2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 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/spinlock.h>
11 #include <linux/completion.h>
12 #include <linux/buffer_head.h>
13 #include <linux/gfs2_ondisk.h>
14 #include <linux/bio.h>
15 #include <linux/posix_acl.h>
31 struct workqueue_struct
*gfs2_freeze_wq
;
33 static void gfs2_ail_error(struct gfs2_glock
*gl
, const struct buffer_head
*bh
)
35 fs_err(gl
->gl_name
.ln_sbd
,
36 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
38 bh
, (unsigned long long)bh
->b_blocknr
, bh
->b_state
,
39 bh
->b_page
->mapping
, bh
->b_page
->flags
);
40 fs_err(gl
->gl_name
.ln_sbd
, "AIL glock %u:%llu mapping %p\n",
41 gl
->gl_name
.ln_type
, gl
->gl_name
.ln_number
,
42 gfs2_glock2aspace(gl
));
43 gfs2_lm_withdraw(gl
->gl_name
.ln_sbd
, "AIL error\n");
47 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
49 * @fsync: set when called from fsync (not all buffers will be clean)
51 * None of the buffers should be dirty, locked, or pinned.
54 static void __gfs2_ail_flush(struct gfs2_glock
*gl
, bool fsync
,
55 unsigned int nr_revokes
)
57 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
58 struct list_head
*head
= &gl
->gl_ail_list
;
59 struct gfs2_bufdata
*bd
, *tmp
;
60 struct buffer_head
*bh
;
61 const unsigned long b_state
= (1UL << BH_Dirty
)|(1UL << BH_Pinned
)|(1UL << BH_Lock
);
64 spin_lock(&sdp
->sd_ail_lock
);
65 list_for_each_entry_safe_reverse(bd
, tmp
, head
, bd_ail_gl_list
) {
69 if (bh
->b_state
& b_state
) {
72 gfs2_ail_error(gl
, bh
);
74 gfs2_trans_add_revoke(sdp
, bd
);
77 GLOCK_BUG_ON(gl
, !fsync
&& atomic_read(&gl
->gl_ail_count
));
78 spin_unlock(&sdp
->sd_ail_lock
);
83 static void gfs2_ail_empty_gl(struct gfs2_glock
*gl
)
85 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
88 memset(&tr
, 0, sizeof(tr
));
89 INIT_LIST_HEAD(&tr
.tr_buf
);
90 INIT_LIST_HEAD(&tr
.tr_databuf
);
91 tr
.tr_revokes
= atomic_read(&gl
->gl_ail_count
);
96 /* A shortened, inline version of gfs2_trans_begin()
97 * tr->alloced is not set since the transaction structure is
99 tr
.tr_reserved
= 1 + gfs2_struct2blk(sdp
, tr
.tr_revokes
, sizeof(u64
));
101 if (gfs2_log_reserve(sdp
, tr
.tr_reserved
) < 0)
103 WARN_ON_ONCE(current
->journal_info
);
104 current
->journal_info
= &tr
;
106 __gfs2_ail_flush(gl
, 0, tr
.tr_revokes
);
109 gfs2_log_flush(sdp
, NULL
, NORMAL_FLUSH
);
112 void gfs2_ail_flush(struct gfs2_glock
*gl
, bool fsync
)
114 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
115 unsigned int revokes
= atomic_read(&gl
->gl_ail_count
);
116 unsigned int max_revokes
= (sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_log_descriptor
)) / sizeof(u64
);
122 while (revokes
> max_revokes
)
123 max_revokes
+= (sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
125 ret
= gfs2_trans_begin(sdp
, 0, max_revokes
);
128 __gfs2_ail_flush(gl
, fsync
, max_revokes
);
130 gfs2_log_flush(sdp
, NULL
, NORMAL_FLUSH
);
134 * rgrp_go_sync - sync out the metadata for this glock
137 * Called when demoting or unlocking an EX glock. We must flush
138 * to disk all dirty buffers/pages relating to this glock, and must not
139 * not return to caller to demote/unlock the glock until I/O is complete.
