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/signal.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>
29 #include <linux/kernel.h>
49 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
91 static const match_table_t tokens
= {
92 {Opt_lockproto
, "lockproto=%s"},
93 {Opt_locktable
, "locktable=%s"},
94 {Opt_hostdata
, "hostdata=%s"},
95 {Opt_spectator
, "spectator"},
96 {Opt_spectator
, "norecovery"},
97 {Opt_ignore_local_fs
, "ignore_local_fs"},
98 {Opt_localflocks
, "localflocks"},
99 {Opt_localcaching
, "localcaching"},
100 {Opt_debug
, "debug"},
101 {Opt_nodebug
, "nodebug"},
102 {Opt_upgrade
, "upgrade"},
104 {Opt_noacl
, "noacl"},
105 {Opt_quota_off
, "quota=off"},
106 {Opt_quota_account
, "quota=account"},
107 {Opt_quota_on
, "quota=on"},
108 {Opt_quota
, "quota"},
109 {Opt_noquota
, "noquota"},
110 {Opt_suiddir
, "suiddir"},
111 {Opt_nosuiddir
, "nosuiddir"},
112 {Opt_data_writeback
, "data=writeback"},
113 {Opt_data_ordered
, "data=ordered"},
115 {Opt_discard
, "discard"},
116 {Opt_nodiscard
, "nodiscard"},
117 {Opt_commit
, "commit=%d"},
118 {Opt_err_withdraw
, "errors=withdraw"},
119 {Opt_err_panic
, "errors=panic"},
120 {Opt_statfs_quantum
, "statfs_quantum=%d"},
121 {Opt_statfs_percent
, "statfs_percent=%d"},
122 {Opt_quota_quantum
, "quota_quantum=%d"},
123 {Opt_barrier
, "barrier"},
124 {Opt_nobarrier
, "nobarrier"},
125 {Opt_rgrplvb
, "rgrplvb"},
126 {Opt_norgrplvb
, "norgrplvb"},
127 {Opt_loccookie
, "loccookie"},
128 {Opt_noloccookie
, "noloccookie"},
133 * gfs2_mount_args - Parse mount options
134 * @args: The structure into which the parsed options will be written
135 * @options: The options to parse
140 int gfs2_mount_args(struct gfs2_args
*args
, char *options
)
144 substring_t tmp
[MAX_OPT_ARGS
];
147 /* Split the options into tokens with the "," character and
151 o
= strsep(&options
, ",");
157 token
= match_token(o
, tokens
, tmp
);
160 match_strlcpy(args
->ar_lockproto
, &tmp
[0],
164 match_strlcpy(args
->ar_locktable
, &tmp
[0],
168 match_strlcpy(args
->ar_hostdata
, &tmp
[0],
172 args
->ar_spectator
= 1;
174 case Opt_ignore_local_fs
:
175 /* Retained for backwards compat only */
177 case Opt_localflocks
:
178 args
->ar_localflocks
= 1;
180 case Opt_localcaching
:
181 /* Retained for backwards compat only */
184 if (args
->ar_errors
== GFS2_ERRORS_PANIC
) {
185 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
194 /* Retained for backwards compat only */
197 args
->ar_posix_acl
= 1;
200 args
->ar_posix_acl
= 0;
204 args
->ar_quota
= GFS2_QUOTA_OFF
;
206 case Opt_quota_account
:
207 args
->ar_quota
= GFS2_QUOTA_ACCOUNT
;
211 args
->ar_quota
= GFS2_QUOTA_ON
;
214 args
->ar_suiddir
= 1;
217 args
->ar_suiddir
= 0;
219 case Opt_data_writeback
:
220 args
->ar_data
= GFS2_DATA_WRITEBACK
;
222 case Opt_data_ordered
:
223 args
->ar_data
= GFS2_DATA_ORDERED
;
229 args
->ar_discard
= 1;
232 args
->ar_discard
= 0;
235 rv
= match_int(&tmp
[0], &args
->ar_commit
);
236 if (rv
|| args
->ar_commit
<= 0) {
237 pr_warn("commit mount option requires a positive numeric argument\n");
238 return rv
? rv
: -EINVAL
;
241 case Opt_statfs_quantum
:
242 rv
= match_int(&tmp
[0], &args
->ar_statfs_quantum
);
243 if (rv
|| args
->ar_statfs_quantum
< 0) {
244 pr_warn("statfs_quantum mount option requires a non-negative numeric argument\n");
245 return rv
? rv
: -EINVAL
;
248 case Opt_quota_quantum
:
249 rv
= match_int(&tmp
[0], &args
->ar_quota_quantum
);
250 if (rv
|| args
->ar_quota_quantum
<= 0) {
251 pr_warn("quota_quantum mount option requires a positive numeric argument\n");
252 return rv
? rv
: -EINVAL
;
255 case Opt_statfs_percent
:
256 rv
= match_int(&tmp
[0], &args
->ar_statfs_percent
);
257 if (rv
|| args
->ar_statfs_percent
< 0 ||
258 args
->ar_statfs_percent
> 100) {
259 pr_warn("statfs_percent mount option requires a numeric argument between 0 and 100\n");
260 return rv
? rv
: -EINVAL
;
263 case Opt_err_withdraw
:
264 args
->ar_errors
= GFS2_ERRORS_WITHDRAW
;
267 if (args
->ar_debug
) {
268 pr_warn("-o debug and -o errors=panic are mutually exclusive\n");
271 args
->ar_errors
= GFS2_ERRORS_PANIC
;
274 args
->ar_nobarrier
= 0;
277 args
->ar_nobarrier
= 1;
280 args
->ar_rgrplvb
= 1;
283 args
->ar_rgrplvb
= 0;
286 args
->ar_loccookie
= 1;
288 case Opt_noloccookie
:
289 args
->ar_loccookie
= 0;
293 pr_warn("invalid mount option: %s\n", o
);
302 * gfs2_jindex_free - Clear all the journal index information
303 * @sdp: The GFS2 superblock
307 void gfs2_jindex_free(struct gfs2_sbd
*sdp
)
309 struct list_head list
;
310 struct gfs2_jdesc
*jd
;
312 spin_lock(&sdp
->sd_jindex_spin
);
313 list_add(&list
, &sdp
->sd_jindex_list
);
314 list_del_init(&sdp
->sd_jindex_list
);
315 sdp
->sd_journals
= 0;
