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)
84 static const match_table_t tokens = {
85 {Opt_lockproto, "lockproto=%s"},
86 {Opt_locktable, "locktable=%s"},
87 {Opt_hostdata, "hostdata=%s"},
88 {Opt_spectator, "spectator"},
89 {Opt_spectator, "norecovery"},
90 {Opt_ignore_local_fs, "ignore_local_fs"},
91 {Opt_localflocks, "localflocks"},
92 {Opt_localcaching, "localcaching"},
94 {Opt_nodebug, "nodebug"},
95 {Opt_upgrade, "upgrade"},
98 {Opt_quota_off, "quota=off"},
99 {Opt_quota_account, "quota=account"},
100 {Opt_quota_on, "quota=on"},
101 {Opt_quota, "quota"},
102 {Opt_noquota, "noquota"},
103 {Opt_suiddir, "suiddir"},
104 {Opt_nosuiddir, "nosuiddir"},
105 {Opt_data_writeback, "data=writeback"},
106 {Opt_data_ordered, "data=ordered"},
108 {Opt_discard, "discard"},
109 {Opt_nodiscard, "nodiscard"},
110 {Opt_commit, "commit=%d"},
111 {Opt_err_withdraw, "errors=withdraw"},
112 {Opt_err_panic, "errors=panic"},
113 {Opt_statfs_quantum, "statfs_quantum=%d"},
114 {Opt_statfs_percent, "statfs_percent=%d"},
115 {Opt_quota_quantum, "quota_quantum=%d"},
116 {Opt_barrier, "barrier"},
117 {Opt_nobarrier, "nobarrier"},
122 * gfs2_mount_args - Parse mount options
123 * @args: The structure into which the parsed options will be written
124 * @options: The options to parse
129 int gfs2_mount_args(struct gfs2_args *args, char *options)
133 substring_t tmp[MAX_OPT_ARGS];
136 /* Split the options into tokens with the "," character and
140 o = strsep(&options, ",");
146 token = match_token(o, tokens, tmp);
149 match_strlcpy(args->ar_lockproto, &tmp[0],
153 match_strlcpy(args->ar_locktable, &tmp[0],
157 match_strlcpy(args->ar_hostdata, &tmp[0],
161 args->ar_spectator = 1;
163 case Opt_ignore_local_fs:
164 /* Retained for backwards compat only */
166 case Opt_localflocks:
167 args->ar_localflocks = 1;
169 case Opt_localcaching:
170 /* Retained for backwards compat only */
173 if (args->ar_errors == GFS2_ERRORS_PANIC) {
174 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
175 "are mutually exclusive.\n");
184 /* Retained for backwards compat only */
187 args->ar_posix_acl = 1;
190 args->ar_posix_acl = 0;
194 args->ar_quota = GFS2_QUOTA_OFF;
196 case Opt_quota_account:
197 args->ar_quota = GFS2_QUOTA_ACCOUNT;
201 args->ar_quota = GFS2_QUOTA_ON;
204 args->ar_suiddir = 1;
207 args->ar_suiddir = 0;
209 case Opt_data_writeback:
210 args->ar_data = GFS2_DATA_WRITEBACK;
212 case Opt_data_ordered:
213 args->ar_data = GFS2_DATA_ORDERED;
219 args->ar_discard = 1;
222 args->ar_discard = 0;
225 rv = match_int(&tmp[0], &args->ar_commit);
226 if (rv || args->ar_commit <= 0) {
227 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
228 return rv ? rv : -EINVAL;
231 case Opt_statfs_quantum:
232 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
233 if (rv || args->ar_statfs_quantum < 0) {
234 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
235 return rv ? rv : -EINVAL;
238 case Opt_quota_quantum:
239 rv = match_int(&tmp[0], &args->ar_quota_quantum);
240 if (rv || args->ar_quota_quantum <= 0) {
241 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
242 return rv ? rv : -EINVAL;
245 case Opt_statfs_percent:
246 rv = match_int(&tmp[0], &args->ar_statfs_percent);
247 if (rv || args->ar_statfs_percent < 0 ||
248 args->ar_statfs_percent > 100) {
249 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
250 return rv ? rv : -EINVAL;
253 case Opt_err_withdraw:
254 args->ar_errors = GFS2_ERRORS_WITHDRAW;
257 if (args->ar_debug) {
258 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
259 "are mutually exclusive.\n");
262 args->ar_errors = GFS2_ERRORS_PANIC;
265 args->ar_nobarrier = 0;
268 args->ar_nobarrier = 1;
272 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
281 * gfs2_jindex_free - Clear all the journal index information
282 * @sdp: The GFS2 superblock
286 void gfs2_jindex_free(struct gfs2_sbd *sdp)
288 struct list_head list, *head;
289 struct gfs2_jdesc *jd;
290 struct gfs2_journal_extent *jext;
292 spin_lock(&sdp->sd_jindex_spin);
