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 struct gfs2_dinode *str = buf;
384 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
385 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
386 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
387 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
388 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
389 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
390 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
391 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
392 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
393 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
394 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
395 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
396 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
397 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
399 str->di_goal_meta = cpu_to_be64(ip->i_goal);
400 str->di_goal_data = cpu_to_be64(ip->i_goal);
401 str->di_generation = cpu_to_be64(ip->i_generation);
403 str->di_flags = cpu_to_be32(ip->i_diskflags);
404 str->di_height = cpu_to_be16(ip->i_height);
405 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
406 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
408 str->di_depth = cpu_to_be16(ip->i_depth);
409 str->di_entries = cpu_to_be32(ip->i_entries);
411 str->di_eattr = cpu_to_be64(ip->i_eattr);
412 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
413 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
414 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
418 * gfs2_write_inode - Make sure the inode is stable on the disk
420 * @wbc: The writeback control structure
425 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
427 struct gfs2_inode *ip = GFS2_I(inode);
428 struct gfs2_sbd *sdp = GFS2_SB(inode);
429 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
430 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
432 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
435 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
436 GFS2_LOG_HEAD_FLUSH_NORMAL |
437 GFS2_LFC_WRITE_INODE);
438 if (bdi->wb.dirty_exceeded)
439 gfs2_ail1_flush(sdp, wbc);
441 filemap_fdatawrite(metamapping);
443 ret = filemap_fdatawait(metamapping);
445 mark_inode_dirty_sync(inode);
447 spin_lock(&inode->i_lock);
448 if (!(inode->i_flags & I_DIRTY))
449 gfs2_ordered_del_inode(ip);
450 spin_unlock(&inode->i_lock);
456 * gfs2_dirty_inode - check for atime updates
457 * @inode: The inode in question
458 * @flags: The type of dirty
460 * Unfortunately it can be called under any combination of inode
461 * glock and transaction lock, so we have to check carefully.
463 * At the moment this deals only with atime - it should be possible
464 * to expand that role in future, once a review of the locking has
468 static void gfs2_dirty_inode(struct inode *inode, int flags)
470 struct gfs2_inode *ip = GFS2_I(inode);
471 struct gfs2_sbd *sdp = GFS2_SB(inode);
472 struct buffer_head *bh;
473 struct gfs2_holder gh;
475 int need_endtrans = 0;
478 if (unlikely(gfs2_withdrawn(sdp)))
480 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
481 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
483 fs_err(sdp, "dirty_inode: glock %d\n", ret);
484 gfs2_dump_glock(NULL, ip->i_gl, true);
488 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
491 if (current->journal_info == NULL) {
492 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
494 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
500 ret = gfs2_meta_inode_buffer(ip, &bh);
502 gfs2_trans_add_meta(ip->i_gl, bh);
503 gfs2_dinode_out(ip, bh->b_data);
511 gfs2_glock_dq_uninit(&gh);
515 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
516 * @sdp: the filesystem
521 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
523 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
525 gfs2_flush_delete_work(sdp);
526 if (!log_write_allowed && current == sdp->sd_quotad_process)
527 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
528 else if (sdp->sd_quotad_process)
529 kthread_stop(sdp->sd_quotad_process);
530 sdp->sd_quotad_process = NULL;
532 if (!log_write_allowed && current == sdp->sd_logd_process)
533 fs_warn(sdp, "The logd daemon is withdrawing.\n");
534 else if (sdp->sd_logd_process)
535 kthread_stop(sdp->sd_logd_process);
536 sdp->sd_logd_process = NULL;
538 if (log_write_allowed) {
539 gfs2_quota_sync(sdp->sd_vfs, 0);
540 gfs2_statfs_sync(sdp->sd_vfs, 0);
542 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
543 GFS2_LFC_MAKE_FS_RO);
544 wait_event_timeout(sdp->sd_log_waitq,
545 gfs2_log_is_empty(sdp),
547 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
549 wait_event_timeout(sdp->sd_log_waitq,
550 gfs2_log_is_empty(sdp),
553 gfs2_quota_cleanup(sdp);
555 if (!log_write_allowed)
556 sdp->sd_vfs->s_flags |= SB_RDONLY;
