1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/sched/signal.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/statfs.h>
16 #include <linux/seq_file.h>
17 #include <linux/mount.h>
18 #include <linux/kthread.h>
19 #include <linux/delay.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/crc32.h>
22 #include <linux/time.h>
23 #include <linux/wait.h>
24 #include <linux/writeback.h>
25 #include <linux/backing-dev.h>
26 #include <linux/kernel.h>
48 * gfs2_jindex_free - Clear all the journal index information
49 * @sdp: The GFS2 superblock
53 void gfs2_jindex_free(struct gfs2_sbd
*sdp
)
55 struct list_head list
;
56 struct gfs2_jdesc
*jd
;
58 spin_lock(&sdp
->sd_jindex_spin
);
59 list_add(&list
, &sdp
->sd_jindex_list
);
60 list_del_init(&sdp
->sd_jindex_list
);
62 spin_unlock(&sdp
->sd_jindex_spin
);
64 while (!list_empty(&list
)) {
65 jd
= list_entry(list
.next
, struct gfs2_jdesc
, jd_list
);
66 gfs2_free_journal_extents(jd
);
67 list_del(&jd
->jd_list
);
73 static struct gfs2_jdesc
*jdesc_find_i(struct list_head
*head
, unsigned int jid
)
75 struct gfs2_jdesc
*jd
;
78 list_for_each_entry(jd
, head
, jd_list
) {
79 if (jd
->jd_jid
== jid
) {
91 struct gfs2_jdesc
*gfs2_jdesc_find(struct gfs2_sbd
*sdp
, unsigned int jid
)
93 struct gfs2_jdesc
*jd
;
95 spin_lock(&sdp
->sd_jindex_spin
);
96 jd
= jdesc_find_i(&sdp
->sd_jindex_list
, jid
);
97 spin_unlock(&sdp
->sd_jindex_spin
);
102 int gfs2_jdesc_check(struct gfs2_jdesc
*jd
)
104 struct gfs2_inode
*ip
= GFS2_I(jd
->jd_inode
);
105 struct gfs2_sbd
*sdp
= GFS2_SB(jd
->jd_inode
);
106 u64 size
= i_size_read(jd
->jd_inode
);
108 if (gfs2_check_internal_file_size(jd
->jd_inode
, 8 << 20, BIT(30)))
111 jd
->jd_blocks
= size
>> sdp
->sd_sb
.sb_bsize_shift
;
113 if (gfs2_write_alloc_required(ip
, 0, size
)) {
114 gfs2_consist_inode(ip
);
121 static int init_threads(struct gfs2_sbd
*sdp
)
123 struct task_struct
*p
;
126 p
= kthread_run(gfs2_logd
, sdp
, "gfs2_logd");
129 fs_err(sdp
, "can't start logd thread: %d\n", error
);
132 sdp
->sd_logd_process
= p
;
134 p
= kthread_run(gfs2_quotad
, sdp
, "gfs2_quotad");
137 fs_err(sdp
, "can't start quotad thread: %d\n", error
);
140 sdp
->sd_quotad_process
= p
;
144 kthread_stop(sdp
->sd_logd_process
);
145 sdp
->sd_logd_process
= NULL
;
150 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
151 * @sdp: the filesystem
156 int gfs2_make_fs_rw(struct gfs2_sbd
*sdp
)
158 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
159 struct gfs2_glock
*j_gl
= ip
->i_gl
;
160 struct gfs2_holder freeze_gh
;
161 struct gfs2_log_header_host head
;
164 error
= init_threads(sdp
);
168 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_SHARED
, 0,
173 j_gl
->gl_ops
->go_inval(j_gl
, DIO_METADATA
);
175 error
= gfs2_find_jhead(sdp
->sd_jdesc
, &head
, false);
179 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
185 /* Initialize some head of the log stuff */
186 sdp
->sd_log_sequence
= head
.lh_sequence
+ 1;
187 gfs2_log_pointers_init(sdp
, head
.lh_blkno
);
189 error
= gfs2_quota_init(sdp
);
193 set_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
);
195 gfs2_glock_dq_uninit(&freeze_gh
);
200 freeze_gh
.gh_flags
|= GL_NOCACHE
;
201 gfs2_glock_dq_uninit(&freeze_gh
);
203 if (sdp
->sd_quotad_process
)
204 kthread_stop(sdp
->sd_quotad_process
);
205 sdp
->sd_quotad_process
= NULL
;
206 if (sdp
->sd_logd_process
)
207 kthread_stop(sdp
->sd_logd_process
);
208 sdp
->sd_logd_process
= NULL
;
212 void gfs2_statfs_change_in(struct gfs2_statfs_change_host
*sc
, const void *buf
)
214 const struct gfs2_statfs_change
*str
= buf
;
216 sc
->sc_total
= be64_to_cpu(str
->sc_total
);
217 sc
->sc_free
= be64_to_cpu(str
->sc_free
);
218 sc
->sc_dinodes
= be64_to_cpu(str
->sc_dinodes
);
221 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host
*sc
, void *buf
)
223 struct gfs2_statfs_change
*str
= buf
;
225 str
->sc_total
= cpu_to_be64(sc
->sc_total
);
226 str
->sc_free
= cpu_to_be64(sc
->sc_free
);
227 str
->sc_dinodes
