2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/bio.h>
11 #include <linux/sched.h>
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.h>
16 #include <linux/statfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/mount.h>
19 #include <linux/kthread.h>
20 #include <linux/delay.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/crc32.h>
23 #include <linux/time.h>
24 #include <linux/wait.h>
25 #include <linux/writeback.h>
26 #include <linux/backing-dev.h>
46 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
86 static const match_table_t tokens = {
87 {Opt_lockproto, "lockproto=%s"},
88 {Opt_locktable, "locktable=%s"},
89 {Opt_hostdata, "hostdata=%s"},
90 {Opt_spectator, "spectator"},
91 {Opt_spectator, "norecovery"},
92 {Opt_ignore_local_fs, "ignore_local_fs"},
93 {Opt_localflocks, "localflocks"},
94 {Opt_localcaching, "localcaching"},
96 {Opt_nodebug, "nodebug"},
97 {Opt_upgrade, "upgrade"},
100 {Opt_quota_off, "quota=off"},
101 {Opt_quota_account, "quota=account"},
102 {Opt_quota_on, "quota=on"},
103 {Opt_quota, "quota"},
104 {Opt_noquota, "noquota"},
105 {Opt_suiddir, "suiddir"},
106 {Opt_nosuiddir, "nosuiddir"},
107 {Opt_data_writeback, "data=writeback"},
108 {Opt_data_ordered, "data=ordered"},
110 {Opt_discard, "discard"},
111 {Opt_nodiscard, "nodiscard"},
112 {Opt_commit, "commit=%d"},
113 {Opt_err_withdraw, "errors=withdraw"},
114 {Opt_err_panic, "errors=panic"},
115 {Opt_statfs_quantum, "statfs_quantum=%d"},
116 {Opt_statfs_percent, "statfs_percent=%d"},
117 {Opt_quota_quantum, "quota_quantum=%d"},
118 {Opt_barrier, "barrier"},
119 {Opt_nobarrier, "nobarrier"},
120 {Opt_rgrplvb, "rgrplvb"},
121 {Opt_norgrplvb, "norgrplvb"},
126 * gfs2_mount_args - Parse mount options
127 * @args: The structure into which the parsed options will be written
128 * @options: The options to parse
133 int gfs2_mount_args(struct gfs2_args *args, char *options)
137 substring_t tmp[MAX_OPT_ARGS];
140 /* Split the options into tokens with the "," character and
144 o = strsep(&options, ",");
150 token = match_token(o, tokens, tmp);
153 match_strlcpy(args->ar_lockproto, &tmp[0],
157 match_strlcpy(args->ar_locktable, &tmp[0],
161 match_strlcpy(args->ar_hostdata, &tmp[0],
165 args->ar_spectator = 1;
167 case Opt_ignore_local_fs:
168 /* Retained for backwards compat only */
170 case Opt_localflocks:
171 args->ar_localflocks = 1;
173 case Opt_localcaching:
174 /* Retained for backwards compat only */
177 if (args->ar_errors == GFS2_ERRORS_PANIC) {
178 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
179 "are mutually exclusive.\n");
188 /* Retained for backwards compat only */
191 args->ar_posix_acl = 1;
194 args->ar_posix_acl = 0;
198 args->ar_quota = GFS2_QUOTA_OFF;
200 case Opt_quota_account:
201 args->ar_quota = GFS2_QUOTA_ACCOUNT;
205 args->ar_quota = GFS2_QUOTA_ON;
208 args->ar_suiddir = 1;
211 args->ar_suiddir = 0;
213 case Opt_data_writeback:
214 args->ar_data = GFS2_DATA_WRITEBACK;
216 case Opt_data_ordered:
217 args->ar_data = GFS2_DATA_ORDERED;
223 args->ar_discard = 1;
226 args->ar_discard = 0;
229 rv = match_int(&tmp[0], &args->ar_commit);
230 if (rv || args->ar_commit <= 0) {
231 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
232 return rv ? rv : -EINVAL;
235 case Opt_statfs_quantum:
236 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
237 if (rv || args->ar_statfs_quantum < 0) {
238 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
239 return rv ? rv : -EINVAL;
242 case Opt_quota_quantum:
243 rv = match_int(&tmp[0], &args->ar_quota_quantum);
244 if (rv || args->ar_quota_quantum <= 0) {
245 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
246 return rv ? rv : -EINVAL;
249 case Opt_statfs_percent:
250 rv = match_int(&tmp[0], &args->ar_statfs_percent);
251 if (rv || args->ar_statfs_percent < 0 ||
252 args->ar_statfs_percent > 100) {
253 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
254 return rv ? rv : -EINVAL;
257 case Opt_err_withdraw:
258 args->ar_errors = GFS2_ERRORS_WITHDRAW;
261 if (args->ar_debug) {
262 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
263 "are mutually exclusive.\n");
266 args->ar_errors = GFS2_ERRORS_PANIC;
269 args->ar_nobarrier = 0;
272 args->ar_nobarrier = 1;
275 args->ar_rgrplvb = 1;