560 * gfs2_put_super - Unmount the filesystem
561 * @sb: The VFS superblock
565 static void gfs2_put_super(struct super_block *sb)
567 struct gfs2_sbd *sdp = sb->s_fs_info;
568 struct gfs2_jdesc *jd;
570 /* No more recovery requests */
571 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
574 /* Wait on outstanding recovery */
576 spin_lock(&sdp->sd_jindex_spin);
577 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
578 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
580 spin_unlock(&sdp->sd_jindex_spin);
581 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
582 TASK_UNINTERRUPTIBLE);
585 spin_unlock(&sdp->sd_jindex_spin);
587 if (!sb_rdonly(sb)) {
588 gfs2_make_fs_ro(sdp);
590 WARN_ON(gfs2_withdrawing(sdp));
592 /* At this point, we're through modifying the disk */
596 iput(sdp->sd_jindex);
597 iput(sdp->sd_statfs_inode);
598 iput(sdp->sd_rindex);
599 iput(sdp->sd_quota_inode);
601 gfs2_glock_put(sdp->sd_rename_gl);
602 gfs2_glock_put(sdp->sd_freeze_gl);
604 if (!sdp->sd_args.ar_spectator) {
605 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
606 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
607 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
608 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
609 brelse(sdp->sd_sc_bh);
610 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
611 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
612 free_local_statfs_inodes(sdp);
613 iput(sdp->sd_qc_inode);
616 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
617 gfs2_clear_rgrpd(sdp);
618 gfs2_jindex_free(sdp);
619 /* Take apart glock structures and buffer lists */
620 gfs2_gl_hash_clear(sdp);
621 truncate_inode_pages_final(&sdp->sd_aspace);
622 gfs2_delete_debugfs_file(sdp);
623 /* Unmount the locking protocol */
624 gfs2_lm_unmount(sdp);
626 /* At this point, we're through participating in the lockspace */
627 gfs2_sys_fs_del(sdp);
632 * gfs2_sync_fs - sync the filesystem
633 * @sb: the superblock
634 * @wait: true to wait for completion
636 * Flushes the log to disk.
639 static int gfs2_sync_fs(struct super_block *sb, int wait)
641 struct gfs2_sbd *sdp = sb->s_fs_info;
643 gfs2_quota_sync(sb, -1);
645 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
647 return sdp->sd_log_error;
650 void gfs2_freeze_func(struct work_struct *work)
653 struct gfs2_holder freeze_gh;
654 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
655 struct super_block *sb = sdp->sd_vfs;
657 atomic_inc(&sb->s_active);
658 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
660 gfs2_assert_withdraw(sdp, 0);
662 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
663 error = thaw_super(sb);
665 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
667 gfs2_assert_withdraw(sdp, 0);
669 gfs2_freeze_unlock(&freeze_gh);
671 deactivate_super(sb);
672 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
673 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
678 * gfs2_freeze - prevent further writes to the filesystem
679 * @sb: the VFS structure for the filesystem
683 static int gfs2_freeze(struct super_block *sb)
685 struct gfs2_sbd *sdp = sb->s_fs_info;
688 mutex_lock(&sdp->sd_freeze_mutex);
689 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
695 if (gfs2_withdrawn(sdp)) {
700 error = gfs2_lock_fs_check_clean(sdp);
705 fs_err(sdp, "waiting for recovery before freeze\n");
706 else if (error == -EIO) {
707 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
708 "to recovery error.\n");
711 fs_err(sdp, "error freezing FS: %d\n", error);
713 fs_err(sdp, "retrying...\n");
716 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
718 mutex_unlock(&sdp->sd_freeze_mutex);
723 * gfs2_unfreeze - reallow writes to the filesystem
724 * @sb: the VFS structure for the filesystem
728 static int gfs2_unfreeze(struct super_block *sb)
730 struct gfs2_sbd *sdp = sb->s_fs_info;
732 mutex_lock(&sdp->sd_freeze_mutex);
733 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
734 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
735 mutex_unlock(&sdp->sd_freeze_mutex);
739 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
740 mutex_unlock(&sdp->sd_freeze_mutex);
741 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
745 * statfs_slow_fill - fill in the sg for a given RG
747 * @sc: the sc structure
749 * Returns: 0 on success, -ESTALE if the LVB is invalid
752 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
753 struct gfs2_statfs_change_host *sc)
755 gfs2_rgrp_verify(rgd);
756 sc->sc_total += rgd->rd_data;
757 sc->sc_free += rgd->rd_free;
758 sc->sc_dinodes += rgd->rd_dinodes;
763 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
764 * @sdp: the filesystem
765 * @sc: the sc info that will be returned
767 * Any error (other than a signal) will cause this routine to fall back
768 * to the synchronous version.