293 list_add(&list, &sdp->sd_jindex_list);
294 list_del_init(&sdp->sd_jindex_list);
295 sdp->sd_journals = 0;
296 spin_unlock(&sdp->sd_jindex_spin);
298 while (!list_empty(&list)) {
299 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
300 head = &jd->extent_list;
301 while (!list_empty(head)) {
302 jext = list_entry(head->next,
303 struct gfs2_journal_extent,
305 list_del(&jext->extent_list);
308 list_del(&jd->jd_list);
314 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
316 struct gfs2_jdesc *jd;
319 list_for_each_entry(jd, head, jd_list) {
320 if (jd->jd_jid == jid) {
332 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
334 struct gfs2_jdesc *jd;
336 spin_lock(&sdp->sd_jindex_spin);
337 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
338 spin_unlock(&sdp->sd_jindex_spin);
343 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
345 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
346 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
347 u64 size = i_size_read(jd->jd_inode);
349 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
352 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
354 if (gfs2_write_alloc_required(ip, 0, size)) {
355 gfs2_consist_inode(ip);
363 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
364 * @sdp: the filesystem
369 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
371 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
372 struct gfs2_glock *j_gl = ip->i_gl;
373 struct gfs2_holder t_gh;
374 struct gfs2_log_header_host head;
377 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
381 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
383 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
387 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
393 /* Initialize some head of the log stuff */
394 sdp->sd_log_sequence = head.lh_sequence + 1;
395 gfs2_log_pointers_init(sdp, head.lh_blkno);
397 error = gfs2_quota_init(sdp);
401 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
403 gfs2_glock_dq_uninit(&t_gh);
408 t_gh.gh_flags |= GL_NOCACHE;
409 gfs2_glock_dq_uninit(&t_gh);
414 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
416 const struct gfs2_statfs_change *str = buf;
418 sc->sc_total = be64_to_cpu(str->sc_total);
419 sc->sc_free = be64_to_cpu(str->sc_free);
420 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
423 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
425 struct gfs2_statfs_change *str = buf;
427 str->sc_total = cpu_to_be64(sc->sc_total);
428 str->sc_free = cpu_to_be64(sc->sc_free);
429 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
432 int gfs2_statfs_init(struct gfs2_sbd *sdp)
434 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
435 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
436 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
437 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
438 struct buffer_head *m_bh, *l_bh;
439 struct gfs2_holder gh;
442 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
447 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
451 if (sdp->sd_args.ar_spectator) {
452 spin_lock(&sdp->sd_statfs_spin);
453 gfs2_statfs_change_in(m_sc, m_bh->b_data +
454 sizeof(struct gfs2_dinode));
455 spin_unlock(&sdp->sd_statfs_spin);
457 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
461 spin_lock(&sdp->sd_statfs_spin);
462 gfs2_statfs_change_in(m_sc, m_bh->b_data +
463 sizeof(struct gfs2_dinode));
464 gfs2_statfs_change_in(l_sc, l_bh->b_data +
465 sizeof(struct gfs2_dinode));
466 spin_unlock(&sdp->sd_statfs_spin);
474 gfs2_glock_dq_uninit(&gh);
478 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
481 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
482 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
483 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
484 struct buffer_head *l_bh;
489 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
493 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
495 spin_lock(&sdp->sd_statfs_spin);
496 l_sc->sc_total += total;
497 l_sc->sc_free += free;
