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);
146 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
151 /* Initialize some head of the log stuff */
152 sdp->sd_log_sequence = head.lh_sequence + 1;
153 gfs2_log_pointers_init(sdp, head.lh_blkno);
155 error = gfs2_quota_init(sdp);
156 if (!error && gfs2_withdrawn(sdp))
159 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
163 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
165 const struct gfs2_statfs_change *str = buf;
167 sc->sc_total = be64_to_cpu(str->sc_total);
168 sc->sc_free = be64_to_cpu(str->sc_free);
169 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
172 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
174 struct gfs2_statfs_change *str = buf;
176 str->sc_total = cpu_to_be64(sc->sc_total);
177 str->sc_free = cpu_to_be64(sc->sc_free);
178 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
181 int gfs2_statfs_init(struct gfs2_sbd *sdp)
183 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
184 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
185 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
186 struct buffer_head *m_bh;
187 struct gfs2_holder gh;
190 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
195 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
199 if (sdp->sd_args.ar_spectator) {
200 spin_lock(&sdp->sd_statfs_spin);
201 gfs2_statfs_change_in(m_sc, m_bh->b_data +
202 sizeof(struct gfs2_dinode));
203 spin_unlock(&sdp->sd_statfs_spin);
205 spin_lock(&sdp->sd_statfs_spin);
206 gfs2_statfs_change_in(m_sc, m_bh->b_data +
207 sizeof(struct gfs2_dinode));
208 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
209 sizeof(struct gfs2_dinode));
210 spin_unlock(&sdp->sd_statfs_spin);
216 gfs2_glock_dq_uninit(&gh);
220 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
223 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
224 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
225 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
229 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
231 spin_lock(&sdp->sd_statfs_spin);
232 l_sc->sc_total += total;
233 l_sc->sc_free += free;
234 l_sc->sc_dinodes += dinodes;
235 gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
236 sizeof(struct gfs2_dinode));
237 if (sdp->sd_args.ar_statfs_percent) {
238 x = 100 * l_sc->sc_free;
239 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
240 if (x >= y || x <= -y)
243 spin_unlock(&sdp->sd_statfs_spin);
246 gfs2_wake_up_statfs(sdp);
249 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
251 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
252 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
253 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
254 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
256 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
257 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
259 spin_lock(&sdp->sd_statfs_spin);
260 m_sc->sc_total += l_sc->sc_total;
261 m_sc->sc_free += l_sc->sc_free;
262 m_sc->sc_dinodes += l_sc->sc_dinodes;
263 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
264 memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
265 0, sizeof(struct gfs2_statfs_change));
266 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
267 spin_unlock(&sdp->sd_statfs_spin);
270 int gfs2_statfs_sync(struct super_block *sb, int type)
272 struct gfs2_sbd *sdp = sb->s_fs_info;
273 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
274 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
275 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
276 struct gfs2_holder gh;
277 struct buffer_head *m_bh;
280 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
285 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
289 spin_lock(&sdp->sd_statfs_spin);
290 gfs2_statfs_change_in(m_sc, m_bh->b_data +
291 sizeof(struct gfs2_dinode));
292 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
293 spin_unlock(&sdp->sd_statfs_spin);
296 spin_unlock(&sdp->sd_statfs_spin);
298 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
302 update_statfs(sdp, m_bh);
