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 thaw_gh;
403 struct gfs2_log_header_host head;
406 error = init_threads(sdp);
410 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
415 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
417 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
421 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
427 /* Initialize some head of the log stuff */
428 sdp->sd_log_sequence = head.lh_sequence + 1;
429 gfs2_log_pointers_init(sdp, head.lh_blkno);
431 error = gfs2_quota_init(sdp);
435 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
437 gfs2_glock_dq_uninit(&thaw_gh);
442 thaw_gh.gh_flags |= GL_NOCACHE;
443 gfs2_glock_dq_uninit(&thaw_gh);
445 kthread_stop(sdp->sd_quotad_process);
446 kthread_stop(sdp->sd_logd_process);
450 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
452 const struct gfs2_statfs_change *str = buf;
454 sc->sc_total = be64_to_cpu(str->sc_total);
455 sc->sc_free = be64_to_cpu(str->sc_free);
456 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
459 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
461 struct gfs2_statfs_change *str = buf;
463 str->sc_total = cpu_to_be64(sc->sc_total);
464 str->sc_free = cpu_to_be64(sc->sc_free);
465 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
468 int gfs2_statfs_init(struct gfs2_sbd *sdp)
470 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
471 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
472 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
473 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
474 struct buffer_head *m_bh, *l_bh;
475 struct gfs2_holder gh;
478 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
483 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
487 if (sdp->sd_args.ar_spectator) {
488 spin_lock(&sdp->sd_statfs_spin);
489 gfs2_statfs_change_in(m_sc, m_bh->b_data +
490 sizeof(struct gfs2_dinode));
491 spin_unlock(&sdp->sd_statfs_spin);
493 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
497 spin_lock(&sdp->sd_statfs_spin);
498 gfs2_statfs_change_in(m_sc, m_bh->b_data +
499 sizeof(struct gfs2_dinode));
500 gfs2_statfs_change_in(l_sc, l_bh->b_data +
501 sizeof(struct gfs2_dinode));
502 spin_unlock(&sdp->sd_statfs_spin);
510 gfs2_glock_dq_uninit(&gh);
514 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
517 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
518 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
519 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
520 struct buffer_head *l_bh;
525 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
529 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
531 spin_lock(&sdp->sd_statfs_spin);
532 l_sc->sc_total += total;
533 l_sc->sc_free += free;
534 l_sc->sc_dinodes += dinodes;
535 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
536 if (sdp->sd_args.ar_statfs_percent) {
537 x = 100 * l_sc->sc_free;
538 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
539 if (x >= y || x <= -y)
542 spin_unlock(&sdp->sd_statfs_spin);
546 gfs2_wake_up_statfs(sdp);
549 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
550 struct buffer_head *l_bh)
552 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
553 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
554 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
555 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
557 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
559 spin_lock(&sdp->sd_statfs_spin);
560 m_sc->sc_total += l_sc->sc_total;
561 m_sc->sc_free += l_sc->sc_free;
562 m_sc->sc_dinodes += l_sc->sc_dinodes;
563 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
564 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
565 0, sizeof(struct gfs2_statfs_change));
566 spin_unlock(&sdp->sd_statfs_spin);
568 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
569 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
572 int gfs2_statfs_sync(struct super_block *sb, int type)
574 struct gfs2_sbd *sdp = sb->s_fs_info;
575 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
576 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
577 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
578 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
579 struct gfs2_holder gh;
580 struct buffer_head *m_bh, *l_bh;
583 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
588 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
592 spin_lock(&sdp->sd_statfs_spin);
593 gfs2_statfs_change_in(m_sc, m_bh->b_data +
594 sizeof(struct gfs2_dinode));
595 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
596 spin_unlock(&sdp->sd_statfs_spin);
599 spin_unlock(&sdp->sd_statfs_spin);
601 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