498 l_sc->sc_dinodes += dinodes;
499 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
500 if (sdp->sd_args.ar_statfs_percent) {
501 x = 100 * l_sc->sc_free;
502 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
503 if (x >= y || x <= -y)
506 spin_unlock(&sdp->sd_statfs_spin);
510 gfs2_wake_up_statfs(sdp);
513 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
514 struct buffer_head *l_bh)
516 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
517 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
518 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
519 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
521 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
523 spin_lock(&sdp->sd_statfs_spin);
524 m_sc->sc_total += l_sc->sc_total;
525 m_sc->sc_free += l_sc->sc_free;
526 m_sc->sc_dinodes += l_sc->sc_dinodes;
527 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
528 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
529 0, sizeof(struct gfs2_statfs_change));
530 spin_unlock(&sdp->sd_statfs_spin);
532 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
533 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
536 int gfs2_statfs_sync(struct super_block *sb, int type)
538 struct gfs2_sbd *sdp = sb->s_fs_info;
539 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
540 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
541 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
542 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
543 struct gfs2_holder gh;
544 struct buffer_head *m_bh, *l_bh;
547 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
552 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
556 spin_lock(&sdp->sd_statfs_spin);
557 gfs2_statfs_change_in(m_sc, m_bh->b_data +
558 sizeof(struct gfs2_dinode));
559 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
560 spin_unlock(&sdp->sd_statfs_spin);
563 spin_unlock(&sdp->sd_statfs_spin);
565 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
569 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
573 update_statfs(sdp, m_bh, l_bh);
574 sdp->sd_statfs_force_sync = 0;
583 gfs2_glock_dq_uninit(&gh);
588 struct list_head list;
589 struct gfs2_holder gh;
593 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
595 * @sdp: the file system
596 * @state: the state to put the transaction lock into
597 * @t_gh: the hold on the transaction lock
602 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
603 struct gfs2_holder *t_gh)
605 struct gfs2_inode *ip;
606 struct gfs2_jdesc *jd;
609 struct gfs2_log_header_host lh;
612 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
613 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
618 ip = GFS2_I(jd->jd_inode);
619 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
624 list_add(&lfcc->list, &list);
627 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
630 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
631 error = gfs2_jdesc_check(jd);
634 error = gfs2_find_jhead(jd, &lh);
637 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
644 gfs2_glock_dq_uninit(t_gh);
647 while (!list_empty(&list)) {
648 lfcc = list_entry(list.next, struct lfcc, list);
649 list_del(&lfcc->list);
650 gfs2_glock_dq_uninit(&lfcc->gh);
657 * gfs2_freeze_fs - freezes the file system
658 * @sdp: the file system
660 * This function flushes data and meta data for all machines by
661 * acquiring the transaction log exclusively. All journals are
662 * ensured to be in a clean state as well.
667 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
671 mutex_lock(&sdp->sd_freeze_lock);
673 if (!sdp->sd_freeze_count++) {
674 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
676 sdp->sd_freeze_count--;
679 mutex_unlock(&sdp->sd_freeze_lock);
685 * gfs2_unfreeze_fs - unfreezes the file system
686 * @sdp: the file system
688 * This function allows the file system to proceed by unlocking
689 * the exclusively held transaction lock. Other GFS2 nodes are
690 * now free to acquire the lock shared and go on with their lives.