142 static void rgrp_go_sync(struct gfs2_glock
*gl
)
144 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
145 struct address_space
*mapping
= &sdp
->sd_aspace
;
146 struct gfs2_rgrpd
*rgd
;
149 spin_lock(&gl
->gl_lockref
.lock
);
152 gfs2_rgrp_brelse(rgd
);
153 spin_unlock(&gl
->gl_lockref
.lock
);
155 if (!test_and_clear_bit(GLF_DIRTY
, &gl
->gl_flags
))
157 GLOCK_BUG_ON(gl
, gl
->gl_state
!= LM_ST_EXCLUSIVE
);
159 gfs2_log_flush(sdp
, gl
, NORMAL_FLUSH
);
160 filemap_fdatawrite_range(mapping
, gl
->gl_vm
.start
, gl
->gl_vm
.end
);
161 error
= filemap_fdatawait_range(mapping
, gl
->gl_vm
.start
, gl
->gl_vm
.end
);
162 mapping_set_error(mapping
, error
);
163 gfs2_ail_empty_gl(gl
);
165 spin_lock(&gl
->gl_lockref
.lock
);
168 gfs2_free_clones(rgd
);
169 spin_unlock(&gl
->gl_lockref
.lock
);
173 * rgrp_go_inval - invalidate the metadata for this glock
177 * We never used LM_ST_DEFERRED with resource groups, so that we
178 * should always see the metadata flag set here.
182 static void rgrp_go_inval(struct gfs2_glock
*gl
, int flags
)
184 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
185 struct address_space
*mapping
= &sdp
->sd_aspace
;
186 struct gfs2_rgrpd
*rgd
= gl
->gl_object
;
189 gfs2_rgrp_brelse(rgd
);
191 WARN_ON_ONCE(!(flags
& DIO_METADATA
));
192 gfs2_assert_withdraw(sdp
, !atomic_read(&gl
->gl_ail_count
));
193 truncate_inode_pages_range(mapping
, gl
->gl_vm
.start
, gl
->gl_vm
.end
);
196 rgd
->rd_flags
&= ~GFS2_RDF_UPTODATE
;
200 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
201 * @gl: the glock protecting the inode
205 static void inode_go_sync(struct gfs2_glock
*gl
)
207 struct gfs2_inode
*ip
= gl
->gl_object
;
208 struct address_space
*metamapping
= gfs2_glock2aspace(gl
);
211 if (ip
&& !S_ISREG(ip
->i_inode
.i_mode
))
214 if (test_and_clear_bit(GIF_SW_PAGED
, &ip
->i_flags
))
215 unmap_shared_mapping_range(ip
->i_inode
.i_mapping
, 0, 0);
216 inode_dio_wait(&ip
->i_inode
);
218 if (!test_and_clear_bit(GLF_DIRTY
, &gl
->gl_flags
))
221 GLOCK_BUG_ON(gl
, gl
->gl_state
!= LM_ST_EXCLUSIVE
);
223 gfs2_log_flush(gl
->gl_name
.ln_sbd
, gl
, NORMAL_FLUSH
);
224 filemap_fdatawrite(metamapping
);
226 struct address_space
*mapping
= ip
->i_inode
.i_mapping
;
227 filemap_fdatawrite(mapping
);
228 error
= filemap_fdatawait(mapping
);
229 mapping_set_error(mapping
, error
);
231 error
= filemap_fdatawait(metamapping
);
232 mapping_set_error(metamapping
, error
);
233 gfs2_ail_empty_gl(gl
);
235 * Writeback of the data mapping may cause the dirty flag to be set
236 * so we have to clear it again here.
238 smp_mb__before_atomic();
239 clear_bit(GLF_DIRTY
, &gl
->gl_flags
);
243 * inode_go_inval - prepare a inode glock to be released
247 * Normally we invalidate everything, but if we are moving into
248 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
249 * can keep hold of the metadata, since it won't have changed.