316 spin_unlock(&sdp
->sd_jindex_spin
);
318 while (!list_empty(&list
)) {
319 jd
= list_entry(list
.next
, struct gfs2_jdesc
, jd_list
);
320 gfs2_free_journal_extents(jd
);
321 list_del(&jd
->jd_list
);
327 static struct gfs2_jdesc
*jdesc_find_i(struct list_head
*head
, unsigned int jid
)
329 struct gfs2_jdesc
*jd
;
332 list_for_each_entry(jd
, head
, jd_list
) {
333 if (jd
->jd_jid
== jid
) {
345 struct gfs2_jdesc
*gfs2_jdesc_find(struct gfs2_sbd
*sdp
, unsigned int jid
)
347 struct gfs2_jdesc
*jd
;
349 spin_lock(&sdp
->sd_jindex_spin
);
350 jd
= jdesc_find_i(&sdp
->sd_jindex_list
, jid
);
351 spin_unlock(&sdp
->sd_jindex_spin
);
356 int gfs2_jdesc_check(struct gfs2_jdesc
*jd
)
358 struct gfs2_inode
*ip
= GFS2_I(jd
->jd_inode
);
359 struct gfs2_sbd
*sdp
= GFS2_SB(jd
->jd_inode
);
360 u64 size
= i_size_read(jd
->jd_inode
);
362 if (gfs2_check_internal_file_size(jd
->jd_inode
, 8 << 20, BIT(30)))
365 jd
->jd_blocks
= size
>> sdp
->sd_sb
.sb_bsize_shift
;
367 if (gfs2_write_alloc_required(ip
, 0, size
)) {
368 gfs2_consist_inode(ip
);
375 static int init_threads(struct gfs2_sbd
*sdp
)
377 struct task_struct
*p
;
380 p
= kthread_run(gfs2_logd
, sdp
, "gfs2_logd");
383 fs_err(sdp
, "can't start logd thread: %d\n", error
);
386 sdp
->sd_logd_process
= p
;
388 p
= kthread_run(gfs2_quotad
, sdp
, "gfs2_quotad");
391 fs_err(sdp
, "can't start quotad thread: %d\n", error
);
394 sdp
->sd_quotad_process
= p
;
398 kthread_stop(sdp
->sd_logd_process
);
403 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
404 * @sdp: the filesystem
409 int gfs2_make_fs_rw(struct gfs2_sbd
*sdp
)
411 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
412 struct gfs2_glock
*j_gl
= ip
->i_gl
;
413 struct gfs2_holder freeze_gh
;
414 struct gfs2_log_header_host head
;
417 error
= init_threads(sdp
);
421 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_SHARED
, 0,
426 j_gl
->gl_ops
->go_inval(j_gl
, DIO_METADATA
);
428 error
= gfs2_find_jhead(sdp
->sd_jdesc
, &head
);
432 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
438 /* Initialize some head of the log stuff */
439 sdp
->sd_log_sequence
= head
.lh_sequence
+ 1;
440 gfs2_log_pointers_init(sdp
, head
.lh_blkno
);
442 error
= gfs2_quota_init(sdp
);
446 set_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
);
448 gfs2_glock_dq_uninit(&freeze_gh
);
453 freeze_gh
.gh_flags
|= GL_NOCACHE
;
454 gfs2_glock_dq_uninit(&freeze_gh
);
456 kthread_stop(sdp
->sd_quotad_process
);
457 kthread_stop(sdp
->sd_logd_process
);
461 void gfs2_statfs_change_in(struct gfs2_statfs_change_host
*sc
, const void *buf
)
463 const struct gfs2_statfs_change
*str
= buf
;
465 sc
->sc_total
= be64_to_cpu(str
->sc_total
);
466 sc
->sc_free
= be64_to_cpu(str
->sc_free
);
467 sc
->sc_dinodes
= be64_to_cpu(str
->sc_dinodes
);
470 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host
*sc
, void *buf
)
472 struct gfs2_statfs_change
*str
= buf
;
474 str
->sc_total
= cpu_to_be64(sc
->sc_total
);
475 str
->sc_free
= cpu_to_be64(sc
->sc_free
);
476 str
->sc_dinodes
= cpu_to_be64(sc
->sc_dinodes
);
479 int gfs2_statfs_init(struct gfs2_sbd
*sdp
)
481 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
482 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
483 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
484 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
485 struct buffer_head
*m_bh
, *l_bh
;
486 struct gfs2_holder gh
;
489 error
= gfs2_glock_nq_init(m_ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_NOCACHE
,
494 error
= gfs2_meta_inode_buffer(m_ip
, &m_bh
);
498 if (sdp
->sd_args
.ar_spectator
) {
499 spin_lock(&sdp
->sd_statfs_spin
);
500 gfs2_statfs_change_in(m_sc
, m_bh
->b_data
+
501 sizeof(struct gfs2_dinode
));
502 spin_unlock(&sdp
->sd_statfs_spin
);
504 error
= gfs2_meta_inode_buffer(l_ip
, &l_bh
);
508 spin_lock(&sdp
->sd_statfs_spin
);
509 gfs2_statfs_change_in(m_sc
, m_bh
->b_data
+
510 sizeof(struct gfs2_dinode
));
511 gfs2_statfs_change_in(l_sc
, l_bh
->b_data
+
512 sizeof(struct gfs2_dinode
));
513 spin_unlock(&sdp
->sd_statfs_spin
);
521 gfs2_glock_dq_uninit(&gh
);
525 void gfs2_statfs_change(struct gfs2_sbd
*sdp
, s64 total
, s64 free
,
528 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
529 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
530 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
531 struct buffer_head
*l_bh
;
536 error
= gfs2_meta_inode_buffer(l_ip
, &l_bh
);
540 gfs2_trans_add_meta(l_ip
->i_gl
, l_bh
);
542 spin_lock(&sdp
->sd_statfs_spin
);
543 l_sc
->sc_total
+= total
;
544 l_sc
->sc_free
+= free
;
545 l_sc
->sc_dinodes
+= dinodes
;
546 gfs2_statfs_change_out(l_sc
, l_bh
->b_data
+ sizeof(struct gfs2_dinode
));
547 if (sdp
->sd_args
