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);
122 static int init_threads(struct gfs2_sbd *sdp)
124 struct task_struct *p;
127 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
130 fs_err(sdp, "can't start logd thread: %d\n", error);
133 sdp->sd_logd_process = p;
135 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
138 fs_err(sdp, "can't start quotad thread: %d\n", error);
141 sdp->sd_quotad_process = p;
145 kthread_stop(sdp->sd_logd_process);
146 sdp->sd_logd_process = NULL;
151 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
152 * @sdp: the filesystem
157 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
159 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
160 struct gfs2_glock *j_gl = ip->i_gl;
161 struct gfs2_log_header_host head;
164 error = init_threads(sdp);
168 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
169 if (gfs2_withdrawn(sdp)) {
174 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
175 if (error || gfs2_withdrawn(sdp))
178 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
184 /* Initialize some head of the log stuff */
185 sdp->sd_log_sequence = head.lh_sequence + 1;
186 gfs2_log_pointers_init(sdp, head.lh_blkno);
188 error = gfs2_quota_init(sdp);
189 if (error || gfs2_withdrawn(sdp))
192 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
197 if (sdp->sd_quotad_process)
198 kthread_stop(sdp->sd_quotad_process);
199 sdp->sd_quotad_process = NULL;
200 if (sdp->sd_logd_process)
201 kthread_stop(sdp->sd_logd_process);
202 sdp->sd_logd_process = NULL;
206 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
208 const struct gfs2_statfs_change *str = buf;
210 sc->sc_total = be64_to_cpu(str->sc_total);
211 sc->sc_free = be64_to_cpu(str->sc_free);
212 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
215 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
217 struct gfs2_statfs_change *str = buf;
219 str->sc_total = cpu_to_be64(sc->sc_total);
220 str->sc_free = cpu_to_be64(sc->sc_free);
221 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
224 int gfs2_statfs_init(struct gfs2_sbd *sdp)
226 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
227 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
228 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
229 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
230 struct buffer_head *m_bh, *l_bh;
231 struct gfs2_holder gh;
234 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
239 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
243 if (sdp->sd_args.ar_spectator) {
244 spin_lock(&sdp->sd_statfs_spin);
245 gfs2_statfs_change_in(m_sc, m_bh->b_data +
246 sizeof(struct gfs2_dinode));
247 spin_unlock(&sdp->sd_statfs_spin);
249 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
253 spin_lock(&sdp->sd_statfs_spin);
254 gfs2_statfs_change_in(m_sc, m_bh->b_data +
255 sizeof(struct gfs2_dinode));
256 gfs2_statfs_change_in(l_sc, l_bh->b_data +
257 sizeof(struct gfs2_dinode));
258 spin_unlock(&sdp->sd_statfs_spin);
266 gfs2_glock_dq_uninit(&gh);
270 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
273 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
274 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
275 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
276 struct buffer_head *l_bh;
281 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
285 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
287 spin_lock(&sdp->sd_statfs_spin);
288 l_sc->sc_total += total;
289 l_sc->sc_free += free;
290 l_sc->sc_dinodes += dinodes;
291 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
292 if (sdp->sd_args.ar_statfs_percent) {
293 x = 100 * l_sc->sc_free;
294 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
295 if (x >= y || x <= -y)
298 spin_unlock(&sdp->sd_statfs_spin);
302 gfs2_wake_up_statfs(sdp);
305 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
306 struct buffer_head *l_bh)
308 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
309 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
310 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
311 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
313 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
314 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
316 spin_lock(&sdp->sd_statfs_spin);
317 m_sc->sc_total += l_sc->sc_total;
318 m_sc->sc_free += l_sc->sc_free;
319 m_sc->sc_dinodes += l_sc->sc_dinodes;
320 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
321 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
322 0, sizeof(struct gfs2_statfs_change));
323 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
324 spin_unlock(&sdp->sd_statfs_spin);
327 int gfs2_statfs_sync(struct super_block *sb, int type)
329 struct gfs2_sbd *sdp = sb->s_fs_info;