303 sdp->sd_statfs_force_sync = 0;
310 gfs2_glock_dq_uninit(&gh);
316 struct list_head list;
317 struct gfs2_holder gh;
321 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
323 * @sdp: the file system
328 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
330 struct gfs2_inode *ip;
331 struct gfs2_jdesc *jd;
334 struct gfs2_log_header_host lh;
337 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
338 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
343 ip = GFS2_I(jd->jd_inode);
344 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
349 list_add(&lfcc->list, &list);
352 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
353 LM_FLAG_NOEXP | GL_NOPID,
358 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
359 error = gfs2_jdesc_check(jd);
362 error = gfs2_find_jhead(jd, &lh, false);
365 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
372 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
375 while (!list_empty(&list)) {
376 lfcc = list_first_entry(&list, struct lfcc, list);
377 list_del(&lfcc->list);
378 gfs2_glock_dq_uninit(&lfcc->gh);
384 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
386 const struct inode *inode = &ip->i_inode;
387 struct gfs2_dinode *str = buf;
389 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
390 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
391 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
392 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
393 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
394 str->di_mode = cpu_to_be32(inode->i_mode);
395 str->di_uid = cpu_to_be32(i_uid_read(inode));
396 str->di_gid = cpu_to_be32(i_gid_read(inode));
397 str->di_nlink = cpu_to_be32(inode->i_nlink);
398 str->di_size = cpu_to_be64(i_size_read(inode));
399 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
400 str->di_atime = cpu_to_be64(inode->i_atime.tv_sec);
401 str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec);
402 str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec);
404 str->di_goal_meta = cpu_to_be64(ip->i_goal);
405 str->di_goal_data = cpu_to_be64(ip->i_goal);
406 str->di_generation = cpu_to_be64(ip->i_generation);
408 str->di_flags = cpu_to_be32(ip->i_diskflags);
409 str->di_height = cpu_to_be16(ip->i_height);
410 str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
411 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
413 str->di_depth = cpu_to_be16(ip->i_depth);
414 str->di_entries = cpu_to_be32(ip->i_entries);
416 str->di_eattr = cpu_to_be64(ip->i_eattr);
417 str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec);
418 str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec);
419 str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec);
423 * gfs2_write_inode - Make sure the inode is stable on the disk
425 * @wbc: The writeback control structure
430 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
432 struct gfs2_inode *ip = GFS2_I(inode);
433 struct gfs2_sbd *sdp = GFS2_SB(inode);
434 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
435 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
437 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
440 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
441 GFS2_LOG_HEAD_FLUSH_NORMAL |
442 GFS2_LFC_WRITE_INODE);
443 if (bdi->wb.dirty_exceeded)
444 gfs2_ail1_flush(sdp, wbc);
446 filemap_fdatawrite(metamapping);
448 ret = filemap_fdatawait(metamapping);
450 mark_inode_dirty_sync(inode);
452 spin_lock(&inode->i_lock);
453 if (!(inode->i_flags & I_DIRTY))
454 gfs2_ordered_del_inode(ip);
455 spin_unlock(&inode->i_lock);
461 * gfs2_dirty_inode - check for atime updates
462 * @inode: The inode in question
463 * @flags: The type of dirty
465 * Unfortunately it can be called under any combination of inode
466 * glock and transaction lock, so we have to check carefully.
468 * At the moment this deals only with atime - it should be possible
469 * to expand that role in future, once a review of the locking has
473 static void gfs2_dirty_inode(struct inode *inode, int flags)
475 struct gfs2_inode *ip = GFS2_I(inode);
476 struct gfs2_sbd *sdp = GFS2_SB(inode);