605 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
609 update_statfs(sdp, m_bh, l_bh);
610 sdp->sd_statfs_force_sync = 0;
619 gfs2_glock_dq_uninit(&gh);
624 struct list_head list;
625 struct gfs2_holder gh;
629 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
631 * @sdp: the file system
632 * @state: the state to put the transaction lock into
633 * @t_gh: the hold on the transaction lock
638 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
639 struct gfs2_holder *freeze_gh)
641 struct gfs2_inode *ip;
642 struct gfs2_jdesc *jd;
645 struct gfs2_log_header_host lh;
646 struct gfs2_inode *dip = GFS2_I(sdp->sd_root_dir->d_inode);
649 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0,
650 &sdp->sd_freeze_root_gh);
653 atomic_set(&sdp->sd_frozen_root, 1);
654 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
655 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
660 ip = GFS2_I(jd->jd_inode);
661 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
666 list_add(&lfcc->list, &list);
669 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
670 GL_NOCACHE, freeze_gh);
672 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
673 error = gfs2_jdesc_check(jd);
676 error = gfs2_find_jhead(jd, &lh);
679 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
686 gfs2_glock_dq_uninit(freeze_gh);
689 while (!list_empty(&list)) {
690 lfcc = list_entry(list.next, struct lfcc, list);
691 list_del(&lfcc->list);
692 gfs2_glock_dq_uninit(&lfcc->gh);
696 atomic_dec(&sdp->sd_frozen_root);
697 wait_event(sdp->sd_frozen_root_wait, atomic_read(&sdp->sd_frozen_root) == 0);
698 gfs2_glock_dq_uninit(&sdp->sd_freeze_root_gh);
703 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
705 struct gfs2_dinode *str = buf;
707 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
708 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
709 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
710 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
711 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
712 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
713 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
714 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
715 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
716 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
717 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
718 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
719 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
720 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
722 str->di_goal_meta = cpu_to_be64(ip->i_goal);
723 str->di_goal_data = cpu_to_be64(ip->i_goal);
724 str->di_generation = cpu_to_be64(ip->i_generation);
726 str->di_flags = cpu_to_be32(ip->i_diskflags);
727 str->di_height = cpu_to_be16(ip->i_height);
728 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
729 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
731 str->di_depth = cpu_to_be16(ip->i_depth);
732 str->di_entries = cpu_to_be32(ip->i_entries);
734 str->di_eattr = cpu_to_be64(ip->i_eattr);
735 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
736 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
737 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
741 * gfs2_write_inode - Make sure the inode is stable on the disk
743 * @wbc: The writeback control structure
748 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
750 struct gfs2_inode *ip = GFS2_I(inode);
751 struct gfs2_sbd *sdp = GFS2_SB(inode);
752 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
753 struct backing_dev_info *bdi = metamapping->backing_dev_info;
756 if (wbc->sync_mode == WB_SYNC_ALL)
757 gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
758 if (bdi->dirty_exceeded)
759 gfs2_ail1_flush(sdp, wbc);
761 filemap_fdatawrite(metamapping);
762 if (wbc->sync_mode == WB_SYNC_ALL)
763 ret = filemap_fdatawait(metamapping);
765 mark_inode_dirty_sync(inode);
770 * gfs2_dirty_inode - check for atime updates
771 * @inode: The inode in question
772 * @flags: The type of dirty
774 * Unfortunately it can be called under any combination of inode
775 * glock and transaction lock, so we have to check carefully.
777 * At the moment this deals only with atime - it should be possible
778 * to expand that role in future, once a review of the locking has
782 static void gfs2_dirty_inode(struct inode *inode, int flags)
784 struct gfs2_inode *ip = GFS2_I(inode);
785 struct gfs2_sbd *sdp = GFS2_SB(inode);
786 struct buffer_head *bh;
787 struct gfs2_holder gh;
789 int need_endtrans = 0;
792 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
795 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
796 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
798 fs_err(sdp, "dirty_inode: glock %d\n", ret);