694 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
696 mutex_lock(&sdp->sd_freeze_lock);
698 if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
699 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
701 mutex_unlock(&sdp->sd_freeze_lock);
704 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
706 struct gfs2_dinode *str = buf;
708 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
709 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
710 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
711 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
712 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
713 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
714 str->di_uid = cpu_to_be32(ip->i_inode.i_uid);
715 str->di_gid = cpu_to_be32(ip->i_inode.i_gid);
716 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
717 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
718 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
719 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
720 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
721 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
723 str->di_goal_meta = cpu_to_be64(ip->i_goal);
724 str->di_goal_data = cpu_to_be64(ip->i_goal);
725 str->di_generation = cpu_to_be64(ip->i_generation);
727 str->di_flags = cpu_to_be32(ip->i_diskflags);
728 str->di_height = cpu_to_be16(ip->i_height);
729 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
730 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
732 str->di_depth = cpu_to_be16(ip->i_depth);
733 str->di_entries = cpu_to_be32(ip->i_entries);
735 str->di_eattr = cpu_to_be64(ip->i_eattr);
736 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
737 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
738 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
742 * gfs2_write_inode - Make sure the inode is stable on the disk
744 * @wbc: The writeback control structure
749 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
751 struct gfs2_inode *ip = GFS2_I(inode);
752 struct gfs2_sbd *sdp = GFS2_SB(inode);
753 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
754 struct backing_dev_info *bdi = metamapping->backing_dev_info;
755 struct gfs2_holder gh;
756 struct buffer_head *bh;
757 struct timespec atime;
758 struct gfs2_dinode *di;
760 int unlock_required = 0;
762 /* Skip timestamp update, if this is from a memalloc */
763 if (current->flags & PF_MEMALLOC)
765 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
766 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
771 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
774 ret = gfs2_meta_inode_buffer(ip, &bh);
776 di = (struct gfs2_dinode *)bh->b_data;
777 atime.tv_sec = be64_to_cpu(di->di_atime);
778 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
779 if (timespec_compare(&inode->i_atime, &atime) > 0) {
780 gfs2_trans_add_bh(ip->i_gl, bh, 1);
781 gfs2_dinode_out(ip, bh->b_data);
788 gfs2_glock_dq_uninit(&gh);
790 if (wbc->sync_mode == WB_SYNC_ALL)
791 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
792 filemap_fdatawrite(metamapping);
793 if (bdi->dirty_exceeded)
794 gfs2_ail1_flush(sdp, wbc);
795 if (!ret && (wbc->sync_mode == WB_SYNC_ALL))
796 ret = filemap_fdatawait(metamapping);
798 mark_inode_dirty_sync(inode);
803 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
804 * @sdp: the filesystem
809 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
811 struct gfs2_holder t_gh;
814 flush_workqueue(gfs2_delete_workqueue);
815 gfs2_quota_sync(sdp->sd_vfs, 0, 1);
816 gfs2_statfs_sync(sdp->sd_vfs, 0);
818 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
820 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
823 gfs2_meta_syncfs(sdp);
824 gfs2_log_shutdown(sdp);
826 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
829 gfs2_glock_dq_uninit(&t_gh);
831 gfs2_quota_cleanup(sdp);
836 static int gfs2_umount_recovery_wait(void *word)
843 * gfs2_put_super - Unmount the filesystem
844 * @sb: The VFS superblock
848 static void gfs2_put_super(struct super_block *sb)
850 struct gfs2_sbd *sdp = sb->s_fs_info;
852 struct gfs2_jdesc *jd;
854 /* Unfreeze the filesystem, if we need to */
856 mutex_lock(&sdp->sd_freeze_lock);
857 if (sdp->sd_freeze_count)
858 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
859 mutex_unlock(&sdp->sd_freeze_lock);
861 /* No more recovery requests */
862 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
865 /* Wait on outstanding recovery */
867 spin_lock(&sdp->sd_jindex_spin);
868 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
869 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
871 spin_unlock(&sdp->sd_jindex_spin);
872 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
873 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
876 spin_unlock(&sdp->sd_jindex_spin);
878 kthread_stop(sdp->sd_quotad_process);
879 kthread_stop(sdp->sd_logd_process);
881 if (!(sb->s_flags & MS_RDONLY)) {
882 error = gfs2_make_fs_ro(sdp);
886 /* At this point, we're through modifying the disk */
890 iput(sdp->sd_jindex);
891 iput(sdp->sd_statfs_inode);
892 iput(sdp->sd_rindex);
893 iput(sdp->sd_quota_inode);
895 gfs2_glock_put(sdp->sd_rename_gl);
896 gfs2_glock_put(sdp->sd_trans_gl);
898 if (!sdp->sd_args.ar_spectator) {
899 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
900 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
901 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
902 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
903 iput(sdp->sd_sc_inode);
904 iput(sdp->sd_qc_inode);
907 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
908 gfs2_clear_rgrpd(sdp);
909 gfs2_jindex_free(sdp);
910 /* Take apart glock structures and buffer lists */
911 gfs2_gl_hash_clear(sdp);
912 /* Unmount the locking protocol */
913 gfs2_lm_unmount(sdp);
915 /* At this point, we're through participating in the lockspace */
916 gfs2_sys_fs_del(sdp);
920 * gfs2_sync_fs - sync the filesystem
921 * @sb: the superblock
923 * Flushes the log to disk.