278 args->ar_rgrplvb = 0;
282 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
291 * gfs2_jindex_free - Clear all the journal index information
292 * @sdp: The GFS2 superblock
296 void gfs2_jindex_free(struct gfs2_sbd *sdp)
298 struct list_head list, *head;
299 struct gfs2_jdesc *jd;
300 struct gfs2_journal_extent *jext;
302 spin_lock(&sdp->sd_jindex_spin);
303 list_add(&list, &sdp->sd_jindex_list);
304 list_del_init(&sdp->sd_jindex_list);
305 sdp->sd_journals = 0;
306 spin_unlock(&sdp->sd_jindex_spin);
308 while (!list_empty(&list)) {
309 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
310 head = &jd->extent_list;
311 while (!list_empty(head)) {
312 jext = list_entry(head->next,
313 struct gfs2_journal_extent,
315 list_del(&jext->extent_list);
318 list_del(&jd->jd_list);
324 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
326 struct gfs2_jdesc *jd;
329 list_for_each_entry(jd, head, jd_list) {
330 if (jd->jd_jid == jid) {
342 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
344 struct gfs2_jdesc *jd;
346 spin_lock(&sdp->sd_jindex_spin);
347 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
348 spin_unlock(&sdp->sd_jindex_spin);
353 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
355 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
356 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
357 u64 size = i_size_read(jd->jd_inode);
359 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
362 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
364 if (gfs2_write_alloc_required(ip, 0, size)) {
365 gfs2_consist_inode(ip);
373 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
374 * @sdp: the filesystem
379 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
381 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
382 struct gfs2_glock *j_gl = ip->i_gl;
383 struct gfs2_holder t_gh;
384 struct gfs2_log_header_host head;
387 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
391 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
393 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
397 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
403 /* Initialize some head of the log stuff */
404 sdp->sd_log_sequence = head.lh_sequence + 1;
405 gfs2_log_pointers_init(sdp, head.lh_blkno);
407 error = gfs2_quota_init(sdp);
411 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
413 gfs2_glock_dq_uninit(&t_gh);
418 t_gh.gh_flags |= GL_NOCACHE;
419 gfs2_glock_dq_uninit(&t_gh);
424 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
426 const struct gfs2_statfs_change *str = buf;
428 sc->sc_total = be64_to_cpu(str->sc_total);
429 sc->sc_free = be64_to_cpu(str->sc_free);
430 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
433 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
435 struct gfs2_statfs_change *str = buf;
437 str->sc_total = cpu_to_be64(sc->sc_total);
438 str->sc_free = cpu_to_be64(sc->sc_free);
439 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
442 int gfs2_statfs_init(struct gfs2_sbd *sdp)
444 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
445 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
446 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
447 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
448 struct buffer_head *m_bh, *l_bh;
449 struct gfs2_holder gh;
452 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
457 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
461 if (sdp->sd_args.ar_spectator) {
462 spin_lock(&sdp->sd_statfs_spin);
463 gfs2_statfs_change_in(m_sc, m_bh->b_data +
464 sizeof(struct gfs2_dinode));
465 spin_unlock(&sdp->sd_statfs_spin);
467 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
471 spin_lock(&sdp->sd_statfs_spin);
472 gfs2_statfs_change_in(m_sc, m_bh->b_data +
473 sizeof(struct gfs2_dinode));
474 gfs2_statfs_change_in(l_sc, l_bh->b_data +
475 sizeof(struct gfs2_dinode));
476 spin_unlock(&sdp->sd_statfs_spin);
484 gfs2_glock_dq_uninit(&gh);
488 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
491 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
492 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
493 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
494 struct buffer_head *l_bh;
