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>
16 #include <linux/security.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 "
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");
48 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
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);
65 spin_lock(&sdp->sd_ail_lock);
66 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
70 if (bh->b_state & b_state) {
73 gfs2_ail_error(gl, bh);
75 gfs2_trans_add_revoke(sdp, bd);
78 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
79 spin_unlock(&sdp->sd_ail_lock);
84 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
86 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
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);
97 /* A shortened, inline version of gfs2_trans_begin()
98 * tr->alloced is not set since the transaction structure is
100 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
102 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
104 WARN_ON_ONCE(current->journal_info);
105 current->journal_info = &tr;
107 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
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);
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);
129 __gfs2_ail_flush(gl, fsync, max_revokes);
131 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
135 * rgrp_go_sync - sync out the metadata for this 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 * 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;
150 spin_lock(&gl->gl_lockref.lock);
153 gfs2_rgrp_brelse(rgd);
154 spin_unlock(&gl->gl_lockref.lock);
156 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
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);
169 gfs2_free_clones(rgd);
170 spin_unlock(&gl->gl_lockref.lock);
174 * rgrp_go_inval - invalidate the metadata for this glock
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 = gfs2_glock2rgrp(gl);
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);
197 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
200 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
202 struct gfs2_inode *ip;
204 spin_lock(&gl->gl_lockref.lock);
207 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
208 spin_unlock(&gl->gl_lockref.lock);
212 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
214 struct gfs2_rgrpd *rgd;
216 spin_lock(&gl->gl_lockref.lock);
218 spin_unlock(&gl->gl_lockref.lock);
223 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
228 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
229 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
233 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
234 * @gl: the glock protecting the inode
238 static void inode_go_sync(struct gfs2_glock *gl)
240 struct gfs2_inode *ip = gfs2_glock2inode(gl);
241 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
242 struct address_space *metamapping = gfs2_glock2aspace(gl);
246 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
247 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
248 inode_dio_wait(&ip->i_inode);
250 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
253 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
255 gfs2_log_flush(gl->gl_name.ln_sbd, gl, NORMAL_FLUSH);
256 filemap_fdatawrite(metamapping);
258 struct address_space *mapping = ip->i_inode.i_mapping;
259 filemap_fdatawrite(mapping);
260 error = filemap_fdatawait(mapping);
261 mapping_set_error(mapping, error);
263 error = filemap_fdatawait(metamapping);
264 mapping_set_error(metamapping, error);
265 gfs2_ail_empty_gl(gl);
267 * Writeback of the data mapping may cause the dirty flag to be set
268 * so we have to clear it again here.
270 smp_mb__before_atomic();
271 clear_bit(GLF_DIRTY, &gl->gl_flags);
274 gfs2_clear_glop_pending(ip);
278 * inode_go_inval - prepare a inode glock to be released
282 * Normally we invalidate everything, but if we are moving into
283 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
284 * can keep hold of the metadata, since it won't have changed.
288 static void inode_go_inval(struct gfs2_glock *gl, int flags)
290 struct gfs2_inode *ip = gfs2_glock2inode(gl);
292 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
294 if (flags & DIO_METADATA) {
295 struct address_space *mapping = gfs2_glock2aspace(gl);
296 truncate_inode_pages(mapping, 0);
298 set_bit(GIF_INVALID, &ip->i_flags);
299 forget_all_cached_acls(&ip->i_inode);
300 security_inode_invalidate_secctx(&ip->i_inode);
301 gfs2_dir_hash_inval(ip);
305 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
306 gfs2_log_flush(gl->gl_name.ln_sbd, NULL, NORMAL_FLUSH);
307 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
309 if (ip && S_ISREG(ip->i_inode.i_mode))
310 truncate_inode_pages(ip->i_inode.i_mapping, 0);
312 gfs2_clear_glop_pending(ip);
316 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
319 * Returns: 1 if it's ok
322 static int inode_go_demote_ok(const struct gfs2_glock *gl)
324 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
326 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
333 * gfs2_set_nlink - Set the inode's link count based on on-disk info
334 * @inode: The inode in question
335 * @nlink: The link count
337 * If the link count has hit zero, it must never be raised, whatever the
338 * on-disk inode might say. When new struct inodes are created the link
339 * count is set to 1, so that we can safely use this test even when reading
340 * in on disk information for the first time.