477 struct buffer_head *bh;
478 struct gfs2_holder gh;
480 int need_endtrans = 0;
483 if (unlikely(!ip->i_gl)) {
484 /* This can only happen during incomplete inode creation. */
485 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
489 if (unlikely(gfs2_withdrawn(sdp)))
491 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
492 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
494 fs_err(sdp, "dirty_inode: glock %d\n", ret);
495 gfs2_dump_glock(NULL, ip->i_gl, true);
499 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
502 if (current->journal_info == NULL) {
503 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
505 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
511 ret = gfs2_meta_inode_buffer(ip, &bh);
513 gfs2_trans_add_meta(ip->i_gl, bh);
514 gfs2_dinode_out(ip, bh->b_data);
522 gfs2_glock_dq_uninit(&gh);
526 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
527 * @sdp: the filesystem
532 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
534 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
536 if (!test_bit(SDF_DEACTIVATING, &sdp->sd_flags))
537 gfs2_flush_delete_work(sdp);
539 if (!log_write_allowed && current == sdp->sd_quotad_process)
540 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
541 else if (sdp->sd_quotad_process)
542 kthread_stop(sdp->sd_quotad_process);
543 sdp->sd_quotad_process = NULL;
545 if (!log_write_allowed && current == sdp->sd_logd_process)
546 fs_warn(sdp, "The logd daemon is withdrawing.\n");
547 else if (sdp->sd_logd_process)
548 kthread_stop(sdp->sd_logd_process);
549 sdp->sd_logd_process = NULL;
551 if (log_write_allowed) {
552 gfs2_quota_sync(sdp->sd_vfs, 0);
553 gfs2_statfs_sync(sdp->sd_vfs, 0);
555 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
556 GFS2_LFC_MAKE_FS_RO);
557 wait_event_timeout(sdp->sd_log_waitq,
558 gfs2_log_is_empty(sdp),
560 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
562 wait_event_timeout(sdp->sd_log_waitq,
563 gfs2_log_is_empty(sdp),
566 gfs2_quota_cleanup(sdp);
568 if (!log_write_allowed)
569 sdp->sd_vfs->s_flags |= SB_RDONLY;
573 * gfs2_put_super - Unmount the filesystem
574 * @sb: The VFS superblock
578 static void gfs2_put_super(struct super_block *sb)
580 struct gfs2_sbd *sdp = sb->s_fs_info;
581 struct gfs2_jdesc *jd;
583 /* No more recovery requests */
584 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
587 /* Wait on outstanding recovery */
589 spin_lock(&sdp->sd_jindex_spin);
590 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
591 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
593 spin_unlock(&sdp->sd_jindex_spin);
594 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
595 TASK_UNINTERRUPTIBLE);
598 spin_unlock(&sdp->sd_jindex_spin);
600 if (!sb_rdonly(sb)) {
601 gfs2_make_fs_ro(sdp);
603 WARN_ON(gfs2_withdrawing(sdp));
605 /* At this point, we're through modifying the disk */
609 iput(sdp->sd_jindex);
610 iput(sdp->sd_statfs_inode);
611 iput(sdp->sd_rindex);
612 iput(sdp->sd_quota_inode);
614 gfs2_glock_put(sdp->sd_rename_gl);
615 gfs2_glock_put(sdp->sd_freeze_gl);
617 if (!sdp->sd_args.ar_spectator) {
618 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
619 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
620 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
621 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
622 brelse(sdp->sd_sc_bh);
623 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
624 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
625 free_local_statfs_inodes(sdp);
626 iput(sdp->sd_qc_inode);
629 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
630 gfs2_clear_rgrpd(sdp);
631 gfs2_jindex_free(sdp);
632 /* Take apart glock structures and buffer lists */
633 gfs2_gl_hash_clear(sdp);
634 truncate_inode_pages_final(&sdp->sd_aspace);
635 gfs2_delete_debugfs_file(sdp);
636 /* Unmount the locking protocol */
637 gfs2_lm_unmount(sdp);
639 /* At this point, we're through participating in the lockspace */
640 gfs2_sys_fs_del(sdp);
645 * gfs2_sync_fs - sync the filesystem
646 * @sb: the superblock
647 * @wait: true to wait for completion
649 * Flushes the log to disk.
