2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/bio.h>
11 #include <linux/sched.h>
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/completion.h>
15 #include <linux/buffer_head.h>
16 #include <linux/statfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/mount.h>
19 #include <linux/kthread.h>
20 #include <linux/delay.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/crc32.h>
23 #include <linux/time.h>
24 #include <linux/wait.h>
25 #include <linux/writeback.h>
26 #include <linux/backing-dev.h>
46 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
86 static const match_table_t tokens = {
87 {Opt_lockproto, "lockproto=%s"},
88 {Opt_locktable, "locktable=%s"},
89 {Opt_hostdata, "hostdata=%s"},
90 {Opt_spectator, "spectator"},
91 {Opt_spectator, "norecovery"},
92 {Opt_ignore_local_fs, "ignore_local_fs"},
93 {Opt_localflocks, "localflocks"},
94 {Opt_localcaching, "localcaching"},
96 {Opt_nodebug, "nodebug"},
97 {Opt_upgrade, "upgrade"},
100 {Opt_quota_off, "quota=off"},
101 {Opt_quota_account, "quota=account"},
102 {Opt_quota_on, "quota=on"},
103 {Opt_quota, "quota"},
104 {Opt_noquota, "noquota"},
105 {Opt_suiddir, "suiddir"},
106 {Opt_nosuiddir, "nosuiddir"},
107 {Opt_data_writeback, "data=writeback"},
108 {Opt_data_ordered, "data=ordered"},
110 {Opt_discard, "discard"},
111 {Opt_nodiscard, "nodiscard"},
112 {Opt_commit, "commit=%d"},
113 {Opt_err_withdraw, "errors=withdraw"},
114 {Opt_err_panic, "errors=panic"},
115 {Opt_statfs_quantum, "statfs_quantum=%d"},
116 {Opt_statfs_percent, "statfs_percent=%d"},
117 {Opt_quota_quantum, "quota_quantum=%d"},
118 {Opt_barrier, "barrier"},
119 {Opt_nobarrier, "nobarrier"},
120 {Opt_rgrplvb, "rgrplvb"},
121 {Opt_norgrplvb, "norgrplvb"},
126 * gfs2_mount_args - Parse mount options
127 * @args: The structure into which the parsed options will be written
128 * @options: The options to parse
133 int gfs2_mount_args(struct gfs2_args *args, char *options)
137 substring_t tmp[MAX_OPT_ARGS];
140 /* Split the options into tokens with the "," character and
144 o = strsep(&options, ",");
150 token = match_token(o, tokens, tmp);
153 match_strlcpy(args->ar_lockproto, &tmp[0],
157 match_strlcpy(args->ar_locktable, &tmp[0],
161 match_strlcpy(args->ar_hostdata, &tmp[0],
165 args->ar_spectator = 1;
167 case Opt_ignore_local_fs:
168 /* Retained for backwards compat only */
170 case Opt_localflocks:
171 args->ar_localflocks = 1;
173 case Opt_localcaching:
174 /* Retained for backwards compat only */
177 if (args->ar_errors == GFS2_ERRORS_PANIC) {
178 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
179 "are mutually exclusive.\n");
188 /* Retained for backwards compat only */
191 args->ar_posix_acl = 1;
194 args->ar_posix_acl = 0;
198 args->ar_quota = GFS2_QUOTA_OFF;
200 case Opt_quota_account:
201 args->ar_quota = GFS2_QUOTA_ACCOUNT;
205 args->ar_quota = GFS2_QUOTA_ON;
208 args->ar_suiddir = 1;
211 args->ar_suiddir = 0;
213 case Opt_data_writeback:
214 args->ar_data = GFS2_DATA_WRITEBACK;
216 case Opt_data_ordered:
217 args->ar_data = GFS2_DATA_ORDERED;
223 args->ar_discard = 1;
226 args->ar_discard = 0;
229 rv = match_int(&tmp[0], &args->ar_commit);
230 if (rv || args->ar_commit <= 0) {
231 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n");
232 return rv ? rv : -EINVAL;
235 case Opt_statfs_quantum:
236 rv = match_int(&tmp[0], &args->ar_statfs_quantum);
237 if (rv || args->ar_statfs_quantum < 0) {
238 printk(KERN_WARNING "GFS2: statfs_quantum mount option requires a non-negative numeric argument\n");
239 return rv ? rv : -EINVAL;
242 case Opt_quota_quantum:
243 rv = match_int(&tmp[0], &args->ar_quota_quantum);
244 if (rv || args->ar_quota_quantum <= 0) {
245 printk(KERN_WARNING "GFS2: quota_quantum mount option requires a positive numeric argument\n");
246 return rv ? rv : -EINVAL;
249 case Opt_statfs_percent:
250 rv = match_int(&tmp[0], &args->ar_statfs_percent);
251 if (rv || args->ar_statfs_percent < 0 ||
252 args->ar_statfs_percent > 100) {
253 printk(KERN_WARNING "statfs_percent mount option requires a numeric argument between 0 and 100\n");
254 return rv ? rv : -EINVAL;
257 case Opt_err_withdraw:
258 args->ar_errors = GFS2_ERRORS_WITHDRAW;
261 if (args->ar_debug) {
262 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic "
263 "are mutually exclusive.\n");
266 args->ar_errors = GFS2_ERRORS_PANIC;
269 args->ar_nobarrier = 0;
272 args->ar_nobarrier = 1;
275 args->ar_rgrplvb = 1;
278 args->ar_rgrplvb = 0;
282 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o);
291 * gfs2_jindex_free - Clear all the journal index information
292 * @sdp: The GFS2 superblock
296 void gfs2_jindex_free(struct gfs2_sbd *sdp)
298 struct list_head list, *head;
299 struct gfs2_jdesc *jd;
300 struct gfs2_journal_extent *jext;
302 spin_lock(&sdp->sd_jindex_spin);
303 list_add(&list, &sdp->sd_jindex_list);
304 list_del_init(&sdp->sd_jindex_list);
305 sdp->sd_journals = 0;
306 spin_unlock(&sdp->sd_jindex_spin);
308 while (!list_empty(&list)) {
309 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
310 head = &jd->extent_list;