.ar_statfs_percent
) {
548 x
= 100 * l_sc
->sc_free
;
549 y
= m_sc
->sc_free
* sdp
->sd_args
.ar_statfs_percent
;
550 if (x
>= y
|| x
<= -y
)
553 spin_unlock(&sdp
->sd_statfs_spin
);
557 gfs2_wake_up_statfs(sdp
);
560 void update_statfs(struct gfs2_sbd
*sdp
, struct buffer_head
*m_bh
,
561 struct buffer_head
*l_bh
)
563 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
564 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
565 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
566 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
568 gfs2_trans_add_meta(l_ip
->i_gl
, l_bh
);
569 gfs2_trans_add_meta(m_ip
->i_gl
, m_bh
);
571 spin_lock(&sdp
->sd_statfs_spin
);
572 m_sc
->sc_total
+= l_sc
->sc_total
;
573 m_sc
->sc_free
+= l_sc
->sc_free
;
574 m_sc
->sc_dinodes
+= l_sc
->sc_dinodes
;
575 memset(l_sc
, 0, sizeof(struct gfs2_statfs_change
));
576 memset(l_bh
->b_data
+ sizeof(struct gfs2_dinode
),
577 0, sizeof(struct gfs2_statfs_change
));
578 gfs2_statfs_change_out(m_sc
, m_bh
->b_data
+ sizeof(struct gfs2_dinode
));
579 spin_unlock(&sdp
->sd_statfs_spin
);
582 int gfs2_statfs_sync(struct super_block
*sb
, int type
)
584 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
585 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
586 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
587 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
588 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
589 struct gfs2_holder gh
;
590 struct buffer_head
*m_bh
, *l_bh
;
594 error
= gfs2_glock_nq_init(m_ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_NOCACHE
,
599 error
= gfs2_meta_inode_buffer(m_ip
, &m_bh
);
603 spin_lock(&sdp
->sd_statfs_spin
);
604 gfs2_statfs_change_in(m_sc
, m_bh
->b_data
+
605 sizeof(struct gfs2_dinode
));
606 if (!l_sc
->sc_total
&& !l_sc
->sc_free
&& !l_sc
->sc_dinodes
) {
607 spin_unlock(&sdp
->sd_statfs_spin
);
610 spin_unlock(&sdp
->sd_statfs_spin
);
612 error
= gfs2_meta_inode_buffer(l_ip
, &l_bh
);
616 error
= gfs2_trans_begin(sdp
, 2 * RES_DINODE
, 0);
620 update_statfs(sdp
, m_bh
, l_bh
);
621 sdp
->sd_statfs_force_sync
= 0;
630 gfs2_glock_dq_uninit(&gh
);
637 struct list_head list
;
638 struct gfs2_holder gh
;
642 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
644 * @sdp: the file system
645 * @state: the state to put the transaction lock into
646 * @t_gh: the hold on the transaction lock
651 static int gfs2_lock_fs_check_clean(struct gfs2_sbd
*sdp
,
652 struct gfs2_holder
*freeze_gh
)
654 struct gfs2_inode
*ip
;
655 struct gfs2_jdesc
*jd
;
658 struct gfs2_log_header_host lh
;
661 list_for_each_entry(jd
, &sdp
->sd_jindex_list
, jd_list
) {
662 lfcc
= kmalloc(sizeof(struct lfcc
), GFP_KERNEL
);
667 ip
= GFS2_I(jd
->jd_inode
);
668 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &lfcc
->gh
);
673 list_add(&lfcc
->list
, &list
);
676 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_EXCLUSIVE
,
677 GL_NOCACHE
, freeze_gh
);
679 list_for_each_entry(jd
, &sdp
->sd_jindex_list
, jd_list
) {
680 error
= gfs2_jdesc_check(jd
);
683 error
= gfs2_find_jhead(jd
, &lh
);
686 if (!(lh
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
693 gfs2_glock_dq_uninit(freeze_gh
);
696 while (!list_empty(&list
)) {
697 lfcc
= list_entry(list
.next
, struct lfcc
, list
);
698 list_del(&lfcc
->list
);
699 gfs2_glock_dq_uninit(&lfcc
->gh
);
705 void gfs2_dinode_out(const struct gfs2_inode
*ip
, void *buf
)
707 struct gfs2_dinode
*str
= buf
;
709 str
->di_header
.mh_magic
= cpu_to_be32(GFS2_MAGIC
);
710 str
->di_header
.mh_type
= cpu_to_be32(GFS2_METATYPE_DI
);
711 str
->di_header
.mh_format
= cpu_to_be32(GFS2_FORMAT_DI
);
712 str
->di_num
.no_addr
= cpu_to_be64(ip
->i_no_addr
);
713 str
->di_num
.no_formal_ino
= cpu_to_be64(ip
->i_no_formal_ino
);
714 str
->di_mode
= cpu_to_be32(ip
->i_inode
.i_mode
);
715 str
->di_uid
= cpu_to_be32(i_uid_read(&ip
->i_inode
));
716 str
->di_gid
= cpu_to_be32(i_gid_read(&ip
->i_inode
));
717 str
->di_nlink
= cpu_to_be32(ip
->i_inode
.i_nlink
);
718 str
->di_size
= cpu_to_be64(i_size_read(&ip
->i_inode
));
719 str
->di_blocks
= cpu_to_be64(gfs2_get_inode_blocks(&ip
->i_inode
));
720 str
->di_atime
= cpu_to_be64(ip
->i_inode
.i_atime
.tv_sec
);
721 str
->di_mtime
= cpu_to_be64(ip
->i_inode
.i_mtime
.tv_sec
);
722 str
->di_ctime
= cpu_to_be64(ip
->i_inode
.i_ctime
.tv_sec
);
724 str
->di_goal_meta
= cpu_to_be64(ip
->i_goal
);
725 str
->di_goal_data
= cpu_to_be64(ip
->i_goal
);
726 str
->di_generation
= cpu_to_be64(ip
->i_generation
);
728 str
->di_flags
= cpu_to_be32(ip
->i_diskflags
);
729 str
->di_height
= cpu_to_be16(ip
->i_height
);
730 str
->di_payload_format
= cpu_to_be32(S_ISDIR(ip
->i_inode
.i_mode
) &&
731 !(ip
->i_diskflags
& GFS2_DIF_EXHASH
) ?