926 static int gfs2_sync_fs(struct super_block *sb, int wait)
928 struct gfs2_sbd *sdp = sb->s_fs_info;
930 gfs2_log_flush(sdp, NULL);
935 * gfs2_freeze - prevent further writes to the filesystem
936 * @sb: the VFS structure for the filesystem
940 static int gfs2_freeze(struct super_block *sb)
942 struct gfs2_sbd *sdp = sb->s_fs_info;
945 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
949 error = gfs2_freeze_fs(sdp);
955 fs_err(sdp, "waiting for recovery before freeze\n");
959 fs_err(sdp, "error freezing FS: %d\n", error);
963 fs_err(sdp, "retrying...\n");
970 * gfs2_unfreeze - reallow writes to the filesystem
971 * @sb: the VFS structure for the filesystem
975 static int gfs2_unfreeze(struct super_block *sb)
977 gfs2_unfreeze_fs(sb->s_fs_info);
982 * statfs_fill - fill in the sg for a given RG
984 * @sc: the sc structure
986 * Returns: 0 on success, -ESTALE if the LVB is invalid
989 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
990 struct gfs2_statfs_change_host *sc)
992 gfs2_rgrp_verify(rgd);
993 sc->sc_total += rgd->rd_data;
994 sc->sc_free += rgd->rd_free;
995 sc->sc_dinodes += rgd->rd_dinodes;
1000 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1001 * @sdp: the filesystem
1002 * @sc: the sc info that will be returned
1004 * Any error (other than a signal) will cause this routine to fall back
1005 * to the synchronous version.
1007 * FIXME: This really shouldn't busy wait like this.