311 while (!list_empty(head)) {
312 jext = list_entry(head->next,
313 struct gfs2_journal_extent,
315 list_del(&jext->extent_list);
318 list_del(&jd->jd_list);
324 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
326 struct gfs2_jdesc *jd;
329 list_for_each_entry(jd, head, jd_list) {
330 if (jd->jd_jid == jid) {
342 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
344 struct gfs2_jdesc *jd;
346 spin_lock(&sdp->sd_jindex_spin);
347 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
348 spin_unlock(&sdp->sd_jindex_spin);
353 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
355 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
356 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
357 u64 size = i_size_read(jd->jd_inode);
359 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, 1 << 30))
362 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
364 if (gfs2_write_alloc_required(ip, 0, size)) {
365 gfs2_consist_inode(ip);
372 static int init_threads(struct gfs2_sbd *sdp)
374 struct task_struct *p;
377 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
380 fs_err(sdp, "can't start logd thread: %d\n", error);
383 sdp->sd_logd_process = p;
385 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
388 fs_err(sdp, "can't start quotad thread: %d\n", error);
391 sdp->sd_quotad_process = p;
395 kthread_stop(sdp->sd_logd_process);
400 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
401 * @sdp: the filesystem
406 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
408 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
409 struct gfs2_glock *j_gl = ip->i_gl;
410 struct gfs2_holder t_gh;
411 struct gfs2_log_header_host head;
414 error = init_threads(sdp);
418 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
422 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
424 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
428 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
434 /* Initialize some head of the log stuff */
435 sdp->sd_log_sequence = head.lh_sequence + 1;
436 gfs2_log_pointers_init(sdp, head.lh_blkno);
438 error = gfs2_quota_init(sdp);
442 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
444 gfs2_glock_dq_uninit(&t_gh);
449 t_gh.gh_flags |= GL_NOCACHE;
450 gfs2_glock_dq_uninit(&t_gh);
452 kthread_stop(sdp->sd_quotad_process);
453 kthread_stop(sdp->sd_logd_process);
457 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
459 const struct gfs2_statfs_change *str = buf;
461 sc->sc_total = be64_to_cpu(str->sc_total);
462 sc->sc_free = be64_to_cpu(str->sc_free);
463 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
466 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
468 struct gfs2_statfs_change *str = buf;
470 str->sc_total = cpu_to_be64(sc->sc_total);
471 str->sc_free = cpu_to_be64(sc->sc_free);
472 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
475 int gfs2_statfs_init(struct gfs2_sbd *sdp)
477 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
478 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
479 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
480 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
481 struct buffer_head *m_bh, *l_bh;
482 struct gfs2_holder gh;
485 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
490 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
494 if (sdp->sd_args.ar_spectator) {
495 spin_lock(&sdp->sd_statfs_spin);
496 gfs2_statfs_change_in(m_sc, m_bh->b_data +
497 sizeof(struct gfs2_dinode));
498 spin_unlock(&sdp->sd_statfs_spin);
500 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
504 spin_lock(&sdp->sd_statfs_spin);
505 gfs2_statfs_change_in(m_sc, m_bh->b_data +
506 sizeof(struct gfs2_dinode));
507 gfs2_statfs_change_in(l_sc, l_bh->b_data +
508 sizeof(struct gfs2_dinode));
509 spin_unlock(&sdp->sd_statfs_spin);
517 gfs2_glock_dq_uninit(&gh);
521 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
524 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
525 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
526 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
527 struct buffer_head *l_bh;
532 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
536 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
538 spin_lock(&sdp->sd_statfs_spin);
539 l_sc->sc_total += total;
540 l_sc->sc_free += free;
541 l_sc->sc_dinodes += dinodes;
542 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
543 if (sdp->sd_args.ar_statfs_percent) {
544 x = 100 * l_sc->sc_free;
545 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
546 if (x >= y || x <= -y)
549 spin_unlock(&sdp->sd_statfs_spin);
553 gfs2_wake_up_statfs(sdp);
556 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
557 struct buffer_head *l_bh)
559 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
560 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
561 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
562 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
564 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
566 spin_lock(&sdp->sd_statfs_spin);