770 * FIXME: This really shouldn't busy wait like this.
775 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
777 struct gfs2_rgrpd *rgd_next;
778 struct gfs2_holder *gha, *gh;
779 unsigned int slots = 64;
784 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
785 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
788 for (x = 0; x < slots; x++)
789 gfs2_holder_mark_uninitialized(gha + x);
791 rgd_next = gfs2_rgrpd_get_first(sdp);
796 for (x = 0; x < slots; x++) {
799 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
800 err = gfs2_glock_wait(gh);
802 gfs2_holder_uninit(gh);
806 struct gfs2_rgrpd *rgd =
807 gfs2_glock2rgrp(gh->gh_gl);
809 error = statfs_slow_fill(rgd, sc);
811 gfs2_glock_dq_uninit(gh);
815 if (gfs2_holder_initialized(gh))
817 else if (rgd_next && !error) {
818 error = gfs2_glock_nq_init(rgd_next->rd_gl,
822 rgd_next = gfs2_rgrpd_get_next(rgd_next);
826 if (signal_pending(current))
827 error = -ERESTARTSYS;
841 * gfs2_statfs_i - Do a statfs
842 * @sdp: the filesystem
843 * @sc: the sc structure
848 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
850 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
851 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
853 spin_lock(&sdp->sd_statfs_spin);
856 sc->sc_total += l_sc->sc_total;
857 sc->sc_free += l_sc->sc_free;
858 sc->sc_dinodes += l_sc->sc_dinodes;
860 spin_unlock(&sdp->sd_statfs_spin);
864 if (sc->sc_free > sc->sc_total)
865 sc->sc_free = sc->sc_total;
866 if (sc->sc_dinodes < 0)
873 * gfs2_statfs - Gather and return stats about the filesystem
874 * @dentry: The name of the link
877 * Returns: 0 on success or error code
880 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
882 struct super_block *sb = dentry->d_sb;
883 struct gfs2_sbd *sdp = sb->s_fs_info;
884 struct gfs2_statfs_change_host sc;
887 error = gfs2_rindex_update(sdp);
891 if (gfs2_tune_get(sdp, gt_statfs_slow))
892 error = gfs2_statfs_slow(sdp, &sc);
894 error = gfs2_statfs_i(sdp, &sc);
899 buf->f_type = GFS2_MAGIC;
900 buf->f_bsize = sdp->sd_sb.sb_bsize;
901 buf->f_blocks = sc.sc_total;
902 buf->f_bfree = sc.sc_free;
903 buf->f_bavail = sc.sc_free;
904 buf->f_files = sc.sc_dinodes + sc.sc_free;
905 buf->f_ffree = sc.sc_free;
906 buf->f_namelen = GFS2_FNAMESIZE;
912 * gfs2_drop_inode - Drop an inode (test for remote unlink)
913 * @inode: The inode to drop
915 * If we've received a callback on an iopen lock then it's because a
916 * remote node tried to deallocate the inode but failed due to this node
917 * still having the inode open. Here we mark the link count zero
918 * since we know that it must have reached zero if the GLF_DEMOTE flag
919 * is set on the iopen glock. If we didn't do a disk read since the
920 * remote node removed the final link then we might otherwise miss
921 * this event. This check ensures that this node will deallocate the
922 * inode's blocks, or alternatively pass the baton on to another
923 * node for later deallocation.
926 static int gfs2_drop_inode(struct inode *inode)
928 struct gfs2_inode *ip = GFS2_I(inode);
930 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
932 gfs2_holder_initialized(&ip->i_iopen_gh)) {
933 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
934 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
939 * When under memory pressure when an inode's link count has dropped to
940 * zero, defer deleting the inode to the delete workqueue. This avoids
941 * calling into DLM under memory pressure, which can deadlock.