1012 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1014 struct gfs2_holder ri_gh;
1015 struct gfs2_rgrpd *rgd_next;
1016 struct gfs2_holder *gha, *gh;
1017 unsigned int slots = 64;
1022 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1023 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1027 error = gfs2_rindex_hold(sdp, &ri_gh);
1031 rgd_next = gfs2_rgrpd_get_first(sdp);
1036 for (x = 0; x < slots; x++) {
1039 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1040 err = gfs2_glock_wait(gh);
1042 gfs2_holder_uninit(gh);
1046 error = statfs_slow_fill(
1047 gh->gh_gl->gl_object, sc);
1048 gfs2_glock_dq_uninit(gh);
1054 else if (rgd_next && !error) {
1055 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1059 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1063 if (signal_pending(current))
1064 error = -ERESTARTSYS;
1073 gfs2_glock_dq_uninit(&ri_gh);
1081 * gfs2_statfs_i - Do a statfs
1082 * @sdp: the filesystem
1083 * @sg: the sg structure
1088 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1090 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1091 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1093 spin_lock(&sdp->sd_statfs_spin);
1096 sc->sc_total += l_sc->sc_total;
1097 sc->sc_free += l_sc->sc_free;
1098 sc->sc_dinodes += l_sc->sc_dinodes;
1100 spin_unlock(&sdp->sd_statfs_spin);
1102 if (sc->sc_free < 0)
1104 if (sc->sc_free > sc->sc_total)
1105 sc->sc_free = sc->sc_total;
1106 if (sc->sc_dinodes < 0)
1113 * gfs2_statfs - Gather and return stats about the filesystem
1114 * @sb: The superblock
1115 * @statfsbuf: The buffer
1117 * Returns: 0 on success or error code
1120 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1122 struct super_block *sb = dentry->d_inode->i_sb;
1123 struct gfs2_sbd *sdp = sb->s_fs_info;
1124 struct gfs2_statfs_change_host sc;
1127 if (gfs2_tune_get(sdp, gt_statfs_slow))
1128 error = gfs2_statfs_slow(sdp, &sc);
1130 error = gfs2_statfs_i(sdp, &sc);
1135 buf->f_type = GFS2_MAGIC;
1136 buf->f_bsize = sdp->sd_sb.sb_bsize;
1137 buf->f_blocks = sc.sc_total;
1138 buf->f_bfree = sc.sc_free;
1139 buf->f_bavail = sc.sc_free;
1140 buf->f_files = sc.sc_dinodes + sc.sc_free;
1141 buf->f_ffree = sc.sc_free;
1142 buf->f_namelen = GFS2_FNAMESIZE;
1148 * gfs2_remount_fs - called when the FS is remounted
1149 * @sb: the filesystem
1150 * @flags: the remount flags
1151 * @data: extra data passed in (not used right now)
1156 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1158 struct gfs2_sbd *sdp = sb->s_fs_info;
1159 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1160 struct gfs2_tune *gt = &sdp->sd_tune;
1163 spin_lock(>->gt_spin);
1164 args.ar_commit = gt->gt_logd_secs;
1165 args.ar_quota_quantum = gt->gt_quota_quantum;
1166 if (gt->gt_statfs_slow)
1167 args.ar_statfs_quantum = 0;
1169 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1170 spin_unlock(>->gt_spin);
1171 error = gfs2_mount_args(&args, data);
1175 /* Not allowed to change locking details */
1176 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1177 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1178 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1181 /* Some flags must not be changed */
1182 if (args_neq(&args, &sdp->sd_args, spectator) ||
1183 args_neq(&args, &sdp->sd_args, localflocks) ||
1184 args_neq(&args, &sdp->sd_args, meta))
1187 if (sdp->sd_args.ar_spectator)
1188 *flags |= MS_RDONLY;
1190 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1191 if (*flags & MS_RDONLY)
1192 error = gfs2_make_fs_ro(sdp);
1194 error = gfs2_make_fs_rw(sdp);
1199 sdp->sd_args = args;
1200 if (sdp->sd_args.ar_posix_acl)
1201 sb->s_flags |= MS_POSIXACL;
1203 sb->s_flags &= ~MS_POSIXACL;
1204 if (sdp->sd_args.ar_nobarrier)
1205 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1207 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1208 spin_lock(>->gt_spin);
1209 gt->gt_logd_secs = args.ar_commit;
1210 gt->gt_quota_quantum = args.ar_quota_quantum;
1211 if (args.ar_statfs_quantum) {
1212 gt->gt_statfs_slow = 0;
1213 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1216 gt->gt_statfs_slow = 1;
1217 gt->gt_statfs_quantum = 30;
1219 spin_unlock(>->gt_spin);
1221 gfs2_online_uevent(sdp);
1226 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1227 * @inode: The inode to drop
1229 * If we've received a callback on an iopen lock then its because a
1230 * remote node tried to deallocate the inode but failed due to this node
1231 * still having the inode open. Here we mark the link count zero
1232 * since we know that it must have reached zero if the GLF_DEMOTE flag
1233 * is set on the iopen glock. If we didn't do a disk read since the
1234 * remote node removed the final link then we might otherwise miss
1235 * this event. This check ensures that this node will deallocate the
1236 * inode's blocks, or alternatively pass the baton on to another
1237 * node for later deallocation.