652 static int gfs2_sync_fs(struct super_block *sb, int wait)
654 struct gfs2_sbd *sdp = sb->s_fs_info;
656 gfs2_quota_sync(sb, -1);
658 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
660 return sdp->sd_log_error;
663 void gfs2_freeze_func(struct work_struct *work)
666 struct gfs2_holder freeze_gh;
667 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
668 struct super_block *sb = sdp->sd_vfs;
670 atomic_inc(&sb->s_active);
671 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
673 gfs2_assert_withdraw(sdp, 0);
675 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
676 error = thaw_super(sb);
678 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
680 gfs2_assert_withdraw(sdp, 0);
682 gfs2_freeze_unlock(&freeze_gh);
684 deactivate_super(sb);
685 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
686 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
691 * gfs2_freeze - prevent further writes to the filesystem
692 * @sb: the VFS structure for the filesystem
696 static int gfs2_freeze(struct super_block *sb)
698 struct gfs2_sbd *sdp = sb->s_fs_info;
701 mutex_lock(&sdp->sd_freeze_mutex);
702 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
708 if (gfs2_withdrawn(sdp)) {
713 error = gfs2_lock_fs_check_clean(sdp);
718 fs_err(sdp, "waiting for recovery before freeze\n");
719 else if (error == -EIO) {
720 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
721 "to recovery error.\n");
724 fs_err(sdp, "error freezing FS: %d\n", error);
726 fs_err(sdp, "retrying...\n");
729 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
731 mutex_unlock(&sdp->sd_freeze_mutex);
736 * gfs2_unfreeze - reallow writes to the filesystem
737 * @sb: the VFS structure for the filesystem
741 static int gfs2_unfreeze(struct super_block *sb)
743 struct gfs2_sbd *sdp = sb->s_fs_info;
745 mutex_lock(&sdp->sd_freeze_mutex);
746 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
747 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
748 mutex_unlock(&sdp->sd_freeze_mutex);
752 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
753 mutex_unlock(&sdp->sd_freeze_mutex);
754 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
758 * statfs_slow_fill - fill in the sg for a given RG
760 * @sc: the sc structure
762 * Returns: 0 on success, -ESTALE if the LVB is invalid
765 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
766 struct gfs2_statfs_change_host *sc)
768 gfs2_rgrp_verify(rgd);
769 sc->sc_total += rgd->rd_data;
770 sc->sc_free += rgd->rd_free;
771 sc->sc_dinodes += rgd->rd_dinodes;
776 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
777 * @sdp: the filesystem
778 * @sc: the sc info that will be returned
780 * Any error (other than a signal) will cause this routine to fall back
781 * to the synchronous version.
783 * FIXME: This really shouldn't busy wait like this.
788 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
790 struct gfs2_rgrpd *rgd_next;
791 struct gfs2_holder *gha, *gh;
792 unsigned int slots = 64;
797 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
798 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
801 for (x = 0; x < slots; x++)
802 gfs2_holder_mark_uninitialized(gha + x);
804 rgd_next = gfs2_rgrpd_get_first(sdp);
809 for (x = 0; x < slots; x++) {
812 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
813 err = gfs2_glock_wait(gh);
815 gfs2_holder_uninit(gh);
819 struct gfs2_rgrpd *rgd =
820 gfs2_glock2rgrp(gh->gh_gl);
822 error = statfs_slow_fill(rgd, sc);
824 gfs2_glock_dq_uninit(gh);
828 if (gfs2_holder_initialized(gh))
830 else if (rgd_next && !error) {
831 error = gfs2_glock_nq_init(rgd_next->rd_gl,
835 rgd_next = gfs2_rgrpd_get_next(rgd_next);
839 if (signal_pending(current))
840 error = -ERESTARTSYS;
854 * gfs2_statfs_i - Do a statfs
855 * @sdp: the filesystem
856 * @sc: the sc structure
861 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
863 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
