Merge 4.9-rc5 into char-misc-next
[linux-2.6/btrfs-unstable.git] / fs / gfs2 / glops.c
blob5db59d44483809fa027196910a90d10536843c37
1 /*
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.
8 */
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>
16 #include <linux/security.h>
18 #include "gfs2.h"
19 #include "incore.h"
20 #include "bmap.h"
21 #include "glock.h"
22 #include "glops.h"
23 #include "inode.h"
24 #include "log.h"
25 #include "meta_io.h"
26 #include "recovery.h"
27 #include "rgrp.h"
28 #include "util.h"
29 #include "trans.h"
30 #include "dir.h"
32 struct workqueue_struct *gfs2_freeze_wq;
34 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
36 fs_err(gl->gl_name.ln_sbd,
37 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
38 "state 0x%lx\n",
39 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
40 bh->b_page->mapping, bh->b_page->flags);
41 fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
42 gl->gl_name.ln_type, gl->gl_name.ln_number,
43 gfs2_glock2aspace(gl));
44 gfs2_lm_withdraw(gl->gl_name.ln_sbd, "AIL error\n");
47 /**
48 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
49 * @gl: the glock
50 * @fsync: set when called from fsync (not all buffers will be clean)
52 * None of the buffers should be dirty, locked, or pinned.
55 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
56 unsigned int nr_revokes)
58 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
59 struct list_head *head = &gl->gl_ail_list;
60 struct gfs2_bufdata *bd, *tmp;
61 struct buffer_head *bh;
62 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
64 gfs2_log_lock(sdp);
65 spin_lock(&sdp->sd_ail_lock);
66 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
67 if (nr_revokes == 0)
68 break;
69 bh = bd->bd_bh;
70 if (bh->b_state & b_state) {
71 if (fsync)
72 continue;
73 gfs2_ail_error(gl, bh);
75 gfs2_trans_add_revoke(sdp, bd);
76 nr_revokes--;
78 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
79 spin_unlock(&sdp->sd_ail_lock);
80 gfs2_log_unlock(sdp);
84 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
86 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
87 struct gfs2_trans tr;
89 memset(&tr, 0, sizeof(tr));
90 INIT_LIST_HEAD(&tr.tr_buf);
91 INIT_LIST_HEAD(&tr.tr_databuf);
92 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
94 if (!tr.tr_revokes)
95 return;
97 /* A shortened, inline version of gfs2_trans_begin()
98 * tr->alloced is not set since the transaction structure is
99 * on the stack */
100 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
101 tr.tr_ip = _RET_IP_;
102 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
103 return;
104 WARN_ON_ONCE(current->journal_info);
105 current->journal_info = &tr;
107 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
109 gfs2_trans_end(sdp);
110 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
113 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
115 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
116 unsigned int revokes = atomic_read(&gl->gl_ail_count);
117 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
118 int ret;
120 if (!revokes)
121 return;
123 while (revokes > max_revokes)
124 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
126 ret = gfs2_trans_begin(sdp, 0, max_revokes);
127 if (ret)
128 return;
129 __gfs2_ail_flush(gl, fsync, max_revokes);
130 gfs2_trans_end(sdp);
131 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
135 * rgrp_go_sync - sync out the metadata for this glock
136 * @gl: the glock
138 * Called when demoting or unlocking an EX glock. We must flush
139 * to disk all dirty buffers/pages relating to this glock, and must not
140 * not return to caller to demote/unlock the glock until I/O is complete.