330 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
331 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
332 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
333 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
334 struct gfs2_holder gh;
335 struct buffer_head *m_bh, *l_bh;
338 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
343 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
347 spin_lock(&sdp->sd_statfs_spin);
348 gfs2_statfs_change_in(m_sc, m_bh->b_data +
349 sizeof(struct gfs2_dinode));
350 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
351 spin_unlock(&sdp->sd_statfs_spin);
354 spin_unlock(&sdp->sd_statfs_spin);
356 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
360 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
364 update_statfs(sdp, m_bh, l_bh);
365 sdp->sd_statfs_force_sync = 0;
374 gfs2_glock_dq_uninit(&gh);
380 struct list_head list;
381 struct gfs2_holder gh;
385 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
387 * @sdp: the file system
388 * @state: the state to put the transaction lock into
389 * @t_gh: the hold on the transaction lock
394 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
396 struct gfs2_inode *ip;
397 struct gfs2_jdesc *jd;
400 struct gfs2_log_header_host lh;
403 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
404 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
409 ip = GFS2_I(jd->jd_inode);
410 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
415 list_add(&lfcc->list, &list);
418 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
419 LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
423 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
424 error = gfs2_jdesc_check(jd);
427 error = gfs2_find_jhead(jd, &lh, false);
430 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
437 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
440 while (!list_empty(&list)) {
441 lfcc = list_first_entry(&list, struct lfcc, list);
442 list_del(&lfcc->list);
443 gfs2_glock_dq_uninit(&lfcc->gh);
449 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
451 struct gfs2_dinode *str = buf;
453 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
454 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
455 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
456 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
457 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
458 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
459 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
460 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
461 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
462 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
463 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
464 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
465 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
466 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
468 str->di_goal_meta = cpu_to_be64(ip->i_goal);
469 str->di_goal_data = cpu_to_be64(ip->i_goal);
470 str->di_generation = cpu_to_be64(ip->i_generation);
472 str->di_flags = cpu_to_be32(ip->i_diskflags);
473 str->di_height = cpu_to_be16(ip->i_height);
474 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
475 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
477 str->di_depth = cpu_to_be16(ip->i_depth);
478 str->di_entries = cpu_to_be32(ip->i_entries);
480 str->di_eattr = cpu_to_be64(ip->i_eattr);
481 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
482 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
483 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
487 * gfs2_write_inode - Make sure the inode is stable on the disk
489 * @wbc: The writeback control structure
494 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
496 struct gfs2_inode *ip = GFS2_I(inode);
497 struct gfs2_sbd *sdp = GFS2_SB(inode);
498 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
499 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
501 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
504 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
505 GFS2_LOG_HEAD_FLUSH_NORMAL |
506 GFS2_LFC_WRITE_INODE);
507 if (bdi->wb.dirty_exceeded)
508 gfs2_ail1_flush(sdp, wbc);
510 filemap_fdatawrite(metamapping);
512 ret = filemap_fdatawait(metamapping);
514 mark_inode_dirty_sync(inode);
516 spin_lock(&inode->i_lock);
517 if (!(inode->i_flags & I_DIRTY))
518 gfs2_ordered_del_inode(ip);
519 spin_unlock(&inode->i_lock);
525 * gfs2_dirty_inode - check for atime updates
526 * @inode: The inode in question
527 * @flags: The type of dirty
529 * Unfortunately it can be called under any combination of inode
530 * glock and transaction lock, so we have to check carefully.