864 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
866 spin_lock(&sdp->sd_statfs_spin);
869 sc->sc_total += l_sc->sc_total;
870 sc->sc_free += l_sc->sc_free;
871 sc->sc_dinodes += l_sc->sc_dinodes;
873 spin_unlock(&sdp->sd_statfs_spin);
877 if (sc->sc_free > sc->sc_total)
878 sc->sc_free = sc->sc_total;
879 if (sc->sc_dinodes < 0)
886 * gfs2_statfs - Gather and return stats about the filesystem
887 * @dentry: The name of the link
890 * Returns: 0 on success or error code
893 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
895 struct super_block *sb = dentry->d_sb;
896 struct gfs2_sbd *sdp = sb->s_fs_info;
897 struct gfs2_statfs_change_host sc;
900 error = gfs2_rindex_update(sdp);
904 if (gfs2_tune_get(sdp, gt_statfs_slow))
905 error = gfs2_statfs_slow(sdp, &sc);
907 error = gfs2_statfs_i(sdp, &sc);
912 buf->f_type = GFS2_MAGIC;
913 buf->f_bsize = sdp->sd_sb.sb_bsize;
914 buf->f_blocks = sc.sc_total;
915 buf->f_bfree = sc.sc_free;
916 buf->f_bavail = sc.sc_free;
917 buf->f_files = sc.sc_dinodes + sc.sc_free;
918 buf->f_ffree = sc.sc_free;
919 buf->f_namelen = GFS2_FNAMESIZE;
925 * gfs2_drop_inode - Drop an inode (test for remote unlink)
926 * @inode: The inode to drop
928 * If we've received a callback on an iopen lock then it's because a
929 * remote node tried to deallocate the inode but failed due to this node
930 * still having the inode open. Here we mark the link count zero
931 * since we know that it must have reached zero if the GLF_DEMOTE flag
932 * is set on the iopen glock. If we didn't do a disk read since the
933 * remote node removed the final link then we might otherwise miss
934 * this event. This check ensures that this node will deallocate the
935 * inode's blocks, or alternatively pass the baton on to another
936 * node for later deallocation.
939 static int gfs2_drop_inode(struct inode *inode)
941 struct gfs2_inode *ip = GFS2_I(inode);
942 struct gfs2_sbd *sdp = GFS2_SB(inode);
944 if (inode->i_nlink &&
945 gfs2_holder_initialized(&ip->i_iopen_gh)) {
946 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
947 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
952 * When under memory pressure when an inode's link count has dropped to
953 * zero, defer deleting the inode to the delete workqueue. This avoids
954 * calling into DLM under memory pressure, which can deadlock.
956 if (!inode->i_nlink &&
957 unlikely(current->flags & PF_MEMALLOC) &&
958 gfs2_holder_initialized(&ip->i_iopen_gh)) {
959 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
962 if (!gfs2_queue_try_to_evict(gl))
963 gfs2_glock_queue_put(gl);
968 * No longer cache inodes when trying to evict them all.
970 if (test_bit(SDF_EVICTING, &sdp->sd_flags))
973 return generic_drop_inode(inode);
976 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
982 } while (!IS_ROOT(d1));
987 * gfs2_show_options - Show mount options for /proc/mounts
988 * @s: seq_file structure
989 * @root: root of this (sub)tree
991 * Returns: 0 on success or error code
994 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
996 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
997 struct gfs2_args *args = &sdp->sd_args;
1000 if (is_ancestor(root, sdp->sd_master_dir))
1001 seq_puts(s, ",meta");
1002 if (args->ar_lockproto[0])
1003 seq_show_option(s, "lockproto", args->ar_lockproto);
1004 if (args->ar_locktable[0])
1005 seq_show_option(s, "locktable", args->ar_locktable);
1006 if (args->ar_hostdata[0])
1007 seq_show_option(s, "hostdata", args->ar_hostdata);
1008 if (args->ar_spectator)
1009 seq_puts(s, ",spectator");
1010 if (args->ar_localflocks)
1011 seq_puts(s, ",localflocks");
1013 seq_puts(s, ",debug");
1014 if (args->ar_posix_acl)
1015 seq_puts(s, ",acl");
1016 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1018 switch (args->ar_quota) {
1019 case GFS2_QUOTA_OFF:
1022 case GFS2_QUOTA_ACCOUNT:
1032 seq_printf(s, ",quota=%s", state);
1034 if (args->ar_suiddir)
1035 seq_puts(s, ",suiddir");