= cpu_to_be64(sc
->sc_dinodes
);
230 int gfs2_statfs_init(struct gfs2_sbd
*sdp
)
232 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
233 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
234 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
235 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
236 struct buffer_head
*m_bh
, *l_bh
;
237 struct gfs2_holder gh
;
240 error
= gfs2_glock_nq_init(m_ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_NOCACHE
,
245 error
= gfs2_meta_inode_buffer(m_ip
, &m_bh
);
249 if (sdp
->sd_args
.ar_spectator
) {
250 spin_lock(&sdp
->sd_statfs_spin
);
251 gfs2_statfs_change_in(m_sc
, m_bh
->b_data
+
252 sizeof(struct gfs2_dinode
));
253 spin_unlock(&sdp
->sd_statfs_spin
);
255 error
= gfs2_meta_inode_buffer(l_ip
, &l_bh
);
259 spin_lock(&sdp
->sd_statfs_spin
);
260 gfs2_statfs_change_in(m_sc
, m_bh
->b_data
+
261 sizeof(struct gfs2_dinode
));
262 gfs2_statfs_change_in(l_sc
, l_bh
->b_data
+
263 sizeof(struct gfs2_dinode
));
264 spin_unlock(&sdp
->sd_statfs_spin
);
272 gfs2_glock_dq_uninit(&gh
);
276 void gfs2_statfs_change(struct gfs2_sbd
*sdp
, s64 total
, s64 free
,
279 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
280 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
281 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
282 struct buffer_head
*l_bh
;
287 error
= gfs2_meta_inode_buffer(l_ip
, &l_bh
);
291 gfs2_trans_add_meta(l_ip
->i_gl
, l_bh
);
293 spin_lock(&sdp
->sd_statfs_spin
);
294 l_sc
->sc_total
+= total
;
295 l_sc
->sc_free
+= free
;
296 l_sc
->sc_dinodes
+= dinodes
;
297 gfs2_statfs_change_out(l_sc
, l_bh
->b_data
+ sizeof(struct gfs2_dinode
));
298 if (sdp
->sd_args
.ar_statfs_percent
) {
299 x
= 100 * l_sc
->sc_free
;
300 y
= m_sc
->sc_free
* sdp
->sd_args
.ar_statfs_percent
;
301 if (x
>= y
|| x
<= -y
)
304 spin_unlock(&sdp
->sd_statfs_spin
);
308 gfs2_wake_up_statfs(sdp
);
311 void update_statfs(struct gfs2_sbd
*sdp
, struct buffer_head
*m_bh
,
312 struct buffer_head
*l_bh
)
314 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
315 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
316 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
317 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
319 gfs2_trans_add_meta(l_ip
->i_gl
, l_bh
);
320 gfs2_trans_add_meta(m_ip
->i_gl
, m_bh
);
322 spin_lock(&sdp
->sd_statfs_spin
);
323 m_sc
->sc_total
+= l_sc
->sc_total
;
324 m_sc
->sc_free
+= l_sc
->sc_free
;
325 m_sc
->sc_dinodes
+= l_sc
->sc_dinodes
;
326 memset(l_sc
, 0, sizeof(struct gfs2_statfs_change
));
327 memset(l_bh
->b_data
+ sizeof(struct gfs2_dinode
),
328 0, sizeof(struct gfs2_statfs_change
));
329 gfs2_statfs_change_out(m_sc
, m_bh
->b_data
+ sizeof(struct gfs2_dinode
));
330 spin_unlock(&sdp
->sd_statfs_spin
);
333 int gfs2_statfs_sync(struct super_block
*sb
, int type
)
335 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
336 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
337 struct gfs2_inode
*l_ip
= GFS2_I(sdp
->sd_sc_inode
);
338 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
339 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
340 struct gfs2_holder gh
;
341 struct buffer_head
*m_bh
, *l_bh
;
345 error
= gfs2_glock_nq_init(m_ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_NOCACHE
,
350 error
= gfs2_meta_inode_buffer(m_ip
, &m_bh
);
354 spin_lock(&sdp
->sd_statfs_spin
);
355 gfs2_statfs_change_in(m_sc
, m_bh
->b_data
+
356 sizeof(struct gfs2_dinode
));
357 if (!l_sc
->sc_total
&& !l_sc
->sc_free
&& !l_sc
->sc_dinodes
) {
358 spin_unlock(&sdp
->sd_statfs_spin
);
361 spin_unlock(&sdp
->sd_statfs_spin
);
363 error
= gfs2_meta_inode_buffer(l_ip
, &l_bh
);
367 error
= gfs2_trans_begin(sdp
, 2 * RES_DINODE
, 0);
371 update_statfs(sdp
, m_bh
, l_bh
);
372 sdp
->sd_statfs_force_sync
= 0;
381 gfs2_glock_dq_uninit(&gh
);
388 struct list_head list
;
389 struct gfs2_holder gh
;
393 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
395 * @sdp: the file system