253 static void inode_go_inval(struct gfs2_glock
*gl
, int flags
)
255 struct gfs2_inode
*ip
= gl
->gl_object
;
257 gfs2_assert_withdraw(gl
->gl_name
.ln_sbd
, !atomic_read(&gl
->gl_ail_count
));
259 if (flags
& DIO_METADATA
) {
260 struct address_space
*mapping
= gfs2_glock2aspace(gl
);
261 truncate_inode_pages(mapping
, 0);
263 set_bit(GIF_INVALID
, &ip
->i_flags
);
264 forget_all_cached_acls(&ip
->i_inode
);
265 gfs2_dir_hash_inval(ip
);
269 if (ip
== GFS2_I(gl
->gl_name
.ln_sbd
->sd_rindex
)) {
270 gfs2_log_flush(gl
->gl_name
.ln_sbd
, NULL
, NORMAL_FLUSH
);
271 gl
->gl_name
.ln_sbd
->sd_rindex_uptodate
= 0;
273 if (ip
&& S_ISREG(ip
->i_inode
.i_mode
))
274 truncate_inode_pages(ip
->i_inode
.i_mapping
, 0);
278 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
281 * Returns: 1 if it's ok
284 static int inode_go_demote_ok(const struct gfs2_glock
*gl
)
286 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
287 struct gfs2_holder
*gh
;
289 if (sdp
->sd_jindex
== gl
->gl_object
|| sdp
->sd_rindex
== gl
->gl_object
)
292 if (!list_empty(&gl
->gl_holders
)) {
293 gh
= list_entry(gl
->gl_holders
.next
, struct gfs2_holder
, gh_list
);
294 if (gh
->gh_list
.next
!= &gl
->gl_holders
)
302 * gfs2_set_nlink - Set the inode's link count based on on-disk info
303 * @inode: The inode in question
304 * @nlink: The link count
306 * If the link count has hit zero, it must never be raised, whatever the
307 * on-disk inode might say. When new struct inodes are created the link
308 * count is set to 1, so that we can safely use this test even when reading
309 * in on disk information for the first time.
312 static void gfs2_set_nlink(struct inode
*inode
, u32 nlink
)
315 * We will need to review setting the nlink count here in the
316 * light of the forthcoming ro bind mount work. This is a reminder
319 if ((inode
->i_nlink
!= nlink
) && (inode
->i_nlink
!= 0)) {
323 set_nlink(inode
, nlink
);
327 static int gfs2_dinode_in(struct gfs2_inode
*ip
, const void *buf
)
329 const struct gfs2_dinode
*str
= buf
;
330 struct timespec atime
;
333 if (unlikely(ip
->i_no_addr
!= be64_to_cpu(str
->di_num
.no_addr
)))
335 ip
->i_no_formal_ino
= be64_to_cpu(str
->di_num
.no_formal_ino
);
336 ip
->i_inode
.i_mode
= be32_to_cpu(str
->di_mode
);
337 ip
->i_inode
.i_rdev
= 0;
338 switch (ip
->i_inode
.i_mode
& S_IFMT
) {
341 ip
->i_inode
.i_rdev
= MKDEV(be32_to_cpu(str
->di_major
),
342 be32_to_cpu(str
->di_minor
));
346 i_uid_write(&ip
->i_inode
, be32_to_cpu(str
->di_uid
));
347 i_gid_write(&ip
->i_inode
, be32_to_cpu(str
->di_gid
));
348 gfs2_set_nlink(&ip
->i_inode
, be32_to_cpu(str
->di_nlink
));
349 i_size_write(&ip
->i_inode
, be64_to_cpu(str
->di_size
));
350 gfs2_set_inode_blocks(&ip
->i_inode
, be64_to_cpu(str
->di_blocks
));
351 atime
.tv_sec
= be64_to_cpu(str
->di_atime
);
352 atime
.tv_nsec
= be32_to_cpu(str
->di_atime_nsec
);
353 if (timespec_compare(&ip
->i_inode
.i_atime
, &atime
) < 0)
354 ip
->i_inode
.i_atime
= atime
;
355 ip
->i_inode
.i_mtime
.tv_sec
= be64_to_cpu(str
->di_mtime
);
356 ip
->i_inode
.i_mtime
.tv_nsec
= be32_to_cpu(str
->di_mtime_nsec
);
357 ip
->i_inode
.i_ctime
.tv_sec
= be64_to_cpu(str
->di_ctime
);
358 ip
->i_inode
.i_ctime