1036 if (args->ar_data != GFS2_DATA_DEFAULT) {
1038 switch (args->ar_data) {
1039 case GFS2_DATA_WRITEBACK:
1040 state = "writeback";
1042 case GFS2_DATA_ORDERED:
1049 seq_printf(s, ",data=%s", state);
1051 if (args->ar_discard)
1052 seq_puts(s, ",discard");
1053 val = sdp->sd_tune.gt_logd_secs;
1055 seq_printf(s, ",commit=%d", val);
1056 val = sdp->sd_tune.gt_statfs_quantum;
1058 seq_printf(s, ",statfs_quantum=%d", val);
1059 else if (sdp->sd_tune.gt_statfs_slow)
1060 seq_puts(s, ",statfs_quantum=0");
1061 val = sdp->sd_tune.gt_quota_quantum;
1063 seq_printf(s, ",quota_quantum=%d", val);
1064 if (args->ar_statfs_percent)
1065 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1066 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1069 switch (args->ar_errors) {
1070 case GFS2_ERRORS_WITHDRAW:
1073 case GFS2_ERRORS_PANIC:
1080 seq_printf(s, ",errors=%s", state);
1082 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1083 seq_puts(s, ",nobarrier");
1084 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1085 seq_puts(s, ",demote_interface_used");
1086 if (args->ar_rgrplvb)
1087 seq_puts(s, ",rgrplvb");
1088 if (args->ar_loccookie)
1089 seq_puts(s, ",loccookie");
1093 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1095 struct inode *inode = &ip->i_inode;
1096 struct gfs2_glock *gl = ip->i_gl;
1098 if (unlikely(!gl)) {
1099 /* This can only happen during incomplete inode creation. */
1100 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1104 truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1105 truncate_inode_pages(&inode->i_data, 0);
1107 if (atomic_read(&gl->gl_revokes) == 0) {
1108 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1109 clear_bit(GLF_DIRTY, &gl->gl_flags);
1113 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1115 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1116 struct gfs2_rgrpd *rgd;
1117 struct gfs2_holder gh;
1120 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1121 gfs2_consist_inode(ip);
1125 error = gfs2_rindex_update(sdp);
1129 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1133 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1135 gfs2_consist_inode(ip);
1140 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1141 LM_FLAG_NODE_SCOPE, &gh);
1145 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1146 sdp->sd_jdesc->jd_blocks);
1148 goto out_rg_gunlock;
1150 gfs2_free_di(rgd, ip);
1152 gfs2_final_release_pages(ip);
1154 gfs2_trans_end(sdp);
1157 gfs2_glock_dq_uninit(&gh);
1159 gfs2_quota_unhold(ip);
1164 * gfs2_glock_put_eventually
1165 * @gl: The glock to put
1167 * When under memory pressure, trigger a deferred glock put to make sure we
1168 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1171 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1173 if (current->flags & PF_MEMALLOC)
1174 gfs2_glock_queue_put(gl);
1179 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1181 struct gfs2_inode *ip = GFS2_I(inode);
1182 struct gfs2_sbd *sdp = GFS2_SB(inode);
1183 struct gfs2_holder *gh = &ip->i_iopen_gh;
1184 long timeout = 5 * HZ;
1187 gh->gh_flags |= GL_NOCACHE;
1188 gfs2_glock_dq_wait(gh);
1191 * If there are no other lock holders, we will immediately get
1192 * exclusive access to the iopen glock here.
1194 * Otherwise, the other nodes holding the lock will be notified about
1195 * our locking request. If they do not have the inode open, they are
1196 * expected to evict the cached inode and release the lock, allowing us
1199 * Otherwise, if they cannot evict the inode, they are expected to poke
1200 * the inode glock (note: not the iopen glock). We will notice that
1201 * and stop waiting for the iopen glock immediately. The other node(s)
1202 * are then expected to take care of deleting the inode when they no
1205 * As a last resort, if another node keeps holding the iopen glock
1206 * without showing any activity on the inode glock, we will eventually
1207 * time out and fail the iopen glock upgrade.