343 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
346 * We will need to review setting the nlink count here in the
347 * light of the forthcoming ro bind mount work. This is a reminder
350 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
354 set_nlink(inode, nlink);
358 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
360 const struct gfs2_dinode *str = buf;
361 struct timespec atime;
364 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
366 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
367 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
368 ip->i_inode.i_rdev = 0;
369 switch (ip->i_inode.i_mode & S_IFMT) {
372 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
373 be32_to_cpu(str->di_minor));
377 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
378 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
379 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
380 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
381 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
382 atime.tv_sec = be64_to_cpu(str->di_atime);
383 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
384 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
385 ip->i_inode.i_atime = atime;
386 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
387 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
388 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
389 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
391 ip->i_goal = be64_to_cpu(str->di_goal_meta);
392 ip->i_generation = be64_to_cpu(str->di_generation);
394 ip->i_diskflags = be32_to_cpu(str->di_flags);
395 ip->i_eattr = be64_to_cpu(str->di_eattr);
396 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
397 gfs2_set_inode_flags(&ip->i_inode);
398 height = be16_to_cpu(str->di_height);
399 if (unlikely(height > GFS2_MAX_META_HEIGHT))
401 ip->i_height = (u8)height;
403 depth = be16_to_cpu(str->di_depth);
404 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
406 ip->i_depth = (u8)depth;
407 ip->i_entries = be32_to_cpu(str->di_entries);
409 if (S_ISREG(ip->i_inode.i_mode))
410 gfs2_set_aops(&ip->i_inode);
414 gfs2_consist_inode(ip);
419 * gfs2_inode_refresh - Refresh the incore copy of the dinode
420 * @ip: The GFS2 inode
425 int gfs2_inode_refresh(struct gfs2_inode *ip)
427 struct buffer_head *dibh;
430 error = gfs2_meta_inode_buffer(ip, &dibh);
434 error = gfs2_dinode_in(ip, dibh->b_data);
436 clear_bit(GIF_INVALID, &ip->i_flags);
442 * inode_go_lock - operation done after an inode lock is locked by a process
449 static int inode_go_lock(struct gfs2_holder *gh)
451 struct gfs2_glock *gl = gh->gh_gl;
452 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
453 struct gfs2_inode *ip = gl->gl_object;
456 if (!ip || (gh->gh_flags & GL_SKIP))
459 if (test_bit(GIF_INVALID, &ip->i_flags)) {
460 error = gfs2_inode_refresh(ip);
465 if (gh->gh_state != LM_ST_DEFERRED)
466 inode_dio_wait(&ip->i_inode);
468 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
469 (gl->gl_state == LM_ST_EXCLUSIVE) &&
470 (gh->gh_state == LM_ST_EXCLUSIVE)) {
471 spin_lock(&sdp->sd_trunc_lock);
472 if (list_empty(&ip->i_trunc_list))
473 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
474 spin_unlock(&sdp->sd_trunc_lock);
475 wake_up(&sdp->sd_quota_wait);
483 * inode_go_dump - print information about an inode
489 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
491 const struct gfs2_inode *ip = gl->gl_object;
494 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
495 (unsigned long long)ip->i_no_formal_ino,
496 (unsigned long long)ip->i_no_addr,
497 IF2DT(ip->i_inode.i_mode), ip->i_flags,
498 (unsigned int)ip->i_diskflags,
499 (unsigned long long)i_size_read(&ip->i_inode));
503 * freeze_go_sync - promote/demote the freeze glock
505 * @state: the requested state
510 static void freeze_go_sync(struct gfs2_glock *gl)
513 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
515 if (gl->gl_state == LM_ST_SHARED &&
516 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
517 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
518 error = freeze_super(sdp->sd_vfs);
520 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
521 gfs2_assert_withdraw(sdp, 0);
523 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
524 gfs2_log_flush(sdp, NULL, FREEZE_FLUSH);
529 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
534 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
536 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
537 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
538 struct gfs2_glock *j_gl = ip->i_gl;
539 struct gfs2_log_header_host head;
542 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
543 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
545 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
548 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
551 /* Initialize some head of the log stuff */
552 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
553 sdp->sd_log_sequence = head.lh_sequence + 1;
554 gfs2_log_pointers_init(sdp, head.lh_blkno);
567 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
573 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
576 * gl_lockref.lock lock is held while calling this
578 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
580 struct gfs2_inode *ip = gl->gl_object;
581 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
583 if (!remote || (sdp->sd_vfs->s_flags & MS_RDONLY))
586 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
587 gl->gl_state == LM_ST_SHARED && ip) {
588 gl->gl_lockref.count++;
589 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
590 gl->gl_lockref.count--;
594 const struct gfs2_glock_operations gfs2_meta_glops = {
595 .go_type = LM_TYPE_META,
598 const struct gfs2_glock_operations gfs2_inode_glops = {
599 .go_sync = inode_go_sync,
600 .go_inval = inode_go_inval,
601 .go_demote_ok = inode_go_demote_ok,
602 .go_lock = inode_go_lock,
603 .go_dump = inode_go_dump,
604 .go_type = LM_TYPE_INODE,
605 .go_flags = GLOF_ASPACE | GLOF_LRU,
608 const struct gfs2_glock_operations gfs2_rgrp_glops = {
609 .go_sync = rgrp_go_sync,
610 .go_inval = rgrp_go_inval,
611 .go_lock = gfs2_rgrp_go_lock,
612 .go_unlock = gfs2_rgrp_go_unlock,
613 .go_dump = gfs2_rgrp_dump,
614 .go_type = LM_TYPE_RGRP,
615 .go_flags = GLOF_LVB,
618 const struct gfs2_glock_operations gfs2_freeze_glops = {
619 .go_sync = freeze_go_sync,
620 .go_xmote_bh = freeze_go_xmote_bh,
621 .go_demote_ok = freeze_go_demote_ok,
622 .go_type = LM_TYPE_NONDISK,
625 const struct gfs2_glock_operations gfs2_iopen_glops = {
626 .go_type = LM_TYPE_IOPEN,
627 .go_callback = iopen_go_callback,
628 .go_flags = GLOF_LRU,
631 const struct gfs2_glock_operations gfs2_flock_glops = {
632 .go_type = LM_TYPE_FLOCK,
633 .go_flags = GLOF_LRU,
636 const struct gfs2_glock_operations gfs2_nondisk_glops = {
637 .go_type = LM_TYPE_NONDISK,
640 const struct gfs2_glock_operations gfs2_quota_glops = {
641 .go_type = LM_TYPE_QUOTA,
642 .go_flags = GLOF_LVB | GLOF_LRU,
645 const struct gfs2_glock_operations gfs2_journal_glops = {
646 .go_type = LM_TYPE_JOURNAL,
649 const struct gfs2_glock_operations *gfs2_glops_list[] = {
650 [LM_TYPE_META] = &gfs2_meta_glops,
651 [LM_TYPE_INODE] = &gfs2_inode_glops,
652 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
653 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
654 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
655 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
656 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
657 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,