240 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1241 return SHOULD_DEFER_EVICTION;
1243 /* Must not read inode block until block type has been verified */
1244 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1245 if (unlikely(ret)) {
1246 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1247 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1248 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1249 return SHOULD_DEFER_EVICTION;
1252 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1253 return SHOULD_NOT_DELETE_DINODE;
1254 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1256 return SHOULD_NOT_DELETE_DINODE;
1258 ret = gfs2_instantiate(gh);
1260 return SHOULD_NOT_DELETE_DINODE;
1263 * The inode may have been recreated in the meantime.
1266 return SHOULD_NOT_DELETE_DINODE;
1269 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1270 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1271 if (!gfs2_upgrade_iopen_glock(inode)) {
1272 gfs2_holder_uninit(&ip->i_iopen_gh);
1273 return SHOULD_NOT_DELETE_DINODE;
1276 return SHOULD_DELETE_DINODE;
1280 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1281 * @inode: The inode to evict
1283 static int evict_unlinked_inode(struct inode *inode)
1285 struct gfs2_inode *ip = GFS2_I(inode);
1288 if (S_ISDIR(inode->i_mode) &&
1289 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1290 ret = gfs2_dir_exhash_dealloc(ip);
1296 ret = gfs2_ea_dealloc(ip);
1301 if (!gfs2_is_stuffed(ip)) {
1302 ret = gfs2_file_dealloc(ip);
1308 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1311 * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1312 * can get called to recreate it, or even gfs2_inode_lookup() if the
1313 * inode was recreated on another node in the meantime.
1315 * However, inserting the new inode into the inode hash table will not
1316 * succeed until the old inode is removed, and that only happens after
1317 * ->evict_inode() returns. The new inode is attached to its inode and
1318 * iopen glocks after inserting it into the inode hash table, so at
1319 * that point we can be sure that both glocks are unused.
1322 ret = gfs2_dinode_dealloc(ip);
1328 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1329 * @inode: The inode to evict
1331 static int evict_linked_inode(struct inode *inode)
1333 struct super_block *sb = inode->i_sb;
1334 struct gfs2_sbd *sdp = sb->s_fs_info;
1335 struct gfs2_inode *ip = GFS2_I(inode);
1336 struct address_space *metamapping;
1339 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1340 GFS2_LFC_EVICT_INODE);
1341 metamapping = gfs2_glock2aspace(ip->i_gl);
1342 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1343 filemap_fdatawrite(metamapping);
1344 filemap_fdatawait(metamapping);
1346 write_inode_now(inode, 1);
1347 gfs2_ail_flush(ip->i_gl, 0);
1349 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1353 /* Needs to be done before glock release & also in a transaction */
1354 truncate_inode_pages(&inode->i_data, 0);
1355 truncate_inode_pages(metamapping, 0);
1356 gfs2_trans_end(sdp);
1361 * gfs2_evict_inode - Remove an inode from cache
1362 * @inode: The inode to evict
1364 * There are three cases to consider:
1365 * 1. i_nlink == 0, we are final opener (and must deallocate)
1366 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1369 * If the fs is read only, then we have to treat all cases as per #3
1370 * since we are unable to do any deallocation. The inode will be
1371 * deallocated by the next read/write node to attempt an allocation
1372 * in the same resource group
1374 * We have to (at the moment) hold the inodes main lock to cover
1375 * the gap between unlocking the shared lock on the iopen lock and
1376 * taking the exclusive lock. I'd rather do a shared -> exclusive
1377 * conversion on the iopen lock, but we can change that later. This
1378 * is safe, just less efficient.
1381 static void gfs2_evict_inode(struct inode *inode)
1383 struct super_block *sb = inode->i_sb;
1384 struct gfs2_sbd *sdp = sb->s_fs_info;
1385 struct gfs2_inode *ip = GFS2_I(inode);
1386 struct gfs2_holder gh;
1389 if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1392 gfs2_holder_mark_uninitialized(&gh);
1393 ret = evict_should_delete(inode, &gh);
1394 if (ret == SHOULD_DEFER_EVICTION)
1396 if (ret == SHOULD_DELETE_DINODE)
1397 ret = evict_unlinked_inode(inode);
1399 ret = evict_linked_inode(inode);
1401 if (gfs2_rs_active(&ip->i_res))
1402 gfs2_rs_deltree(&ip->i_res);
1404 if (gfs2_holder_initialized(&gh)) {
1405 glock_clear_object(ip->i_gl, ip);
1406 gfs2_glock_dq_uninit(&gh);
1408 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1409 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1411 truncate_inode_pages_final(&inode->i_data);
1413 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1414 gfs2_rs_deltree(&ip->i_res);
1415 gfs2_ordered_del_inode(ip);
1417 gfs2_dir_hash_inval(ip);
1418 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1419 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1421 glock_clear_object(gl, ip);
1422 gfs2_glock_hold(gl);
1423 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1424 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1425 gfs2_glock_put_eventually(gl);
1428 glock_clear_object(ip->i_gl, ip);
1429 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1430 gfs2_glock_add_to_lru(ip->i_gl);
1431 gfs2_glock_put_eventually(ip->i_gl);
1436 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1438 struct gfs2_inode *ip;
1440 ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1446 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1447 memset(&ip->i_res, 0, sizeof(ip->i_res));
1448 RB_CLEAR_NODE(&ip->i_res.rs_node);
1450 return &ip->i_inode;
1453 static void gfs2_free_inode(struct inode *inode)
1455 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1458 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1460 struct local_statfs_inode *lsi, *safe;
1462 /* Run through the statfs inodes list to iput and free memory */
1463 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1464 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1465 sdp->sd_sc_inode = NULL; /* belongs to this node */
1466 if (lsi->si_sc_inode)
1467 iput(lsi->si_sc_inode);
1468 list_del(&lsi->si_list);
1473 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1476 struct local_statfs_inode *lsi;
1478 /* Return the local (per node) statfs inode in the
1479 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1480 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1481 if (lsi->si_jid == index)
1482 return lsi->si_sc_inode;
1487 const struct super_operations gfs2_super_ops = {
1488 .alloc_inode = gfs2_alloc_inode,
1489 .free_inode = gfs2_free_inode,
1490 .write_inode = gfs2_write_inode,
1491 .dirty_inode = gfs2_dirty_inode,
1492 .evict_inode = gfs2_evict_inode,
1493 .put_super = gfs2_put_super,
1494 .sync_fs = gfs2_sync_fs,
1495 .freeze_super = gfs2_freeze,
1496 .thaw_super = gfs2_unfreeze,
1497 .statfs = gfs2_statfs,
1498 .drop_inode = gfs2_drop_inode,
1499 .show_options = gfs2_show_options,