943 if (!inode->i_nlink &&
944 unlikely(current->flags & PF_MEMALLOC) &&
945 gfs2_holder_initialized(&ip->i_iopen_gh)) {
946 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
949 if (!gfs2_queue_delete_work(gl, 0))
950 gfs2_glock_queue_put(gl);
954 return generic_drop_inode(inode);
957 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
963 } while (!IS_ROOT(d1));
968 * gfs2_show_options - Show mount options for /proc/mounts
969 * @s: seq_file structure
970 * @root: root of this (sub)tree
972 * Returns: 0 on success or error code
975 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
977 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
978 struct gfs2_args *args = &sdp->sd_args;
981 if (is_ancestor(root, sdp->sd_master_dir))
982 seq_puts(s, ",meta");
983 if (args->ar_lockproto[0])
984 seq_show_option(s, "lockproto", args->ar_lockproto);
985 if (args->ar_locktable[0])
986 seq_show_option(s, "locktable", args->ar_locktable);
987 if (args->ar_hostdata[0])
988 seq_show_option(s, "hostdata", args->ar_hostdata);
989 if (args->ar_spectator)
990 seq_puts(s, ",spectator");
991 if (args->ar_localflocks)
992 seq_puts(s, ",localflocks");
994 seq_puts(s, ",debug");
995 if (args->ar_posix_acl)
997 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
999 switch (args->ar_quota) {
1000 case GFS2_QUOTA_OFF:
1003 case GFS2_QUOTA_ACCOUNT:
1013 seq_printf(s, ",quota=%s", state);
1015 if (args->ar_suiddir)
1016 seq_puts(s, ",suiddir");
1017 if (args->ar_data != GFS2_DATA_DEFAULT) {
1019 switch (args->ar_data) {
1020 case GFS2_DATA_WRITEBACK:
1021 state = "writeback";
1023 case GFS2_DATA_ORDERED:
1030 seq_printf(s, ",data=%s", state);
1032 if (args->ar_discard)
1033 seq_puts(s, ",discard");
1034 val = sdp->sd_tune.gt_logd_secs;
1036 seq_printf(s, ",commit=%d", val);
1037 val = sdp->sd_tune.gt_statfs_quantum;
1039 seq_printf(s, ",statfs_quantum=%d", val);
1040 else if (sdp->sd_tune.gt_statfs_slow)
1041 seq_puts(s, ",statfs_quantum=0");
1042 val = sdp->sd_tune.gt_quota_quantum;
1044 seq_printf(s, ",quota_quantum=%d", val);
1045 if (args->ar_statfs_percent)
1046 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1047 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1050 switch (args->ar_errors) {
1051 case GFS2_ERRORS_WITHDRAW:
1054 case GFS2_ERRORS_PANIC:
1061 seq_printf(s, ",errors=%s", state);
1063 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1064 seq_puts(s, ",nobarrier");
1065 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1066 seq_puts(s, ",demote_interface_used");
1067 if (args->ar_rgrplvb)
1068 seq_puts(s, ",rgrplvb");
1069 if (args->ar_loccookie)
1070 seq_puts(s, ",loccookie");
1074 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1076 struct inode *inode = &ip->i_inode;
1077 struct gfs2_glock *gl = ip->i_gl;
1079 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1080 truncate_inode_pages(&inode->i_data, 0);
1082 if (atomic_read(&gl->gl_revokes) == 0) {
1083 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1084 clear_bit(GLF_DIRTY, &gl->gl_flags);
1088 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1090 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1091 struct gfs2_rgrpd *rgd;
1092 struct gfs2_holder gh;
1095 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1096 gfs2_consist_inode(ip);
1100 error = gfs2_rindex_update(sdp);
1104 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1108 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1110 gfs2_consist_inode(ip);
1115 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1116 LM_FLAG_NODE_SCOPE, &gh);
1120 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1121 sdp->sd_jdesc->jd_blocks);
1123 goto out_rg_gunlock;
1125 gfs2_free_di(rgd, ip);
1127 gfs2_final_release_pages(ip);
1129 gfs2_trans_end(sdp);
1132 gfs2_glock_dq_uninit(&gh);
1134 gfs2_quota_unhold(ip);
1139 * gfs2_glock_put_eventually
1140 * @gl: The glock to put
1142 * When under memory pressure, trigger a deferred glock put to make sure we
1143 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1146 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1148 if (current->flags & PF_MEMALLOC)
1149 gfs2_glock_queue_put(gl);
1154 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1156 struct gfs2_inode *ip = GFS2_I(inode);
1157 struct gfs2_sbd *sdp = GFS2_SB(inode);
1158 struct gfs2_holder *gh = &ip->i_iopen_gh;
1159 long timeout = 5 * HZ;
1162 gh->gh_flags |= GL_NOCACHE;
1163 gfs2_glock_dq_wait(gh);
1166 * If there are no other lock holders, we'll get the lock immediately.