532 * At the moment this deals only with atime - it should be possible
533 * to expand that role in future, once a review of the locking has
537 static void gfs2_dirty_inode(struct inode *inode, int flags)
539 struct gfs2_inode *ip = GFS2_I(inode);
540 struct gfs2_sbd *sdp = GFS2_SB(inode);
541 struct buffer_head *bh;
542 struct gfs2_holder gh;
544 int need_endtrans = 0;
547 if (unlikely(gfs2_withdrawn(sdp)))
549 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
550 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
552 fs_err(sdp, "dirty_inode: glock %d\n", ret);
553 gfs2_dump_glock(NULL, ip->i_gl, true);
557 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
560 if (current->journal_info == NULL) {
561 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
563 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
569 ret = gfs2_meta_inode_buffer(ip, &bh);
571 gfs2_trans_add_meta(ip->i_gl, bh);
572 gfs2_dinode_out(ip, bh->b_data);
580 gfs2_glock_dq_uninit(&gh);
584 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
585 * @sdp: the filesystem
590 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
593 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
595 gfs2_flush_delete_work(sdp);
596 if (!log_write_allowed && current == sdp->sd_quotad_process)
597 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
598 else if (sdp->sd_quotad_process)
599 kthread_stop(sdp->sd_quotad_process);
600 sdp->sd_quotad_process = NULL;
602 if (!log_write_allowed && current == sdp->sd_logd_process)
603 fs_warn(sdp, "The logd daemon is withdrawing.\n");
604 else if (sdp->sd_logd_process)
605 kthread_stop(sdp->sd_logd_process);
606 sdp->sd_logd_process = NULL;
608 if (log_write_allowed) {
609 gfs2_quota_sync(sdp->sd_vfs, 0);
610 gfs2_statfs_sync(sdp->sd_vfs, 0);
612 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
613 GFS2_LFC_MAKE_FS_RO);
614 wait_event_timeout(sdp->sd_log_waitq,
615 gfs2_log_is_empty(sdp),
617 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
619 wait_event_timeout(sdp->sd_log_waitq,
620 gfs2_log_is_empty(sdp),
623 gfs2_quota_cleanup(sdp);
625 if (!log_write_allowed)
626 sdp->sd_vfs->s_flags |= SB_RDONLY;
632 * gfs2_put_super - Unmount the filesystem
633 * @sb: The VFS superblock
637 static void gfs2_put_super(struct super_block *sb)
639 struct gfs2_sbd *sdp = sb->s_fs_info;
641 struct gfs2_jdesc *jd;
643 /* No more recovery requests */
644 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
647 /* Wait on outstanding recovery */
649 spin_lock(&sdp->sd_jindex_spin);
650 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
651 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
653 spin_unlock(&sdp->sd_jindex_spin);
654 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
655 TASK_UNINTERRUPTIBLE);
658 spin_unlock(&sdp->sd_jindex_spin);
660 if (!sb_rdonly(sb)) {
661 error = gfs2_make_fs_ro(sdp);
665 WARN_ON(gfs2_withdrawing(sdp));
667 /* At this point, we're through modifying the disk */
671 iput(sdp->sd_jindex);
672 iput(sdp->sd_statfs_inode);
673 iput(sdp->sd_rindex);
674 iput(sdp->sd_quota_inode);
676 gfs2_glock_put(sdp->sd_rename_gl);
677 gfs2_glock_put(sdp->sd_freeze_gl);
679 if (!sdp->sd_args.ar_spectator) {
680 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
681 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
682 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
683 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
684 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
685 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
686 free_local_statfs_inodes(sdp);
687 iput(sdp->sd_qc_inode);
690 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
691 gfs2_clear_rgrpd(sdp);
692 gfs2_jindex_free(sdp);
693 /* Take apart glock structures and buffer lists */
694 gfs2_gl_hash_clear(sdp);
695 truncate_inode_pages_final(&sdp->sd_aspace);
696 gfs2_delete_debugfs_file(sdp);
697 /* Unmount the locking protocol */
698 gfs2_lm_unmount(sdp);
700 /* At this point, we're through participating in the lockspace */
701 gfs2_sys_fs_del(sdp);
706 * gfs2_sync_fs - sync the filesystem
707 * @sb: the superblock
709 * Flushes the log to disk.
