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>
48 * gfs2_jindex_free - Clear all the journal index information
49 * @sdp: The GFS2 superblock
53 void gfs2_jindex_free(struct gfs2_sbd *sdp)
55 struct list_head list;
56 struct gfs2_jdesc *jd;
58 spin_lock(&sdp->sd_jindex_spin);
59 list_add(&list, &sdp->sd_jindex_list);
60 list_del_init(&sdp->sd_jindex_list);
62 spin_unlock(&sdp->sd_jindex_spin);
65 while (!list_empty(&list)) {
66 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
67 gfs2_free_journal_extents(jd);
68 list_del(&jd->jd_list);
75 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
77 struct gfs2_jdesc *jd;
80 list_for_each_entry(jd, head, jd_list) {
81 if (jd->jd_jid == jid) {
93 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
95 struct gfs2_jdesc *jd;
97 spin_lock(&sdp->sd_jindex_spin);
98 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
99 spin_unlock(&sdp->sd_jindex_spin);
104 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
106 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
107 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
108 u64 size = i_size_read(jd->jd_inode);
110 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
113 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
115 if (gfs2_write_alloc_required(ip, 0, size)) {
116 gfs2_consist_inode(ip);
123 static int init_threads(struct gfs2_sbd *sdp)
125 struct task_struct *p;
128 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
131 fs_err(sdp, "can't start logd thread: %d\n", error);
134 sdp->sd_logd_process = p;
136 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
139 fs_err(sdp, "can't start quotad thread: %d\n", error);
142 sdp->sd_quotad_process = p;
146 kthread_stop(sdp->sd_logd_process);
147 sdp->sd_logd_process = NULL;
152 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
153 * @sdp: the filesystem
158 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
160 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
161 struct gfs2_glock *j_gl = ip->i_gl;
162 struct gfs2_holder freeze_gh;
163 struct gfs2_log_header_host head;
166 error = init_threads(sdp);
170 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
171 LM_FLAG_NOEXP | GL_EXACT,
176 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
177 if (gfs2_withdrawn(sdp)) {
182 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
183 if (error || gfs2_withdrawn(sdp))
186 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
192 /* Initialize some head of the log stuff */
193 sdp->sd_log_sequence = head.lh_sequence + 1;
194 gfs2_log_pointers_init(sdp, head.lh_blkno);
196 error = gfs2_quota_init(sdp);
197 if (error || gfs2_withdrawn(sdp))
200 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
202 gfs2_glock_dq_uninit(&freeze_gh);
207 gfs2_glock_dq_uninit(&freeze_gh);
209 if (sdp->sd_quotad_process)
210 kthread_stop(sdp->sd_quotad_process);
211 sdp->sd_quotad_process = NULL;
212 if (sdp->sd_logd_process)
213 kthread_stop(sdp->sd_logd_process);
214 sdp->sd_logd_process = NULL;
218 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
220 const struct gfs2_statfs_change *str = buf;
222 sc->sc_total = be64_to_cpu(str->sc_total);
223 sc->sc_free = be64_to_cpu(str->sc_free);
224 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
227 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
229 struct gfs2_statfs_change *str = buf;
231 str->sc_total = cpu_to_be64(sc->sc_total);
232 str->sc_free = cpu_to_be64(sc->sc_free);
233 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
236 int gfs2_statfs_init(struct gfs2_sbd *sdp)
238 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
239 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
240 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
241 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
242 struct buffer_head *m_bh, *l_bh;
243 struct gfs2_holder gh;
246 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
251 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
255 if (sdp->sd_args.ar_spectator) {
256 spin_lock(&sdp->sd_statfs_spin);
257 gfs2_statfs_change_in(m_sc, m_bh->b_data +
258 sizeof(struct gfs2_dinode));
259 spin_unlock(&sdp->sd_statfs_spin);
261 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
265 spin_lock(&sdp->sd_statfs_spin);
266 gfs2_statfs_change_in(m_sc, m_bh->b_data +
267 sizeof(struct gfs2_dinode));
268 gfs2_statfs_change_in(l_sc, l_bh->b_data +
269 sizeof(struct gfs2_dinode));
270 spin_unlock(&sdp->sd_statfs_spin);
278 gfs2_glock_dq_uninit(&gh);
282 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
285 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
286 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
287 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
288 struct buffer_head *l_bh;
293 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
297 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
299 spin_lock(&sdp->sd_statfs_spin);
300 l_sc->sc_total += total;
301 l_sc->sc_free += free;
302 l_sc->sc_dinodes += dinodes;
303 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
304 if (sdp->sd_args.ar_statfs_percent) {
305 x = 100 * l_sc->sc_free;
306 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
307 if (x >= y || x <= -y)
310 spin_unlock(&sdp->sd_statfs_spin);
314 gfs2_wake_up_statfs(sdp);
317 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
318 struct buffer_head *l_bh)
320 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
321 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