1167 * Otherwise, the other nodes holding the lock will be notified about
1168 * our locking request. If they don't have the inode open, they'll
1169 * evict the cached inode and release the lock. Otherwise, if they
1170 * poke the inode glock, we'll take this as an indication that they
1171 * still need the iopen glock and that they'll take care of deleting
1172 * the inode when they're done. As a last resort, if another node
1173 * keeps holding the iopen glock without showing any activity on the
1174 * inode glock, we'll eventually time out.
1176 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1177 * locking request as an optimization to notify lock holders as soon as
1178 * possible. Without that flag, they'd be notified implicitly by the
1179 * second locking request.
1182 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1183 error = gfs2_glock_nq(gh);
1184 if (error != GLR_TRYFAILED)
1187 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1188 error = gfs2_glock_nq(gh);
1192 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1193 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1194 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1196 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1200 return gfs2_glock_holder_ready(gh) == 0;
1204 * evict_should_delete - determine whether the inode is eligible for deletion
1205 * @inode: The inode to evict
1206 * @gh: The glock holder structure
1208 * This function determines whether the evicted inode is eligible to be deleted
1209 * and locks the inode glock.
1211 * Returns: the fate of the dinode
1213 static enum dinode_demise evict_should_delete(struct inode *inode,
1214 struct gfs2_holder *gh)
1216 struct gfs2_inode *ip = GFS2_I(inode);
1217 struct super_block *sb = inode->i_sb;
1218 struct gfs2_sbd *sdp = sb->s_fs_info;
1221 if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1222 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1226 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1227 return SHOULD_DEFER_EVICTION;
1229 /* Deletes should never happen under memory pressure anymore. */
1230 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1231 return SHOULD_DEFER_EVICTION;
1233 /* Must not read inode block until block type has been verified */
1234 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1235 if (unlikely(ret)) {
1236 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1237 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1238 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1239 return SHOULD_DEFER_EVICTION;
1242 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1243 return SHOULD_NOT_DELETE_DINODE;
1244 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1246 return SHOULD_NOT_DELETE_DINODE;
1248 ret = gfs2_instantiate(gh);
1250 return SHOULD_NOT_DELETE_DINODE;
1253 * The inode may have been recreated in the meantime.
1256 return SHOULD_NOT_DELETE_DINODE;
1259 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1260 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1261 if (!gfs2_upgrade_iopen_glock(inode)) {
1262 gfs2_holder_uninit(&ip->i_iopen_gh);
1263 return SHOULD_NOT_DELETE_DINODE;
1266 return SHOULD_DELETE_DINODE;
1270 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1271 * @inode: The inode to evict
1273 static int evict_unlinked_inode(struct inode *inode)
1275 struct gfs2_inode *ip = GFS2_I(inode);
1278 if (S_ISDIR(inode->i_mode) &&
1279 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1280 ret = gfs2_dir_exhash_dealloc(ip);
1286 ret = gfs2_ea_dealloc(ip);
1291 if (!gfs2_is_stuffed(ip)) {
1292 ret = gfs2_file_dealloc(ip);
1297 /* We're about to clear the bitmap for the dinode, but as soon as we
1298 do, gfs2_create_inode can create another inode at the same block
1299 location and try to set gl_object again. We clear gl_object here so
1300 that subsequent inode creates don't see an old gl_object. */
1301 glock_clear_object(ip->i_gl, ip);
1302 ret = gfs2_dinode_dealloc(ip);
1303 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1309 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1310 * @inode: The inode to evict
1312 static int evict_linked_inode(struct inode *inode)
1314 struct super_block *sb = inode->i_sb;
1315 struct gfs2_sbd *sdp = sb->s_fs_info;
1316 struct gfs2_inode *ip = GFS2_I(inode);
1317 struct address_space *metamapping;
1320 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1321 GFS2_LFC_EVICT_INODE);
1322 metamapping = gfs2_glock2aspace(ip->i_gl);
1323 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1324 filemap_fdatawrite(metamapping);
1325 filemap_fdatawait(metamapping);
1327 write_inode_now(inode, 1);
1328 gfs2_ail_flush(ip->i_gl, 0);
1330 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1334 /* Needs to be done before glock release & also in a transaction */
1335 truncate_inode_pages(&inode->i_data, 0);
1336 truncate_inode_pages(metamapping, 0);
1337 gfs2_trans_end(sdp);
1342 * gfs2_evict_inode - Remove an inode from cache
1343 * @inode: The inode to evict
1345 * There are three cases to consider:
1346 * 1. i_nlink == 0, we are final opener (and must deallocate)
1347 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1350 * If the fs is read only, then we have to treat all cases as per #3
1351 * since we are unable to do any deallocation. The inode will be
1352 * deallocated by the next read/write node to attempt an allocation
1353 * in the same resource group
1355 * We have to (at the moment) hold the inodes main lock to cover
1356 * the gap between unlocking the shared lock on the iopen lock and
1357 * taking the exclusive lock. I'd rather do a shared -> exclusive
1358 * conversion on the iopen lock, but we can change that later. This
1359 * is safe, just less efficient.