143 static void rgrp_go_sync(struct gfs2_glock *gl)
145 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
146 struct address_space *mapping = &sdp->sd_aspace;
147 struct gfs2_rgrpd *rgd;
148 int error;
150 spin_lock(&gl->gl_lockref.lock);
151 rgd = gl->gl_object;
152 if (rgd)
153 gfs2_rgrp_brelse(rgd);
154 spin_unlock(&gl->gl_lockref.lock);
156 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
157 return;
158 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
160 gfs2_log_flush(sdp, gl, NORMAL_FLUSH);
161 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
162 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
163 mapping_set_error(mapping, error);
164 gfs2_ail_empty_gl(gl);
166 spin_lock(&gl->gl_lockref.lock);
167 rgd = gl->gl_object;
168 if (rgd)
169 gfs2_free_clones(rgd);
170 spin_unlock(&gl->gl_lockref.lock);
174 * rgrp_go_inval - invalidate the metadata for this glock
175 * @gl: the glock
176 * @flags:
178 * We never used LM_ST_DEFERRED with resource groups, so that we
179 * should always see the metadata flag set here.
183 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
185 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
186 struct address_space *mapping = &sdp->sd_aspace;
187 struct gfs2_rgrpd *rgd = gl->gl_object;
189 if (rgd)
190 gfs2_rgrp_brelse(rgd);
192 WARN_ON_ONCE(!(flags & DIO_METADATA));
193 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
194 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
196 if (rgd)
197 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
201 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
202 * @gl: the glock protecting the inode
206 static void inode_go_sync(struct gfs2_glock *gl)
208 struct gfs2_inode *ip = gl->gl_object;
209 struct address_space *metamapping = gfs2_glock2aspace(gl);
210 int error;
212 if (ip && !S_ISREG(ip->i_inode.i_mode))
213 ip = NULL;
214 if (ip) {
215 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
216 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
217 inode_dio_wait(&ip->i_inode);
219 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
220 return;
222 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
224 gfs2_log_flush(gl->gl_name.ln_sbd, gl, NORMAL_FLUSH);
225 filemap_fdatawrite(metamapping);
226 if (ip) {
227 struct address_space *mapping = ip->i_inode.i_mapping;
228 filemap_fdatawrite(mapping);
229 error = filemap_fdatawait(mapping);
230 mapping_set_error(mapping, error);
232 error = filemap_fdatawait(metamapping);
233 mapping_set_error(metamapping, error);
234 gfs2_ail_empty_gl(gl);
236 * Writeback of the data mapping may cause the dirty flag to be set
237 * so we have to clear it again here.
239 smp_mb__before_atomic();
240 clear_bit(GLF_DIRTY, &gl->gl_flags);
244 * inode_go_inval - prepare a inode glock to be released
245 * @gl: the glock
246 * @flags:
248 * Normally we invalidate everything, but if we are moving into
249 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
250 * can keep hold of the metadata, since it won't have changed.
254 static void inode_go_inval(struct gfs2_glock *gl, int flags)
256 struct gfs2_inode *ip = gl->gl_object;
258 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
260 if (flags & DIO_METADATA) {
261 struct address_space *mapping = gfs2_glock2aspace(gl);
262 truncate_inode_pages(mapping, 0);
263 if (ip) {
264 set_bit(GIF_INVALID, &ip->i_flags);
265 forget_all_cached_acls(&ip->i_inode);
266 security_inode_invalidate_secctx(&ip->i_inode);
267 gfs2_dir_hash_inval(ip);
271 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
272 gfs2_log_flush(gl->gl_name.ln_sbd, NULL, NORMAL_FLUSH);
273 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
275 if (ip && S_ISREG(ip->i_inode.i_mode))
276 truncate_inode_pages(ip->i_inode.i_mapping, 0);
280 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
281 * @gl: the glock
283 * Returns: 1 if it's ok
286 static int inode_go_demote_ok(const struct gfs2_glock *gl)
288 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
290 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
291 return 0;
293 return 1;
297 * gfs2_set_nlink - Set the inode's link count based on on-disk info
298 * @inode: The inode in question
299 * @nlink: The link count
301 * If the link count has hit zero, it must never be raised, whatever the
302 * on-disk inode might say. When new struct inodes are created the link
303 * count is set to 1, so that we can safely use this test even when reading
304 * in on disk information for the first time.
307 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
310 * We will need to review setting the nlink count here in the
311 * light of the forthcoming ro bind mount work. This is a reminder
312 * to do that.