802 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
805 if (current->journal_info == NULL) {
806 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
808 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
814 ret = gfs2_meta_inode_buffer(ip, &bh);
816 gfs2_trans_add_meta(ip->i_gl, bh);
817 gfs2_dinode_out(ip, bh->b_data);
825 gfs2_glock_dq_uninit(&gh);
829 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
830 * @sdp: the filesystem
835 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
837 struct gfs2_holder thaw_gh;
840 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
842 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
845 down_write(&sdp->sd_log_flush_lock);
846 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
847 up_write(&sdp->sd_log_flush_lock);
849 kthread_stop(sdp->sd_quotad_process);
850 kthread_stop(sdp->sd_logd_process);
852 flush_workqueue(gfs2_delete_workqueue);
853 gfs2_quota_sync(sdp->sd_vfs, 0);
854 gfs2_statfs_sync(sdp->sd_vfs, 0);
856 gfs2_log_flush(sdp, NULL, SHUTDOWN_FLUSH);
857 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
860 gfs2_glock_dq_uninit(&thaw_gh);
862 gfs2_quota_cleanup(sdp);
868 * gfs2_put_super - Unmount the filesystem
869 * @sb: The VFS superblock
873 static void gfs2_put_super(struct super_block *sb)
875 struct gfs2_sbd *sdp = sb->s_fs_info;
877 struct gfs2_jdesc *jd;
879 /* No more recovery requests */
880 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
883 /* Wait on outstanding recovery */
885 spin_lock(&sdp->sd_jindex_spin);
886 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
887 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
889 spin_unlock(&sdp->sd_jindex_spin);
890 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
891 TASK_UNINTERRUPTIBLE);
894 spin_unlock(&sdp->sd_jindex_spin);
896 if (!(sb->s_flags & MS_RDONLY)) {
897 error = gfs2_make_fs_ro(sdp);
901 /* At this point, we're through modifying the disk */
905 iput(sdp->sd_jindex);
906 iput(sdp->sd_statfs_inode);
907 iput(sdp->sd_rindex);
908 iput(sdp->sd_quota_inode);
910 gfs2_glock_put(sdp->sd_rename_gl);
911 gfs2_glock_put(sdp->sd_freeze_gl);
913 if (!sdp->sd_args.ar_spectator) {
914 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
915 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
916 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
917 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
918 iput(sdp->sd_sc_inode);
919 iput(sdp->sd_qc_inode);
922 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
923 gfs2_clear_rgrpd(sdp);
924 gfs2_jindex_free(sdp);
925 /* Take apart glock structures and buffer lists */
926 gfs2_gl_hash_clear(sdp);
927 /* Unmount the locking protocol */
928 gfs2_lm_unmount(sdp);
930 /* At this point, we're through participating in the lockspace */
931 gfs2_sys_fs_del(sdp);
935 * gfs2_sync_fs - sync the filesystem
936 * @sb: the superblock
938 * Flushes the log to disk.
941 static int gfs2_sync_fs(struct super_block *sb, int wait)
943 struct gfs2_sbd *sdp = sb->s_fs_info;
945 gfs2_quota_sync(sb, -1);
946 if (wait && sdp && !atomic_read(&sdp->sd_log_freeze))
947 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
952 * gfs2_freeze - prevent further writes to the filesystem
953 * @sb: the VFS structure for the filesystem
957 static int gfs2_freeze(struct super_block *sb)
959 struct gfs2_sbd *sdp = sb->s_fs_info;
962 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
966 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
972 fs_err(sdp, "waiting for recovery before freeze\n");
976 fs_err(sdp, "error freezing FS: %d\n", error);
980 fs_err(sdp, "retrying...\n");
987 * gfs2_unfreeze - reallow writes to the filesystem
988 * @sb: the VFS structure for the filesystem
992 static int gfs2_unfreeze(struct super_block *sb)
994 struct gfs2_sbd *sdp = sb->s_fs_info;
996 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
997 atomic_dec(&sdp->sd_frozen_root);
998 wait_event(sdp->sd_frozen_root_wait, atomic_read(&sdp->sd_frozen_root) == 0);
999 gfs2_glock_dq_uninit(&sdp->sd_freeze_root_gh);
1004 * statfs_fill - fill in the sg for a given RG
1006 * @sc: the sc structure
1008 * Returns: 0 on success, -ESTALE if the LVB is invalid
1011 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1012 struct gfs2_statfs_change_host *sc)
1014 gfs2_rgrp_verify(rgd);
1015 sc->sc_total += rgd->rd_data;
1016 sc->sc_free += rgd->rd_free;
1017 sc->sc_dinodes += rgd->rd_dinodes;
1022 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1023 * @sdp: the filesystem
1024 * @sc: the sc info that will be returned
1026 * Any error (other than a signal) will cause this routine to fall back
1027 * to the synchronous version.
1029 * FIXME: This really shouldn't busy wait like this.