322 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
323 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
325 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
326 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
328 spin_lock(&sdp->sd_statfs_spin);
329 m_sc->sc_total += l_sc->sc_total;
330 m_sc->sc_free += l_sc->sc_free;
331 m_sc->sc_dinodes += l_sc->sc_dinodes;
332 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
333 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
334 0, sizeof(struct gfs2_statfs_change));
335 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
336 spin_unlock(&sdp->sd_statfs_spin);
339 int gfs2_statfs_sync(struct super_block *sb, int type)
341 struct gfs2_sbd *sdp = sb->s_fs_info;
342 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
343 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
344 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
345 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
346 struct gfs2_holder gh;
347 struct buffer_head *m_bh, *l_bh;
351 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
356 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
360 spin_lock(&sdp->sd_statfs_spin);
361 gfs2_statfs_change_in(m_sc, m_bh->b_data +
362 sizeof(struct gfs2_dinode));
363 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
364 spin_unlock(&sdp->sd_statfs_spin);
367 spin_unlock(&sdp->sd_statfs_spin);
369 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
373 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
377 update_statfs(sdp, m_bh, l_bh);
378 sdp->sd_statfs_force_sync = 0;
387 gfs2_glock_dq_uninit(&gh);
394 struct list_head list;
395 struct gfs2_holder gh;
399 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
401 * @sdp: the file system
402 * @state: the state to put the transaction lock into
403 * @t_gh: the hold on the transaction lock
408 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
410 struct gfs2_inode *ip;
411 struct gfs2_jdesc *jd;
414 struct gfs2_log_header_host lh;
417 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
418 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
423 ip = GFS2_I(jd->jd_inode);
424 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
429 list_add(&lfcc->list, &list);
432 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
433 LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
437 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
438 error = gfs2_jdesc_check(jd);
441 error = gfs2_find_jhead(jd, &lh, false);
444 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
451 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
454 while (!list_empty(&list)) {
455 lfcc = list_first_entry(&list, struct lfcc, list);
456 list_del(&lfcc->list);
457 gfs2_glock_dq_uninit(&lfcc->gh);
463 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
465 struct gfs2_dinode *str = buf;
467 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
468 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
469 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
470 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
471 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
472 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
473 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
474 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
475 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
476 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
477 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
478 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
479 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
480 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
482 str->di_goal_meta = cpu_to_be64(ip->i_goal);
483 str->di_goal_data = cpu_to_be64(ip->i_goal);
484 str->di_generation = cpu_to_be64(ip->i_generation);
486 str->di_flags = cpu_to_be32(ip->i_diskflags);
487 str->di_height = cpu_to_be16(ip->i_height);
488 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
489 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
491 str->di_depth = cpu_to_be16(ip->i_depth);
492 str->di_entries = cpu_to_be32(ip->i_entries);
494 str->di_eattr = cpu_to_be64(ip->i_eattr);
495 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
496 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
497 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
501 * gfs2_write_inode - Make sure the inode is stable on the disk
503 * @wbc: The writeback control structure
508 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
510 struct gfs2_inode *ip = GFS2_I(inode);
511 struct gfs2_sbd *sdp = GFS2_SB(inode);
512 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
513 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
515 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
518 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
519 GFS2_LOG_HEAD_FLUSH_NORMAL |
520 GFS2_LFC_WRITE_INODE);
521 if (bdi->wb.dirty_exceeded)
522 gfs2_ail1_flush(sdp, wbc);
524 filemap_fdatawrite(metamapping);
526 ret = filemap_fdatawait(metamapping);
528 mark_inode_dirty_sync(inode);
530 spin_lock(&inode->i_lock);
531 if (!(inode->i_flags & I_DIRTY))
532 gfs2_ordered_del_inode(ip);
533 spin_unlock(&inode->i_lock);
539 * gfs2_dirty_inode - check for atime updates
540 * @inode: The inode in question
541 * @flags: The type of dirty
543 * Unfortunately it can be called under any combination of inode
544 * glock and transaction lock, so we have to check carefully.