733 str
->di_depth
= cpu_to_be16(ip
->i_depth
);
734 str
->di_entries
= cpu_to_be32(ip
->i_entries
);
736 str
->di_eattr
= cpu_to_be64(ip
->i_eattr
);
737 str
->di_atime_nsec
= cpu_to_be32(ip
->i_inode
.i_atime
.tv_nsec
);
738 str
->di_mtime_nsec
= cpu_to_be32(ip
->i_inode
.i_mtime
.tv_nsec
);
739 str
->di_ctime_nsec
= cpu_to_be32(ip
->i_inode
.i_ctime
.tv_nsec
);
743 * gfs2_write_inode - Make sure the inode is stable on the disk
745 * @wbc: The writeback control structure
750 static int gfs2_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
752 struct gfs2_inode
*ip
= GFS2_I(inode
);
753 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
754 struct address_space
*metamapping
= gfs2_glock2aspace(ip
->i_gl
);
755 struct backing_dev_info
*bdi
= inode_to_bdi(metamapping
->host
);
758 if (wbc
->sync_mode
== WB_SYNC_ALL
)
759 gfs2_log_flush(GFS2_SB(inode
), ip
->i_gl
, NORMAL_FLUSH
);
760 if (bdi
->wb
.dirty_exceeded
)
761 gfs2_ail1_flush(sdp
, wbc
);
763 filemap_fdatawrite(metamapping
);
764 if (wbc
->sync_mode
== WB_SYNC_ALL
)
765 ret
= filemap_fdatawait(metamapping
);
767 mark_inode_dirty_sync(inode
);
772 * gfs2_dirty_inode - check for atime updates
773 * @inode: The inode in question
774 * @flags: The type of dirty
776 * Unfortunately it can be called under any combination of inode
777 * glock and transaction lock, so we have to check carefully.
779 * At the moment this deals only with atime - it should be possible
780 * to expand that role in future, once a review of the locking has
784 static void gfs2_dirty_inode(struct inode
*inode
, int flags
)
786 struct gfs2_inode
*ip
= GFS2_I(inode
);
787 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
788 struct buffer_head
*bh
;
789 struct gfs2_holder gh
;
791 int need_endtrans
= 0;
794 if (!(flags
& (I_DIRTY_DATASYNC
|I_DIRTY_SYNC
)))
796 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
798 if (!gfs2_glock_is_locked_by_me(ip
->i_gl
)) {
799 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
801 fs_err(sdp
, "dirty_inode: glock %d\n", ret
);
805 } else if (WARN_ON_ONCE(ip
->i_gl
->gl_state
!= LM_ST_EXCLUSIVE
))
808 if (current
->journal_info
== NULL
) {
809 ret
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
811 fs_err(sdp
, "dirty_inode: gfs2_trans_begin %d\n", ret
);
817 ret
= gfs2_meta_inode_buffer(ip
, &bh
);
819 gfs2_trans_add_meta(ip
->i_gl
, bh
);
820 gfs2_dinode_out(ip
, bh
->b_data
);
828 gfs2_glock_dq_uninit(&gh
);
832 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
833 * @sdp: the filesystem
838 static int gfs2_make_fs_ro(struct gfs2_sbd
*sdp
)
840 struct gfs2_holder freeze_gh
;
843 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_SHARED
, GL_NOCACHE
,
845 if (error
&& !test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))
848 kthread_stop(sdp
->sd_quotad_process
);
849 kthread_stop(sdp
->sd_logd_process
);
851 flush_workqueue(gfs2_delete_workqueue
);
852 gfs2_quota_sync(sdp
->sd_vfs
, 0);
853 gfs2_statfs_sync(sdp
->sd_vfs
, 0);
855 gfs2_log_flush(sdp
, NULL
, SHUTDOWN_FLUSH
);
856 wait_event(sdp
->sd_reserving_log_wait
, atomic_read(&sdp
->sd_reserving_log
) == 0);
857 gfs2_assert_warn(sdp
, atomic_read(&sdp
->sd_log_blks_free
) == sdp
->sd_jdesc
->jd_blocks
);
859 if (gfs2_holder_initialized(&freeze_gh
))
860 gfs2_glock_dq_uninit(&freeze_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_rdonly(sb
)) {
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 gfs2_delete_debugfs_file(sdp
);
928 /* Unmount the locking protocol */
929 gfs2_lm_unmount(sdp
);
931 /* At this point, we're through participating in the lockspace */
932 gfs2_sys_fs_del(sdp
);
936 * gfs2_sync_fs - sync the filesystem
937 * @sb: the superblock
939 * Flushes the log to disk.