499 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
503 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
505 spin_lock(&sdp->sd_statfs_spin);
506 l_sc->sc_total += total;
507 l_sc->sc_free += free;
508 l_sc->sc_dinodes += dinodes;
509 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
510 if (sdp->sd_args.ar_statfs_percent) {
511 x = 100 * l_sc->sc_free;
512 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
513 if (x >= y || x <= -y)
516 spin_unlock(&sdp->sd_statfs_spin);
520 gfs2_wake_up_statfs(sdp);
523 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
524 struct buffer_head *l_bh)
526 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
527 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
528 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
529 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
531 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
533 spin_lock(&sdp->sd_statfs_spin);
534 m_sc->sc_total += l_sc->sc_total;
535 m_sc->sc_free += l_sc->sc_free;
536 m_sc->sc_dinodes += l_sc->sc_dinodes;
537 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
538 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
539 0, sizeof(struct gfs2_statfs_change));
540 spin_unlock(&sdp->sd_statfs_spin);
542 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
543 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
546 int gfs2_statfs_sync(struct super_block *sb, int type)
548 struct gfs2_sbd *sdp = sb->s_fs_info;
549 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
550 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
551 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
552 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
553 struct gfs2_holder gh;
554 struct buffer_head *m_bh, *l_bh;
557 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
562 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
566 spin_lock(&sdp->sd_statfs_spin);
567 gfs2_statfs_change_in(m_sc, m_bh->b_data +
568 sizeof(struct gfs2_dinode));
569 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
570 spin_unlock(&sdp->sd_statfs_spin);
573 spin_unlock(&sdp->sd_statfs_spin);
575 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
579 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
583 update_statfs(sdp, m_bh, l_bh);
584 sdp->sd_statfs_force_sync = 0;
593 gfs2_glock_dq_uninit(&gh);
598 struct list_head list;
599 struct gfs2_holder gh;
603 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
605 * @sdp: the file system
606 * @state: the state to put the transaction lock into
607 * @t_gh: the hold on the transaction lock
612 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
613 struct gfs2_holder *t_gh)
615 struct gfs2_inode *ip;
616 struct gfs2_jdesc *jd;
619 struct gfs2_log_header_host lh;
622 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
623 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
628 ip = GFS2_I(jd->jd_inode);
629 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
634 list_add(&lfcc->list, &list);
637 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
640 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
641 error = gfs2_jdesc_check(jd);
644 error = gfs2_find_jhead(jd, &lh);
647 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
654 gfs2_glock_dq_uninit(t_gh);
657 while (!list_empty(&list)) {
658 lfcc = list_entry(list.next, struct lfcc, list);
659 list_del(&lfcc->list);
660 gfs2_glock_dq_uninit(&lfcc->gh);
667 * gfs2_freeze_fs - freezes the file system
668 * @sdp: the file system
670 * This function flushes data and meta data for all machines by
671 * acquiring the transaction log exclusively. All journals are
672 * ensured to be in a clean state as well.
677 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
681 mutex_lock(&sdp->sd_freeze_lock);
683 if (!sdp->sd_freeze_count++) {
684 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
686 sdp->sd_freeze_count--;
689 mutex_unlock(&sdp->sd_freeze_lock);
695 * gfs2_unfreeze_fs - unfreezes the file system
696 * @sdp: the file system
698 * This function allows the file system to proceed by unlocking
699 * the exclusively held transaction lock. Other GFS2 nodes are
700 * now free to acquire the lock shared and go on with their lives.