546 * At the moment this deals only with atime - it should be possible
547 * to expand that role in future, once a review of the locking has
551 static void gfs2_dirty_inode(struct inode *inode, int flags)
553 struct gfs2_inode *ip = GFS2_I(inode);
554 struct gfs2_sbd *sdp = GFS2_SB(inode);
555 struct buffer_head *bh;
556 struct gfs2_holder gh;
558 int need_endtrans = 0;
561 if (!(flags & I_DIRTY_INODE))
563 if (unlikely(gfs2_withdrawn(sdp)))
565 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
566 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
568 fs_err(sdp, "dirty_inode: glock %d\n", ret);
569 gfs2_dump_glock(NULL, ip->i_gl, true);
573 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
576 if (current->journal_info == NULL) {
577 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
579 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
585 ret = gfs2_meta_inode_buffer(ip, &bh);
587 gfs2_trans_add_meta(ip->i_gl, bh);
588 gfs2_dinode_out(ip, bh->b_data);
596 gfs2_glock_dq_uninit(&gh);
600 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
601 * @sdp: the filesystem
606 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
608 struct gfs2_holder freeze_gh;
610 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
612 gfs2_holder_mark_uninitialized(&freeze_gh);
613 if (sdp->sd_freeze_gl &&
614 !gfs2_glock_is_locked_by_me(sdp->sd_freeze_gl)) {
615 if (!log_write_allowed) {
616 error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
617 LM_ST_SHARED, LM_FLAG_TRY |
618 LM_FLAG_NOEXP | GL_EXACT,
620 if (error == GLR_TRYFAILED)
623 error = gfs2_glock_nq_init(sdp->sd_freeze_gl,
625 LM_FLAG_NOEXP | GL_EXACT,
627 if (error && !gfs2_withdrawn(sdp))
632 gfs2_flush_delete_work(sdp);
633 if (!log_write_allowed && current == sdp->sd_quotad_process)
634 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
635 else if (sdp->sd_quotad_process)
636 kthread_stop(sdp->sd_quotad_process);
637 sdp->sd_quotad_process = NULL;
639 if (!log_write_allowed && current == sdp->sd_logd_process)
640 fs_warn(sdp, "The logd daemon is withdrawing.\n");
641 else if (sdp->sd_logd_process)
642 kthread_stop(sdp->sd_logd_process);
643 sdp->sd_logd_process = NULL;
645 if (log_write_allowed) {
646 gfs2_quota_sync(sdp->sd_vfs, 0);
647 gfs2_statfs_sync(sdp->sd_vfs, 0);
649 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
650 GFS2_LFC_MAKE_FS_RO);
651 wait_event(sdp->sd_reserving_log_wait,
652 atomic_read(&sdp->sd_reserving_log) == 0);
653 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) ==
654 sdp->sd_jdesc->jd_blocks);
656 wait_event_timeout(sdp->sd_reserving_log_wait,
657 atomic_read(&sdp->sd_reserving_log) == 0,
660 if (gfs2_holder_initialized(&freeze_gh))
661 gfs2_glock_dq_uninit(&freeze_gh);
663 gfs2_quota_cleanup(sdp);
665 if (!log_write_allowed)
666 sdp->sd_vfs->s_flags |= SB_RDONLY;
672 * gfs2_put_super - Unmount the filesystem
673 * @sb: The VFS superblock
677 static void gfs2_put_super(struct super_block *sb)
679 struct gfs2_sbd *sdp = sb->s_fs_info;
681 struct gfs2_jdesc *jd;
683 /* No more recovery requests */
684 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
687 /* Wait on outstanding recovery */
689 spin_lock(&sdp->sd_jindex_spin);
690 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
691 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
693 spin_unlock(&sdp->sd_jindex_spin);
694 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
695 TASK_UNINTERRUPTIBLE);
698 spin_unlock(&sdp->sd_jindex_spin);
700 if (!sb_rdonly(sb)) {
701 error = gfs2_make_fs_ro(sdp);
705 /* At this point, we're through modifying the disk */
709 iput(sdp->sd_jindex);
710 iput(sdp->sd_statfs_inode);
711 iput(sdp->sd_rindex);
712 iput(sdp->sd_quota_inode);
714 gfs2_glock_put(sdp->sd_rename_gl);
715 gfs2_glock_put(sdp->sd_freeze_gl);