567 m_sc->sc_total += l_sc->sc_total;
568 m_sc->sc_free += l_sc->sc_free;
569 m_sc->sc_dinodes += l_sc->sc_dinodes;
570 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
571 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
572 0, sizeof(struct gfs2_statfs_change));
573 spin_unlock(&sdp->sd_statfs_spin);
575 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
576 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
579 int gfs2_statfs_sync(struct super_block *sb, int type)
581 struct gfs2_sbd *sdp = sb->s_fs_info;
582 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
583 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
584 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
585 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
586 struct gfs2_holder gh;
587 struct buffer_head *m_bh, *l_bh;
590 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
595 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
599 spin_lock(&sdp->sd_statfs_spin);
600 gfs2_statfs_change_in(m_sc, m_bh->b_data +
601 sizeof(struct gfs2_dinode));
602 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
603 spin_unlock(&sdp->sd_statfs_spin);
606 spin_unlock(&sdp->sd_statfs_spin);
608 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
612 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
616 update_statfs(sdp, m_bh, l_bh);
617 sdp->sd_statfs_force_sync = 0;
626 gfs2_glock_dq_uninit(&gh);
631 struct list_head list;
632 struct gfs2_holder gh;
636 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
638 * @sdp: the file system
639 * @state: the state to put the transaction lock into
640 * @t_gh: the hold on the transaction lock
645 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
646 struct gfs2_holder *t_gh)
648 struct gfs2_inode *ip;
649 struct gfs2_jdesc *jd;
652 struct gfs2_log_header_host lh;
655 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
656 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
661 ip = GFS2_I(jd->jd_inode);
662 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
667 list_add(&lfcc->list, &list);
670 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
673 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
674 error = gfs2_jdesc_check(jd);
677 error = gfs2_find_jhead(jd, &lh);
680 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
687 gfs2_glock_dq_uninit(t_gh);
690 while (!list_empty(&list)) {
691 lfcc = list_entry(list.next, struct lfcc, list);
692 list_del(&lfcc->list);
693 gfs2_glock_dq_uninit(&lfcc->gh);
699 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
701 struct gfs2_dinode *str = buf;
703 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
704 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
705 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
706 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
707 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
708 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
709 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
710 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
711 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
712 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
713 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
714 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
715 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
716 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
718 str->di_goal_meta = cpu_to_be64(ip->i_goal);
719 str->di_goal_data = cpu_to_be64(ip->i_goal);
720 str->di_generation = cpu_to_be64(ip->i_generation);
722 str->di_flags = cpu_to_be32(ip->i_diskflags);
723 str->di_height = cpu_to_be16(ip->i_height);
724 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
725 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
727 str->di_depth = cpu_to_be16(ip->i_depth);
728 str->di_entries = cpu_to_be32(ip->i_entries);
730 str->di_eattr = cpu_to_be64(ip->i_eattr);
731 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
732 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
733 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
737 * gfs2_write_inode - Make sure the inode is stable on the disk
739 * @wbc: The writeback control structure
744 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
746 struct gfs2_inode *ip = GFS2_I(inode);
747 struct gfs2_sbd *sdp = GFS2_SB(inode);
748 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
749 struct backing_dev_info *bdi = metamapping->backing_dev_info;
752 if (wbc->sync_mode == WB_SYNC_ALL)
753 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
754 if (bdi->dirty_exceeded)
755 gfs2_ail1_flush(sdp, wbc);
757 filemap_fdatawrite(metamapping);
758 if (wbc->sync_mode == WB_SYNC_ALL)
759 ret = filemap_fdatawait(metamapping);
761 mark_inode_dirty_sync(inode);
766 * gfs2_dirty_inode - check for atime updates
767 * @inode: The inode in question
768 * @flags: The type of dirty
770 * Unfortunately it can be called under any combination of inode
771 * glock and transaction lock, so we have to check carefully.