942 static int gfs2_sync_fs(struct super_block
*sb
, int wait
)
944 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
946 gfs2_quota_sync(sb
, -1);
948 gfs2_log_flush(sdp
, NULL
, NORMAL_FLUSH
);
949 return sdp
->sd_log_error
;
952 void gfs2_freeze_func(struct work_struct
*work
)
955 struct gfs2_holder freeze_gh
;
956 struct gfs2_sbd
*sdp
= container_of(work
, struct gfs2_sbd
, sd_freeze_work
);
957 struct super_block
*sb
= sdp
->sd_vfs
;
959 atomic_inc(&sb
->s_active
);
960 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_SHARED
, 0,
963 printk(KERN_INFO
"GFS2: couln't get freeze lock : %d\n", error
);
964 gfs2_assert_withdraw(sdp
, 0);
967 atomic_set(&sdp
->sd_freeze_state
, SFS_UNFROZEN
);
968 error
= thaw_super(sb
);
970 printk(KERN_INFO
"GFS2: couldn't thaw filesystem: %d\n",
972 gfs2_assert_withdraw(sdp
, 0);
974 if (!test_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
))
975 freeze_gh
.gh_flags
|= GL_NOCACHE
;
976 gfs2_glock_dq_uninit(&freeze_gh
);
978 deactivate_super(sb
);
983 * gfs2_freeze - prevent further writes to the filesystem
984 * @sb: the VFS structure for the filesystem
988 static int gfs2_freeze(struct super_block
*sb
)
990 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
993 mutex_lock(&sdp
->sd_freeze_mutex
);
994 if (atomic_read(&sdp
->sd_freeze_state
) != SFS_UNFROZEN
)
997 if (test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) {
1003 error
= gfs2_lock_fs_check_clean(sdp
, &sdp
->sd_freeze_gh
);
1009 fs_err(sdp
, "waiting for recovery before freeze\n");
1013 fs_err(sdp
, "error freezing FS: %d\n", error
);
1017 fs_err(sdp
, "retrying...\n");
1022 mutex_unlock(&sdp
->sd_freeze_mutex
);
1027 * gfs2_unfreeze - reallow writes to the filesystem
1028 * @sb: the VFS structure for the filesystem
1032 static int gfs2_unfreeze(struct super_block
*sb
)
1034 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1036 mutex_lock(&sdp
->sd_freeze_mutex
);
1037 if (atomic_read(&sdp
->sd_freeze_state
) != SFS_FROZEN
||
1038 !gfs2_holder_initialized(&sdp
->sd_freeze_gh
)) {
1039 mutex_unlock(&sdp
->sd_freeze_mutex
);
1043 gfs2_glock_dq_uninit(&sdp
->sd_freeze_gh
);
1044 mutex_unlock(&sdp
->sd_freeze_mutex
);
1049 * statfs_fill - fill in the sg for a given RG
1051 * @sc: the sc structure
1053 * Returns: 0 on success, -ESTALE if the LVB is invalid
1056 static int statfs_slow_fill(struct gfs2_rgrpd
*rgd
,
1057 struct gfs2_statfs_change_host
*sc
)
1059 gfs2_rgrp_verify(rgd
);
1060 sc
->sc_total
+= rgd
->rd_data
;
1061 sc
->sc_free
+= rgd
->rd_free
;
1062 sc
->sc_dinodes
+= rgd
->rd_dinodes
;
1067 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
1068 * @sdp: the filesystem
1069 * @sc: the sc info that will be returned
1071 * Any error (other than a signal) will cause this routine to fall back
1072 * to the synchronous version.
1074 * FIXME: This really shouldn't busy wait like this.
1079 static int gfs2_statfs_slow(struct gfs2_sbd
*sdp
, struct gfs2_statfs_change_host
*sc
)
1081 struct gfs2_rgrpd
*rgd_next
;
1082 struct gfs2_holder
*gha
, *gh
;
1083 unsigned int slots
= 64;
1088 memset(sc
, 0, sizeof(struct gfs2_statfs_change_host
));
1089 gha
= kmalloc(slots
* sizeof(struct gfs2_holder
), GFP_KERNEL
);
1092 for (x
= 0; x
< slots
; x
++)
1093 gfs2_holder_mark_uninitialized(gha
+ x
);
1095 rgd_next
= gfs2_rgrpd_get_first(sdp
);
1100 for (x
= 0; x
< slots
; x
++) {
1103 if (gfs2_holder_initialized(gh
) && gfs2_glock_poll(gh
)) {
1104 err
= gfs2_glock_wait(gh
);
1106 gfs2_holder_uninit(gh
);
1110 struct gfs2_rgrpd
*rgd
=
1111 gfs2_glock2rgrp(gh
->gh_gl
);
1113 error
= statfs_slow_fill(rgd
, sc
);
1115 gfs2_glock_dq_uninit(gh
);
1119 if (gfs2_holder_initialized(gh
))
1121 else if (rgd_next
&& !error
) {
1122 error
= gfs2_glock_nq_init(rgd_next
->rd_gl
,
1126 rgd_next
= gfs2_rgrpd_get_next(rgd_next
);
1130 if (signal_pending(current
))
1131 error
= -ERESTARTSYS
;
1145 * gfs2_statfs_i - Do a statfs
1146 * @sdp: the filesystem
1147 * @sg: the sg structure
1152 static int gfs2_statfs_i(struct gfs2_sbd
*sdp
, struct gfs2_statfs_change_host
*sc
)
1154 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
1155 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
1157 spin_lock(&sdp
->sd_statfs_spin
);
1160 sc
->sc_total
+= l_sc
->sc_total
;
1161 sc
->sc_free
+= l_sc
->sc_free
;
1162 sc
->sc_dinodes
+= l_sc
->sc_dinodes
;
1164 spin_unlock(&sdp
->sd_statfs_spin
);
1166 if (sc
->sc_free
< 0)
1168 if (sc
->sc_free
> sc
->sc_total
)
1169 sc
->sc_free
= sc
->sc_total
;
1170 if (sc
->sc_dinodes
< 0)
1177 * gfs2_statfs - Gather and return stats about the filesystem
1178 * @sb: The superblock
1179 * @statfsbuf: The buffer
1181 * Returns: 0 on success or error code
1184 static int gfs2_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1186 struct super_block
*sb
= dentry
->d_sb
;
1187 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1188 struct gfs2_statfs_change_host sc
;
1191 error
= gfs2_rindex_update(sdp
);
1195 if (gfs2_tune_get(sdp
, gt_statfs_slow
))
1196 error
= gfs2_statfs_slow(sdp
, &sc
);
1198 error
= gfs2_statfs_i(sdp
, &sc
);
1203 buf
->f_type