712 static int gfs2_sync_fs(struct super_block *sb, int wait)
714 struct gfs2_sbd *sdp = sb->s_fs_info;
716 gfs2_quota_sync(sb, -1);
718 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
720 return sdp->sd_log_error;
723 void gfs2_freeze_func(struct work_struct *work)
726 struct gfs2_holder freeze_gh;
727 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
728 struct super_block *sb = sdp->sd_vfs;
730 atomic_inc(&sb->s_active);
731 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
733 gfs2_assert_withdraw(sdp, 0);
735 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
736 error = thaw_super(sb);
738 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
740 gfs2_assert_withdraw(sdp, 0);
742 gfs2_freeze_unlock(&freeze_gh);
744 deactivate_super(sb);
745 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
746 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
751 * gfs2_freeze - prevent further writes to the filesystem
752 * @sb: the VFS structure for the filesystem
756 static int gfs2_freeze(struct super_block *sb)
758 struct gfs2_sbd *sdp = sb->s_fs_info;
761 mutex_lock(&sdp->sd_freeze_mutex);
762 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
766 if (gfs2_withdrawn(sdp)) {
771 error = gfs2_lock_fs_check_clean(sdp);
776 fs_err(sdp, "waiting for recovery before freeze\n");
777 else if (error == -EIO) {
778 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
779 "to recovery error.\n");
782 fs_err(sdp, "error freezing FS: %d\n", error);
784 fs_err(sdp, "retrying...\n");
787 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
789 mutex_unlock(&sdp->sd_freeze_mutex);
794 * gfs2_unfreeze - reallow writes to the filesystem
795 * @sb: the VFS structure for the filesystem
799 static int gfs2_unfreeze(struct super_block *sb)
801 struct gfs2_sbd *sdp = sb->s_fs_info;
803 mutex_lock(&sdp->sd_freeze_mutex);
804 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
805 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
806 mutex_unlock(&sdp->sd_freeze_mutex);
810 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
811 mutex_unlock(&sdp->sd_freeze_mutex);
812 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
816 * statfs_fill - fill in the sg for a given RG
818 * @sc: the sc structure
820 * Returns: 0 on success, -ESTALE if the LVB is invalid
823 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
824 struct gfs2_statfs_change_host *sc)
826 gfs2_rgrp_verify(rgd);
827 sc->sc_total += rgd->rd_data;
828 sc->sc_free += rgd->rd_free;
829 sc->sc_dinodes += rgd->rd_dinodes;
834 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
835 * @sdp: the filesystem
836 * @sc: the sc info that will be returned
838 * Any error (other than a signal) will cause this routine to fall back
839 * to the synchronous version.
841 * FIXME: This really shouldn't busy wait like this.
846 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
848 struct gfs2_rgrpd *rgd_next;
849 struct gfs2_holder *gha, *gh;
850 unsigned int slots = 64;
855 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
856 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
859 for (x = 0; x < slots; x++)
860 gfs2_holder_mark_uninitialized(gha + x);
862 rgd_next = gfs2_rgrpd_get_first(sdp);
867 for (x = 0; x < slots; x++) {
870 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
871 err = gfs2_glock_wait(gh);
873 gfs2_holder_uninit(gh);
877 struct gfs2_rgrpd *rgd =
878 gfs2_glock2rgrp(gh->gh_gl);
880 error = statfs_slow_fill(rgd, sc);
882 gfs2_glock_dq_uninit(gh);
886 if (gfs2_holder_initialized(gh))
888 else if (rgd_next && !error) {
889 error = gfs2_glock_nq_init(rgd_next->rd_gl,
893 rgd_next = gfs2_rgrpd_get_next(rgd_next);
897 if (signal_pending(current))
898 error = -ERESTARTSYS;
912 * gfs2_statfs_i - Do a statfs
913 * @sdp: the filesystem
914 * @sg: the sg structure