773 * At the moment this deals only with atime - it should be possible
774 * to expand that role in future, once a review of the locking has
778 static void gfs2_dirty_inode(struct inode *inode, int flags)
780 struct gfs2_inode *ip = GFS2_I(inode);
781 struct gfs2_sbd *sdp = GFS2_SB(inode);
782 struct buffer_head *bh;
783 struct gfs2_holder gh;
785 int need_endtrans = 0;
788 if (!(flags & (I_DIRTY_DATASYNC|I_DIRTY_SYNC)))
791 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
792 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
794 fs_err(sdp, "dirty_inode: glock %d\n", ret);
798 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
801 if (current->journal_info == NULL) {
802 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
804 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
810 ret = gfs2_meta_inode_buffer(ip, &bh);
812 gfs2_trans_add_meta(ip->i_gl, bh);
813 gfs2_dinode_out(ip, bh->b_data);
821 gfs2_glock_dq_uninit(&gh);
825 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
826 * @sdp: the filesystem
831 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
833 struct gfs2_holder t_gh;
836 kthread_stop(sdp->sd_quotad_process);
837 kthread_stop(sdp->sd_logd_process);
839 flush_workqueue(gfs2_delete_workqueue);
840 gfs2_quota_sync(sdp->sd_vfs, 0);
841 gfs2_statfs_sync(sdp->sd_vfs, 0);
843 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
845 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
848 gfs2_meta_syncfs(sdp);
849 gfs2_log_shutdown(sdp);
851 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
854 gfs2_glock_dq_uninit(&t_gh);
856 gfs2_quota_cleanup(sdp);
861 static int gfs2_umount_recovery_wait(void *word)
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 gfs2_umount_recovery_wait, 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_trans_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);
947 gfs2_log_flush(sdp, NULL);
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);
1001 * statfs_fill - fill in the sg for a given RG
1003 * @sc: the sc structure
1005 * Returns: 0 on success, -ESTALE if the LVB is invalid
1008 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
1009 struct gfs2_statfs_change_host *sc)
1011 gfs2_rgrp_verify(rgd);
1012 sc->sc_total += rgd->rd_data;
1013 sc->sc_free += rgd->rd_free;
1014 sc->sc_dinodes += rgd->rd_dinodes;
1019 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1020 * @sdp: the filesystem
1021 * @sc: the sc info that will be returned
1023 * Any error (other than a signal) will cause this routine to fall back
1024 * to the synchronous version.
1026 * FIXME: This really shouldn't busy wait like this.