396 * @state: the state to put the transaction lock into
397 * @t_gh: the hold on the transaction lock
402 static int gfs2_lock_fs_check_clean(struct gfs2_sbd
*sdp
)
404 struct gfs2_inode
*ip
;
405 struct gfs2_jdesc
*jd
;
408 struct gfs2_log_header_host lh
;
411 list_for_each_entry(jd
, &sdp
->sd_jindex_list
, jd_list
) {
412 lfcc
= kmalloc(sizeof(struct lfcc
), GFP_KERNEL
);
417 ip
= GFS2_I(jd
->jd_inode
);
418 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &lfcc
->gh
);
423 list_add(&lfcc
->list
, &list
);
426 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_EXCLUSIVE
,
427 GL_NOCACHE
, &sdp
->sd_freeze_gh
);
431 list_for_each_entry(jd
, &sdp
->sd_jindex_list
, jd_list
) {
432 error
= gfs2_jdesc_check(jd
);
435 error
= gfs2_find_jhead(jd
, &lh
, false);
438 if (!(lh
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
445 gfs2_glock_dq_uninit(&sdp
->sd_freeze_gh
);
448 while (!list_empty(&list
)) {
449 lfcc
= list_entry(list
.next
, struct lfcc
, list
);
450 list_del(&lfcc
->list
);
451 gfs2_glock_dq_uninit(&lfcc
->gh
);
457 void gfs2_dinode_out(const struct gfs2_inode
*ip
, void *buf
)
459 struct gfs2_dinode
*str
= buf
;
461 str
->di_header
.mh_magic
= cpu_to_be32(GFS2_MAGIC
);
462 str
->di_header
.mh_type
= cpu_to_be32(GFS2_METATYPE_DI
);
463 str
->di_header
.mh_format
= cpu_to_be32(GFS2_FORMAT_DI
);
464 str
->di_num
.no_addr
= cpu_to_be64(ip
->i_no_addr
);
465 str
->di_num
.no_formal_ino
= cpu_to_be64(ip
->i_no_formal_ino
);
466 str
->di_mode
= cpu_to_be32(ip
->i_inode
.i_mode
);
467 str
->di_uid
= cpu_to_be32(i_uid_read(&ip
->i_inode
));
468 str
->di_gid
= cpu_to_be32(i_gid_read(&ip
->i_inode
));
469 str
->di_nlink
= cpu_to_be32(ip
->i_inode
.i_nlink
);
470 str
->di_size
= cpu_to_be64(i_size_read(&ip
->i_inode
));
471 str
->di_blocks
= cpu_to_be64(gfs2_get_inode_blocks(&ip
->i_inode
));
472 str
->di_atime
= cpu_to_be64(ip
->i_inode
.i_atime
.tv_sec
);
473 str
->di_mtime
= cpu_to_be64(ip
->i_inode
.i_mtime
.tv_sec
);
474 str
->di_ctime
= cpu_to_be64(ip
->i_inode
.i_ctime
.tv_sec
);
476 str
->di_goal_meta
= cpu_to_be64(ip
->i_goal
);
477 str
->di_goal_data
= cpu_to_be64(ip
->i_goal
);
478 str
->di_generation
= cpu_to_be64(ip
->i_generation
);
480 str
->di_flags
= cpu_to_be32(ip
->i_diskflags
);
481 str
->di_height
= cpu_to_be16(ip
->i_height
);
482 str
->di_payload_format
= cpu_to_be32(S_ISDIR(ip
->i_inode
.i_mode
) &&
483 !(ip
->i_diskflags
& GFS2_DIF_EXHASH
) ?
485 str
->di_depth
= cpu_to_be16(ip
->i_depth
);
486 str
->di_entries
= cpu_to_be32(ip
->i_entries
);
488 str
->di_eattr
= cpu_to_be64(ip
->i_eattr
);
489 str
->di_atime_nsec
= cpu_to_be32(ip
->i_inode
.i_atime
.tv_nsec
);
490 str
->di_mtime_nsec
= cpu_to_be32(ip
->i_inode
.i_mtime
.tv_nsec
);
491 str
->di_ctime_nsec
= cpu_to_be32(ip
->i_inode
.i_ctime
.tv_nsec
);
495 * gfs2_write_inode - Make sure the inode is stable on the disk
497 * @wbc: The writeback control structure
502 static int gfs2_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
504 struct gfs2_inode
*ip
= GFS2_I(inode
);
505 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
506 struct address_space
*metamapping
= gfs2_glock2aspace(ip
->i_gl
);
507 struct backing_dev_info
*bdi
= inode_to_bdi(metamapping
->host
);
509 bool flush_all
= (wbc
->sync_mode
== WB_SYNC_ALL
|| gfs2_is_jdata(ip
));
512 gfs2_log_flush(GFS2_SB(inode
), ip
->i_gl
,
513 GFS2_LOG_HEAD_FLUSH_NORMAL
|
514 GFS2_LFC_WRITE_INODE
);
515 if (bdi
->wb
.dirty_exceeded
)
516 gfs2_ail1_flush(sdp
, wbc
);
518 filemap_fdatawrite(metamapping
);
520 ret
= filemap_fdatawait(metamapping
);
522 mark_inode_dirty_sync(inode
);
524 spin_lock(&inode
->i_lock
);
525 if (!(inode
->i_flags
& I_DIRTY
))
526 gfs2_ordered_del_inode(ip
);
527 spin_unlock(&inode
->i_lock
);
533 * gfs2_dirty_inode - check for atime updates
534 * @inode: The inode in question
535 * @flags: The type of dirty
537 * Unfortunately it can be called under any combination of inode
538 * glock and transaction lock, so we have to check carefully.