.tv_nsec
= be32_to_cpu(str
->di_ctime_nsec
);
360 ip
->i_goal
= be64_to_cpu(str
->di_goal_meta
);
361 ip
->i_generation
= be64_to_cpu(str
->di_generation
);
363 ip
->i_diskflags
= be32_to_cpu(str
->di_flags
);
364 ip
->i_eattr
= be64_to_cpu(str
->di_eattr
);
365 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
366 gfs2_set_inode_flags(&ip
->i_inode
);
367 height
= be16_to_cpu(str
->di_height
);
368 if (unlikely(height
> GFS2_MAX_META_HEIGHT
))
370 ip
->i_height
= (u8
)height
;
372 depth
= be16_to_cpu(str
->di_depth
);
373 if (unlikely(depth
> GFS2_DIR_MAX_DEPTH
))
375 ip
->i_depth
= (u8
)depth
;
376 ip
->i_entries
= be32_to_cpu(str
->di_entries
);
378 if (S_ISREG(ip
->i_inode
.i_mode
))
379 gfs2_set_aops(&ip
->i_inode
);
383 gfs2_consist_inode(ip
);
388 * gfs2_inode_refresh - Refresh the incore copy of the dinode
389 * @ip: The GFS2 inode
394 int gfs2_inode_refresh(struct gfs2_inode
*ip
)
396 struct buffer_head
*dibh
;
399 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
403 error
= gfs2_dinode_in(ip
, dibh
->b_data
);
405 clear_bit(GIF_INVALID
, &ip
->i_flags
);
411 * inode_go_lock - operation done after an inode lock is locked by a process
418 static int inode_go_lock(struct gfs2_holder
*gh
)
420 struct gfs2_glock
*gl
= gh
->gh_gl
;
421 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
422 struct gfs2_inode
*ip
= gl
->gl_object
;
425 if (!ip
|| (gh
->gh_flags
& GL_SKIP
))
428 if (test_bit(GIF_INVALID
, &ip
->i_flags
)) {
429 error
= gfs2_inode_refresh(ip
);
434 if (gh
->gh_state
!= LM_ST_DEFERRED
)
435 inode_dio_wait(&ip
->i_inode
);
437 if ((ip
->i_diskflags
& GFS2_DIF_TRUNC_IN_PROG
) &&
438 (gl
->gl_state
== LM_ST_EXCLUSIVE
) &&
439 (gh
->gh_state
== LM_ST_EXCLUSIVE
)) {
440 spin_lock(&sdp
->sd_trunc_lock
);
441 if (list_empty(&ip
->i_trunc_list
))
442 list_add(&sdp
->sd_trunc_list
, &ip
->i_trunc_list
);
443 spin_unlock(&sdp
->sd_trunc_lock
);
444 wake_up(&sdp
->sd_quota_wait
);
452 * inode_go_dump - print information about an inode
458 static void inode_go_dump(struct seq_file
*seq
, const struct gfs2_glock
*gl
)
460 const struct gfs2_inode
*ip
= gl
->gl_object
;
463 gfs2_print_dbg(seq
, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
464 (unsigned long long)ip
->i_no_formal_ino
,
465 (unsigned long long)ip
->i_no_addr
,
466 IF2DT(ip
->i_inode
.i_mode
), ip
->i_flags
,
467 (unsigned int)ip
->i_diskflags
,
468 (unsigned long long)i_size_read(&ip
->i_inode
));
472 * freeze_go_sync - promote/demote the freeze glock
474 * @state: the requested state
479 static void freeze_go_sync(struct gfs2_glock
*gl
)
482 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
484 if (gl
->gl_state
== LM_ST_SHARED
&&
485 test_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
)) {
486 atomic_set(&sdp
->sd_freeze_state
, SFS_STARTING_FREEZE
);
487 error
= freeze_super(sdp
->sd_vfs
);
489 printk(KERN_INFO
"GFS2: couldn't freeze filesystem: %d\n", error
);
490 gfs2_assert_withdraw(sdp
, 0);
492 queue_work(gfs2_freeze_wq
, &sdp
->sd_freeze_work
);
493 gfs2_log_flush(sdp
, NULL
, FREEZE_FLUSH
);
498 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
503 static int freeze_go_xmote_bh(struct gfs2_glock
*gl
, struct gfs2_holder
*gh
)
