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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/bio.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/completion.h>
17 #include <linux/buffer_head.h>
18 #include <linux/statfs.h>
19 #include <linux/seq_file.h>
20 #include <linux/mount.h>
21 #include <linux/kthread.h>
22 #include <linux/delay.h>
23 #include <linux/gfs2_ondisk.h>
24 #include <linux/crc32.h>
25 #include <linux/time.h>
26 #include <linux/wait.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
48 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
88 static const match_table_t tokens = {
89 {Opt_lockproto, "lockproto=%s"},
90 {Opt_locktable, "locktable=%s"},
91 {Opt_hostdata, "hostdata=%s"},
92 {Opt_spectator, "spectator"},
93 {Opt_spectator, "norecovery"},
94 {Opt_ignore_local_fs, "ignore_local_fs"},
95 {Opt_localflocks, "localflocks"},
96 {Opt_localcaching, "localcaching"},
98 {Opt_nodebug, "nodebug"},
99 {Opt_upgrade, "upgrade"},
101 {Opt_noacl, "noacl"},
102 {Opt_quota_off, "quota=off"},
103 {Opt_quota_account, "quota=account"},
104 {Opt_quota_on, "quota=on"},
105 {Opt_quota, "quota"},
106 {Opt_noquota, "noquota"},
107 {Opt_suiddir, "suiddir"},
108 {Opt_nosuiddir, "nosuiddir"},
109 {Opt_data_writeback, "data=writeback"},
110 {Opt_data_ordered, "data=ordered"},
112 {Opt_discard, "discard"},
113 {Opt_nodiscard, "nodiscard"},
114 {Opt_commit, "commit=%d"},
115 {Opt_err_withdraw, "errors=withdraw"},
116 {Opt_err_panic, "errors=panic"},
117 {Opt_statfs_quantum, "statfs_quantum=%d"},
118 {Opt_statfs_percent, "statfs_percent=%d"},
119 {Opt_quota_quantum, "quota_quantum=%d"},
120 {Opt_barrier, "barrier"},
121 {Opt_nobarrier, "nobarrier"},
122 {Opt_rgrplvb, "rgrplvb"},
123 {Opt_norgrplvb, "norgrplvb"},
128 * gfs2_mount_args - Parse mount options
129 * @args: The structure into which the parsed options will be written
130 * @options: The options to parse
135 int gfs2_mount_args(struct gfs2_args *args, char *options)
139 substring_t tmp[MAX_OPT_ARGS];
142 /* Split the options into tokens with the "," character and
146 o = strsep(&options, ",");
152 token = match_token(o, tokens, tmp);
155 match_strlcpy(args->ar_lockproto, &tmp[0],
159 match_strlcpy(args->ar_locktable, &tmp[0],
163 match_strlcpy(args->ar_hostdata, &tmp[0],
167 args->ar_spectator = 1;
169 case Opt_ignore_local_fs:
170 /* Retained for backwards compat only */
172 case Opt_localflocks:
173 args->ar_localflocks = 1;
175 case Opt_localcaching:
176 /* Retained for backwards compat only */
179 if (args->ar_errors == GFS2_ERRORS_PANIC) {
180 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
189 /* Retained for backwards compat only */
192 args->ar_posix_acl = 1;
195 args->ar_posix_acl = 0;
199 args->ar_quota = GFS2_QUOTA_OFF;
201 case Opt_quota_account:
202 args->ar_quota = GFS2_QUOTA_ACCOUNT;
206 args->ar_quota = GFS2_QUOTA_ON;
209 args->ar_suiddir = 1;
212 args->ar_suiddir = 0;
214 case Opt_data_writeback:
215 args->ar_data = GFS2_DATA_WRITEBACK;
217 case Opt_data_ordered:
218 args->ar_data = GFS2_DATA_ORDERED;
224 args->ar_discard = 1;
227 args->ar_discard = 0;
230 rv = match_int(&tmp[0], &args->ar_commit);
231 if (rv || args->ar_commit <= 0) {
232 pr_warn("commit mount option requires a positive numeric argument\n");
233 return rv ? rv : -EINVAL;
236 case Opt_statfs_quantum:
237 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
238 if (rv || args->ar_statfs_quantum < 0) {
239 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
240 return rv ? rv : -EINVAL;
243 case Opt_quota_quantum:
244 rv = match_int(&tmp[0], &args->ar_quota_quantum);
245 if (rv || args->ar_quota_quantum <= 0) {
246 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
247 return rv ? rv : -EINVAL;
250 case Opt_statfs_percent:
251 rv = match_int(&tmp[0], &args->ar_statfs_percent);
252 if (rv || args->ar_statfs_percent < 0 ||
253 args->ar_statfs_percent > 100) {
254 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
255 return rv ? rv : -EINVAL;
258 case Opt_err_withdraw:
259 args->ar_errors = GFS2_ERRORS_WITHDRAW;
262 if (args->ar_debug) {
263 pr_warn("-o debug and -o errors=panic 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 pr_warn("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;
299 struct gfs2_jdesc *jd;
301 spin_lock(&sdp->sd_jindex_spin);
302 list_add(&list, &sdp->sd_jindex_list);
303 list_del_init(&sdp->sd_jindex_list);
304 sdp->sd_journals = 0;
305 spin_unlock(&sdp->sd_jindex_spin);
307 while (!list_empty(&list)) {