540 * At the moment this deals only with atime - it should be possible
541 * to expand that role in future, once a review of the locking has
545 static void gfs2_dirty_inode(struct inode
*inode
, int flags
)
547 struct gfs2_inode
*ip
= GFS2_I(inode
);
548 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
549 struct buffer_head
*bh
;
550 struct gfs2_holder gh
;
552 int need_endtrans
= 0;
555 if (!(flags
& I_DIRTY_INODE
))
557 if (unlikely(gfs2_withdrawn(sdp
)))
559 if (!gfs2_glock_is_locked_by_me(ip
->i_gl
)) {
560 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
562 fs_err(sdp
, "dirty_inode: glock %d\n", ret
);
566 } else if (WARN_ON_ONCE(ip
->i_gl
->gl_state
!= LM_ST_EXCLUSIVE
))
569 if (current
->journal_info
== NULL
) {
570 ret
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
572 fs_err(sdp
, "dirty_inode: gfs2_trans_begin %d\n", ret
);
578 ret
= gfs2_meta_inode_buffer(ip
, &bh
);
580 gfs2_trans_add_meta(ip
->i_gl
, bh
);
581 gfs2_dinode_out(ip
, bh
->b_data
);
589 gfs2_glock_dq_uninit(&gh
);
593 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
594 * @sdp: the filesystem
599 int gfs2_make_fs_ro(struct gfs2_sbd
*sdp
)
601 struct gfs2_holder freeze_gh
;
604 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_SHARED
, GL_NOCACHE
,
606 if (error
&& !gfs2_withdrawn(sdp
))
609 flush_workqueue(gfs2_delete_workqueue
);
610 if (sdp
->sd_quotad_process
)
611 kthread_stop(sdp
->sd_quotad_process
);
612 sdp
->sd_quotad_process
= NULL
;
613 if (sdp
->sd_logd_process
)
614 kthread_stop(sdp
->sd_logd_process
);
615 sdp
->sd_logd_process
= NULL
;
617 gfs2_quota_sync(sdp
->sd_vfs
, 0);
618 gfs2_statfs_sync(sdp
->sd_vfs
, 0);
620 gfs2_log_flush(sdp
, NULL
, GFS2_LOG_HEAD_FLUSH_SHUTDOWN
|
621 GFS2_LFC_MAKE_FS_RO
);
622 wait_event(sdp
->sd_reserving_log_wait
, atomic_read(&sdp
->sd_reserving_log
) == 0);
623 gfs2_assert_warn(sdp
, atomic_read(&sdp
->sd_log_blks_free
) == sdp
->sd_jdesc
->jd_blocks
);
625 if (gfs2_holder_initialized(&freeze_gh
))
626 gfs2_glock_dq_uninit(&freeze_gh
);
628 gfs2_quota_cleanup(sdp
);
634 * gfs2_put_super - Unmount the filesystem
635 * @sb: The VFS superblock
639 static void gfs2_put_super(struct super_block
*sb
)
641 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
643 struct gfs2_jdesc
*jd
;
645 /* No more recovery requests */
646 set_bit(SDF_NORECOVERY
, &sdp
->sd_flags
);
649 /* Wait on outstanding recovery */
651 spin_lock(&sdp
->sd_jindex_spin
);
652 list_for_each_entry(jd
, &sdp
->sd_jindex_list
, jd_list
) {
653 if (!test_bit(JDF_RECOVERY
, &jd
->jd_flags
))
655 spin_unlock(&sdp
->sd_jindex_spin
);
656 wait_on_bit(&jd
->jd_flags
, JDF_RECOVERY
,
657 TASK_UNINTERRUPTIBLE
);
660 spin_unlock(&sdp
->sd_jindex_spin
);
662 if (!sb_rdonly(sb
)) {
663 error
= gfs2_make_fs_ro(sdp
);
667 /* At this point, we're through modifying the disk */
671 iput(sdp
->sd_jindex
);
672 iput(sdp
->sd_statfs_inode
);
673 iput(sdp
->sd_rindex
);
674 iput(sdp
->sd_quota_inode
);
676 gfs2_glock_put(sdp
->sd_rename_gl
);
677 gfs2_glock_put(sdp
->sd_freeze_gl
);
679 if (!sdp
->sd_args
.ar_spectator
) {
680 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
681 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
682 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
683 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
684 iput(sdp
->sd_sc_inode
);
685 iput(sdp
->sd_qc_inode
);
688 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
689 gfs2_clear_rgrpd(sdp
);
690 gfs2_jindex_free(sdp
);
691 /* Take apart glock structures and buffer lists */
692 gfs2_gl_hash_clear(sdp
);
693 gfs2_delete_debugfs_file(sdp
);
694 /* Unmount the locking protocol */
695 gfs2_lm_unmount(sdp
);
697 /* At this point, we're through participating in the lockspace */
698 gfs2_sys_fs_del(sdp
);
702 * gfs2_sync_fs - sync the filesystem
703 * @sb: the superblock
705 * Flushes the log to disk.