717 if (!sdp->sd_args.ar_spectator) {
718 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
719 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
720 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
721 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
722 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
723 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
724 iput(sdp->sd_sc_inode);
725 iput(sdp->sd_qc_inode);
728 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
729 gfs2_clear_rgrpd(sdp);
730 gfs2_jindex_free(sdp);
731 /* Take apart glock structures and buffer lists */
732 gfs2_gl_hash_clear(sdp);
733 gfs2_delete_debugfs_file(sdp);
734 /* Unmount the locking protocol */
735 gfs2_lm_unmount(sdp);
737 /* At this point, we're through participating in the lockspace */
738 gfs2_sys_fs_del(sdp);
742 * gfs2_sync_fs - sync the filesystem
743 * @sb: the superblock
745 * Flushes the log to disk.
748 static int gfs2_sync_fs(struct super_block *sb, int wait)
750 struct gfs2_sbd *sdp = sb->s_fs_info;
752 gfs2_quota_sync(sb, -1);
754 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
756 return sdp->sd_log_error;
759 void gfs2_freeze_func(struct work_struct *work)
762 struct gfs2_holder freeze_gh;
763 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
764 struct super_block *sb = sdp->sd_vfs;
766 atomic_inc(&sb->s_active);
767 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
768 LM_FLAG_NOEXP | GL_EXACT, &freeze_gh);
770 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
771 gfs2_assert_withdraw(sdp, 0);
773 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
774 error = thaw_super(sb);
776 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
778 gfs2_assert_withdraw(sdp, 0);
780 gfs2_glock_dq_uninit(&freeze_gh);
782 deactivate_super(sb);
783 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
784 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
789 * gfs2_freeze - prevent further writes to the filesystem
790 * @sb: the VFS structure for the filesystem
794 static int gfs2_freeze(struct super_block *sb)
796 struct gfs2_sbd *sdp = sb->s_fs_info;
799 mutex_lock(&sdp->sd_freeze_mutex);
800 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN)
804 if (gfs2_withdrawn(sdp)) {
809 error = gfs2_lock_fs_check_clean(sdp);
814 fs_err(sdp, "waiting for recovery before freeze\n");
815 else if (error == -EIO) {
816 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
817 "to recovery error.\n");
820 fs_err(sdp, "error freezing FS: %d\n", error);
822 fs_err(sdp, "retrying...\n");
825 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
827 mutex_unlock(&sdp->sd_freeze_mutex);
832 * gfs2_unfreeze - reallow writes to the filesystem
833 * @sb: the VFS structure for the filesystem
837 static int gfs2_unfreeze(struct super_block *sb)
839 struct gfs2_sbd *sdp = sb->s_fs_info;
841 mutex_lock(&sdp->sd_freeze_mutex);
842 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
843 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
844 mutex_unlock(&sdp->sd_freeze_mutex);
848 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
849 mutex_unlock(&sdp->sd_freeze_mutex);
850 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
854 * statfs_fill - fill in the sg for a given RG
856 * @sc: the sc structure
858 * Returns: 0 on success, -ESTALE if the LVB is invalid
861 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
862 struct gfs2_statfs_change_host *sc)
864 gfs2_rgrp_verify(rgd);
865 sc->sc_total += rgd->rd_data;
866 sc->sc_free += rgd->rd_free;
867 sc->sc_dinodes += rgd->rd_dinodes;
872 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
873 * @sdp: the filesystem
874 * @sc: the sc info that will be returned
876 * Any error (other than a signal) will cause this routine to fall back
877 * to the synchronous version.