1031 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1033 struct gfs2_rgrpd *rgd_next;
1034 struct gfs2_holder *gha, *gh;
1035 unsigned int slots = 64;
1040 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
1041 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
1045 rgd_next = gfs2_rgrpd_get_first(sdp);
1050 for (x = 0; x < slots; x++) {
1053 if (gh->gh_gl && gfs2_glock_poll(gh)) {
1054 err = gfs2_glock_wait(gh);
1056 gfs2_holder_uninit(gh);
1060 error = statfs_slow_fill(
1061 gh->gh_gl->gl_object, sc);
1062 gfs2_glock_dq_uninit(gh);
1068 else if (rgd_next && !error) {
1069 error = gfs2_glock_nq_init(rgd_next->rd_gl,
1073 rgd_next = gfs2_rgrpd_get_next(rgd_next);
1077 if (signal_pending(current))
1078 error = -ERESTARTSYS;
1092 * gfs2_statfs_i - Do a statfs
1093 * @sdp: the filesystem
1094 * @sg: the sg structure
1099 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
1101 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
1102 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
1104 spin_lock(&sdp->sd_statfs_spin);
1107 sc->sc_total += l_sc->sc_total;
1108 sc->sc_free += l_sc->sc_free;
1109 sc->sc_dinodes += l_sc->sc_dinodes;
1111 spin_unlock(&sdp->sd_statfs_spin);
1113 if (sc->sc_free < 0)
1115 if (sc->sc_free > sc->sc_total)
1116 sc->sc_free = sc->sc_total;
1117 if (sc->sc_dinodes < 0)
1124 * gfs2_statfs - Gather and return stats about the filesystem
1125 * @sb: The superblock
1126 * @statfsbuf: The buffer
1128 * Returns: 0 on success or error code
1131 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1133 struct super_block *sb = dentry->d_inode->i_sb;
1134 struct gfs2_sbd *sdp = sb->s_fs_info;
1135 struct gfs2_statfs_change_host sc;
1138 error = gfs2_rindex_update(sdp);
1142 if (gfs2_tune_get(sdp, gt_statfs_slow))
1143 error = gfs2_statfs_slow(sdp, &sc);
1145 error = gfs2_statfs_i(sdp, &sc);
1150 buf->f_type = GFS2_MAGIC;
1151 buf->f_bsize = sdp->sd_sb.sb_bsize;
1152 buf->f_blocks = sc.sc_total;
1153 buf->f_bfree = sc.sc_free;
1154 buf->f_bavail = sc.sc_free;
1155 buf->f_files = sc.sc_dinodes + sc.sc_free;
1156 buf->f_ffree = sc.sc_free;
1157 buf->f_namelen = GFS2_FNAMESIZE;
1163 * gfs2_remount_fs - called when the FS is remounted
1164 * @sb: the filesystem
1165 * @flags: the remount flags
1166 * @data: extra data passed in (not used right now)
1171 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
1173 struct gfs2_sbd *sdp = sb->s_fs_info;
1174 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
1175 struct gfs2_tune *gt = &sdp->sd_tune;
1178 spin_lock(>->gt_spin);
1179 args.ar_commit = gt->gt_logd_secs;
1180 args.ar_quota_quantum = gt->gt_quota_quantum;
1181 if (gt->gt_statfs_slow)
1182 args.ar_statfs_quantum = 0;
1184 args.ar_statfs_quantum = gt->gt_statfs_quantum;
1185 spin_unlock(>->gt_spin);
1186 error = gfs2_mount_args(&args, data);
1190 /* Not allowed to change locking details */
1191 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
1192 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
1193 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
1196 /* Some flags must not be changed */
1197 if (args_neq(&args, &sdp->sd_args, spectator) ||
1198 args_neq(&args, &sdp->sd_args, localflocks) ||
1199 args_neq(&args, &sdp->sd_args, meta))
1202 if (sdp->sd_args.ar_spectator)
1203 *flags |= MS_RDONLY;
1205 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
1206 if (*flags & MS_RDONLY)
1207 error = gfs2_make_fs_ro(sdp);
1209 error = gfs2_make_fs_rw(sdp);
1214 sdp->sd_args = args;
1215 if (sdp->sd_args.ar_posix_acl)
1216 sb->s_flags |= MS_POSIXACL;
1218 sb->s_flags &= ~MS_POSIXACL;
1219 if (sdp->sd_args.ar_nobarrier)
1220 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1222 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1223 spin_lock(>->gt_spin);
1224 gt->gt_logd_secs = args.ar_commit;
1225 gt->gt_quota_quantum = args.ar_quota_quantum;
1226 if (args.ar_statfs_quantum) {
1227 gt->gt_statfs_slow = 0;
1228 gt->gt_statfs_quantum = args.ar_statfs_quantum;
1231 gt->gt_statfs_slow = 1;
1232 gt->gt_statfs_quantum = 30;
1234 spin_unlock(>->gt_spin);
1236 gfs2_online_uevent(sdp);
1241 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1242 * @inode: The inode to drop
1244 * If we've received a callback on an iopen lock then its because a
1245 * remote node tried to deallocate the inode but failed due to this node
1246 * still having the inode open. Here we mark the link count zero
1247 * since we know that it must have reached zero if the GLF_DEMOTE flag
1248 * is set on the iopen glock. If we didn't do a disk read since the
1249 * remote node removed the final link then we might otherwise miss
1250 * this event. This check ensures that this node will deallocate the
1251 * inode's blocks, or alternatively pass the baton on to another
1252 * node for later deallocation.