919 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
921 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
922 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
924 spin_lock(&sdp->sd_statfs_spin);
927 sc->sc_total += l_sc->sc_total;
928 sc->sc_free += l_sc->sc_free;
929 sc->sc_dinodes += l_sc->sc_dinodes;
931 spin_unlock(&sdp->sd_statfs_spin);
935 if (sc->sc_free > sc->sc_total)
936 sc->sc_free = sc->sc_total;
937 if (sc->sc_dinodes < 0)
944 * gfs2_statfs - Gather and return stats about the filesystem
945 * @sb: The superblock
946 * @statfsbuf: The buffer
948 * Returns: 0 on success or error code
951 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
953 struct super_block *sb = dentry->d_sb;
954 struct gfs2_sbd *sdp = sb->s_fs_info;
955 struct gfs2_statfs_change_host sc;
958 error = gfs2_rindex_update(sdp);
962 if (gfs2_tune_get(sdp, gt_statfs_slow))
963 error = gfs2_statfs_slow(sdp, &sc);
965 error = gfs2_statfs_i(sdp, &sc);
970 buf->f_type = GFS2_MAGIC;
971 buf->f_bsize = sdp->sd_sb.sb_bsize;
972 buf->f_blocks = sc.sc_total;
973 buf->f_bfree = sc.sc_free;
974 buf->f_bavail = sc.sc_free;
975 buf->f_files = sc.sc_dinodes + sc.sc_free;
976 buf->f_ffree = sc.sc_free;
977 buf->f_namelen = GFS2_FNAMESIZE;
983 * gfs2_drop_inode - Drop an inode (test for remote unlink)
984 * @inode: The inode to drop
986 * If we've received a callback on an iopen lock then it's because a
987 * remote node tried to deallocate the inode but failed due to this node
988 * still having the inode open. Here we mark the link count zero
989 * since we know that it must have reached zero if the GLF_DEMOTE flag
990 * is set on the iopen glock. If we didn't do a disk read since the
991 * remote node removed the final link then we might otherwise miss
992 * this event. This check ensures that this node will deallocate the
993 * inode's blocks, or alternatively pass the baton on to another
994 * node for later deallocation.
997 static int gfs2_drop_inode(struct inode *inode)
999 struct gfs2_inode *ip = GFS2_I(inode);
1001 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1003 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1004 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1005 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1010 * When under memory pressure when an inode's link count has dropped to
1011 * zero, defer deleting the inode to the delete workqueue. This avoids
1012 * calling into DLM under memory pressure, which can deadlock.
1014 if (!inode->i_nlink &&
1015 unlikely(current->flags & PF_MEMALLOC) &&
1016 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1017 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1019 gfs2_glock_hold(gl);
1020 if (!gfs2_queue_delete_work(gl, 0))
1021 gfs2_glock_queue_put(gl);
1025 return generic_drop_inode(inode);
1028 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1034 } while (!IS_ROOT(d1));
1039 * gfs2_show_options - Show mount options for /proc/mounts
1040 * @s: seq_file structure
1041 * @root: root of this (sub)tree
1043 * Returns: 0 on success or error code
1046 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1048 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1049 struct gfs2_args *args = &sdp->sd_args;
1052 if (is_ancestor(root, sdp->sd_master_dir))
1053 seq_puts(s, ",meta");
1054 if (args->ar_lockproto[0])
1055 seq_show_option(s, "lockproto", args->ar_lockproto);
1056 if (args->ar_locktable[0])
1057 seq_show_option(s, "locktable", args->ar_locktable);
1058 if (args->ar_hostdata[0])
1059 seq_show_option(s, "hostdata", args->ar_hostdata);
1060 if (args->ar_spectator)
1061 seq_puts(s, ",spectator");
1062 if (args->ar_localflocks)
1063 seq_puts(s, ",localflocks");
1065 seq_puts(s, ",debug");
1066 if (args->ar_posix_acl)
1067 seq_puts(s, ",acl");