879 * FIXME: This really shouldn't busy wait like this.
884 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
886 struct gfs2_rgrpd *rgd_next;
887 struct gfs2_holder *gha, *gh;
888 unsigned int slots = 64;
893 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
894 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
897 for (x = 0; x < slots; x++)
898 gfs2_holder_mark_uninitialized(gha + x);
900 rgd_next = gfs2_rgrpd_get_first(sdp);
905 for (x = 0; x < slots; x++) {
908 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
909 err = gfs2_glock_wait(gh);
911 gfs2_holder_uninit(gh);
915 struct gfs2_rgrpd *rgd =
916 gfs2_glock2rgrp(gh->gh_gl);
918 error = statfs_slow_fill(rgd, sc);
920 gfs2_glock_dq_uninit(gh);
924 if (gfs2_holder_initialized(gh))
926 else if (rgd_next && !error) {
927 error = gfs2_glock_nq_init(rgd_next->rd_gl,
931 rgd_next = gfs2_rgrpd_get_next(rgd_next);
935 if (signal_pending(current))
936 error = -ERESTARTSYS;
950 * gfs2_statfs_i - Do a statfs
951 * @sdp: the filesystem
952 * @sg: the sg structure
957 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
959 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
960 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
962 spin_lock(&sdp->sd_statfs_spin);
965 sc->sc_total += l_sc->sc_total;
966 sc->sc_free += l_sc->sc_free;
967 sc->sc_dinodes += l_sc->sc_dinodes;
969 spin_unlock(&sdp->sd_statfs_spin);
973 if (sc->sc_free > sc->sc_total)
974 sc->sc_free = sc->sc_total;
975 if (sc->sc_dinodes < 0)
982 * gfs2_statfs - Gather and return stats about the filesystem
983 * @sb: The superblock
984 * @statfsbuf: The buffer
986 * Returns: 0 on success or error code
989 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
991 struct super_block *sb = dentry->d_sb;
992 struct gfs2_sbd *sdp = sb->s_fs_info;
993 struct gfs2_statfs_change_host sc;
996 error = gfs2_rindex_update(sdp);
1000 if (gfs2_tune_get(sdp, gt_statfs_slow))
1001 error = gfs2_statfs_slow(sdp, &sc);
1003 error = gfs2_statfs_i(sdp, &sc);
1008 buf->f_type = GFS2_MAGIC;
1009 buf->f_bsize = sdp->sd_sb.sb_bsize;
1010 buf->f_blocks = sc.sc_total;
1011 buf->f_bfree = sc.sc_free;
1012 buf->f_bavail = sc.sc_free;
1013 buf->f_files = sc.sc_dinodes + sc.sc_free;
1014 buf->f_ffree = sc.sc_free;
1015 buf->f_namelen = GFS2_FNAMESIZE;
1021 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1022 * @inode: The inode to drop
1024 * If we've received a callback on an iopen lock then it's because a
1025 * remote node tried to deallocate the inode but failed due to this node
1026 * still having the inode open. Here we mark the link count zero
1027 * since we know that it must have reached zero if the GLF_DEMOTE flag
1028 * is set on the iopen glock. If we didn't do a disk read since the
1029 * remote node removed the final link then we might otherwise miss
1030 * this event. This check ensures that this node will deallocate the
1031 * inode's blocks, or alternatively pass the baton on to another
1032 * node for later deallocation.