708 static int gfs2_sync_fs(struct super_block
*sb
, int wait
)
710 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
712 gfs2_quota_sync(sb
, -1);
714 gfs2_log_flush(sdp
, NULL
, GFS2_LOG_HEAD_FLUSH_NORMAL
|
716 return sdp
->sd_log_error
;
719 void gfs2_freeze_func(struct work_struct
*work
)
722 struct gfs2_holder freeze_gh
;
723 struct gfs2_sbd
*sdp
= container_of(work
, struct gfs2_sbd
, sd_freeze_work
);
724 struct super_block
*sb
= sdp
->sd_vfs
;
726 atomic_inc(&sb
->s_active
);
727 error
= gfs2_glock_nq_init(sdp
->sd_freeze_gl
, LM_ST_SHARED
, 0,
730 fs_info(sdp
, "GFS2: couldn't get freeze lock : %d\n", error
);
731 gfs2_assert_withdraw(sdp
, 0);
733 atomic_set(&sdp
->sd_freeze_state
, SFS_UNFROZEN
);
734 error
= thaw_super(sb
);
736 fs_info(sdp
, "GFS2: couldn't thaw filesystem: %d\n",
738 gfs2_assert_withdraw(sdp
, 0);
740 if (!test_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
))
741 freeze_gh
.gh_flags
|= GL_NOCACHE
;
742 gfs2_glock_dq_uninit(&freeze_gh
);
744 deactivate_super(sb
);
745 clear_bit_unlock(SDF_FS_FROZEN
, &sdp
->sd_flags
);
746 wake_up_bit(&sdp
->sd_flags
, SDF_FS_FROZEN
);
751 * gfs2_freeze - prevent further writes to the filesystem
752 * @sb: the VFS structure for the filesystem
756 static int gfs2_freeze(struct super_block
*sb
)
758 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
761 mutex_lock(&sdp
->sd_freeze_mutex
);
762 if (atomic_read(&sdp
->sd_freeze_state
) != SFS_UNFROZEN
)
766 if (gfs2_withdrawn(sdp
)) {
771 error
= gfs2_lock_fs_check_clean(sdp
);
776 fs_err(sdp
, "waiting for recovery before freeze\n");
777 else if (error
== -EIO
) {
778 fs_err(sdp
, "Fatal IO error: cannot freeze gfs2 due "
779 "to recovery error.\n");
782 fs_err(sdp
, "error freezing FS: %d\n", error
);
784 fs_err(sdp
, "retrying...\n");
787 set_bit(SDF_FS_FROZEN
, &sdp
->sd_flags
);
789 mutex_unlock(&sdp
->sd_freeze_mutex
);
794 * gfs2_unfreeze - reallow writes to the filesystem
795 * @sb: the VFS structure for the filesystem
799 static int gfs2_unfreeze(struct super_block
*sb
)
801 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
803 mutex_lock(&sdp
->sd_freeze_mutex
);
804 if (atomic_read(&sdp
->sd_freeze_state
) != SFS_FROZEN
||
805 !gfs2_holder_initialized(&sdp
->sd_freeze_gh
)) {
806 mutex_unlock(&sdp
->sd_freeze_mutex
);
810 gfs2_glock_dq_uninit(&sdp
->sd_freeze_gh
);
811 mutex_unlock(&sdp
->sd_freeze_mutex
);
812 return wait_on_bit(&sdp
->sd_flags
, SDF_FS_FROZEN
, TASK_INTERRUPTIBLE
);
816 * statfs_fill - fill in the sg for a given RG
818 * @sc: the sc structure
820 * Returns: 0 on success, -ESTALE if the LVB is invalid
823 static int statfs_slow_fill(struct gfs2_rgrpd
*rgd
,
824 struct gfs2_statfs_change_host
*sc
)
826 gfs2_rgrp_verify(rgd
);
827 sc
->sc_total
+= rgd
->rd_data
;
828 sc
->sc_free
+= rgd
->rd_free
;
829 sc
->sc_dinodes
+= rgd
->rd_dinodes
;
834 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
835 * @sdp: the filesystem
836 * @sc: the sc info that will be returned
838 * Any error (other than a signal) will cause this routine to fall back
839 * to the synchronous version.
841 * FIXME: This really shouldn't busy wait like this.
846 static int gfs2_statfs_slow(struct gfs2_sbd
*sdp
, struct gfs2_statfs_change_host
*sc
)
848 struct gfs2_rgrpd
*rgd_next
;
849 struct gfs2_holder
*gha
, *gh
;
850 unsigned int slots
= 64;
855 memset(sc
, 0, sizeof(struct gfs2_statfs_change_host
));
856 gha
= kmalloc_array(slots
, sizeof(struct gfs2_holder
), GFP_KERNEL
);
859 for (x
= 0; x
< slots
; x
++)
860 gfs2_holder_mark_uninitialized(gha
+ x
);
862 rgd_next
= gfs2_rgrpd_get_first(sdp
);
867 for (x
= 0; x
< slots
; x
++) {
870 if (gfs2_holder_initialized(gh
) && gfs2_glock_poll(gh
)) {
871 err
= gfs2_glock_wait(gh
);
873 gfs2_holder_uninit(gh
);
877 struct gfs2_rgrpd
*rgd
=
878 gfs2_glock2rgrp(gh
->gh_gl
);
880 error
= statfs_slow_fill(rgd
, sc
);
882 gfs2_glock_dq_uninit(gh
);
886 if (gfs2_holder_initialized(gh
))
888 else if (rgd_next
&& !error
) {
889 error
= gfs2_glock_nq_init(rgd_next
->rd_gl
,
893 rgd_next
= gfs2_rgrpd_get_next(rgd_next
);
897 if (signal_pending(current
))
898 error
= -ERESTARTSYS
;
912 * gfs2_statfs_i - Do a statfs
913 * @sdp: the filesystem
914 * @sg: the sg structure
919 static int gfs2_statfs_i(struct gfs2_sbd
*sdp
, struct gfs2_statfs_change_host
*sc
)
921 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
922 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
924 spin_lock(&sdp