1068 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1070 switch (args->ar_quota) {
1071 case GFS2_QUOTA_OFF:
1074 case GFS2_QUOTA_ACCOUNT:
1084 seq_printf(s, ",quota=%s", state);
1086 if (args->ar_suiddir)
1087 seq_puts(s, ",suiddir");
1088 if (args->ar_data != GFS2_DATA_DEFAULT) {
1090 switch (args->ar_data) {
1091 case GFS2_DATA_WRITEBACK:
1092 state = "writeback";
1094 case GFS2_DATA_ORDERED:
1101 seq_printf(s, ",data=%s", state);
1103 if (args->ar_discard)
1104 seq_puts(s, ",discard");
1105 val = sdp->sd_tune.gt_logd_secs;
1107 seq_printf(s, ",commit=%d", val);
1108 val = sdp->sd_tune.gt_statfs_quantum;
1110 seq_printf(s, ",statfs_quantum=%d", val);
1111 else if (sdp->sd_tune.gt_statfs_slow)
1112 seq_puts(s, ",statfs_quantum=0");
1113 val = sdp->sd_tune.gt_quota_quantum;
1115 seq_printf(s, ",quota_quantum=%d", val);
1116 if (args->ar_statfs_percent)
1117 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1118 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1121 switch (args->ar_errors) {
1122 case GFS2_ERRORS_WITHDRAW:
1125 case GFS2_ERRORS_PANIC:
1132 seq_printf(s, ",errors=%s", state);
1134 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1135 seq_puts(s, ",nobarrier");
1136 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1137 seq_puts(s, ",demote_interface_used");
1138 if (args->ar_rgrplvb)
1139 seq_puts(s, ",rgrplvb");
1140 if (args->ar_loccookie)
1141 seq_puts(s, ",loccookie");
1145 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1147 struct inode *inode = &ip->i_inode;
1148 struct gfs2_glock *gl = ip->i_gl;
1150 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1151 truncate_inode_pages(&inode->i_data, 0);
1153 if (atomic_read(&gl->gl_revokes) == 0) {
1154 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1155 clear_bit(GLF_DIRTY, &gl->gl_flags);
1159 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1161 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1162 struct gfs2_rgrpd *rgd;
1163 struct gfs2_holder gh;
1166 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1167 gfs2_consist_inode(ip);
1171 error = gfs2_rindex_update(sdp);
1175 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1179 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1181 gfs2_consist_inode(ip);
1186 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1187 LM_FLAG_NODE_SCOPE, &gh);
1191 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1192 sdp->sd_jdesc->jd_blocks);
1194 goto out_rg_gunlock;
1196 gfs2_free_di(rgd, ip);
1198 gfs2_final_release_pages(ip);
1200 gfs2_trans_end(sdp);
1203 gfs2_glock_dq_uninit(&gh);
1205 gfs2_quota_unhold(ip);
1210 * gfs2_glock_put_eventually
1211 * @gl: The glock to put
1213 * When under memory pressure, trigger a deferred glock put to make sure we
1214 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1217 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1219 if (current->flags & PF_MEMALLOC)
1220 gfs2_glock_queue_put(gl);
1225 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1227 struct gfs2_inode *ip = GFS2_I(inode);
1228 struct gfs2_sbd *sdp = GFS2_SB(inode);
1229 struct gfs2_holder *gh = &ip->i_iopen_gh;
1230 long timeout = 5 * HZ;
1233 gh->gh_flags |= GL_NOCACHE;
1234 gfs2_glock_dq_wait(gh);
1237 * If there are no other lock holders, we'll get the lock immediately.
1238 * Otherwise, the other nodes holding the lock will be notified about
1239 * our locking request. If they don't have the inode open, they'll
1240 * evict the cached inode and release the lock. Otherwise, if they
1241 * poke the inode glock, we'll take this as an indication that they
1242 * still need the iopen glock and that they'll take care of deleting
1243 * the inode when they're done. As a last resort, if another node
1244 * keeps holding the iopen glock without showing any activity on the
1245 * inode glock, we'll eventually time out.