314 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
315 if (nlink == 0)
316 clear_nlink(inode);
317 else
318 set_nlink(inode, nlink);
322 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
324 const struct gfs2_dinode *str = buf;
325 struct timespec atime;
326 u16 height, depth;
328 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
329 goto corrupt;
330 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
331 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
332 ip->i_inode.i_rdev = 0;
333 switch (ip->i_inode.i_mode & S_IFMT) {
334 case S_IFBLK:
335 case S_IFCHR:
336 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
337 be32_to_cpu(str->di_minor));
338 break;
341 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
342 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
343 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
344 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
345 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
346 atime.tv_sec = be64_to_cpu(str->di_atime);
347 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
348 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
349 ip->i_inode.i_atime = atime;
350 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
351 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
352 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
353 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
355 ip->i_goal = be64_to_cpu(str->di_goal_meta);
356 ip->i_generation = be64_to_cpu(str->di_generation);
358 ip->i_diskflags = be32_to_cpu(str->di_flags);
359 ip->i_eattr = be64_to_cpu(str->di_eattr);
360 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
361 gfs2_set_inode_flags(&ip->i_inode);
362 height = be16_to_cpu(str->di_height);
363 if (unlikely(height > GFS2_MAX_META_HEIGHT))
364 goto corrupt;
365 ip->i_height = (u8)height;
367 depth = be16_to_cpu(str->di_depth);
368 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
369 goto corrupt;
370 ip->i_depth = (u8)depth;
371 ip->i_entries = be32_to_cpu(str->di_entries);
373 if (S_ISREG(ip->i_inode.i_mode))
374 gfs2_set_aops(&ip->i_inode);
376 return 0;
377 corrupt:
378 gfs2_consist_inode(ip);
379 return -EIO;
383 * gfs2_inode_refresh - Refresh the incore copy of the dinode
384 * @ip: The GFS2 inode
386 * Returns: errno
389 int gfs2_inode_refresh(struct gfs2_inode *ip)
391 struct buffer_head *dibh;
392 int error;
394 error = gfs2_meta_inode_buffer(ip, &dibh);
395 if (error)
396 return error;
398 error = gfs2_dinode_in(ip, dibh->b_data);
399 brelse(dibh);
400 clear_bit(GIF_INVALID, &ip->i_flags);
402 return error;
406 * inode_go_lock - operation done after an inode lock is locked by a process
407 * @gl: the glock
408 * @flags:
410 * Returns: errno
413 static int inode_go_lock(struct gfs2_holder *gh)
415 struct gfs2_glock *gl = gh->gh_gl;
416 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
417 struct gfs2_inode *ip = gl->gl_object;
418 int error = 0;
420 if (!ip || (gh->gh_flags & GL_SKIP))
421 return 0;
423 if (test_bit(GIF_INVALID, &ip->i_flags)) {
424 error = gfs2_inode_refresh(ip);
425 if (error)
426 return error;
429 if (gh->gh_state != LM_ST_DEFERRED)
430 inode_dio_wait(&ip->i_inode);
432 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
433 (gl->gl_state == LM_ST_EXCLUSIVE) &&
434 (gh->gh_state == LM_ST_EXCLUSIVE)) {
435 spin_lock(&sdp->sd_trunc_lock);
436 if (list_empty(&ip->i_trunc_list))
437 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
438 spin_unlock(&sdp->sd_trunc_lock);
439 wake_up(&sdp->sd_quota_wait);
440 return 1;
443 return error;
447 * inode_go_dump - print information about an inode
448 * @seq: The iterator
449 * @ip: the inode
453 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
455 const struct gfs2_inode *ip = gl->gl_object;
456 if (ip == NULL)
457 return;
458 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
459 (unsigned long long)ip->i_no_formal_ino,
460 (unsigned long long)ip->i_no_addr,
461 IF2DT(ip->i_inode.i_mode), ip->i_flags,
462 (unsigned int)ip->i_diskflags,
463 (unsigned long long)i_size_read(&ip->i_inode));
467 * freeze_go_sync - promote/demote the freeze glock