->sd_statfs_spin
);
927 sc
->sc_total
+= l_sc
->sc_total
;
928 sc
->sc_free
+= l_sc
->sc_free
;
929 sc
->sc_dinodes
+= l_sc
->sc_dinodes
;
931 spin_unlock(&sdp
->sd_statfs_spin
);
935 if (sc
->sc_free
> sc
->sc_total
)
936 sc
->sc_free
= sc
->sc_total
;
937 if (sc
->sc_dinodes
< 0)
944 * gfs2_statfs - Gather and return stats about the filesystem
945 * @sb: The superblock
946 * @statfsbuf: The buffer
948 * Returns: 0 on success or error code
951 static int gfs2_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
953 struct super_block
*sb
= dentry
->d_sb
;
954 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
955 struct gfs2_statfs_change_host sc
;
958 error
= gfs2_rindex_update(sdp
);
962 if (gfs2_tune_get(sdp
, gt_statfs_slow
))
963 error
= gfs2_statfs_slow(sdp
, &sc
);
965 error
= gfs2_statfs_i(sdp
, &sc
);
970 buf
->f_type
= GFS2_MAGIC
;
971 buf
->f_bsize
= sdp
->sd_sb
.sb_bsize
;
972 buf
->f_blocks
= sc
.sc_total
;
973 buf
->f_bfree
= sc
.sc_free
;
974 buf
->f_bavail
= sc
.sc_free
;
975 buf
->f_files
= sc
.sc_dinodes
+ sc
.sc_free
;
976 buf
->f_ffree
= sc
.sc_free
;
977 buf
->f_namelen
= GFS2_FNAMESIZE
;
983 * gfs2_drop_inode - Drop an inode (test for remote unlink)
984 * @inode: The inode to drop
986 * If we've received a callback on an iopen lock then it's because a
987 * remote node tried to deallocate the inode but failed due to this node
988 * still having the inode open. Here we mark the link count zero
989 * since we know that it must have reached zero if the GLF_DEMOTE flag
990 * is set on the iopen glock. If we didn't do a disk read since the
991 * remote node removed the final link then we might otherwise miss
992 * this event. This check ensures that this node will deallocate the
993 * inode's blocks, or alternatively pass the baton on to another
994 * node for later deallocation.
997 static int gfs2_drop_inode(struct inode
*inode
)
999 struct gfs2_inode
*ip
= GFS2_I(inode
);
1001 if (!test_bit(GIF_FREE_VFS_INODE
, &ip
->i_flags
) &&
1003 gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1004 struct gfs2_glock
*gl
= ip
->i_iopen_gh
.gh_gl
;
1005 if (test_bit(GLF_DEMOTE
, &gl
->gl_flags
))
1010 * When under memory pressure when an inode's link count has dropped to
1011 * zero, defer deleting the inode to the delete workqueue. This avoids
1012 * calling into DLM under memory pressure, which can deadlock.
1014 if (!inode
->i_nlink
&&
1015 unlikely(current
->flags
& PF_MEMALLOC
) &&
1016 gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1017 struct gfs2_glock
*gl
= ip
->i_iopen_gh
.gh_gl
;
1019 gfs2_glock_hold(gl
);
1020 if (queue_work(gfs2_delete_workqueue
, &gl
->gl_delete
) == 0)
1021 gfs2_glock_queue_put(gl
);
1025 return generic_drop_inode(inode
);
1028 static int is_ancestor(const struct dentry
*d1
, const struct dentry
*d2
)
1034 } while (!IS_ROOT(d1
));
1039 * gfs2_show_options - Show mount options for /proc/mounts
1040 * @s: seq_file structure
1041 * @root: root of this (sub)tree
1043 * Returns: 0 on success or error code
1046 static int gfs2_show_options(struct seq_file
*s
, struct dentry
*root
)
1048 struct gfs2_sbd
*sdp
= root
->d_sb
->s_fs_info
;
1049 struct gfs2_args
*args
= &sdp
->sd_args
;
1052 if (is_ancestor(root
, sdp
->sd_master_dir
))
1053 seq_puts(s
, ",meta");
1054 if (args
->ar_lockproto
[0])
1055 seq_show_option(s
, "lockproto", args
->ar_lockproto
);
1056 if (args
->ar_locktable
[0])
1057 seq_show_option(s
, "locktable", args
->ar_locktable
);
1058 if (args
->ar_hostdata
[0])
1059 seq_show_option(s
, "hostdata", args
->ar_hostdata
);
1060 if (args
->ar_spectator
)
1061 seq_puts(s
, ",spectator");
1062 if (args
->ar_localflocks
)
1063 seq_puts(s
, ",localflocks");
1065 seq_puts(s
, ",debug");
1066 if (args
->ar_posix_acl
)
1067 seq_puts(s
, ",acl");
1068 if (args
->ar_quota
!= GFS2_QUOTA_DEFAULT
) {
1070 switch (args
->ar_quota
) {
1071 case GFS2_QUOTA_OFF
:
1074 case GFS2_QUOTA_ACCOUNT
:
1084 seq_printf(s
, ",quota=%s", state
);
1086 if (args
->ar_suiddir
)
1087 seq_puts(s
, ",suiddir");
1088 if (args
->ar_data
!= GFS2_DATA_DEFAULT
) {
1090 switch (args
->ar_data
) {
1091 case GFS2_DATA_WRITEBACK
:
1092 state
= "writeback";
1094 case GFS2_DATA_ORDERED
:
1101 seq_printf(s
, ",data=%s", state
);
1103 if (args
->ar_discard
)
1104 seq_puts(s
, ",discard");
1105 val
= sdp
->sd_tune
.gt_logd_secs
;
1107 seq_printf(s
, ",commit=%d", val
);
1108 val
= sdp
->sd_tune
.gt_statfs_quantum
;
1110 seq_printf(s
, ",statfs_quantum=%d", val