308 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
309 gfs2_free_journal_extents(jd);
310 list_del(&jd->jd_list);
316 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
318 struct gfs2_jdesc *jd;
321 list_for_each_entry(jd, head, jd_list) {
322 if (jd->jd_jid == jid) {
334 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
336 struct gfs2_jdesc *jd;
338 spin_lock(&sdp->sd_jindex_spin);
339 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
340 spin_unlock(&sdp->sd_jindex_spin);
345 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
347 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
348 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
349 u64 size = i_size_read(jd->jd_inode);
351 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
354 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
356 if (gfs2_write_alloc_required(ip, 0, size)) {
357 gfs2_consist_inode(ip);
364 static int init_threads(struct gfs2_sbd *sdp)
366 struct task_struct *p;
369 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
372 fs_err(sdp, "can't start logd thread: %d\n", error);
375 sdp->sd_logd_process = p;
377 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
380 fs_err(sdp, "can't start quotad thread: %d\n", error);
383 sdp->sd_quotad_process = p;
387 kthread_stop(sdp->sd_logd_process);
392 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
393 * @sdp: the filesystem
398 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
400 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
401 struct gfs2_glock *j_gl = ip->i_gl;
402 struct gfs2_holder t_gh;
403 struct gfs2_log_header_host head;
406 error = init_threads(sdp);
410 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
414 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
416 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
420 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
426 /* Initialize some head of the log stuff */
427 sdp->sd_log_sequence = head.lh_sequence + 1;
428 gfs2_log_pointers_init(sdp, head.lh_blkno);
430 error = gfs2_quota_init(sdp);
434 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
436 gfs2_glock_dq_uninit(&t_gh);
441 t_gh.gh_flags |= GL_NOCACHE;
442 gfs2_glock_dq_uninit(&t_gh);
444 kthread_stop(sdp->sd_quotad_process);
445 kthread_stop(sdp->sd_logd_process);
449 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
451 const struct gfs2_statfs_change *str = buf;
453 sc->sc_total = be64_to_cpu(str->sc_total);
454 sc->sc_free = be64_to_cpu(str->sc_free);
455 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
458 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
460 struct gfs2_statfs_change *str = buf;
462 str->sc_total = cpu_to_be64(sc->sc_total);
463 str->sc_free = cpu_to_be64(sc->sc_free);
464 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
467 int gfs2_statfs_init(struct gfs2_sbd *sdp)
469 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
470 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
471 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
472 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
473 struct buffer_head *m_bh, *l_bh;
474 struct gfs2_holder gh;
477 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
482 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
486 if (sdp->sd_args.ar_spectator) {
487 spin_lock(&sdp->sd_statfs_spin);
488 gfs2_statfs_change_in(m_sc, m_bh->b_data +
489 sizeof(struct gfs2_dinode));
490 spin_unlock(&sdp->sd_statfs_spin);
492 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
496 spin_lock(&sdp->sd_statfs_spin);
497 gfs2_statfs_change_in(m_sc, m_bh->b_data +
498 sizeof(struct gfs2_dinode));
499 gfs2_statfs_change_in(l_sc, l_bh->b_data +
500 sizeof(struct gfs2_dinode));
501 spin_unlock(&sdp->sd_statfs_spin);
509 gfs2_glock_dq_uninit(&gh);
513 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
516 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
517 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
518 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
519 struct buffer_head *l_bh;
524 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
528 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
530 spin_lock(&sdp->sd_statfs_spin);
531 l_sc->sc_total += total;
532 l_sc->sc_free += free;
533 l_sc->sc_dinodes += dinodes;
534 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
535 if (sdp->sd_args.ar_statfs_percent) {
536 x = 100 * l_sc->sc_free;
537 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
538 if (x >= y || x <= -y)