1034 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1036 struct gfs2_rgrpd *rgd_next;
1037 struct gfs2_holder *gha, *gh;
1038 unsigned int slots = 64;
1043 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1044 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1048 rgd_next = gfs2_rgrpd_get_first(sdp);
1053 for (x = 0; x < slots; x++) {
1056 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1057 err = gfs2_glock_wait(gh);
1059 gfs2_holder_uninit(gh);
1063 error = statfs_slow_fill(
1064 gh->gh_gl->gl_object, sc);
1065 gfs2_glock_dq_uninit(gh);
1071 else if (rgd_next && !error) {
1072 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1076 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1080 if (signal_pending(current))
1081 error = -ERESTARTSYS;
1095 * gfs2_statfs_i - Do a statfs
1096 * @sdp: the filesystem
1097 * @sg: the sg structure
1102 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1104 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1105 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1107 spin_lock(&sdp->sd_statfs_spin);
1110 sc->sc_total += l_sc->sc_total;
1111 sc->sc_free += l_sc->sc_free;
1112 sc->sc_dinodes += l_sc->sc_dinodes;
1114 spin_unlock(&sdp->sd_statfs_spin);
1116 if (sc->sc_free < 0)
1118 if (sc->sc_free > sc->sc_total)
1119 sc->sc_free = sc->sc_total;
1120 if (sc->sc_dinodes < 0)
1127 * gfs2_statfs - Gather and return stats about the filesystem
1128 * @sb: The superblock
1129 * @statfsbuf: The buffer
1131 * Returns: 0 on success or error code
1134 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1136 struct super_block *sb = dentry->d_inode->i_sb;
1137 struct gfs2_sbd *sdp = sb->s_fs_info;
1138 struct gfs2_statfs_change_host sc;
1141 error = gfs2_rindex_update(sdp);
1145 if (gfs2_tune_get(sdp, gt_statfs_slow))
1146 error = gfs2_statfs_slow(sdp, &sc);
1148 error = gfs2_statfs_i(sdp, &sc);
1153 buf->f_type = GFS2_MAGIC;
1154 buf->f_bsize = sdp->sd_sb.sb_bsize;
1155 buf->f_blocks = sc.sc_total;
1156 buf->f_bfree = sc.sc_free;
1157 buf->f_bavail = sc.sc_free;
1158 buf->f_files = sc.sc_dinodes + sc.sc_free;
1159 buf->f_ffree = sc.sc_free;
1160 buf->f_namelen = GFS2_FNAMESIZE;
1166 * gfs2_remount_fs - called when the FS is remounted
1167 * @sb: the filesystem
1168 * @flags: the remount flags
1169 * @data: extra data passed in (not used right now)
1174 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1176 struct gfs2_sbd *sdp = sb->s_fs_info;
1177 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1178 struct gfs2_tune *gt = &sdp->sd_tune;
1181 sync_filesystem(sb);
1183 spin_lock(>->gt_spin);
1184 args.ar_commit = gt->gt_logd_secs;
1185 args.ar_quota_quantum = gt->gt_quota_quantum;
1186 if (gt->gt_statfs_slow)
1187 args.ar_statfs_quantum = 0;
1189 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1190 spin_unlock(>->gt_spin);
1191 error = gfs2_mount_args(&args, data);
1195 /* Not allowed to change locking details */
1196 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1197 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1198 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1201 /* Some flags must not be changed */
1202 if (args_neq(&args, &sdp->sd_args, spectator) ||
1203 args_neq(&args, &sdp->sd_args, localflocks) ||
1204 args_neq(&args, &sdp->sd_args, meta))
1207 if (sdp->sd_args.ar_spectator)
1208 *flags |= MS_RDONLY;
1210 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1211 if (*flags & MS_RDONLY)
1212 error = gfs2_make_fs_ro(sdp);
1214 error = gfs2_make_fs_rw(sdp);
1219 sdp->sd_args = args;
1220 if (sdp->sd_args.ar_posix_acl)
1221 sb->s_flags |= MS_POSIXACL;
1223 sb->s_flags &= ~MS_POSIXACL;
1224 if (sdp->sd_args.ar_nobarrier)
1225 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1227 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1228 spin_lock(>->gt_spin);
1229 gt->gt_logd_secs = args.ar_commit;
1230 gt->gt_quota_quantum = args.ar_quota_quantum;
1231 if (args.ar_statfs_quantum) {
1232 gt->gt_statfs_slow = 0;
1233 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1236 gt->gt_statfs_slow = 1;
1237 gt->gt_statfs_quantum = 30;
1239 spin_unlock(>->gt_spin);
1241 gfs2_online_uevent(sdp);
1246 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1247 * @inode: The inode to drop
1249 * If we've received a callback on an iopen lock then its because a
1250 * remote node tried to deallocate the inode but failed due to this node
1251 * still having the inode open. Here we mark the link count zero
1252 * since we know that it must have reached zero if the GLF_DEMOTE flag
1253 * is set on the iopen glock. If we didn't do a disk read since the
1254 * remote node removed the final link then we might otherwise miss
1255 * this event. This check ensures that this node will deallocate the
1256 * inode's blocks, or alternatively pass the baton on to another
1257 * node for later deallocation.