704 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
706 mutex_lock(&sdp->sd_freeze_lock);
708 if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
709 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
711 mutex_unlock(&sdp->sd_freeze_lock);
714 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
716 struct gfs2_dinode *str = buf;
718 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
719 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
720 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
721 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
722 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
723 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
724 str->di_uid = cpu_to_be32(ip->i_inode.i_uid);
725 str->di_gid = cpu_to_be32(ip->i_inode.i_gid);
726 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
727 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
728 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
729 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
730 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
731 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
733 str->di_goal_meta = cpu_to_be64(ip->i_goal);
734 str->di_goal_data = cpu_to_be64(ip->i_goal);
735 str->di_generation = cpu_to_be64(ip->i_generation);
737 str->di_flags = cpu_to_be32(ip->i_diskflags);
738 str->di_height = cpu_to_be16(ip->i_height);
739 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
740 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
742 str->di_depth = cpu_to_be16(ip->i_depth);
743 str->di_entries = cpu_to_be32(ip->i_entries);
745 str->di_eattr = cpu_to_be64(ip->i_eattr);
746 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
747 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
748 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
752 * gfs2_write_inode - Make sure the inode is stable on the disk
754 * @wbc: The writeback control structure
759 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
761 struct gfs2_inode *ip = GFS2_I(inode);
762 struct gfs2_sbd *sdp = GFS2_SB(inode);
763 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
764 struct backing_dev_info *bdi = metamapping->backing_dev_info;
767 if (wbc->sync_mode == WB_SYNC_ALL)
768 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
769 if (bdi->dirty_exceeded)
770 gfs2_ail1_flush(sdp, wbc);
772 filemap_fdatawrite(metamapping);
773 if (wbc->sync_mode == WB_SYNC_ALL)
774 ret = filemap_fdatawait(metamapping);
776 mark_inode_dirty_sync(inode);
781 * gfs2_dirty_inode - check for atime updates
782 * @inode: The inode in question
783 * @flags: The type of dirty
785 * Unfortunately it can be called under any combination of inode
786 * glock and transaction lock, so we have to check carefully.
788 * At the moment this deals only with atime - it should be possible
789 * to expand that role in future, once a review of the locking has
793 static void gfs2_dirty_inode(struct inode *inode, int flags)
795 struct gfs2_inode *ip = GFS2_I(inode);
796 struct gfs2_sbd *sdp = GFS2_SB(inode);
797 struct buffer_head *bh;
798 struct gfs2_holder gh;
800 int need_endtrans = 0;
803 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
806 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
807 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
809 fs_err(sdp, "dirty_inode: glock %d\n", ret);
813 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
816 if (current->journal_info == NULL) {
817 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
819 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
825 ret = gfs2_meta_inode_buffer(ip, &bh);
827 gfs2_trans_add_bh(ip->i_gl, bh, 1);
828 gfs2_dinode_out(ip, bh->b_data);
836 gfs2_glock_dq_uninit(&gh);
840 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
841 * @sdp: the filesystem
846 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
848 struct gfs2_holder t_gh;
851 flush_workqueue(gfs2_delete_workqueue);