505 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
506 struct gfs2_inode
*ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
507 struct gfs2_glock
*j_gl
= ip
->i_gl
;
508 struct gfs2_log_header_host head
;
511 if (test_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
)) {
512 j_gl
->gl_ops
->go_inval(j_gl
, DIO_METADATA
);
514 error
= gfs2_find_jhead(sdp
->sd_jdesc
, &head
);
517 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
))
520 /* Initialize some head of the log stuff */
521 if (!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) {
522 sdp
->sd_log_sequence
= head
.lh_sequence
+ 1;
523 gfs2_log_pointers_init(sdp
, head
.lh_blkno
);
536 static int freeze_go_demote_ok(const struct gfs2_glock
*gl
)
542 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
545 * gl_lockref.lock lock is held while calling this
547 static void iopen_go_callback(struct gfs2_glock
*gl
, bool remote
)
549 struct gfs2_inode
*ip
= (struct gfs2_inode
*)gl
->gl_object
;
550 struct gfs2_sbd
*sdp
= gl
->gl_name
.ln_sbd
;
552 if (!remote
|| (sdp
->sd_vfs
->s_flags
& MS_RDONLY
))
555 if (gl
->gl_demote_state
== LM_ST_UNLOCKED
&&
556 gl
->gl_state
== LM_ST_SHARED
&& ip
) {
557 gl
->gl_lockref
.count
++;
558 if (queue_work(gfs2_delete_workqueue
, &gl
->gl_delete
) == 0)
559 gl
->gl_lockref
.count
--;
563 const struct gfs2_glock_operations gfs2_meta_glops
= {
564 .go_type
= LM_TYPE_META
,
567 const struct gfs2_glock_operations gfs2_inode_glops
= {
568 .go_sync
= inode_go_sync
,
569 .go_inval
= inode_go_inval
,
570 .go_demote_ok
= inode_go_demote_ok
,
571 .go_lock
= inode_go_lock
,
572 .go_dump
= inode_go_dump
,
573 .go_type
= LM_TYPE_INODE
,
574 .go_flags
= GLOF_ASPACE
| GLOF_LRU
,
577 const struct gfs2_glock_operations gfs2_rgrp_glops
= {
578 .go_sync
= rgrp_go_sync
,
579 .go_inval
= rgrp_go_inval
,
580 .go_lock
= gfs2_rgrp_go_lock
,
581 .go_unlock
= gfs2_rgrp_go_unlock
,
582 .go_dump
= gfs2_rgrp_dump
,
583 .go_type
= LM_TYPE_RGRP
,
584 .go_flags
= GLOF_LVB
,
587 const struct gfs2_glock_operations gfs2_freeze_glops
= {
588 .go_sync
= freeze_go_sync
,
589 .go_xmote_bh
= freeze_go_xmote_bh
,
590 .go_demote_ok
= freeze_go_demote_ok
,
591 .go_type
= LM_TYPE_NONDISK
,
594 const struct gfs2_glock_operations gfs2_iopen_glops
= {
595 .go_type
= LM_TYPE_IOPEN
,
596 .go_callback
= iopen_go_callback
,
597 .go_flags
= GLOF_LRU
,
600 const struct gfs2_glock_operations gfs2_flock_glops
= {
601 .go_type
= LM_TYPE_FLOCK
,
602 .go_flags
= GLOF_LRU
,
605 const struct gfs2_glock_operations gfs2_nondisk_glops
= {
606 .go_type
= LM_TYPE_NONDISK
,
609 const struct gfs2_glock_operations gfs2_quota_glops
= {
610 .go_type
= LM_TYPE_QUOTA
,
611 .go_flags
= GLOF_LVB
| GLOF_LRU
,
614 const struct gfs2_glock_operations gfs2_journal_glops
= {
615 .go_type
= LM_TYPE_JOURNAL
,
618 const struct gfs2_glock_operations
*gfs2_glops_list
[] = {
619 [LM_TYPE_META
] = &gfs2_meta_glops
,
620 [LM_TYPE_INODE
] = &gfs2_inode_glops
,
621 [LM_TYPE_RGRP
] = &gfs2_rgrp_glops
,
622 [LM_TYPE_IOPEN
] = &gfs2_iopen_glops
,
623 [LM_TYPE_FLOCK
] = &gfs2_flock_glops
,
624 [LM_TYPE_NONDISK
] = &gfs2_nondisk_glops
,
625 [LM_TYPE_QUOTA
] = &gfs2_quota_glops
,
626 [LM_TYPE_JOURNAL
] = &gfs2_journal_glops
,