1260 static int gfs2_drop_inode(struct inode *inode)
1262 struct gfs2_inode *ip = GFS2_I(inode);
1264 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) && inode->i_nlink) {
1265 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1266 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1269 return generic_drop_inode(inode);
1272 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1278 } while (!IS_ROOT(d1));
1283 * gfs2_show_options - Show mount options for /proc/mounts
1284 * @s: seq_file structure
1285 * @root: root of this (sub)tree
1287 * Returns: 0 on success or error code
1290 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1292 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1293 struct gfs2_args *args = &sdp->sd_args;
1296 if (is_ancestor(root, sdp->sd_master_dir))
1297 seq_puts(s, ",meta");
1298 if (args->ar_lockproto[0])
1299 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1300 if (args->ar_locktable[0])
1301 seq_printf(s, ",locktable=%s", args->ar_locktable);
1302 if (args->ar_hostdata[0])
1303 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1304 if (args->ar_spectator)
1305 seq_puts(s, ",spectator");
1306 if (args->ar_localflocks)
1307 seq_puts(s, ",localflocks");
1309 seq_puts(s, ",debug");
1310 if (args->ar_posix_acl)
1311 seq_puts(s, ",acl");
1312 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1314 switch (args->ar_quota) {
1315 case GFS2_QUOTA_OFF:
1318 case GFS2_QUOTA_ACCOUNT:
1328 seq_printf(s, ",quota=%s", state);
1330 if (args->ar_suiddir)
1331 seq_puts(s, ",suiddir");
1332 if (args->ar_data != GFS2_DATA_DEFAULT) {
1334 switch (args->ar_data) {
1335 case GFS2_DATA_WRITEBACK:
1336 state = "writeback";
1338 case GFS2_DATA_ORDERED:
1345 seq_printf(s, ",data=%s", state);
1347 if (args->ar_discard)
1348 seq_puts(s, ",discard");
1349 val = sdp->sd_tune.gt_logd_secs;
1351 seq_printf(s, ",commit=%d", val);
1352 val = sdp->sd_tune.gt_statfs_quantum;
1354 seq_printf(s, ",statfs_quantum=%d", val);
1355 else if (sdp->sd_tune.gt_statfs_slow)
1356 seq_puts(s, ",statfs_quantum=0");
1357 val = sdp->sd_tune.gt_quota_quantum;
1359 seq_printf(s, ",quota_quantum=%d", val);
1360 if (args->ar_statfs_percent)
1361 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1362 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1365 switch (args->ar_errors) {
1366 case GFS2_ERRORS_WITHDRAW:
1369 case GFS2_ERRORS_PANIC:
1376 seq_printf(s, ",errors=%s", state);
1378 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1379 seq_puts(s, ",nobarrier");
1380 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1381 seq_puts(s, ",demote_interface_used");
1382 if (args->ar_rgrplvb)
1383 seq_puts(s, ",rgrplvb");
1387 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1389 struct inode *inode = &ip->i_inode;
1390 struct gfs2_glock *gl = ip->i_gl;
1392 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1393 truncate_inode_pages(&inode->i_data, 0);
1395 if (atomic_read(&gl->gl_revokes) == 0) {
1396 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1397 clear_bit(GLF_DIRTY, &gl->gl_flags);
1401 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1403 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1404 struct gfs2_rgrpd *rgd;
1405 struct gfs2_holder gh;
1408 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1409 gfs2_consist_inode(ip);
1413 error = gfs2_rindex_update(sdp);
1417 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1421 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1423 gfs2_consist_inode(ip);
1428 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1432 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1433 sdp->sd_jdesc->jd_blocks);
1435 goto out_rg_gunlock;
1437 gfs2_free_di(rgd, ip);
1439 gfs2_final_release_pages(ip);
1441 gfs2_trans_end(sdp);
1444 gfs2_glock_dq_uninit(&gh);
1446 gfs2_quota_unhold(ip);
1451 * gfs2_evict_inode - Remove an inode from cache
1452 * @inode: The inode to evict
1454 * There are three cases to consider:
1455 * 1. i_nlink == 0, we are final opener (and must deallocate)
1456 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1459 * If the fs is read only, then we have to treat all cases as per #3
1460 * since we are unable to do any deallocation. The inode will be
1461 * deallocated by the next read/write node to attempt an allocation
1462 * in the same resource group
1464 * We have to (at the moment) hold the inodes main lock to cover
1465 * the gap between unlocking the shared lock on the iopen lock and
1466 * taking the exclusive lock. I'd rather do a shared -> exclusive
1467 * conversion on the iopen lock, but we can change that later. This
1468 * is safe, just less efficient.