= GFS2_MAGIC
;
1204 buf
->f_bsize
= sdp
->sd_sb
.sb_bsize
;
1205 buf
->f_blocks
= sc
.sc_total
;
1206 buf
->f_bfree
= sc
.sc_free
;
1207 buf
->f_bavail
= sc
.sc_free
;
1208 buf
->f_files
= sc
.sc_dinodes
+ sc
.sc_free
;
1209 buf
->f_ffree
= sc
.sc_free
;
1210 buf
->f_namelen
= GFS2_FNAMESIZE
;
1216 * gfs2_remount_fs - called when the FS is remounted
1217 * @sb: the filesystem
1218 * @flags: the remount flags
1219 * @data: extra data passed in (not used right now)
1224 static int gfs2_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
1226 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1227 struct gfs2_args args
= sdp
->sd_args
; /* Default to current settings */
1228 struct gfs2_tune
*gt
= &sdp
->sd_tune
;
1231 sync_filesystem(sb
);
1233 spin_lock(>
->gt_spin
);
1234 args
.ar_commit
= gt
->gt_logd_secs
;
1235 args
.ar_quota_quantum
= gt
->gt_quota_quantum
;
1236 if (gt
->gt_statfs_slow
)
1237 args
.ar_statfs_quantum
= 0;
1239 args
.ar_statfs_quantum
= gt
->gt_statfs_quantum
;
1240 spin_unlock(>
->gt_spin
);
1241 error
= gfs2_mount_args(&args
, data
);
1245 /* Not allowed to change locking details */
1246 if (strcmp(args
.ar_lockproto
, sdp
->sd_args
.ar_lockproto
) ||
1247 strcmp(args
.ar_locktable
, sdp
->sd_args
.ar_locktable
) ||
1248 strcmp(args
.ar_hostdata
, sdp
->sd_args
.ar_hostdata
))
1251 /* Some flags must not be changed */
1252 if (args_neq(&args
, &sdp
->sd_args
, spectator
) ||
1253 args_neq(&args
, &sdp
->sd_args
, localflocks
) ||
1254 args_neq(&args
, &sdp
->sd_args
, meta
))
1257 if (sdp
->sd_args
.ar_spectator
)
1258 *flags
|= MS_RDONLY
;
1260 if ((sb
->s_flags
^ *flags
) & MS_RDONLY
) {
1261 if (*flags
& MS_RDONLY
)
1262 error
= gfs2_make_fs_ro(sdp
);
1264 error
= gfs2_make_fs_rw(sdp
);
1269 sdp
->sd_args
= args
;
1270 if (sdp
->sd_args
.ar_posix_acl
)
1271 sb
->s_flags
|= MS_POSIXACL
;
1273 sb
->s_flags
&= ~MS_POSIXACL
;
1274 if (sdp
->sd_args
.ar_nobarrier
)
1275 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1277 clear_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1278 spin_lock(>
->gt_spin
);
1279 gt
->gt_logd_secs
= args
.ar_commit
;
1280 gt
->gt_quota_quantum
= args
.ar_quota_quantum
;
1281 if (args
.ar_statfs_quantum
) {
1282 gt
->gt_statfs_slow
= 0;
1283 gt
->gt_statfs_quantum
= args
.ar_statfs_quantum
;
1286 gt
->gt_statfs_slow
= 1;
1287 gt
->gt_statfs_quantum
= 30;
1289 spin_unlock(>
->gt_spin
);
1291 gfs2_online_uevent(sdp
);
1296 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1297 * @inode: The inode to drop
1299 * If we've received a callback on an iopen lock then it's because a
1300 * remote node tried to deallocate the inode but failed due to this node
1301 * still having the inode open. Here we mark the link count zero
1302 * since we know that it must have reached zero if the GLF_DEMOTE flag
1303 * is set on the iopen glock. If we didn't do a disk read since the
1304 * remote node removed the final link then we might otherwise miss
1305 * this event. This check ensures that this node will deallocate the
1306 * inode's blocks, or alternatively pass the baton on to another
1307 * node for later deallocation.
1310 static int gfs2_drop_inode(struct inode
*inode
)
1312 struct gfs2_inode
*ip
= GFS2_I(inode
);
1314 if (!test_bit(GIF_FREE_VFS_INODE
, &ip
->i_flags
) &&
1316 gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1317 struct gfs2_glock
*gl
= ip
->i_iopen_gh
.gh_gl
;
1318 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1323 * When under memory pressure when an inode's link count has dropped to
1324 * zero, defer deleting the inode to the delete workqueue. This avoids
1325 * calling into DLM under memory pressure, which can deadlock.
1327 if (!inode
->i_nlink
&&
1328 unlikely(current
->flags
& PF_MEMALLOC
) &&
1329 gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1330 struct gfs2_glock
*gl
= ip
->i_iopen_gh
.gh_gl
;
1332 gfs2_glock_hold(gl
);
1333 if (queue_work(gfs2_delete_workqueue
, &gl
->gl_delete
) == 0)
1334 gfs2_glock_queue_put(gl
);
1338 return generic_drop_inode(inode
);
1341 static int is_ancestor(const struct dentry
*d1
, const struct dentry
*d2
)
1347 } while (!IS_ROOT(d1
));
1352 * gfs2_show_options - Show mount options for /proc/mounts
1353 * @s: seq_file structure
1354 * @root: root of this (sub)tree
1356 * Returns: 0 on success or error code
1359 static int gfs2_show_options(struct seq_file
*s
, struct dentry
*root
)
1361 struct gfs2_sbd
*sdp
= root
->d_sb
->s_fs_info
;
1362 struct gfs2_args
*args
= &sdp
->sd_args
;
1365 if (is_ancestor(root
, sdp
->sd_master_dir
))
1366 seq_puts(s
, ",meta");
1367 if (args
->ar_lockproto
[0])
1368 seq_show_option(s
, "lockproto", args
->ar_lockproto
);
1369 if (args
->ar_locktable
[0])
1370 seq_show_option(s
, "locktable", args
->ar_locktable
);
1371 if (args
->ar_hostdata
[0])
1372 seq_show_option(s
, "hostdata", args
->ar_hostdata
);
1373 if (args
->ar_spectator
)
1374 seq_puts(s
, ",spectator");
1375 if (args
->ar_localflocks
)
1376 seq_puts(s
, ",localflocks");
1378 seq_puts(s
, ",debug");
1379 if (args
->ar_posix_acl
)
1380 seq_puts(s
, ",acl");