1240 static int gfs2_drop_inode(struct inode *inode)
1242 struct gfs2_inode *ip = GFS2_I(inode);
1244 if (inode->i_nlink) {
1245 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1246 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1249 return generic_drop_inode(inode);
1252 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1258 } while (!IS_ROOT(d1));
1263 * gfs2_show_options - Show mount options for /proc/mounts
1264 * @s: seq_file structure
1267 * Returns: 0 on success or error code
1270 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1272 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
1273 struct gfs2_args *args = &sdp->sd_args;
1276 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
1277 seq_printf(s, ",meta");
1278 if (args->ar_lockproto[0])
1279 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1280 if (args->ar_locktable[0])
1281 seq_printf(s, ",locktable=%s", args->ar_locktable);
1282 if (args->ar_hostdata[0])
1283 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1284 if (args->ar_spectator)
1285 seq_printf(s, ",spectator");
1286 if (args->ar_localflocks)
1287 seq_printf(s, ",localflocks");
1289 seq_printf(s, ",debug");
1290 if (args->ar_posix_acl)
1291 seq_printf(s, ",acl");
1292 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1294 switch (args->ar_quota) {
1295 case GFS2_QUOTA_OFF:
1298 case GFS2_QUOTA_ACCOUNT:
1308 seq_printf(s, ",quota=%s", state);
1310 if (args->ar_suiddir)
1311 seq_printf(s, ",suiddir");
1312 if (args->ar_data != GFS2_DATA_DEFAULT) {
1314 switch (args->ar_data) {
1315 case GFS2_DATA_WRITEBACK:
1316 state = "writeback";
1318 case GFS2_DATA_ORDERED:
1325 seq_printf(s, ",data=%s", state);
1327 if (args->ar_discard)
1328 seq_printf(s, ",discard");
1329 val = sdp->sd_tune.gt_logd_secs;
1331 seq_printf(s, ",commit=%d", val);
1332 val = sdp->sd_tune.gt_statfs_quantum;
1334 seq_printf(s, ",statfs_quantum=%d", val);
1335 val = sdp->sd_tune.gt_quota_quantum;
1337 seq_printf(s, ",quota_quantum=%d", val);
1338 if (args->ar_statfs_percent)
1339 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1340 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1343 switch (args->ar_errors) {
1344 case GFS2_ERRORS_WITHDRAW:
1347 case GFS2_ERRORS_PANIC:
1354 seq_printf(s, ",errors=%s", state);
1356 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1357 seq_printf(s, ",nobarrier");
1358 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1359 seq_printf(s, ",demote_interface_used");
1363 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1365 struct inode *inode = &ip->i_inode;
1366 struct gfs2_glock *gl = ip->i_gl;
1368 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1369 truncate_inode_pages(&inode->i_data, 0);
1371 if (atomic_read(&gl->gl_revokes) == 0) {
1372 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1373 clear_bit(GLF_DIRTY, &gl->gl_flags);
1377 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1379 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1380 struct gfs2_alloc *al;
1381 struct gfs2_rgrpd *rgd;
1384 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1385 gfs2_consist_inode(ip);
1389 al = gfs2_alloc_get(ip);
1393 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1397 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
1401 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr);
1403 gfs2_consist_inode(ip);
1405 goto out_rindex_relse;
1408 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
1411 goto out_rindex_relse;
1413 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1414 sdp->sd_jdesc->jd_blocks);
1416 goto out_rg_gunlock;
1418 gfs2_free_di(rgd, ip);
1420 gfs2_final_release_pages(ip);
1422 gfs2_trans_end(sdp);
1425 gfs2_glock_dq_uninit(&al->al_rgd_gh);
1427 gfs2_glock_dq_uninit(&al->al_ri_gh);
1429 gfs2_quota_unhold(ip);
1436 * gfs2_evict_inode - Remove an inode from cache
1437 * @inode: The inode to evict
1439 * There are three cases to consider:
1440 * 1. i_nlink == 0, we are final opener (and must deallocate)
1441 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1444 * If the fs is read only, then we have to treat all cases as per #3
1445 * since we are unable to do any deallocation. The inode will be
1446 * deallocated by the next read/write node to attempt an allocation
1447 * in the same resource group
1449 * We have to (at the moment) hold the inodes main lock to cover
1450 * the gap between unlocking the shared lock on the iopen lock and
1451 * taking the exclusive lock. I'd rather do a shared -> exclusive
1452 * conversion on the iopen lock, but we can change that later. This
1453 * is safe, just less efficient.