1362 static void gfs2_evict_inode(struct inode *inode)
1364 struct super_block *sb = inode->i_sb;
1365 struct gfs2_sbd *sdp = sb->s_fs_info;
1366 struct gfs2_inode *ip = GFS2_I(inode);
1367 struct gfs2_holder gh;
1370 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1375 if (inode->i_nlink || sb_rdonly(sb))
1378 gfs2_holder_mark_uninitialized(&gh);
1379 ret = evict_should_delete(inode, &gh);
1380 if (ret == SHOULD_DEFER_EVICTION)
1382 if (ret == SHOULD_DELETE_DINODE)
1383 ret = evict_unlinked_inode(inode);
1385 ret = evict_linked_inode(inode);
1387 if (gfs2_rs_active(&ip->i_res))
1388 gfs2_rs_deltree(&ip->i_res);
1390 if (gfs2_holder_initialized(&gh)) {
1391 glock_clear_object(ip->i_gl, ip);
1392 gfs2_glock_dq_uninit(&gh);
1394 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1395 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1397 truncate_inode_pages_final(&inode->i_data);
1399 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1400 gfs2_rs_deltree(&ip->i_res);
1401 gfs2_ordered_del_inode(ip);
1403 gfs2_dir_hash_inval(ip);
1404 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1405 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1407 glock_clear_object(gl, ip);
1408 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1409 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1410 gfs2_glock_dq(&ip->i_iopen_gh);
1412 gfs2_glock_hold(gl);
1413 gfs2_holder_uninit(&ip->i_iopen_gh);
1414 gfs2_glock_put_eventually(gl);
1417 glock_clear_object(ip->i_gl, ip);
1418 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1419 gfs2_glock_add_to_lru(ip->i_gl);
1420 gfs2_glock_put_eventually(ip->i_gl);
1425 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1427 struct gfs2_inode *ip;
1429 ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1434 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1435 memset(&ip->i_res, 0, sizeof(ip->i_res));
1436 RB_CLEAR_NODE(&ip->i_res.rs_node);
1438 return &ip->i_inode;
1441 static void gfs2_free_inode(struct inode *inode)
1443 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1446 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1448 struct local_statfs_inode *lsi, *safe;
1450 /* Run through the statfs inodes list to iput and free memory */
1451 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1452 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1453 sdp->sd_sc_inode = NULL; /* belongs to this node */
1454 if (lsi->si_sc_inode)
1455 iput(lsi->si_sc_inode);
1456 list_del(&lsi->si_list);
1461 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1464 struct local_statfs_inode *lsi;
1466 /* Return the local (per node) statfs inode in the
1467 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1468 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1469 if (lsi->si_jid == index)
1470 return lsi->si_sc_inode;
1475 const struct super_operations gfs2_super_ops = {
1476 .alloc_inode = gfs2_alloc_inode,
1477 .free_inode = gfs2_free_inode,
1478 .write_inode = gfs2_write_inode,
1479 .dirty_inode = gfs2_dirty_inode,
1480 .evict_inode = gfs2_evict_inode,
1481 .put_super = gfs2_put_super,
1482 .sync_fs = gfs2_sync_fs,
1483 .freeze_super = gfs2_freeze,
1484 .thaw_super = gfs2_unfreeze,
1485 .statfs = gfs2_statfs,
1486 .drop_inode = gfs2_drop_inode,
1487 .show_options = gfs2_show_options,