1381 if (args
->ar_quota
!= GFS2_QUOTA_DEFAULT
) {
1383 switch (args
->ar_quota
) {
1384 case GFS2_QUOTA_OFF
:
1387 case GFS2_QUOTA_ACCOUNT
:
1397 seq_printf(s
, ",quota=%s", state
);
1399 if (args
->ar_suiddir
)
1400 seq_puts(s
, ",suiddir");
1401 if (args
->ar_data
!= GFS2_DATA_DEFAULT
) {
1403 switch (args
->ar_data
) {
1404 case GFS2_DATA_WRITEBACK
:
1405 state
= "writeback";
1407 case GFS2_DATA_ORDERED
:
1414 seq_printf(s
, ",data=%s", state
);
1416 if (args
->ar_discard
)
1417 seq_puts(s
, ",discard");
1418 val
= sdp
->sd_tune
.gt_logd_secs
;
1420 seq_printf(s
, ",commit=%d", val
);
1421 val
= sdp
->sd_tune
.gt_statfs_quantum
;
1423 seq_printf(s
, ",statfs_quantum=%d", val
);
1424 else if (sdp
->sd_tune
.gt_statfs_slow
)
1425 seq_puts(s
, ",statfs_quantum=0");
1426 val
= sdp
->sd_tune
.gt_quota_quantum
;
1428 seq_printf(s
, ",quota_quantum=%d", val
);
1429 if (args
->ar_statfs_percent
)
1430 seq_printf(s
, ",statfs_percent=%d", args
->ar_statfs_percent
);
1431 if (args
->ar_errors
!= GFS2_ERRORS_DEFAULT
) {
1434 switch (args
->ar_errors
) {
1435 case GFS2_ERRORS_WITHDRAW
:
1438 case GFS2_ERRORS_PANIC
:
1445 seq_printf(s
, ",errors=%s", state
);
1447 if (test_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
))
1448 seq_puts(s
, ",nobarrier");
1449 if (test_bit(SDF_DEMOTE
, &sdp
->sd_flags
))
1450 seq_puts(s
, ",demote_interface_used");
1451 if (args
->ar_rgrplvb
)
1452 seq_puts(s
, ",rgrplvb");
1453 if (args
->ar_loccookie
)
1454 seq_puts(s
, ",loccookie");
1458 static void gfs2_final_release_pages(struct gfs2_inode
*ip
)
1460 struct inode
*inode
= &ip
->i_inode
;
1461 struct gfs2_glock
*gl
= ip
->i_gl
;
1463 truncate_inode_pages(gfs2_glock2aspace(ip
->i_gl
), 0);
1464 truncate_inode_pages(&inode
->i_data
, 0);
1466 if (atomic_read(&gl
->gl_revokes
) == 0) {
1467 clear_bit(GLF_LFLUSH
, &gl
->gl_flags
);
1468 clear_bit(GLF_DIRTY
, &gl
->gl_flags
);
1472 static int gfs2_dinode_dealloc(struct gfs2_inode
*ip
)
1474 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1475 struct gfs2_rgrpd
*rgd
;
1476 struct gfs2_holder gh
;
1479 if (gfs2_get_inode_blocks(&ip
->i_inode
) != 1) {
1480 gfs2_consist_inode(ip
);
1484 error
= gfs2_rindex_update(sdp
);
1488 error
= gfs2_quota_hold(ip
, NO_UID_QUOTA_CHANGE
, NO_GID_QUOTA_CHANGE
);
1492 rgd
= gfs2_blk2rgrpd(sdp
, ip
->i_no_addr
, 1);
1494 gfs2_consist_inode(ip
);
1499 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
1503 error
= gfs2_trans_begin(sdp
, RES_RG_BIT
+ RES_STATFS
+ RES_QUOTA
,
1504 sdp
->sd_jdesc
->jd_blocks
);
1506 goto out_rg_gunlock
;
1508 gfs2_free_di(rgd
, ip
);
1510 gfs2_final_release_pages(ip
);
1512 gfs2_trans_end(sdp
);
1515 gfs2_glock_dq_uninit(&gh
);
1517 gfs2_quota_unhold(ip
);
1522 * gfs2_glock_put_eventually
1523 * @gl: The glock to put
1525 * When under memory pressure, trigger a deferred glock put to make sure we
1526 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1529 static void gfs2_glock_put_eventually(struct gfs2_glock
*gl
)
1531 if (current
->flags
& PF_MEMALLOC
)
1532 gfs2_glock_queue_put(gl
);
1538 * gfs2_evict_inode - Remove an inode from cache
1539 * @inode: The inode to evict
1541 * There are three cases to consider:
1542 * 1. i_nlink == 0, we are final opener (and must deallocate)
1543 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1546 * If the fs is read only, then we have to treat all cases as per #3
1547 * since we are unable to do any deallocation. The inode will be
1548 * deallocated by the next read/write node to attempt an allocation
1549 * in the same resource group
1551 * We have to (at the moment) hold the inodes main lock to cover
1552 * the gap between unlocking the shared lock on the iopen lock and
1553 * taking the exclusive lock. I'd rather do a shared -> exclusive
1554 * conversion on the iopen lock, but we can change that later. This
1555 * is safe, just less efficient.
1558 static void gfs2_evict_inode(struct inode
*inode
)
1560 struct super_block
*sb
= inode
->i_sb
;
1561 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1562 struct gfs2_inode
*ip
= GFS2_I(inode
);
1563 struct gfs2_holder gh
;
1564 struct address_space
*metamapping
;
1567 if (test_bit(GIF_FREE_VFS_INODE
, &ip
->i_flags
)) {
1572 if (inode
->i_nlink
|| sb_rdonly(sb
))
1575 if (test_bit(GIF_ALLOC_FAILED
, &ip
->i_flags
)) {
1576 BUG_ON(!gfs2_glock_is_locked_by_me(ip
->i_gl
));
1577 gfs2_holder_mark_uninitialized(&gh
);
1581 /* Deletes should never happen under memory pressure anymore. */
1582 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
1585 /* Must not read inode block until block type has been verified */
1586 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_SKIP
, &gh
);
1587 if (unlikely(error
)) {
1588 glock_clear_object(ip
->i_iopen_gh
.gh_gl
, ip
);
1589 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1590 gfs2_glock_dq_uninit(&ip
->i_iopen_gh
);
1594 error
= gfs2_check_blk_type(sdp
, ip
->i_no_addr
, GFS2_BLKST_UNLINKED
);
1598 if (test_bit(GIF_INVALID
, &ip
->i_flags
)) {
1599 error
= gfs2_inode_refresh(ip
);
1605 * The inode may have been recreated in the meantime.