541 spin_unlock(&sdp->sd_statfs_spin);
545 gfs2_wake_up_statfs(sdp);
548 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
549 struct buffer_head *l_bh)
551 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
552 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
553 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
554 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
556 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
558 spin_lock(&sdp->sd_statfs_spin);
559 m_sc->sc_total += l_sc->sc_total;
560 m_sc->sc_free += l_sc->sc_free;
561 m_sc->sc_dinodes += l_sc->sc_dinodes;
562 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
563 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
564 0, sizeof(struct gfs2_statfs_change));
565 spin_unlock(&sdp->sd_statfs_spin);
567 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
568 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
571 int gfs2_statfs_sync(struct super_block *sb, int type)
573 struct gfs2_sbd *sdp = sb->s_fs_info;
574 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
575 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
576 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
577 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
578 struct gfs2_holder gh;
579 struct buffer_head *m_bh, *l_bh;
582 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
587 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
591 spin_lock(&sdp->sd_statfs_spin);
592 gfs2_statfs_change_in(m_sc, m_bh->b_data +
593 sizeof(struct gfs2_dinode));
594 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
595 spin_unlock(&sdp->sd_statfs_spin);
598 spin_unlock(&sdp->sd_statfs_spin);
600 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
604 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
608 update_statfs(sdp, m_bh, l_bh);
609 sdp->sd_statfs_force_sync = 0;
618 gfs2_glock_dq_uninit(&gh);
623 struct list_head list;
624 struct gfs2_holder gh;
628 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
630 * @sdp: the file system
631 * @state: the state to put the transaction lock into
632 * @t_gh: the hold on the transaction lock
637 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
638 struct gfs2_holder *t_gh)
640 struct gfs2_inode *ip;
641 struct gfs2_jdesc *jd;
644 struct gfs2_log_header_host lh;
647 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
648 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
653 ip = GFS2_I(jd->jd_inode);
654 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
659 list_add(&lfcc->list, &list);
662 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
665 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
666 error = gfs2_jdesc_check(jd);
669 error = gfs2_find_jhead(jd, &lh);
672 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
679 gfs2_glock_dq_uninit(t_gh);
682 while (!list_empty(&list)) {
683 lfcc = list_entry(list.next, struct lfcc, list);
684 list_del(&lfcc->list);
685 gfs2_glock_dq_uninit(&lfcc->gh);
691 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
693 struct gfs2_dinode *str = buf;
695 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
696 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
697 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
698 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
699 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
700 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
701 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
702 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
703 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
704 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
705 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
706 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
707 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
708 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
710 str->di_goal_meta = cpu_to_be64(ip->i_goal);
711 str->di_goal_data = cpu_to_be64(ip->i_goal);
712 str->di_generation = cpu_to_be64(ip->i_generation);
714 str->di_flags = cpu_to_be32(ip->i_diskflags);
715 str->di_height = cpu_to_be16(ip->i_height);
716 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
717 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
719 str->di_depth = cpu_to_be16(ip->i_depth);
720 str->di_entries = cpu_to_be32(ip->i_entries);
722 str->di_eattr = cpu_to_be64(ip->i_eattr);
723 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