);
1111 else if (sdp
->sd_tune
.gt_statfs_slow
)
1112 seq_puts(s
, ",statfs_quantum=0");
1113 val
= sdp
->sd_tune
.gt_quota_quantum
;
1115 seq_printf(s
, ",quota_quantum=%d", val
);
1116 if (args
->ar_statfs_percent
)
1117 seq_printf(s
, ",statfs_percent=%d", args
->ar_statfs_percent
);
1118 if (args
->ar_errors
!= GFS2_ERRORS_DEFAULT
) {
1121 switch (args
->ar_errors
) {
1122 case GFS2_ERRORS_WITHDRAW
:
1125 case GFS2_ERRORS_PANIC
:
1132 seq_printf(s
, ",errors=%s", state
);
1134 if (test_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
))
1135 seq_puts(s
, ",nobarrier");
1136 if (test_bit(SDF_DEMOTE
, &sdp
->sd_flags
))
1137 seq_puts(s
, ",demote_interface_used");
1138 if (args
->ar_rgrplvb
)
1139 seq_puts(s
, ",rgrplvb");
1140 if (args
->ar_loccookie
)
1141 seq_puts(s
, ",loccookie");
1145 static void gfs2_final_release_pages(struct gfs2_inode
*ip
)
1147 struct inode
*inode
= &ip
->i_inode
;
1148 struct gfs2_glock
*gl
= ip
->i_gl
;
1150 truncate_inode_pages(gfs2_glock2aspace(ip
->i_gl
), 0);
1151 truncate_inode_pages(&inode
->i_data
, 0);
1153 if (atomic_read(&gl
->gl_revokes
) == 0) {
1154 clear_bit(GLF_LFLUSH
, &gl
->gl_flags
);
1155 clear_bit(GLF_DIRTY
, &gl
->gl_flags
);
1159 static int gfs2_dinode_dealloc(struct gfs2_inode
*ip
)
1161 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1162 struct gfs2_rgrpd
*rgd
;
1163 struct gfs2_holder gh
;
1166 if (gfs2_get_inode_blocks(&ip
->i_inode
) != 1) {
1167 gfs2_consist_inode(ip
);
1171 error
= gfs2_rindex_update(sdp
);
1175 error
= gfs2_quota_hold(ip
, NO_UID_QUOTA_CHANGE
, NO_GID_QUOTA_CHANGE
);
1179 rgd
= gfs2_blk2rgrpd(sdp
, ip
->i_no_addr
, 1);
1181 gfs2_consist_inode(ip
);
1186 error
= gfs2_glock_nq_init(rgd
->rd_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
1190 error
= gfs2_trans_begin(sdp
, RES_RG_BIT
+ RES_STATFS
+ RES_QUOTA
,
1191 sdp
->sd_jdesc
->jd_blocks
);
1193 goto out_rg_gunlock
;
1195 gfs2_free_di(rgd
, ip
);
1197 gfs2_final_release_pages(ip
);
1199 gfs2_trans_end(sdp
);
1202 gfs2_glock_dq_uninit(&gh
);
1204 gfs2_quota_unhold(ip
);
1209 * gfs2_glock_put_eventually
1210 * @gl: The glock to put
1212 * When under memory pressure, trigger a deferred glock put to make sure we
1213 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1216 static void gfs2_glock_put_eventually(struct gfs2_glock
*gl
)
1218 if (current
->flags
& PF_MEMALLOC
)
1219 gfs2_glock_queue_put(gl
);
1225 * gfs2_evict_inode - Remove an inode from cache
1226 * @inode: The inode to evict
1228 * There are three cases to consider:
1229 * 1. i_nlink == 0, we are final opener (and must deallocate)
1230 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1233 * If the fs is read only, then we have to treat all cases as per #3
1234 * since we are unable to do any deallocation. The inode will be
1235 * deallocated by the next read/write node to attempt an allocation
1236 * in the same resource group
1238 * We have to (at the moment) hold the inodes main lock to cover
1239 * the gap between unlocking the shared lock on the iopen lock and
1240 * taking the exclusive lock. I'd rather do a shared -> exclusive
1241 * conversion on the iopen lock, but we can change that later. This
1242 * is safe, just less efficient.
1245 static void gfs2_evict_inode(struct inode
*inode
)
1247 struct super_block
*sb
= inode
->i_sb
;
1248 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1249 struct gfs2_inode
*ip
= GFS2_I(inode
);
1250 struct gfs2_holder gh
;
1251 struct address_space
*metamapping
;
1254 if (test_bit(GIF_FREE_VFS_INODE
, &ip
->i_flags
)) {
1259 if (inode
->i_nlink
|| sb_rdonly(sb
))
1262 if (test_bit(GIF_ALLOC_FAILED
, &ip
->i_flags
)) {
1263 BUG_ON(!gfs2_glock_is_locked_by_me(ip
->i_gl
));
1264 gfs2_holder_mark_uninitialized(&gh
);
1268 /* Deletes should never happen under memory pressure anymore. */
1269 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
1272 /* Must not read inode block until block type has been verified */
1273 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_SKIP
, &gh
);
1274 if (unlikely(error
)) {
1275 glock_clear_object(ip
->i_iopen_gh
.gh_gl
, ip
);
1276 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1277 gfs2_glock_dq_uninit(&ip
->i_iopen_gh
);
1281 error
= gfs2_check_blk_type(sdp
, ip
->i_no_addr
, GFS2_BLKST_UNLINKED
);
1285 if (test_bit(GIF_INVALID
, &ip
->i_flags
)) {
1286 error
= gfs2_inode_refresh(ip
);
1292 * The inode may have been recreated in the meantime.