852 gfs2_quota_sync(sdp->sd_vfs, 0);
853 gfs2_statfs_sync(sdp->sd_vfs, 0);
855 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
857 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
860 gfs2_meta_syncfs(sdp);
861 gfs2_log_shutdown(sdp);
863 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
866 gfs2_glock_dq_uninit(&t_gh);
868 gfs2_quota_cleanup(sdp);
873 static int gfs2_umount_recovery_wait(void *word)
880 * gfs2_put_super - Unmount the filesystem
881 * @sb: The VFS superblock
885 static void gfs2_put_super(struct super_block *sb)
887 struct gfs2_sbd *sdp = sb->s_fs_info;
889 struct gfs2_jdesc *jd;
891 /* Unfreeze the filesystem, if we need to */
893 mutex_lock(&sdp->sd_freeze_lock);
894 if (sdp->sd_freeze_count)
895 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
896 mutex_unlock(&sdp->sd_freeze_lock);
898 /* No more recovery requests */
899 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
902 /* Wait on outstanding recovery */
904 spin_lock(&sdp->sd_jindex_spin);
905 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
906 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
908 spin_unlock(&sdp->sd_jindex_spin);
909 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
910 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
913 spin_unlock(&sdp->sd_jindex_spin);
915 kthread_stop(sdp->sd_quotad_process);
916 kthread_stop(sdp->sd_logd_process);
918 if (!(sb->s_flags & MS_RDONLY)) {
919 error = gfs2_make_fs_ro(sdp);
923 /* At this point, we're through modifying the disk */
927 iput(sdp->sd_jindex);
928 iput(sdp->sd_statfs_inode);
929 iput(sdp->sd_rindex);
930 iput(sdp->sd_quota_inode);
932 gfs2_glock_put(sdp->sd_rename_gl);
933 gfs2_glock_put(sdp->sd_trans_gl);
935 if (!sdp->sd_args.ar_spectator) {
936 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
937 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
938 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
939 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
940 iput(sdp->sd_sc_inode);
941 iput(sdp->sd_qc_inode);
944 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
945 gfs2_clear_rgrpd(sdp);
946 gfs2_jindex_free(sdp);
947 /* Take apart glock structures and buffer lists */
948 gfs2_gl_hash_clear(sdp);
949 /* Unmount the locking protocol */
950 gfs2_lm_unmount(sdp);
952 /* At this point, we're through participating in the lockspace */
953 gfs2_sys_fs_del(sdp);
957 * gfs2_sync_fs - sync the filesystem
958 * @sb: the superblock
960 * Flushes the log to disk.
963 static int gfs2_sync_fs(struct super_block *sb, int wait)
965 struct gfs2_sbd *sdp = sb->s_fs_info;
967 gfs2_quota_sync(sb, -1);
969 gfs2_log_flush(sdp, NULL);
974 * gfs2_freeze - prevent further writes to the filesystem
975 * @sb: the VFS structure for the filesystem
979 static int gfs2_freeze(struct super_block *sb)
981 struct gfs2_sbd *sdp = sb->s_fs_info;
984 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
988 error = gfs2_freeze_fs(sdp);
994 fs_err(sdp, "waiting for recovery before freeze\n");
998 fs_err(sdp, "error freezing FS: %d\n", error);
1002 fs_err(sdp, "retrying...\n");
1009 * gfs2_unfreeze - reallow writes to the filesystem
1010 * @sb: the VFS structure for the filesystem
1014 static int gfs2_unfreeze(struct super_block *sb)
1016 gfs2_unfreeze_fs(sb->s_fs_info);
1021 * statfs_fill - fill in the sg for a given RG
1023 * @sc: the sc structure
1025 * Returns: 0 on success, -ESTALE if the LVB is invalid
1028 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1029 struct gfs2_statfs_change_host *sc)
1031 gfs2_rgrp_verify(rgd);
1032 sc->sc_total += rgd->rd_data;
1033 sc->sc_free += rgd->rd_free;
1034 sc->sc_dinodes += rgd->rd_dinodes;
1039 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1040 * @sdp: the filesystem
1041 * @sc: the sc info that will be returned
1043 * Any error (other than a signal) will cause this routine to fall back
1044 * to the synchronous version.
1046 * FIXME: This really shouldn't busy wait like this.
