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_sbd, "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page state 0x%lx\n",
36 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
37 bh->b_page->mapping, bh->b_page->flags);
38 fs_err(gl->gl_sbd, "AIL glock %u:%llu mapping %p\n",
39 gl->gl_name.ln_type, gl->gl_name.ln_number,
40 gfs2_glock2aspace(gl));
41 gfs2_lm_withdraw(gl->gl_sbd, "AIL error\n");
45 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
47 * @fsync: set when called from fsync (not all buffers will be clean)
49 * None of the buffers should be dirty, locked, or pinned.
52 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
53 unsigned int nr_revokes)
55 struct gfs2_sbd *sdp = gl->gl_sbd;
56 struct list_head *head = &gl->gl_ail_list;
57 struct gfs2_bufdata *bd, *tmp;
58 struct buffer_head *bh;
59 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
62 spin_lock(&sdp->sd_ail_lock);
63 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
67 if (bh->b_state & b_state) {
70 gfs2_ail_error(gl, bh);
72 gfs2_trans_add_revoke(sdp, bd);
75 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
76 spin_unlock(&sdp->sd_ail_lock);
81 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
83 struct gfs2_sbd *sdp = gl->gl_sbd;
86 memset(&tr, 0, sizeof(tr));
87 INIT_LIST_HEAD(&tr.tr_buf);
88 INIT_LIST_HEAD(&tr.tr_databuf);
89 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
94 /* A shortened, inline version of gfs2_trans_begin()
95 * tr->alloced is not set since the transaction structure is
97 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
99 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
101 WARN_ON_ONCE(current->journal_info);
102 current->journal_info = &tr;
104 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
107 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
110 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
112 struct gfs2_sbd *sdp = gl->gl_sbd;
113 unsigned int revokes = atomic_read(&gl->gl_ail_count);
114 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
120 while (revokes > max_revokes)
121 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
123 ret = gfs2_trans_begin(sdp, 0, max_revokes);
126 __gfs2_ail_flush(gl, fsync, max_revokes);
128 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
132 * rgrp_go_sync - sync out the metadata for this glock
135 * Called when demoting or unlocking an EX glock. We must flush
136 * to disk all dirty buffers/pages relating to this glock, and must not
137 * not return to caller to demote/unlock the glock until I/O is complete.
140 static void rgrp_go_sync(struct gfs2_glock *gl)
142 struct gfs2_sbd *sdp = gl->gl_sbd;
143 struct address_space *mapping = &sdp->sd_aspace;
144 struct gfs2_rgrpd *rgd;
147 spin_lock(&gl->gl_spin);
150 gfs2_rgrp_brelse(rgd);
151 spin_unlock(&gl->gl_spin);
153 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
155 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
157 gfs2_log_flush(sdp, gl, NORMAL_FLUSH);
158 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
159 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
160 mapping_set_error(mapping, error);
161 gfs2_ail_empty_gl(gl);
163 spin_lock(&gl->gl_spin);
166 gfs2_free_clones(rgd);
167 spin_unlock(&gl->gl_spin);
171 * rgrp_go_inval - invalidate the metadata for this glock
175 * We never used LM_ST_DEFERRED with resource groups, so that we
176 * should always see the metadata flag set here.
180 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
182 struct gfs2_sbd *sdp = gl->gl_sbd;
183 struct address_space *mapping = &sdp->sd_aspace;
184 struct gfs2_rgrpd *rgd = gl->gl_object;
187 gfs2_rgrp_brelse(rgd);
189 WARN_ON_ONCE(!(flags & DIO_METADATA));
190 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
191 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
194 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
198 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
199 * @gl: the glock protecting the inode
203 static void inode_go_sync(struct gfs2_glock *gl)
205 struct gfs2_inode *ip = gl->gl_object;
206 struct address_space *metamapping = gfs2_glock2aspace(gl);
209 if (ip && !S_ISREG(ip->i_inode.i_mode))
212 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
213 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
214 inode_dio_wait(&ip->i_inode);
216 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
219 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
221 gfs2_log_flush(gl->gl_sbd, gl, NORMAL_FLUSH);
222 filemap_fdatawrite(metamapping);
224 struct address_space *mapping = ip->i_inode.i_mapping;
225 filemap_fdatawrite(mapping);
226 error = filemap_fdatawait(mapping);
227 mapping_set_error(mapping, error);
229 error = filemap_fdatawait(metamapping);
230 mapping_set_error(metamapping, error);
231 gfs2_ail_empty_gl(gl);
233 * Writeback of the data mapping may cause the dirty flag to be set
234 * so we have to clear it again here.
236 smp_mb__before_atomic();
237 clear_bit(GLF_DIRTY, &gl->gl_flags);
241 * inode_go_inval - prepare a inode glock to be released
245 * Normally we invalidate everything, but if we are moving into
246 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
247 * can keep hold of the metadata, since it won't have changed.