1471 static void gfs2_evict_inode(struct inode *inode)
1473 struct super_block *sb = inode->i_sb;
1474 struct gfs2_sbd *sdp = sb->s_fs_info;
1475 struct gfs2_inode *ip = GFS2_I(inode);
1476 struct gfs2_holder gh;
1479 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1484 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1487 /* Must not read inode block until block type has been verified */
1488 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1489 if (unlikely(error)) {
1490 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1491 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1495 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1496 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1501 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1502 error = gfs2_inode_refresh(ip);
1507 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1508 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1509 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1510 error = gfs2_glock_nq(&ip->i_iopen_gh);
1514 /* Case 1 starts here */
1516 if (S_ISDIR(inode->i_mode) &&
1517 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1518 error = gfs2_dir_exhash_dealloc(ip);
1524 error = gfs2_ea_dealloc(ip);
1529 if (!gfs2_is_stuffed(ip)) {
1530 error = gfs2_file_dealloc(ip);
1535 error = gfs2_dinode_dealloc(ip);
1539 gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
1540 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1541 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1542 filemap_fdatawrite(metamapping);
1543 filemap_fdatawait(metamapping);
1545 write_inode_now(inode, 1);
1546 gfs2_ail_flush(ip->i_gl, 0);
1548 /* Case 2 starts here */
1549 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1552 /* Needs to be done before glock release & also in a transaction */
1553 truncate_inode_pages(&inode->i_data, 0);
1554 gfs2_trans_end(sdp);
1557 /* Error path for case 1 */
1558 if (gfs2_rs_active(ip->i_res))
1559 gfs2_rs_deltree(ip->i_res);
1561 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1562 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1563 gfs2_glock_dq(&ip->i_iopen_gh);
1565 gfs2_holder_uninit(&ip->i_iopen_gh);
1566 gfs2_glock_dq_uninit(&gh);
1567 if (error && error != GLR_TRYFAILED && error != -EROFS)
1568 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1570 /* Case 3 starts here */
1571 truncate_inode_pages_final(&inode->i_data);
1572 gfs2_rs_delete(ip, NULL);
1573 gfs2_ordered_del_inode(ip);
1575 gfs2_dir_hash_inval(ip);
1576 ip->i_gl->gl_object = NULL;
1577 flush_delayed_work(&ip->i_gl->gl_work);
1578 gfs2_glock_add_to_lru(ip->i_gl);
1579 gfs2_glock_put(ip->i_gl);
1581 if (ip->i_iopen_gh.gh_gl) {
1582 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1583 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1584 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1588 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1590 struct gfs2_inode *ip;
1592 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1599 return &ip->i_inode;
1602 static void gfs2_i_callback(struct rcu_head *head)
1604 struct inode *inode = container_of(head, struct inode, i_rcu);
1605 kmem_cache_free(gfs2_inode_cachep, inode);
1608 static void gfs2_destroy_inode(struct inode *inode)
1610 call_rcu(&inode->i_rcu, gfs2_i_callback);
1613 const struct super_operations gfs2_super_ops = {
1614 .alloc_inode = gfs2_alloc_inode,
1615 .destroy_inode = gfs2_destroy_inode,
1616 .write_inode = gfs2_write_inode,
1617 .dirty_inode = gfs2_dirty_inode,
1618 .evict_inode = gfs2_evict_inode,
1619 .put_super = gfs2_put_super,
1620 .sync_fs = gfs2_sync_fs,
1621 .freeze_fs = gfs2_freeze,
1622 .unfreeze_fs = gfs2_unfreeze,
1623 .statfs = gfs2_statfs,
1624 .remount_fs = gfs2_remount_fs,
1625 .drop_inode = gfs2_drop_inode,
1626 .show_options = gfs2_show_options,