1255 static int gfs2_drop_inode(struct inode *inode)
1257 struct gfs2_inode *ip = GFS2_I(inode);
1259 if (inode->i_nlink) {
1260 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1261 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
1264 return generic_drop_inode(inode);
1267 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1273 } while (!IS_ROOT(d1));
1278 * gfs2_show_options - Show mount options for /proc/mounts
1279 * @s: seq_file structure
1280 * @root: root of this (sub)tree
1282 * Returns: 0 on success or error code
1285 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1287 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1288 struct gfs2_args *args = &sdp->sd_args;
1291 if (is_ancestor(root, sdp->sd_master_dir))
1292 seq_printf(s, ",meta");
1293 if (args->ar_lockproto[0])
1294 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
1295 if (args->ar_locktable[0])
1296 seq_printf(s, ",locktable=%s", args->ar_locktable);
1297 if (args->ar_hostdata[0])
1298 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
1299 if (args->ar_spectator)
1300 seq_printf(s, ",spectator");
1301 if (args->ar_localflocks)
1302 seq_printf(s, ",localflocks");
1304 seq_printf(s, ",debug");
1305 if (args->ar_posix_acl)
1306 seq_printf(s, ",acl");
1307 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1309 switch (args->ar_quota) {
1310 case GFS2_QUOTA_OFF:
1313 case GFS2_QUOTA_ACCOUNT:
1323 seq_printf(s, ",quota=%s", state);
1325 if (args->ar_suiddir)
1326 seq_printf(s, ",suiddir");
1327 if (args->ar_data != GFS2_DATA_DEFAULT) {
1329 switch (args->ar_data) {
1330 case GFS2_DATA_WRITEBACK:
1331 state = "writeback";
1333 case GFS2_DATA_ORDERED:
1340 seq_printf(s, ",data=%s", state);
1342 if (args->ar_discard)
1343 seq_printf(s, ",discard");
1344 val = sdp->sd_tune.gt_logd_secs;
1346 seq_printf(s, ",commit=%d", val);
1347 val = sdp->sd_tune.gt_statfs_quantum;
1349 seq_printf(s, ",statfs_quantum=%d", val);
1350 else if (sdp->sd_tune.gt_statfs_slow)
1351 seq_puts(s, ",statfs_quantum=0");
1352 val = sdp->sd_tune.gt_quota_quantum;
1354 seq_printf(s, ",quota_quantum=%d", val);
1355 if (args->ar_statfs_percent)
1356 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1357 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1360 switch (args->ar_errors) {
1361 case GFS2_ERRORS_WITHDRAW:
1364 case GFS2_ERRORS_PANIC:
1371 seq_printf(s, ",errors=%s", state);
1373 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1374 seq_printf(s, ",nobarrier");
1375 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1376 seq_printf(s, ",demote_interface_used");
1377 if (args->ar_rgrplvb)
1378 seq_printf(s, ",rgrplvb");
1382 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1384 struct inode *inode = &ip->i_inode;
1385 struct gfs2_glock *gl = ip->i_gl;
1387 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1388 truncate_inode_pages(&inode->i_data, 0);
1390 if (atomic_read(&gl->gl_revokes) == 0) {
1391 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1392 clear_bit(GLF_DIRTY, &gl->gl_flags);
1396 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1398 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1399 struct gfs2_rgrpd *rgd;
1400 struct gfs2_holder gh;
1403 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1404 gfs2_consist_inode(ip);
1408 error = gfs2_rindex_update(sdp);
1412 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1416 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1418 gfs2_consist_inode(ip);
1423 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1427 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1428 sdp->sd_jdesc->jd_blocks);
1430 goto out_rg_gunlock;
1432 gfs2_free_di(rgd, ip);
1434 gfs2_final_release_pages(ip);
1436 gfs2_trans_end(sdp);
1439 gfs2_glock_dq_uninit(&gh);
1441 gfs2_quota_unhold(ip);
1446 * gfs2_evict_inode - Remove an inode from cache
1447 * @inode: The inode to evict
1449 * There are three cases to consider:
1450 * 1. i_nlink == 0, we are final opener (and must deallocate)
1451 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1454 * If the fs is read only, then we have to treat all cases as per #3
1455 * since we are unable to do any deallocation. The inode will be
1456 * deallocated by the next read/write node to attempt an allocation
1457 * in the same resource group
1459 * We have to (at the moment) hold the inodes main lock to cover
1460 * the gap between unlocking the shared lock on the iopen lock and
1461 * taking the exclusive lock. I'd rather do a shared -> exclusive
1462 * conversion on the iopen lock, but we can change that later. This
1463 * is safe, just less efficient.