468 * @gl: the glock
469 * @state: the requested state
470 * @flags:
474 static void freeze_go_sync(struct gfs2_glock *gl)
476 int error = 0;
477 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
479 if (gl->gl_state == LM_ST_SHARED &&
480 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
481 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
482 error = freeze_super(sdp->sd_vfs);
483 if (error) {
484 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
485 gfs2_assert_withdraw(sdp, 0);
487 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
488 gfs2_log_flush(sdp, NULL, FREEZE_FLUSH);
493 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
494 * @gl: the glock
498 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
500 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
501 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
502 struct gfs2_glock *j_gl = ip->i_gl;
503 struct gfs2_log_header_host head;
504 int error;
506 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
507 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
509 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
510 if (error)
511 gfs2_consist(sdp);
512 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
513 gfs2_consist(sdp);
515 /* Initialize some head of the log stuff */
516 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
517 sdp->sd_log_sequence = head.lh_sequence + 1;
518 gfs2_log_pointers_init(sdp, head.lh_blkno);
521 return 0;
525 * trans_go_demote_ok
526 * @gl: the glock
528 * Always returns 0
531 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
533 return 0;
537 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
538 * @gl: the glock
540 * gl_lockref.lock lock is held while calling this
542 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
544 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
545 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
547 if (!remote || (sdp->sd_vfs->s_flags & MS_RDONLY))
548 return;
550 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
551 gl->gl_state == LM_ST_SHARED && ip) {
552 gl->gl_lockref.count++;
553 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
554 gl->gl_lockref.count--;
558 const struct gfs2_glock_operations gfs2_meta_glops = {
559 .go_type = LM_TYPE_META,
562 const struct gfs2_glock_operations gfs2_inode_glops = {
563 .go_sync = inode_go_sync,
564 .go_inval = inode_go_inval,
565 .go_demote_ok = inode_go_demote_ok,
566 .go_lock = inode_go_lock,
567 .go_dump = inode_go_dump,
568 .go_type = LM_TYPE_INODE,
569 .go_flags = GLOF_ASPACE | GLOF_LRU,
572 const struct gfs2_glock_operations gfs2_rgrp_glops = {
573 .go_sync = rgrp_go_sync,
574 .go_inval = rgrp_go_inval,
575 .go_lock = gfs2_rgrp_go_lock,
576 .go_unlock = gfs2_rgrp_go_unlock,
577 .go_dump = gfs2_rgrp_dump,
578 .go_type = LM_TYPE_RGRP,
579 .go_flags = GLOF_LVB,
582 const struct gfs2_glock_operations gfs2_freeze_glops = {
583 .go_sync = freeze_go_sync,
584 .go_xmote_bh = freeze_go_xmote_bh,
585 .go_demote_ok = freeze_go_demote_ok,
586 .go_type = LM_TYPE_NONDISK,
589 const struct gfs2_glock_operations gfs2_iopen_glops = {
590 .go_type = LM_TYPE_IOPEN,
591 .go_callback = iopen_go_callback,
592 .go_flags = GLOF_LRU,
595 const struct gfs2_glock_operations gfs2_flock_glops = {
596 .go_type = LM_TYPE_FLOCK,
597 .go_flags = GLOF_LRU,
600 const struct gfs2_glock_operations gfs2_nondisk_glops = {
601 .go_type = LM_TYPE_NONDISK,
604 const struct gfs2_glock_operations gfs2_quota_glops = {
605 .go_type = LM_TYPE_QUOTA,
606 .go_flags = GLOF_LVB | GLOF_LRU,
609 const struct gfs2_glock_operations gfs2_journal_glops = {
610 .go_type = LM_TYPE_JOURNAL,
613 const struct gfs2_glock_operations *gfs2_glops_list[] = {
614 [LM_TYPE_META] = &gfs2_meta_glops,
615 [LM_TYPE_INODE] = &gfs2_inode_glops,
616 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
617 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
618 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
619 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
620 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
621 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,