1051 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1053 struct gfs2_rgrpd *rgd_next;
1054 struct gfs2_holder *gha, *gh;
1055 unsigned int slots = 64;
1060 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1061 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1065 rgd_next = gfs2_rgrpd_get_first(sdp);
1070 for (x = 0; x < slots; x++) {
1073 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1074 err = gfs2_glock_wait(gh);
1076 gfs2_holder_uninit(gh);
1080 error = statfs_slow_fill(
1081 gh->gh_gl->gl_object, sc);
1082 gfs2_glock_dq_uninit(gh);
1088 else if (rgd_next && !error) {
1089 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1093 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1097 if (signal_pending(current))
1098 error = -ERESTARTSYS;
1112 * gfs2_statfs_i - Do a statfs
1113 * @sdp: the filesystem
1114 * @sg: the sg structure
1119 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1121 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1122 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1124 spin_lock(&sdp->sd_statfs_spin);
1127 sc->sc_total += l_sc->sc_total;
1128 sc->sc_free += l_sc->sc_free;
1129 sc->sc_dinodes += l_sc->sc_dinodes;
1131 spin_unlock(&sdp->sd_statfs_spin);
1133 if (sc->sc_free < 0)
1135 if (sc->sc_free > sc->sc_total)
1136 sc->sc_free = sc->sc_total;
1137 if (sc->sc_dinodes < 0)
1144 * gfs2_statfs - Gather and return stats about the filesystem
1145 * @sb: The superblock
1146 * @statfsbuf: The buffer
1148 * Returns: 0 on success or error code
1151 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1153 struct super_block *sb = dentry->d_inode->i_sb;
1154 struct gfs2_sbd *sdp = sb->s_fs_info;
1155 struct gfs2_statfs_change_host sc;
1158 error = gfs2_rindex_update(sdp);
1162 if (gfs2_tune_get(sdp, gt_statfs_slow))
1163 error = gfs2_statfs_slow(sdp, &sc);
1165 error = gfs2_statfs_i(sdp, &sc);
1170 buf->f_type = GFS2_MAGIC;
1171 buf->f_bsize = sdp->sd_sb.sb_bsize;
1172 buf->f_blocks = sc.sc_total;
1173 buf->f_bfree = sc.sc_free;
1174 buf->f_bavail = sc.sc_free;
1175 buf->f_files = sc.sc_dinodes + sc.sc_free;
1176 buf->f_ffree = sc.sc_free;
1177 buf->f_namelen = GFS2_FNAMESIZE;
1183 * gfs2_remount_fs - called when the FS is remounted
1184 * @sb: the filesystem
1185 * @flags: the remount flags
1186 * @data: extra data passed in (not used right now)
1191 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1193 struct gfs2_sbd *sdp = sb->s_fs_info;
1194 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1195 struct gfs2_tune *gt = &sdp->sd_tune;
1198 spin_lock(>->gt_spin);
1199 args.ar_commit = gt->gt_logd_secs;
1200 args.ar_quota_quantum = gt->gt_quota_quantum;
1201 if (gt->gt_statfs_slow)
1202 args.ar_statfs_quantum = 0;
1204 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1205 spin_unlock(>->gt_spin);
1206 error = gfs2_mount_args(&args, data);
1210 /* Not allowed to change locking details */
1211 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1212 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1213 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1216 /* Some flags must not be changed */
1217 if (args_neq(&args, &sdp->sd_args, spectator) ||
1218 args_neq(&args, &sdp->sd_args, localflocks) ||
1219 args_neq(&args, &sdp->sd_args, meta))
1222 if (sdp->sd_args.ar_spectator)
1223 *flags |= MS_RDONLY;
1225 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1226 if (*flags & MS_RDONLY)
1227 error = gfs2_make_fs_ro(sdp);
1229 error = gfs2_make_fs_rw(sdp);
1234 sdp->sd_args = args;
1235 if (sdp->sd_args.ar_posix_acl)
1236 sb->s_flags |= MS_POSIXACL;
1238 sb->s_flags &= ~MS_POSIXACL;
1239 if (sdp->sd_args.ar_nobarrier)
1240 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1242 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1243 spin_lock(>->gt_spin);
1244 gt->gt_logd_secs = args.ar_commit;
1245 gt->gt_quota_quantum = args.ar_quota_quantum;
1246 if (args.ar_statfs_quantum) {
1247 gt->gt_statfs_slow = 0;
1248 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1251 gt->gt_statfs_slow = 1;
1252 gt->gt_statfs_quantum = 30;
1254 spin_unlock(>->gt_spin);
1256 gfs2_online_uevent(sdp);
1261 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1262 * @inode: The inode to drop
1264 * If we've received a callback on an iopen lock then its because a
1265 * remote node tried to deallocate the inode but failed due to this node
1266 * still having the inode open. Here we mark the link count zero
1267 * since we know that it must have reached zero if the GLF_DEMOTE flag
1268 * is set on the iopen glock. If we didn't do a disk read since the
1269 * remote node removed the final link then we might otherwise miss
1270 * this event. This check ensures that this node will deallocate the
1271 * inode's blocks, or alternatively pass the baton on to another
1272 * node for later deallocation.