1466 static void gfs2_evict_inode(struct inode *inode)
1468 struct super_block *sb = inode->i_sb;
1469 struct gfs2_sbd *sdp = sb->s_fs_info;
1470 struct gfs2_inode *ip = GFS2_I(inode);
1471 struct gfs2_holder gh;
1474 if (inode->i_nlink || (sb->s_flags & MS_RDONLY))
1477 /* Must not read inode block until block type has been verified */
1478 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1479 if (unlikely(error)) {
1480 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1481 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1485 if (!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1486 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1491 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1492 error = gfs2_inode_refresh(ip);
1497 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1498 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1499 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
1500 error = gfs2_glock_nq(&ip->i_iopen_gh);
1504 /* Case 1 starts here */
1506 if (S_ISDIR(inode->i_mode) &&
1507 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1508 error = gfs2_dir_exhash_dealloc(ip);
1514 error = gfs2_ea_dealloc(ip);
1519 if (!gfs2_is_stuffed(ip)) {
1520 error = gfs2_file_dealloc(ip);
1525 error = gfs2_dinode_dealloc(ip);
1529 gfs2_log_flush(sdp, ip->i_gl);
1530 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1531 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
1532 filemap_fdatawrite(metamapping);
1533 filemap_fdatawait(metamapping);
1535 write_inode_now(inode, 1);
1536 gfs2_ail_flush(ip->i_gl, 0);
1538 /* Case 2 starts here */
1539 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1542 /* Needs to be done before glock release & also in a transaction */
1543 truncate_inode_pages(&inode->i_data, 0);
1544 gfs2_trans_end(sdp);
1547 /* Error path for case 1 */
1548 if (gfs2_rs_active(ip->i_res))
1549 gfs2_rs_deltree(ip->i_res);
1551 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1552 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1553 gfs2_glock_dq(&ip->i_iopen_gh);
1555 gfs2_holder_uninit(&ip->i_iopen_gh);
1556 gfs2_glock_dq_uninit(&gh);
1557 if (error && error != GLR_TRYFAILED && error != -EROFS)
1558 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1560 /* Case 3 starts here */
1561 truncate_inode_pages(&inode->i_data, 0);
1562 gfs2_rs_delete(ip, NULL);
1563 gfs2_ordered_del_inode(ip);
1565 gfs2_dir_hash_inval(ip);
1566 ip->i_gl->gl_object = NULL;
1567 flush_delayed_work(&ip->i_gl->gl_work);
1568 gfs2_glock_add_to_lru(ip->i_gl);
1569 gfs2_glock_put(ip->i_gl);
1571 if (ip->i_iopen_gh.gh_gl) {
1572 ip->i_iopen_gh.gh_gl->gl_object = NULL;
1573 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1574 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1578 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1580 struct gfs2_inode *ip;
1582 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1589 return &ip->i_inode;
1592 static void gfs2_i_callback(struct rcu_head *head)
1594 struct inode *inode = container_of(head, struct inode, i_rcu);
1595 kmem_cache_free(gfs2_inode_cachep, inode);
1598 static void gfs2_destroy_inode(struct inode *inode)
1600 call_rcu(&inode->i_rcu, gfs2_i_callback);
1603 const struct super_operations gfs2_super_ops = {
1604 .alloc_inode = gfs2_alloc_inode,
1605 .destroy_inode = gfs2_destroy_inode,
1606 .write_inode = gfs2_write_inode,
1607 .dirty_inode = gfs2_dirty_inode,
1608 .evict_inode = gfs2_evict_inode,
1609 .put_super = gfs2_put_super,
1610 .sync_fs = gfs2_sync_fs,
1611 .freeze_fs = gfs2_freeze,
1612 .unfreeze_fs = gfs2_unfreeze,
1613 .statfs = gfs2_statfs,
1614 .remount_fs = gfs2_remount_fs,
1615 .drop_inode = gfs2_drop_inode,
1616 .show_options = gfs2_show_options,