724 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
725 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
729 * gfs2_write_inode - Make sure the inode is stable on the disk
731 * @wbc: The writeback control structure
736 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
738 struct gfs2_inode *ip = GFS2_I(inode);
739 struct gfs2_sbd *sdp = GFS2_SB(inode);
740 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
741 struct backing_dev_info *bdi = metamapping->backing_dev_info;
744 if (wbc->sync_mode == WB_SYNC_ALL)
745 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
746 if (bdi->dirty_exceeded)
747 gfs2_ail1_flush(sdp, wbc);
749 filemap_fdatawrite(metamapping);
750 if (wbc->sync_mode == WB_SYNC_ALL)
751 ret = filemap_fdatawait(metamapping);
753 mark_inode_dirty_sync(inode);
758 * gfs2_dirty_inode - check for atime updates
759 * @inode: The inode in question
760 * @flags: The type of dirty
762 * Unfortunately it can be called under any combination of inode
763 * glock and transaction lock, so we have to check carefully.
765 * At the moment this deals only with atime - it should be possible
766 * to expand that role in future, once a review of the locking has
770 static void gfs2_dirty_inode(struct inode *inode, int flags)
772 struct gfs2_inode *ip = GFS2_I(inode);
773 struct gfs2_sbd *sdp = GFS2_SB(inode);
774 struct buffer_head *bh;
775 struct gfs2_holder gh;
777 int need_endtrans = 0;
780 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
783 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
784 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
786 fs_err(sdp, "dirty_inode: glock %d\n", ret);
790 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
793 if (current->journal_info == NULL) {
794 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
796 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
802 ret = gfs2_meta_inode_buffer(ip, &bh);
804 gfs2_trans_add_meta(ip->i_gl, bh);
805 gfs2_dinode_out(ip, bh->b_data);
813 gfs2_glock_dq_uninit(&gh);
817 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
818 * @sdp: the filesystem
823 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
825 struct gfs2_holder t_gh;
828 kthread_stop(sdp->sd_quotad_process);
829 kthread_stop(sdp->sd_logd_process);
831 flush_workqueue(gfs2_delete_workqueue);
832 gfs2_quota_sync(sdp->sd_vfs, 0);
833 gfs2_statfs_sync(sdp->sd_vfs, 0);
835 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
837 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
840 gfs2_meta_syncfs(sdp);
841 gfs2_log_shutdown(sdp);
843 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
846 gfs2_glock_dq_uninit(&t_gh);
848 gfs2_quota_cleanup(sdp);
853 static int gfs2_umount_recovery_wait(void *word)
860 * gfs2_put_super - Unmount the filesystem
861 * @sb: The VFS superblock
865 static void gfs2_put_super(struct super_block *sb)
867 struct gfs2_sbd *sdp = sb->s_fs_info;
869 struct gfs2_jdesc *jd;
871 /* No more recovery requests */
872 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
875 /* Wait on outstanding recovery */
877 spin_lock(&sdp->sd_jindex_spin);
878 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
879 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
881 spin_unlock(&sdp->sd_jindex_spin);
882 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
883 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE);
886 spin_unlock(&sdp->sd_jindex_spin);
888 if (!(sb->s_flags & MS_RDONLY)) {
889 error = gfs2_make_fs_ro(sdp);
893 /* At this point, we're through modifying the disk */
897 iput(sdp->sd_jindex);
898 iput(sdp->sd_statfs_inode);
899 iput(sdp->sd_rindex);
900 iput(sdp->sd_quota_inode);
902 gfs2_glock_put(sdp->sd_rename_gl);
903 gfs2_glock_put(sdp->sd_trans_gl);
905 if (!sdp->sd_args.ar_spectator) {
906 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
907 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
908 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
909 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
910 iput(sdp->sd_sc_inode);
911 iput(sdp->sd_qc_inode);
914 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
915 gfs2_clear_rgrpd(sdp);
916 gfs2_jindex_free(sdp);
917 /* Take apart glock structures and buffer lists */
918 gfs2_gl_hash_clear(sdp);
919 /* Unmount the locking protocol */
920 gfs2_lm_unmount(sdp);
922 /* At this point, we're through participating in the lockspace */
923 gfs2_sys_fs_del(sdp);
927 * gfs2_sync_fs - sync the filesystem
928 * @sb: the superblock
930 * Flushes the log to disk.