1275 static int gfs2_drop_inode(struct inode *inode)
1277 struct gfs2_inode *ip = GFS2_I(inode);
1279 if (inode->i_nlink) {
1280 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1281 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1284 return generic_drop_inode(inode);
1287 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1293 } while (!IS_ROOT(d1));
1298 * gfs2_show_options - Show mount options for /proc/mounts
1299 * @s: seq_file structure
1300 * @root: root of this (sub)tree
1302 * Returns: 0 on success or error code
1305 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1307 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1308 struct gfs2_args *args = &sdp->sd_args;
1311 if (is_ancestor(root, sdp->sd_master_dir))
1312 seq_printf(s, ",meta");
1313 if (args->ar_lockproto[0])
1314 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1315 if (args->ar_locktable[0])
1316 seq_printf(s, ",locktable=%s", args->ar_locktable);
1317 if (args->ar_hostdata[0])
1318 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1319 if (args->ar_spectator)
1320 seq_printf(s, ",spectator");
1321 if (args->ar_localflocks)
1322 seq_printf(s, ",localflocks");
1324 seq_printf(s, ",debug");
1325 if (args->ar_posix_acl)
1326 seq_printf(s, ",acl");
1327 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1329 switch (args->ar_quota) {
1330 case GFS2_QUOTA_OFF:
1333 case GFS2_QUOTA_ACCOUNT:
1343 seq_printf(s, ",quota=%s", state);
1345 if (args->ar_suiddir)
1346 seq_printf(s, ",suiddir");
1347 if (args->ar_data != GFS2_DATA_DEFAULT) {
1349 switch (args->ar_data) {
1350 case GFS2_DATA_WRITEBACK:
1351 state = "writeback";
1353 case GFS2_DATA_ORDERED:
1360 seq_printf(s, ",data=%s", state);
1362 if (args->ar_discard)
1363 seq_printf(s, ",discard");
1364 val = sdp->sd_tune.gt_logd_secs;
1366 seq_printf(s, ",commit=%d", val);
1367 val = sdp->sd_tune.gt_statfs_quantum;
1369 seq_printf(s, ",statfs_quantum=%d", val);
1370 else if (sdp->sd_tune.gt_statfs_slow)
1371 seq_puts(s, ",statfs_quantum=0");
1372 val = sdp->sd_tune.gt_quota_quantum;
1374 seq_printf(s, ",quota_quantum=%d", val);
1375 if (args->ar_statfs_percent)
1376 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1377 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1380 switch (args->ar_errors) {
1381 case GFS2_ERRORS_WITHDRAW:
1384 case GFS2_ERRORS_PANIC:
1391 seq_printf(s, ",errors=%s", state);
1393 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1394 seq_printf(s, ",nobarrier");
1395 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1396 seq_printf(s, ",demote_interface_used");
1397 if (args->ar_rgrplvb)
1398 seq_printf(s, ",rgrplvb");
1402 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1404 struct inode *inode = &ip->i_inode;
1405 struct gfs2_glock *gl = ip->i_gl;
1407 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1408 truncate_inode_pages(&inode->i_data, 0);
1410 if (atomic_read(&gl->gl_revokes) == 0) {
1411 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1412 clear_bit(GLF_DIRTY, &gl->gl_flags);
1416 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1418 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1419 struct gfs2_rgrpd *rgd;
1420 struct gfs2_holder gh;
1423 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1424 gfs2_consist_inode(ip);
1428 error = gfs2_rindex_update(sdp);
1432 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1436 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1438 gfs2_consist_inode(ip);
1443 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1447 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1448 sdp->sd_jdesc->jd_blocks);
1450 goto out_rg_gunlock;
1452 gfs2_free_di(rgd, ip);
1454 gfs2_final_release_pages(ip);
1456 gfs2_trans_end(sdp);
1459 gfs2_glock_dq_uninit(&gh);
1461 gfs2_quota_unhold(ip);
1466 * gfs2_evict_inode - Remove an inode from cache
1467 * @inode: The inode to evict
1469 * There are three cases to consider:
1470 * 1. i_nlink == 0, we are final opener (and must deallocate)
1471 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1474 * If the fs is read only, then we have to treat all cases as per #3
1475 * since we are unable to do any deallocation. The inode will be
1476 * deallocated by the next read/write node to attempt an allocation
1477 * in the same resource group
1479 * We have to (at the moment) hold the inodes main lock to cover
1480 * the gap between unlocking the shared lock on the iopen lock and
1481 * taking the exclusive lock. I'd rather do a shared -> exclusive
1482 * conversion on the iopen lock, but we can change that later. This
1483 * is safe, just less efficient.