1035 static int gfs2_drop_inode(struct inode *inode)
1037 struct gfs2_inode *ip = GFS2_I(inode);
1039 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1041 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1042 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1043 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1048 * When under memory pressure when an inode's link count has dropped to
1049 * zero, defer deleting the inode to the delete workqueue. This avoids
1050 * calling into DLM under memory pressure, which can deadlock.
1052 if (!inode->i_nlink &&
1053 unlikely(current->flags & PF_MEMALLOC) &&
1054 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1055 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1057 gfs2_glock_hold(gl);
1058 if (!gfs2_queue_delete_work(gl, 0))
1059 gfs2_glock_queue_put(gl);
1063 return generic_drop_inode(inode);
1066 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1072 } while (!IS_ROOT(d1));
1077 * gfs2_show_options - Show mount options for /proc/mounts
1078 * @s: seq_file structure
1079 * @root: root of this (sub)tree
1081 * Returns: 0 on success or error code
1084 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1086 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1087 struct gfs2_args *args = &sdp->sd_args;
1090 if (is_ancestor(root, sdp->sd_master_dir))
1091 seq_puts(s, ",meta");
1092 if (args->ar_lockproto[0])
1093 seq_show_option(s, "lockproto", args->ar_lockproto);
1094 if (args->ar_locktable[0])
1095 seq_show_option(s, "locktable", args->ar_locktable);
1096 if (args->ar_hostdata[0])
1097 seq_show_option(s, "hostdata", args->ar_hostdata);
1098 if (args->ar_spectator)
1099 seq_puts(s, ",spectator");
1100 if (args->ar_localflocks)
1101 seq_puts(s, ",localflocks");
1103 seq_puts(s, ",debug");
1104 if (args->ar_posix_acl)
1105 seq_puts(s, ",acl");
1106 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1108 switch (args->ar_quota) {
1109 case GFS2_QUOTA_OFF:
1112 case GFS2_QUOTA_ACCOUNT:
1122 seq_printf(s, ",quota=%s", state);
1124 if (args->ar_suiddir)
1125 seq_puts(s, ",suiddir");
1126 if (args->ar_data != GFS2_DATA_DEFAULT) {
1128 switch (args->ar_data) {
1129 case GFS2_DATA_WRITEBACK:
1130 state = "writeback";
1132 case GFS2_DATA_ORDERED:
1139 seq_printf(s, ",data=%s", state);
1141 if (args->ar_discard)
1142 seq_puts(s, ",discard");
1143 val = sdp->sd_tune.gt_logd_secs;
1145 seq_printf(s, ",commit=%d", val);
1146 val = sdp->sd_tune.gt_statfs_quantum;
1148 seq_printf(s, ",statfs_quantum=%d", val);
1149 else if (sdp->sd_tune.gt_statfs_slow)
1150 seq_puts(s, ",statfs_quantum=0");
1151 val = sdp->sd_tune.gt_quota_quantum;
1153 seq_printf(s, ",quota_quantum=%d", val);
1154 if (args->ar_statfs_percent)
1155 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1156 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1159 switch (args->ar_errors) {
1160 case GFS2_ERRORS_WITHDRAW:
1163 case GFS2_ERRORS_PANIC:
1170 seq_printf(s, ",errors=%s", state);
1172 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1173 seq_puts(s, ",nobarrier");
1174 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1175 seq_puts(s, ",demote_interface_used");
1176 if (args->ar_rgrplvb)
1177 seq_puts(s, ",rgrplvb");
1178 if (args->ar_loccookie)
1179 seq_puts(s, ",loccookie");
1183 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1185 struct inode *inode = &ip->i_inode;
1186 struct gfs2_glock *gl = ip->i_gl;
1188 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1189 truncate_inode_pages(&inode->i_data, 0);
1191 if (atomic_read(&gl->gl_revokes) == 0) {
1192 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1193 clear_bit(GLF_DIRTY, &gl->gl_flags);
1197 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1199 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1200 struct gfs2_rgrpd *rgd;
1201 struct gfs2_holder gh;
1204 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1205 gfs2_consist_inode(ip);
1209 error = gfs2_rindex_update(sdp);
1213 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1217 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1219 gfs2_consist_inode(ip);
1224 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1228 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1229 sdp->sd_jdesc->jd_blocks);