1298 if (gfs2_holder_initialized(&ip
->i_iopen_gh
) &&
1299 test_bit(HIF_HOLDER
, &ip
->i_iopen_gh
.gh_iflags
)) {
1300 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1301 gfs2_glock_dq_wait(&ip
->i_iopen_gh
);
1302 gfs2_holder_reinit(LM_ST_EXCLUSIVE
, LM_FLAG_TRY_1CB
| GL_NOCACHE
,
1304 error
= gfs2_glock_nq(&ip
->i_iopen_gh
);
1309 if (S_ISDIR(inode
->i_mode
) &&
1310 (ip
->i_diskflags
& GFS2_DIF_EXHASH
)) {
1311 error
= gfs2_dir_exhash_dealloc(ip
);
1317 error
= gfs2_ea_dealloc(ip
);
1322 if (!gfs2_is_stuffed(ip
)) {
1323 error
= gfs2_file_dealloc(ip
);
1328 /* We're about to clear the bitmap for the dinode, but as soon as we
1329 do, gfs2_create_inode can create another inode at the same block
1330 location and try to set gl_object again. We clear gl_object here so
1331 that subsequent inode creates don't see an old gl_object. */
1332 glock_clear_object(ip
->i_gl
, ip
);
1333 error
= gfs2_dinode_dealloc(ip
);
1337 gfs2_log_flush(sdp
, ip
->i_gl
, GFS2_LOG_HEAD_FLUSH_NORMAL
|
1338 GFS2_LFC_EVICT_INODE
);
1339 metamapping
= gfs2_glock2aspace(ip
->i_gl
);
1340 if (test_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
)) {
1341 filemap_fdatawrite(metamapping
);
1342 filemap_fdatawait(metamapping
);
1344 write_inode_now(inode
, 1);
1345 gfs2_ail_flush(ip
->i_gl
, 0);
1347 error
= gfs2_trans_begin(sdp
, 0, sdp
->sd_jdesc
->jd_blocks
);
1350 /* Needs to be done before glock release & also in a transaction */
1351 truncate_inode_pages(&inode
->i_data
, 0);
1352 truncate_inode_pages(metamapping
, 0);
1353 gfs2_trans_end(sdp
);
1356 if (gfs2_rs_active(&ip
->i_res
))
1357 gfs2_rs_deltree(&ip
->i_res
);
1359 if (gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1360 glock_clear_object(ip
->i_iopen_gh
.gh_gl
, ip
);
1361 if (test_bit(HIF_HOLDER
, &ip
->i_iopen_gh
.gh_iflags
)) {
1362 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1363 gfs2_glock_dq(&ip
->i_iopen_gh
);
1365 gfs2_holder_uninit(&ip
->i_iopen_gh
);
1367 if (gfs2_holder_initialized(&gh
)) {
1368 glock_clear_object(ip
->i_gl
, ip
);
1369 gfs2_glock_dq_uninit(&gh
);
1371 if (error
&& error
!= GLR_TRYFAILED
&& error
!= -EROFS
)
1372 fs_warn(sdp
, "gfs2_evict_inode: %d\n", error
);
1374 truncate_inode_pages_final(&inode
->i_data
);
1375 gfs2_rsqa_delete(ip
, NULL
);
1376 gfs2_ordered_del_inode(ip
);
1378 gfs2_dir_hash_inval(ip
);
1379 glock_clear_object(ip
->i_gl
, ip
);
1380 wait_on_bit_io(&ip
->i_flags
, GIF_GLOP_PENDING
, TASK_UNINTERRUPTIBLE
);
1381 gfs2_glock_add_to_lru(ip
->i_gl
);
1382 gfs2_glock_put_eventually(ip
->i_gl
);
1384 if (gfs2_holder_initialized(&ip
->i_iopen_gh
)) {
1385 struct gfs2_glock
*gl
= ip
->i_iopen_gh
.gh_gl
;
1387 glock_clear_object(gl
, ip
);
1388 ip
->i_iopen_gh
.gh_flags
|= GL_NOCACHE
;
1389 gfs2_glock_hold(gl
);
1390 gfs2_glock_dq_uninit(&ip
->i_iopen_gh
);
1391 gfs2_glock_put_eventually(gl
);
1395 static struct inode
*gfs2_alloc_inode(struct super_block
*sb
)
1397 struct gfs2_inode
*ip
;
1399 ip
= kmem_cache_alloc(gfs2_inode_cachep
, GFP_KERNEL
);
1404 memset(&ip
->i_res
, 0, sizeof(ip
->i_res
));
1405 RB_CLEAR_NODE(&ip
->i_res
.rs_node
);
1407 return &ip
->i_inode
;
1410 static void gfs2_free_inode(struct inode
*inode
)
1412 kmem_cache_free(gfs2_inode_cachep
, GFS2_I(inode
));
1415 const struct super_operations gfs2_super_ops
= {
1416 .alloc_inode
= gfs2_alloc_inode
,
1417 .free_inode
= gfs2_free_inode
,
1418 .write_inode
= gfs2_write_inode
,
1419 .dirty_inode
= gfs2_dirty_inode
,
1420 .evict_inode
= gfs2_evict_inode
,
1421 .put_super
= gfs2_put_super
,
1422 .sync_fs
= gfs2_sync_fs
,
1423 .freeze_super
= gfs2_freeze
,
1424 .thaw_super
= gfs2_unfreeze
,
1425 .statfs
= gfs2_statfs
,
1426 .drop_inode
= gfs2_drop_inode
,
1427 .show_options
= gfs2_show_options
,