1486 static void gfs2_evict_inode(struct inode *inode)
1488 struct super_block *sb = inode->i_sb;
1489 struct gfs2_sbd *sdp = sb->s_fs_info;
1490 struct gfs2_inode *ip = GFS2_I(inode);
1491 struct gfs2_holder gh;
1494 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1497 /* Must not read inode block until block type has been verified */
1498 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1499 if (unlikely(error)) {
1500 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1504 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1505 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1510 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1511 error = gfs2_inode_refresh(ip);
1516 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1517 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1518 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1519 error = gfs2_glock_nq(&ip->i_iopen_gh);
1523 /* Case 1 starts here */
1525 if (S_ISDIR(inode->i_mode) &&
1526 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1527 error = gfs2_dir_exhash_dealloc(ip);
1533 error = gfs2_ea_dealloc(ip);
1538 if (!gfs2_is_stuffed(ip)) {
1539 error = gfs2_file_dealloc(ip);
1544 error = gfs2_dinode_dealloc(ip);
1548 gfs2_log_flush(sdp, ip->i_gl);
1549 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1550 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1551 filemap_fdatawrite(metamapping);
1552 filemap_fdatawait(metamapping);
1554 write_inode_now(inode, 1);
1555 gfs2_ail_flush(ip->i_gl, 0);
1557 /* Case 2 starts here */
1558 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1561 /* Needs to be done before glock release & also in a transaction */
1562 truncate_inode_pages(&inode->i_data, 0);
1563 gfs2_trans_end(sdp);
1566 /* Error path for case 1 */
1567 if (gfs2_rs_active(ip->i_res))
1568 gfs2_rs_deltree(ip, ip->i_res);
1570 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1571 gfs2_glock_dq(&ip->i_iopen_gh);
1572 gfs2_holder_uninit(&ip->i_iopen_gh);
1573 gfs2_glock_dq_uninit(&gh);
1574 if (error && error != GLR_TRYFAILED && error != -EROFS)
1575 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1577 /* Case 3 starts here */
1578 truncate_inode_pages(&inode->i_data, 0);
1581 gfs2_dir_hash_inval(ip);
1582 ip->i_gl->gl_object = NULL;
1583 flush_delayed_work(&ip->i_gl->gl_work);
1584 gfs2_glock_add_to_lru(ip->i_gl);
1585 gfs2_glock_put(ip->i_gl);
1587 if (ip->i_iopen_gh.gh_gl) {
1588 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1589 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1593 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1595 struct gfs2_inode *ip;
1597 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1604 return &ip->i_inode;
1607 static void gfs2_i_callback(struct rcu_head *head)
1609 struct inode *inode = container_of(head, struct inode, i_rcu);
1610 kmem_cache_free(gfs2_inode_cachep, inode);
1613 static void gfs2_destroy_inode(struct inode *inode)
1615 call_rcu(&inode->i_rcu, gfs2_i_callback);
1618 const struct super_operations gfs2_super_ops = {
1619 .alloc_inode = gfs2_alloc_inode,
1620 .destroy_inode = gfs2_destroy_inode,
1621 .write_inode = gfs2_write_inode,
1622 .dirty_inode = gfs2_dirty_inode,
1623 .evict_inode = gfs2_evict_inode,
1624 .put_super = gfs2_put_super,
1625 .sync_fs = gfs2_sync_fs,
1626 .freeze_fs = gfs2_freeze,
1627 .unfreeze_fs = gfs2_unfreeze,
1628 .statfs = gfs2_statfs,
1629 .remount_fs = gfs2_remount_fs,
1630 .drop_inode = gfs2_drop_inode,
1631 .show_options = gfs2_show_options,