1231 goto out_rg_gunlock;
1233 gfs2_free_di(rgd, ip);
1235 gfs2_final_release_pages(ip);
1237 gfs2_trans_end(sdp);
1240 gfs2_glock_dq_uninit(&gh);
1242 gfs2_quota_unhold(ip);
1247 * gfs2_glock_put_eventually
1248 * @gl: The glock to put
1250 * When under memory pressure, trigger a deferred glock put to make sure we
1251 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1254 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1256 if (current->flags & PF_MEMALLOC)
1257 gfs2_glock_queue_put(gl);
1262 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1264 struct gfs2_inode *ip = GFS2_I(inode);
1265 struct gfs2_sbd *sdp = GFS2_SB(inode);
1266 struct gfs2_holder *gh = &ip->i_iopen_gh;
1267 long timeout = 5 * HZ;
1270 gh->gh_flags |= GL_NOCACHE;
1271 gfs2_glock_dq_wait(gh);
1274 * If there are no other lock holders, we'll get the lock immediately.
1275 * Otherwise, the other nodes holding the lock will be notified about
1276 * our locking request. If they don't have the inode open, they'll
1277 * evict the cached inode and release the lock. Otherwise, if they
1278 * poke the inode glock, we'll take this as an indication that they
1279 * still need the iopen glock and that they'll take care of deleting
1280 * the inode when they're done. As a last resort, if another node
1281 * keeps holding the iopen glock without showing any activity on the
1282 * inode glock, we'll eventually time out.
1284 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1285 * locking request as an optimization to notify lock holders as soon as
1286 * possible. Without that flag, they'd be notified implicitly by the
1287 * second locking request.
1290 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1291 error = gfs2_glock_nq(gh);
1292 if (error != GLR_TRYFAILED)
1295 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1296 error = gfs2_glock_nq(gh);
1300 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1301 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1302 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1304 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1312 * gfs2_evict_inode - Remove an inode from cache
1313 * @inode: The inode to evict
1315 * There are three cases to consider:
1316 * 1. i_nlink == 0, we are final opener (and must deallocate)
1317 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1320 * If the fs is read only, then we have to treat all cases as per #3
1321 * since we are unable to do any deallocation. The inode will be
1322 * deallocated by the next read/write node to attempt an allocation
1323 * in the same resource group
1325 * We have to (at the moment) hold the inodes main lock to cover
1326 * the gap between unlocking the shared lock on the iopen lock and
1327 * taking the exclusive lock. I'd rather do a shared -> exclusive
1328 * conversion on the iopen lock, but we can change that later. This
1329 * is safe, just less efficient.
1332 static void gfs2_evict_inode(struct inode *inode)
1334 struct super_block *sb = inode->i_sb;
1335 struct gfs2_sbd *sdp = sb->s_fs_info;
1336 struct gfs2_inode *ip = GFS2_I(inode);
1337 struct gfs2_holder gh;
1338 struct address_space *metamapping;
1341 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1346 if (inode->i_nlink || sb_rdonly(sb))
1349 if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1350 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1351 gfs2_holder_mark_uninitialized(&gh);
1355 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1358 /* Deletes should never happen under memory pressure anymore. */
1359 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1362 /* Must not read inode block until block type has been verified */
1363 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1364 if (unlikely(error)) {
1365 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1366 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1367 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1371 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1373 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1377 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1378 error = gfs2_inode_refresh(ip);
1384 * The inode may have been recreated in the meantime.