1209 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1210 * locking request as an optimization to notify lock holders as soon as
1211 * possible. Without that flag, they'd be notified implicitly by the
1212 * second locking request.
1215 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1216 error = gfs2_glock_nq(gh);
1217 if (error != GLR_TRYFAILED)
1220 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1221 error = gfs2_glock_nq(gh);
1225 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1226 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1227 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1229 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1233 return gfs2_glock_holder_ready(gh) == 0;
1237 * evict_should_delete - determine whether the inode is eligible for deletion
1238 * @inode: The inode to evict
1239 * @gh: The glock holder structure
1241 * This function determines whether the evicted inode is eligible to be deleted
1242 * and locks the inode glock.
1244 * Returns: the fate of the dinode
1246 static enum dinode_demise evict_should_delete(struct inode *inode,
1247 struct gfs2_holder *gh)
1249 struct gfs2_inode *ip = GFS2_I(inode);
1250 struct super_block *sb = inode->i_sb;
1251 struct gfs2_sbd *sdp = sb->s_fs_info;
1254 if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1257 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1258 return SHOULD_DEFER_EVICTION;
1260 /* Deletes should never happen under memory pressure anymore. */
1261 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1262 return SHOULD_DEFER_EVICTION;
1264 /* Must not read inode block until block type has been verified */
1265 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1266 if (unlikely(ret)) {
1267 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1268 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1269 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1270 return SHOULD_DEFER_EVICTION;
1273 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1274 return SHOULD_NOT_DELETE_DINODE;
1275 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1277 return SHOULD_NOT_DELETE_DINODE;
1279 ret = gfs2_instantiate(gh);
1281 return SHOULD_NOT_DELETE_DINODE;
1284 * The inode may have been recreated in the meantime.
1287 return SHOULD_NOT_DELETE_DINODE;
1290 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1291 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1292 if (!gfs2_upgrade_iopen_glock(inode)) {
1293 gfs2_holder_uninit(&ip->i_iopen_gh);
1294 return SHOULD_NOT_DELETE_DINODE;
1297 return SHOULD_DELETE_DINODE;
1301 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1302 * @inode: The inode to evict
1304 static int evict_unlinked_inode(struct inode *inode)
1306 struct gfs2_inode *ip = GFS2_I(inode);
1309 if (S_ISDIR(inode->i_mode) &&
1310 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1311 ret = gfs2_dir_exhash_dealloc(ip);
1317 ret = gfs2_ea_dealloc(ip);
1322 if (!gfs2_is_stuffed(ip)) {
1323 ret = gfs2_file_dealloc(ip);
1329 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1332 * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1333 * can get called to recreate it, or even gfs2_inode_lookup() if the
1334 * inode was recreated on another node in the meantime.
1336 * However, inserting the new inode into the inode hash table will not
1337 * succeed until the old inode is removed, and that only happens after
1338 * ->evict_inode() returns. The new inode is attached to its inode and
1339 * iopen glocks after inserting it into the inode hash table, so at
1340 * that point we can be sure that both glocks are unused.
1343 ret = gfs2_dinode_dealloc(ip);
1349 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1350 * @inode: The inode to evict
1352 static int evict_linked_inode(struct inode *inode)
1354 struct super_block *sb = inode->i_sb;
1355 struct gfs2_sbd *sdp = sb->s_fs_info;
1356 struct gfs2_inode *ip = GFS2_I(inode);
1357 struct address_space *metamapping;
1360 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1361 GFS2_LFC_EVICT_INODE);
1362 metamapping = gfs2_glock2aspace(ip->i_gl);
1363 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1364 filemap_fdatawrite(metamapping);
1365 filemap_fdatawait(metamapping);
1367 write_inode_now(inode, 1);
1368 gfs2_ail_flush(ip->i_gl, 0);
1370 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1374 /* Needs to be done before glock release & also in a transaction */
1375 truncate_inode_pages(&inode->i_data, 0);
1376 truncate_inode_pages(metamapping, 0);
1377 gfs2_trans_end(sdp);
1382 * gfs2_evict_inode - Remove an inode from cache
1383 * @inode: The inode to evict
1385 * There are three cases to consider:
1386 * 1. i_nlink == 0, we are final opener (and must deallocate)
1387 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1390 * If the fs is read only, then we have to treat all cases as per #3
1391 * since we are unable to do any deallocation. The inode will be
1392 * deallocated by the next read/write node to attempt an allocation
1393 * in the same resource group
1395 * We have to (at the moment) hold the inodes main lock to cover
1396 * the gap between unlocking the shared lock on the iopen lock and
1397 * taking the exclusive lock. I'd rather do a shared -> exclusive
1398 * conversion on the iopen lock, but we can change that later. This
1399 * is safe, just less efficient.