1247 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1248 * locking request as an optimization to notify lock holders as soon as
1249 * possible. Without that flag, they'd be notified implicitly by the
1250 * second locking request.
1253 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1254 error = gfs2_glock_nq(gh);
1255 if (error != GLR_TRYFAILED)
1258 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1259 error = gfs2_glock_nq(gh);
1263 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1264 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1265 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1267 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1275 * evict_should_delete - determine whether the inode is eligible for deletion
1276 * @inode: The inode to evict
1278 * This function determines whether the evicted inode is eligible to be deleted
1279 * and locks the inode glock.
1281 * Returns: the fate of the dinode
1283 static enum dinode_demise evict_should_delete(struct inode *inode,
1284 struct gfs2_holder *gh)
1286 struct gfs2_inode *ip = GFS2_I(inode);
1287 struct super_block *sb = inode->i_sb;
1288 struct gfs2_sbd *sdp = sb->s_fs_info;
1291 if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1292 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1296 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1297 return SHOULD_DEFER_EVICTION;
1299 /* Deletes should never happen under memory pressure anymore. */
1300 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1301 return SHOULD_DEFER_EVICTION;
1303 /* Must not read inode block until block type has been verified */
1304 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1305 if (unlikely(ret)) {
1306 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1307 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1308 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1309 return SHOULD_DEFER_EVICTION;
1312 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1313 return SHOULD_NOT_DELETE_DINODE;
1314 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1316 return SHOULD_NOT_DELETE_DINODE;
1318 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1319 ret = gfs2_inode_refresh(ip);
1321 return SHOULD_NOT_DELETE_DINODE;
1325 * The inode may have been recreated in the meantime.
1328 return SHOULD_NOT_DELETE_DINODE;
1331 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1332 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1333 if (!gfs2_upgrade_iopen_glock(inode)) {
1334 gfs2_holder_uninit(&ip->i_iopen_gh);
1335 return SHOULD_NOT_DELETE_DINODE;
1338 return SHOULD_DELETE_DINODE;
1342 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1343 * @inode: The inode to evict
1345 static int evict_unlinked_inode(struct inode *inode)
1347 struct gfs2_inode *ip = GFS2_I(inode);
1350 if (S_ISDIR(inode->i_mode) &&
1351 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1352 ret = gfs2_dir_exhash_dealloc(ip);
1358 ret = gfs2_ea_dealloc(ip);
1363 if (!gfs2_is_stuffed(ip)) {
1364 ret = gfs2_file_dealloc(ip);
1369 /* We're about to clear the bitmap for the dinode, but as soon as we
1370 do, gfs2_create_inode can create another inode at the same block
1371 location and try to set gl_object again. We clear gl_object here so
1372 that subsequent inode creates don't see an old gl_object. */
1373 glock_clear_object(ip->i_gl, ip);
1374 ret = gfs2_dinode_dealloc(ip);
1375 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1381 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1382 * @inode: The inode to evict
1384 static int evict_linked_inode(struct inode *inode)
1386 struct super_block *sb = inode->i_sb;
1387 struct gfs2_sbd *sdp = sb->s_fs_info;
1388 struct gfs2_inode *ip = GFS2_I(inode);
1389 struct address_space *metamapping;
1392 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1393 GFS2_LFC_EVICT_INODE);
1394 metamapping = gfs2_glock2aspace(ip->i_gl);
1395 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1396 filemap_fdatawrite(metamapping);
1397 filemap_fdatawait(metamapping);
1399 write_inode_now(inode, 1);
1400 gfs2_ail_flush(ip->i_gl, 0);
1402 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1406 /* Needs to be done before glock release & also in a transaction */
1407 truncate_inode_pages(&inode->i_data, 0);
1408 truncate_inode_pages(metamapping, 0);
1409 gfs2_trans_end(sdp);
1414 * gfs2_evict_inode - Remove an inode from cache
1415 * @inode: The inode to evict
1417 * There are three cases to consider:
1418 * 1. i_nlink == 0, we are final opener (and must deallocate)
1419 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1422 * If the fs is read only, then we have to treat all cases as per #3
1423 * since we are unable to do any deallocation. The inode will be
1424 * deallocated by the next read/write node to attempt an allocation
1425 * in the same resource group
1427 * We have to (at the moment) hold the inodes main lock to cover
1428 * the gap between unlocking the shared lock on the iopen lock and
1429 * taking the exclusive lock. I'd rather do a shared -> exclusive
1430 * conversion on the iopen lock, but we can change that later. This
1431 * is safe, just less efficient.