933 static int gfs2_sync_fs(struct super_block *sb, int wait)
935 struct gfs2_sbd *sdp = sb->s_fs_info;
937 gfs2_quota_sync(sb, -1);
939 gfs2_log_flush(sdp, NULL);
944 * gfs2_freeze - prevent further writes to the filesystem
945 * @sb: the VFS structure for the filesystem
949 static int gfs2_freeze(struct super_block *sb)
951 struct gfs2_sbd *sdp = sb->s_fs_info;
954 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
958 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
964 fs_err(sdp, "waiting for recovery before freeze\n");
968 fs_err(sdp, "error freezing FS: %d\n", error);
972 fs_err(sdp, "retrying...\n");
979 * gfs2_unfreeze - reallow writes to the filesystem
980 * @sb: the VFS structure for the filesystem
984 static int gfs2_unfreeze(struct super_block *sb)
986 struct gfs2_sbd *sdp = sb->s_fs_info;
988 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
993 * statfs_fill - fill in the sg for a given RG
995 * @sc: the sc structure
997 * Returns: 0 on success, -ESTALE if the LVB is invalid
1000 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1001 struct gfs2_statfs_change_host *sc)
1003 gfs2_rgrp_verify(rgd);
1004 sc->sc_total += rgd->rd_data;
1005 sc->sc_free += rgd->rd_free;
1006 sc->sc_dinodes += rgd->rd_dinodes;
1011 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1012 * @sdp: the filesystem
1013 * @sc: the sc info that will be returned
1015 * Any error (other than a signal) will cause this routine to fall back
1016 * to the synchronous version.
1018 * FIXME: This really shouldn't busy wait like this.
1023 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1025 struct gfs2_rgrpd *rgd_next;
1026 struct gfs2_holder *gha, *gh;
1027 unsigned int slots = 64;
1032 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1033 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1037 rgd_next = gfs2_rgrpd_get_first(sdp);
1042 for (x = 0; x < slots; x++) {
1045 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1046 err = gfs2_glock_wait(gh);
1048 gfs2_holder_uninit(gh);
1052 error = statfs_slow_fill(
1053 gh->gh_gl->gl_object, sc);
1054 gfs2_glock_dq_uninit(gh);
1060 else if (rgd_next && !error) {
1061 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1065 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1069 if (signal_pending(current))
1070 error = -ERESTARTSYS;
1084 * gfs2_statfs_i - Do a statfs
1085 * @sdp: the filesystem
1086 * @sg: the sg structure
1091 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1093 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1094 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1096 spin_lock(&sdp->sd_statfs_spin);
1099 sc->sc_total += l_sc->sc_total;
1100 sc->sc_free += l_sc->sc_free;
1101 sc->sc_dinodes += l_sc->sc_dinodes;
1103 spin_unlock(&sdp->sd_statfs_spin);
1105 if (sc->sc_free < 0)
1107 if (sc->sc_free > sc->sc_total)
1108 sc->sc_free = sc->sc_total;
1109 if (sc->sc_dinodes < 0)
1116 * gfs2_statfs - Gather and return stats about the filesystem
1117 * @sb: The superblock
1118 * @statfsbuf: The buffer
1120 * Returns: 0 on success or error code
1123 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1125 struct super_block *sb = dentry->d_inode->i_sb;