1402 static void gfs2_evict_inode(struct inode *inode)
1404 struct super_block *sb = inode->i_sb;
1405 struct gfs2_sbd *sdp = sb->s_fs_info;
1406 struct gfs2_inode *ip = GFS2_I(inode);
1407 struct gfs2_holder gh;
1410 if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1413 gfs2_holder_mark_uninitialized(&gh);
1414 ret = evict_should_delete(inode, &gh);
1415 if (ret == SHOULD_DEFER_EVICTION)
1417 if (ret == SHOULD_DELETE_DINODE)
1418 ret = evict_unlinked_inode(inode);
1420 ret = evict_linked_inode(inode);
1422 if (gfs2_rs_active(&ip->i_res))
1423 gfs2_rs_deltree(&ip->i_res);
1425 if (gfs2_holder_initialized(&gh))
1426 gfs2_glock_dq_uninit(&gh);
1427 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1428 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1430 truncate_inode_pages_final(&inode->i_data);
1432 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1433 gfs2_rs_deltree(&ip->i_res);
1434 gfs2_ordered_del_inode(ip);
1436 gfs2_dir_hash_inval(ip);
1437 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1438 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1440 glock_clear_object(gl, ip);
1441 gfs2_glock_hold(gl);
1442 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1443 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1444 gfs2_glock_put_eventually(gl);
1447 glock_clear_object(ip->i_gl, ip);
1448 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1449 gfs2_glock_add_to_lru(ip->i_gl);
1450 gfs2_glock_put_eventually(ip->i_gl);
1455 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1457 struct gfs2_inode *ip;
1459 ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1465 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1466 memset(&ip->i_res, 0, sizeof(ip->i_res));
1467 RB_CLEAR_NODE(&ip->i_res.rs_node);
1469 return &ip->i_inode;
1472 static void gfs2_free_inode(struct inode *inode)
1474 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1477 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1479 struct local_statfs_inode *lsi, *safe;
1481 /* Run through the statfs inodes list to iput and free memory */
1482 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1483 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1484 sdp->sd_sc_inode = NULL; /* belongs to this node */
1485 if (lsi->si_sc_inode)
1486 iput(lsi->si_sc_inode);
1487 list_del(&lsi->si_list);
1492 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1495 struct local_statfs_inode *lsi;
1497 /* Return the local (per node) statfs inode in the
1498 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1499 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1500 if (lsi->si_jid == index)
1501 return lsi->si_sc_inode;
1506 const struct super_operations gfs2_super_ops = {
1507 .alloc_inode = gfs2_alloc_inode,
1508 .free_inode = gfs2_free_inode,
1509 .write_inode = gfs2_write_inode,
1510 .dirty_inode = gfs2_dirty_inode,
1511 .evict_inode = gfs2_evict_inode,
1512 .put_super = gfs2_put_super,
1513 .sync_fs = gfs2_sync_fs,
1514 .freeze_super = gfs2_freeze,
1515 .thaw_super = gfs2_unfreeze,
1516 .statfs = gfs2_statfs,
1517 .drop_inode = gfs2_drop_inode,
1518 .show_options = gfs2_show_options,