1390 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1391 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1392 if (!gfs2_upgrade_iopen_glock(inode)) {
1393 gfs2_holder_uninit(&ip->i_iopen_gh);
1398 if (S_ISDIR(inode->i_mode) &&
1399 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1400 error = gfs2_dir_exhash_dealloc(ip);
1406 error = gfs2_ea_dealloc(ip);
1411 if (!gfs2_is_stuffed(ip)) {
1412 error = gfs2_file_dealloc(ip);
1417 /* We're about to clear the bitmap for the dinode, but as soon as we
1418 do, gfs2_create_inode can create another inode at the same block
1419 location and try to set gl_object again. We clear gl_object here so
1420 that subsequent inode creates don't see an old gl_object. */
1421 glock_clear_object(ip->i_gl, ip);
1422 error = gfs2_dinode_dealloc(ip);
1423 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1427 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1428 GFS2_LFC_EVICT_INODE);
1429 metamapping = gfs2_glock2aspace(ip->i_gl);
1430 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1431 filemap_fdatawrite(metamapping);
1432 filemap_fdatawait(metamapping);
1434 write_inode_now(inode, 1);
1435 gfs2_ail_flush(ip->i_gl, 0);
1437 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1440 /* Needs to be done before glock release & also in a transaction */
1441 truncate_inode_pages(&inode->i_data, 0);
1442 truncate_inode_pages(metamapping, 0);
1443 gfs2_trans_end(sdp);
1446 if (gfs2_rs_active(&ip->i_res))
1447 gfs2_rs_deltree(&ip->i_res);
1449 if (gfs2_holder_initialized(&gh)) {
1450 glock_clear_object(ip->i_gl, ip);
1451 gfs2_glock_dq_uninit(&gh);
1453 if (error && error != GLR_TRYFAILED && error != -EROFS)
1454 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1456 truncate_inode_pages_final(&inode->i_data);
1458 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1459 gfs2_rs_delete(ip, NULL);
1460 gfs2_ordered_del_inode(ip);
1462 gfs2_dir_hash_inval(ip);
1464 glock_clear_object(ip->i_gl, ip);
1465 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1466 gfs2_glock_add_to_lru(ip->i_gl);
1467 gfs2_glock_put_eventually(ip->i_gl);
1470 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1471 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1473 glock_clear_object(gl, ip);
1474 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1475 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1476 gfs2_glock_dq(&ip->i_iopen_gh);
1478 gfs2_glock_hold(gl);
1479 gfs2_holder_uninit(&ip->i_iopen_gh);
1480 gfs2_glock_put_eventually(gl);
1484 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1486 struct gfs2_inode *ip;
1488 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1493 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1494 memset(&ip->i_res, 0, sizeof(ip->i_res));
1495 RB_CLEAR_NODE(&ip->i_res.rs_node);
1497 return &ip->i_inode;
1500 static void gfs2_free_inode(struct inode *inode)
1502 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1505 const struct super_operations gfs2_super_ops = {
1506 .alloc_inode = gfs2_alloc_inode,
1507 .free_inode = gfs2_free_inode,
1508 .write_inode = gfs2_write_inode,
1509 .dirty_inode = gfs2_dirty_inode,
1510 .evict_inode = gfs2_evict_inode,
1511 .put_super = gfs2_put_super,
1512 .sync_fs = gfs2_sync_fs,
1513 .freeze_super = gfs2_freeze,
1514 .thaw_super = gfs2_unfreeze,
1515 .statfs = gfs2_statfs,
1516 .drop_inode = gfs2_drop_inode,
1517 .show_options = gfs2_show_options,