1126 struct gfs2_sbd *sdp = sb->s_fs_info;
1127 struct gfs2_statfs_change_host sc;
1130 error = gfs2_rindex_update(sdp);
1134 if (gfs2_tune_get(sdp, gt_statfs_slow))
1135 error = gfs2_statfs_slow(sdp, &sc);
1137 error = gfs2_statfs_i(sdp, &sc);
1142 buf->f_type = GFS2_MAGIC;
1143 buf->f_bsize = sdp->sd_sb.sb_bsize;
1144 buf->f_blocks = sc.sc_total;
1145 buf->f_bfree = sc.sc_free;
1146 buf->f_bavail = sc.sc_free;
1147 buf->f_files = sc.sc_dinodes + sc.sc_free;
1148 buf->f_ffree = sc.sc_free;
1149 buf->f_namelen = GFS2_FNAMESIZE;
1155 * gfs2_remount_fs - called when the FS is remounted
1156 * @sb: the filesystem
1157 * @flags: the remount flags
1158 * @data: extra data passed in (not used right now)
1163 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1165 struct gfs2_sbd *sdp = sb->s_fs_info;
1166 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1167 struct gfs2_tune *gt = &sdp->sd_tune;
1170 spin_lock(>->gt_spin);
1171 args.ar_commit = gt->gt_logd_secs;
1172 args.ar_quota_quantum = gt->gt_quota_quantum;
1173 if (gt->gt_statfs_slow)
1174 args.ar_statfs_quantum = 0;
1176 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1177 spin_unlock(>->gt_spin);
1178 error = gfs2_mount_args(&args, data);
1182 /* Not allowed to change locking details */
1183 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1184 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1185 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1188 /* Some flags must not be changed */
1189 if (args_neq(&args, &sdp->sd_args, spectator) ||
1190 args_neq(&args, &sdp->sd_args, localflocks) ||
1191 args_neq(&args, &sdp->sd_args, meta))
1194 if (sdp->sd_args.ar_spectator)
1195 *flags |= MS_RDONLY;
1197 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1198 if (*flags & MS_RDONLY)
1199 error = gfs2_make_fs_ro(sdp);
1201 error = gfs2_make_fs_rw(sdp);
1206 sdp->sd_args = args;
1207 if (sdp->sd_args.ar_posix_acl)
1208 sb->s_flags |= MS_POSIXACL;
1210 sb->s_flags &= ~MS_POSIXACL;
1211 if (sdp->sd_args.ar_nobarrier)
1212 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1214 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1215 spin_lock(>->gt_spin);
1216 gt->gt_logd_secs = args.ar_commit;
1217 gt->gt_quota_quantum = args.ar_quota_quantum;
1218 if (args.ar_statfs_quantum) {
1219 gt->gt_statfs_slow = 0;
1220 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1223 gt->gt_statfs_slow = 1;
1224 gt->gt_statfs_quantum = 30;
1226 spin_unlock(>->gt_spin);
1228 gfs2_online_uevent(sdp);
1233 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1234 * @inode: The inode to drop
1236 * If we've received a callback on an iopen lock then its because a
1237 * remote node tried to deallocate the inode but failed due to this node
1238 * still having the inode open. Here we mark the link count zero
1239 * since we know that it must have reached zero if the GLF_DEMOTE flag
1240 * is set on the iopen glock. If we didn't do a disk read since the
1241 * remote node removed the final link then we might otherwise miss
1242 * this event. This check ensures that this node will deallocate the
1243 * inode's blocks, or alternatively pass the baton on to another
1244 * node for later deallocation.