1434 static void gfs2_evict_inode(struct inode *inode)
1436 struct super_block *sb = inode->i_sb;
1437 struct gfs2_sbd *sdp = sb->s_fs_info;
1438 struct gfs2_inode *ip = GFS2_I(inode);
1439 struct gfs2_holder gh;
1442 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1447 if (inode->i_nlink || sb_rdonly(sb))
1450 gfs2_holder_mark_uninitialized(&gh);
1451 ret = evict_should_delete(inode, &gh);
1452 if (ret == SHOULD_DEFER_EVICTION)
1454 if (ret == SHOULD_DELETE_DINODE)
1455 ret = evict_unlinked_inode(inode);
1457 ret = evict_linked_inode(inode);
1459 if (gfs2_rs_active(&ip->i_res))
1460 gfs2_rs_deltree(&ip->i_res);
1462 if (gfs2_holder_initialized(&gh)) {
1463 glock_clear_object(ip->i_gl, ip);
1464 gfs2_glock_dq_uninit(&gh);
1466 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1467 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1469 truncate_inode_pages_final(&inode->i_data);
1471 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1472 gfs2_rs_delete(ip, NULL);
1473 gfs2_ordered_del_inode(ip);
1475 gfs2_dir_hash_inval(ip);
1477 glock_clear_object(ip->i_gl, ip);
1478 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1479 gfs2_glock_add_to_lru(ip->i_gl);
1480 gfs2_glock_put_eventually(ip->i_gl);
1483 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1484 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1486 glock_clear_object(gl, ip);
1487 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1488 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1489 gfs2_glock_dq(&ip->i_iopen_gh);
1491 gfs2_glock_hold(gl);
1492 gfs2_holder_uninit(&ip->i_iopen_gh);
1493 gfs2_glock_put_eventually(gl);
1497 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1499 struct gfs2_inode *ip;
1501 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1506 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1507 memset(&ip->i_res, 0, sizeof(ip->i_res));
1508 RB_CLEAR_NODE(&ip->i_res.rs_node);
1510 return &ip->i_inode;
1513 static void gfs2_free_inode(struct inode *inode)
1515 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1518 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1520 struct local_statfs_inode *lsi, *safe;
1522 /* Run through the statfs inodes list to iput and free memory */
1523 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1524 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1525 sdp->sd_sc_inode = NULL; /* belongs to this node */
1526 if (lsi->si_sc_inode)
1527 iput(lsi->si_sc_inode);
1528 list_del(&lsi->si_list);
1533 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1536 struct local_statfs_inode *lsi;
1538 /* Return the local (per node) statfs inode in the
1539 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1540 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1541 if (lsi->si_jid == index)
1542 return lsi->si_sc_inode;
1547 const struct super_operations gfs2_super_ops = {
1548 .alloc_inode = gfs2_alloc_inode,
1549 .free_inode = gfs2_free_inode,
1550 .write_inode = gfs2_write_inode,
1551 .dirty_inode = gfs2_dirty_inode,
1552 .evict_inode = gfs2_evict_inode,
1553 .put_super = gfs2_put_super,
1554 .sync_fs = gfs2_sync_fs,
1555 .freeze_super = gfs2_freeze,
1556 .thaw_super = gfs2_unfreeze,
1557 .statfs = gfs2_statfs,
1558 .drop_inode = gfs2_drop_inode,
1559 .show_options = gfs2_show_options,