1456 static void gfs2_evict_inode(struct inode *inode)
1458 struct super_block *sb = inode->i_sb;
1459 struct gfs2_sbd *sdp = sb->s_fs_info;
1460 struct gfs2_inode *ip = GFS2_I(inode);
1461 struct gfs2_holder gh;
1464 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1467 /* Must not read inode block until block type has been verified */
1468 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1469 if (unlikely(error)) {
1470 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1474 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1478 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1479 error = gfs2_inode_refresh(ip);
1484 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1485 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1486 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1487 error = gfs2_glock_nq(&ip->i_iopen_gh);
1491 /* Case 1 starts here */
1493 if (S_ISDIR(inode->i_mode) &&
1494 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1495 error = gfs2_dir_exhash_dealloc(ip);
1501 error = gfs2_ea_dealloc(ip);
1506 if (!gfs2_is_stuffed(ip)) {
1507 error = gfs2_file_dealloc(ip);
1512 error = gfs2_dinode_dealloc(ip);
1516 /* Case 2 starts here */
1517 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1520 /* Needs to be done before glock release & also in a transaction */
1521 truncate_inode_pages(&inode->i_data, 0);
1522 gfs2_trans_end(sdp);
1525 /* Error path for case 1 */
1526 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
1527 gfs2_glock_dq(&ip->i_iopen_gh);
1528 gfs2_holder_uninit(&ip->i_iopen_gh);
1529 gfs2_glock_dq_uninit(&gh);
1530 if (error && error != GLR_TRYFAILED && error != -EROFS)
1531 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1533 /* Case 3 starts here */
1534 truncate_inode_pages(&inode->i_data, 0);
1535 end_writeback(inode);
1536 gfs2_dir_hash_inval(ip);
1537 ip->i_gl->gl_object = NULL;
1538 gfs2_glock_add_to_lru(ip->i_gl);
1539 gfs2_glock_put(ip->i_gl);
1541 if (ip->i_iopen_gh.gh_gl) {
1542 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1543 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1547 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1549 struct gfs2_inode *ip;
1551 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1556 return &ip->i_inode;
1559 static void gfs2_i_callback(struct rcu_head *head)
1561 struct inode *inode = container_of(head, struct inode, i_rcu);
1562 INIT_LIST_HEAD(&inode->i_dentry);
1563 kmem_cache_free(gfs2_inode_cachep, inode);
1566 static void gfs2_destroy_inode(struct inode *inode)
1568 call_rcu(&inode->i_rcu, gfs2_i_callback);
1571 const struct super_operations gfs2_super_ops = {
1572 .alloc_inode = gfs2_alloc_inode,
1573 .destroy_inode = gfs2_destroy_inode,
1574 .write_inode = gfs2_write_inode,
1575 .evict_inode = gfs2_evict_inode,
1576 .put_super = gfs2_put_super,
1577 .sync_fs = gfs2_sync_fs,
1578 .freeze_fs = gfs2_freeze,
1579 .unfreeze_fs = gfs2_unfreeze,
1580 .statfs = gfs2_statfs,
1581 .remount_fs = gfs2_remount_fs,
1582 .drop_inode = gfs2_drop_inode,
1583 .show_options = gfs2_show_options,