1247 static int gfs2_drop_inode(struct inode *inode)
1249 struct gfs2_inode *ip = GFS2_I(inode);
1251 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && inode->i_nlink) {
1252 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1253 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1256 return generic_drop_inode(inode);
1259 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1265 } while (!IS_ROOT(d1));
1270 * gfs2_show_options - Show mount options for /proc/mounts
1271 * @s: seq_file structure
1272 * @root: root of this (sub)tree
1274 * Returns: 0 on success or error code
1277 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1279 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1280 struct gfs2_args *args = &sdp->sd_args;
1283 if (is_ancestor(root, sdp->sd_master_dir))
1284 seq_printf(s, ",meta");
1285 if (args->ar_lockproto[0])
1286 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1287 if (args->ar_locktable[0])
1288 seq_printf(s, ",locktable=%s", args->ar_locktable);
1289 if (args->ar_hostdata[0])
1290 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1291 if (args->ar_spectator)
1292 seq_printf(s, ",spectator");
1293 if (args->ar_localflocks)
1294 seq_printf(s, ",localflocks");
1296 seq_printf(s, ",debug");
1297 if (args->ar_posix_acl)
1298 seq_printf(s, ",acl");
1299 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1301 switch (args->ar_quota) {
1302 case GFS2_QUOTA_OFF:
1305 case GFS2_QUOTA_ACCOUNT:
1315 seq_printf(s, ",quota=%s", state);
1317 if (args->ar_suiddir)
1318 seq_printf(s, ",suiddir");
1319 if (args->ar_data != GFS2_DATA_DEFAULT) {
1321 switch (args->ar_data) {
1322 case GFS2_DATA_WRITEBACK:
1323 state = "writeback";
1325 case GFS2_DATA_ORDERED:
1332 seq_printf(s, ",data=%s", state);
1334 if (args->ar_discard)
1335 seq_printf(s, ",discard");
1336 val = sdp->sd_tune.gt_logd_secs;
1338 seq_printf(s, ",commit=%d", val);
1339 val = sdp->sd_tune.gt_statfs_quantum;
1341 seq_printf(s, ",statfs_quantum=%d", val);
1342 else if (sdp->sd_tune.gt_statfs_slow)
1343 seq_puts(s, ",statfs_quantum=0");
1344 val = sdp->sd_tune.gt_quota_quantum;
1346 seq_printf(s, ",quota_quantum=%d", val);
1347 if (args->ar_statfs_percent)
1348 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1349 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1352 switch (args->ar_errors) {
1353 case GFS2_ERRORS_WITHDRAW:
1356 case GFS2_ERRORS_PANIC:
1363 seq_printf(s, ",errors=%s", state);
1365 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1366 seq_printf(s, ",nobarrier");
1367 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1368 seq_printf(s, ",demote_interface_used");
1369 if (args->ar_rgrplvb)
1370 seq_printf(s, ",rgrplvb");
1374 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1376 struct inode *inode = &ip->i_inode;
1377 struct gfs2_glock *gl = ip->i_gl;
1379 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1380 truncate_inode_pages(&inode->i_data, 0);
1382 if (atomic_read(&gl->gl_revokes) == 0) {
1383 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1384 clear_bit(GLF_DIRTY, &gl->gl_flags);
1388 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1390 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1391 struct gfs2_rgrpd *rgd;
1392 struct gfs2_holder gh;
1395 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1396 gfs2_consist_inode(ip);
1400 error = gfs2_rindex_update(sdp);
1404 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1408 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1410 gfs2_consist_inode(ip);
1415 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1419 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1420 sdp->sd_jdesc->jd_blocks);
1422 goto out_rg_gunlock;
1424 gfs2_free_di(rgd, ip);
1426 gfs2_final_release_pages(ip);
1428 gfs2_trans_end(sdp);
1431 gfs2_glock_dq_uninit(&gh);
1433 gfs2_quota_unhold(ip);
1438 * gfs2_evict_inode - Remove an inode from cache
1439 * @inode: The inode to evict
1441 * There are three cases to consider:
1442 * 1. i_nlink == 0, we are final opener (and must deallocate)
1443 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1446 * If the fs is read only, then we have to treat all cases as per #3
1447 * since we are unable to do any deallocation. The inode will be
1448 * deallocated by the next read/write node to attempt an allocation
1449 * in the same resource group
1451 * We have to (at the moment) hold the inodes main lock to cover
1452 * the gap between unlocking the shared lock on the iopen lock and
1453 * taking the exclusive lock. I'd rather do a shared -> exclusive
1454 * conversion on the iopen lock, but we can change that later. This
1455 * is safe, just less efficient.