1611 if (gfs2_holder_initialized(&ip
->i_iopen_gh
) &&
1612 test_bit(HIF_HOLDER
, &ip
->i_iopen_gh
.gh_iflags
)) {
1613 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1614 gfs2_glock_dq_wait(&ip
->i_iopen_gh
);
1615 gfs2_holder_reinit(LM_ST_EXCLUSIVE
, LM_FLAG_TRY_1CB
| GL_NOCACHE
,
1617 error
= gfs2_glock_nq(&ip
->i_iopen_gh
);
1622 /* Case 1 starts here */
1624 if (S_ISDIR(inode
->i_mode
) &&
1625 (ip
->i_diskflags
& GFS2_DIF_EXHASH
)) {
1626 error
= gfs2_dir_exhash_dealloc(ip
);
1632 error
= gfs2_ea_dealloc(ip
);
1637 if (!gfs2_is_stuffed(ip
)) {
1638 error
= gfs2_file_dealloc(ip
);
1643 /* We're about to clear the bitmap for the dinode, but as soon as we
1644 do, gfs2_create_inode can create another inode at the same block
1645 location and try to set gl_object again. We clear gl_object here so
1646 that subsequent inode creates don't see an old gl_object. */
1647 glock_clear_object(ip
->i_gl
, ip
);
1648 error
= gfs2_dinode_dealloc(ip
);
1652 gfs2_log_flush(sdp
, ip
->i_gl
, NORMAL_FLUSH
);
1653 metamapping
= gfs2_glock2aspace(ip
->i_gl
);
1654 if (test_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
)) {
1655 filemap_fdatawrite(metamapping
);
1656 filemap_fdatawait(metamapping
);
1658 write_inode_now(inode
, 1);
1659 gfs2_ail_flush(ip
->i_gl
, 0);
1661 /* Case 2 starts here */
1662 error
= gfs2_trans_begin(sdp
, 0, sdp
->sd_jdesc
->jd_blocks
);
1665 /* Needs to be done before glock release & also in a transaction */
1666 truncate_inode_pages(&inode
->i_data
, 0);
1667 truncate_inode_pages(metamapping
, 0);
1668 gfs2_trans_end(sdp
);
1671 /* Error path for case 1 */
1672 if (gfs2_rs_active(&ip
->i_res
))
1673 gfs2_rs_deltree(&ip
->i_res
);
1675 if (gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1676 glock_clear_object(ip
->i_iopen_gh
.gh_gl
, ip
);
1677 if (test_bit(HIF_HOLDER
, &ip
->i_iopen_gh
.gh_iflags
)) {
1678 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1679 gfs2_glock_dq(&ip
->i_iopen_gh
);
1681 gfs2_holder_uninit(&ip
->i_iopen_gh
);
1683 if (gfs2_holder_initialized(&gh
)) {
1684 glock_clear_object(ip
->i_gl
, ip
);
1685 gfs2_glock_dq_uninit(&gh
);
1687 if (error
&& error
!= GLR_TRYFAILED
&& error
!= -EROFS
)
1688 fs_warn(sdp
, "gfs2_evict_inode: %d\n", error
);
1690 /* Case 3 starts here */
1691 truncate_inode_pages_final(&inode
->i_data
);
1692 gfs2_rsqa_delete(ip
, NULL
);
1693 gfs2_ordered_del_inode(ip
);
1695 gfs2_dir_hash_inval(ip
);
1696 glock_clear_object(ip
->i_gl
, ip
);
1697 wait_on_bit_io(&ip
->i_flags
, GIF_GLOP_PENDING
, TASK_UNINTERRUPTIBLE
);
1698 gfs2_glock_add_to_lru(ip
->i_gl
);
1699 gfs2_glock_put_eventually(ip
->i_gl
);
1701 if (gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1702 struct gfs2_glock
*gl
= ip
->i_iopen_gh
.gh_gl
;
1704 glock_clear_object(gl
, ip
);
1705 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1706 gfs2_glock_hold(gl
);
1707 gfs2_glock_dq_uninit(&ip
->i_iopen_gh
);
1708 gfs2_glock_put_eventually(gl
);
1712 static struct inode
*gfs2_alloc_inode(struct super_block
*sb
)
1714 struct gfs2_inode
*ip
;
1716 ip
= kmem_cache_alloc(gfs2_inode_cachep
, GFP_KERNEL
);
1721 memset(&ip
->i_res
, 0, sizeof(ip
->i_res
));
1722 RB_CLEAR_NODE(&ip
->i_res
.rs_node
);
1725 return &ip
->i_inode
;
1728 static void gfs2_i_callback(struct rcu_head
*head
)
1730 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
1731 kmem_cache_free(gfs2_inode_cachep
, inode
);
1734 static void gfs2_destroy_inode(struct inode
*inode
)
1736 call_rcu(&inode
->i_rcu
, gfs2_i_callback
);
1739 const struct super_operations gfs2_super_ops
= {
1740 .alloc_inode
= gfs2_alloc_inode
,
1741 .destroy_inode
= gfs2_destroy_inode
,
1742 .write_inode
= gfs2_write_inode
,
1743 .dirty_inode
= gfs2_dirty_inode
,
1744 .evict_inode
= gfs2_evict_inode
,
1745 .put_super
= gfs2_put_super
,
1746 .sync_fs
= gfs2_sync_fs
,
1747 .freeze_super
= gfs2_freeze
,
1748 .thaw_super
= gfs2_unfreeze
,
1749 .statfs
= gfs2_statfs
,
1750 .remount_fs
= gfs2_remount_fs
,
1751 .drop_inode
= gfs2_drop_inode
,
1752 .show_options
= gfs2_show_options
,