1458 static void gfs2_evict_inode(struct inode *inode)
1460 struct super_block *sb = inode->i_sb;
1461 struct gfs2_sbd *sdp = sb->s_fs_info;
1462 struct gfs2_inode *ip = GFS2_I(inode);
1463 struct gfs2_holder gh;
1466 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1471 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1474 /* Must not read inode block until block type has been verified */
1475 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1476 if (unlikely(error)) {
1477 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1478 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1482 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1483 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1488 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1489 error = gfs2_inode_refresh(ip);
1494 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1495 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1496 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1497 error = gfs2_glock_nq(&ip->i_iopen_gh);
1501 /* Case 1 starts here */
1503 if (S_ISDIR(inode->i_mode) &&
1504 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1505 error = gfs2_dir_exhash_dealloc(ip);
1511 error = gfs2_ea_dealloc(ip);
1516 if (!gfs2_is_stuffed(ip)) {
1517 error = gfs2_file_dealloc(ip);
1522 error = gfs2_dinode_dealloc(ip);
1526 gfs2_log_flush(sdp, ip->i_gl);
1527 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1528 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1529 filemap_fdatawrite(metamapping);
1530 filemap_fdatawait(metamapping);
1532 write_inode_now(inode, 1);
1533 gfs2_ail_flush(ip->i_gl, 0);
1535 /* Case 2 starts here */
1536 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1539 /* Needs to be done before glock release & also in a transaction */
1540 truncate_inode_pages(&inode->i_data, 0);
1541 gfs2_trans_end(sdp);
1544 /* Error path for case 1 */
1545 if (gfs2_rs_active(ip->i_res))
1546 gfs2_rs_deltree(ip->i_res);
1548 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1549 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1550 gfs2_glock_dq(&ip->i_iopen_gh);
1552 gfs2_holder_uninit(&ip->i_iopen_gh);
1553 gfs2_glock_dq_uninit(&gh);
1554 if (error && error != GLR_TRYFAILED && error != -EROFS)
1555 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1557 /* Case 3 starts here */
1558 truncate_inode_pages_final(&inode->i_data);
1559 gfs2_rs_delete(ip, NULL);
1560 gfs2_ordered_del_inode(ip);
1562 gfs2_dir_hash_inval(ip);
1563 ip->i_gl->gl_object = NULL;
1564 flush_delayed_work(&ip->i_gl->gl_work);
1565 gfs2_glock_add_to_lru(ip->i_gl);
1566 gfs2_glock_put(ip->i_gl);
1568 if (ip->i_iopen_gh.gh_gl) {
1569 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1570 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1571 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1575 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1577 struct gfs2_inode *ip;
1579 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1586 return &ip->i_inode;
1589 static void gfs2_i_callback(struct rcu_head *head)
1591 struct inode *inode = container_of(head, struct inode, i_rcu);
1592 kmem_cache_free(gfs2_inode_cachep, inode);
1595 static void gfs2_destroy_inode(struct inode *inode)
1597 call_rcu(&inode->i_rcu, gfs2_i_callback);
1600 const struct super_operations gfs2_super_ops = {
1601 .alloc_inode = gfs2_alloc_inode,
1602 .destroy_inode = gfs2_destroy_inode,
1603 .write_inode = gfs2_write_inode,
1604 .dirty_inode = gfs2_dirty_inode,
1605 .evict_inode = gfs2_evict_inode,
1606 .put_super = gfs2_put_super,
1607 .sync_fs = gfs2_sync_fs,
1608 .freeze_fs = gfs2_freeze,
1609 .unfreeze_fs = gfs2_unfreeze,
1610 .statfs = gfs2_statfs,
1611 .remount_fs = gfs2_remount_fs,
1612 .drop_inode = gfs2_drop_inode,
1613 .show_options = gfs2_show_options,