251 static void inode_go_inval(struct gfs2_glock *gl, int flags)
253 struct gfs2_inode *ip = gl->gl_object;
255 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
257 if (flags & DIO_METADATA) {
258 struct address_space *mapping = gfs2_glock2aspace(gl);
259 truncate_inode_pages(mapping, 0);
261 set_bit(GIF_INVALID, &ip->i_flags);
262 forget_all_cached_acls(&ip->i_inode);
263 gfs2_dir_hash_inval(ip);
267 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
268 gfs2_log_flush(gl->gl_sbd, NULL, NORMAL_FLUSH);
269 gl->gl_sbd->sd_rindex_uptodate = 0;
271 if (ip && S_ISREG(ip->i_inode.i_mode))
272 truncate_inode_pages(ip->i_inode.i_mapping, 0);
276 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
279 * Returns: 1 if it's ok
282 static int inode_go_demote_ok(const struct gfs2_glock *gl)
284 struct gfs2_sbd *sdp = gl->gl_sbd;
285 struct gfs2_holder *gh;
287 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
290 if (!list_empty(&gl->gl_holders)) {
291 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
292 if (gh->gh_list.next != &gl->gl_holders)
300 * gfs2_set_nlink - Set the inode's link count based on on-disk info
301 * @inode: The inode in question
302 * @nlink: The link count
304 * If the link count has hit zero, it must never be raised, whatever the
305 * on-disk inode might say. When new struct inodes are created the link
306 * count is set to 1, so that we can safely use this test even when reading
307 * in on disk information for the first time.
310 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
313 * We will need to review setting the nlink count here in the
314 * light of the forthcoming ro bind mount work. This is a reminder
317 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
321 set_nlink(inode, nlink);
325 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
327 const struct gfs2_dinode *str = buf;
328 struct timespec atime;
331 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
333 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
334 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
335 ip->i_inode.i_rdev = 0;
336 switch (ip->i_inode.i_mode & S_IFMT) {
339 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
340 be32_to_cpu(str->di_minor));
344 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
345 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
346 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
347 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
348 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
349 atime.tv_sec = be64_to_cpu(str->di_atime);
350 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
351 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
352 ip->i_inode.i_atime = atime;
353 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
354 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
355 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
356 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
358 ip->i_goal = be64_to_cpu(str->di_goal_meta);
359 ip->i_generation = be64_to_cpu(str->di_generation);
361 ip->i_diskflags = be32_to_cpu(str->di_flags);
362 ip->i_eattr = be64_to_cpu(str->di_eattr);
363 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
364 gfs2_set_inode_flags(&ip->i_inode);
365 height = be16_to_cpu(str->di_height);
366 if (unlikely(height > GFS2_MAX_META_HEIGHT))
368 ip->i_height = (u8)height;
370 depth = be16_to_cpu(str->di_depth);
371 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
373 ip->i_depth = (u8)depth;
374 ip->i_entries = be32_to_cpu(str->di_entries);
376 if (S_ISREG(ip->i_inode.i_mode))
377 gfs2_set_aops(&ip->i_inode);
381 gfs2_consist_inode(ip);
386 * gfs2_inode_refresh - Refresh the incore copy of the dinode
387 * @ip: The GFS2 inode
392 int gfs2_inode_refresh(struct gfs2_inode *ip)
394 struct buffer_head *dibh;
397 error = gfs2_meta_inode_buffer(ip, &dibh);
401 error = gfs2_dinode_in(ip, dibh->b_data);
403 clear_bit(GIF_INVALID, &ip->i_flags);
409 * inode_go_lock - operation done after an inode lock is locked by a process
416 static int inode_go_lock(struct gfs2_holder *gh)
418 struct gfs2_glock *gl = gh->gh_gl;
419 struct gfs2_sbd *sdp = gl->gl_sbd;
420 struct gfs2_inode *ip = gl->gl_object;
423 if (!ip || (gh->gh_flags & GL_SKIP))
426 if (test_bit(GIF_INVALID, &ip->i_flags)) {
427 error = gfs2_inode_refresh(ip);
432 if (gh->gh_state != LM_ST_DEFERRED)
433 inode_dio_wait(&ip->i_inode);
435 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
436 (gl->gl_state == LM_ST_EXCLUSIVE) &&
437 (gh->gh_state == LM_ST_EXCLUSIVE)) {
438 spin_lock(&sdp->sd_trunc_lock);
439 if (list_empty(&ip->i_trunc_list))
440 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
441 spin_unlock(&sdp->sd_trunc_lock);
442 wake_up(&sdp->sd_quota_wait);
450 * inode_go_dump - print information about an inode
456 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
458 const struct gfs2_inode *ip = gl->gl_object;
461 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
462 (unsigned long long)ip->i_no_formal_ino,
463 (unsigned long long)ip->i_no_addr,
464 IF2DT(ip->i_inode.i_mode), ip->i_flags,
465 (unsigned int)ip->i_diskflags,
466 (unsigned long long)i_size_read(&ip->i_inode));
470 * freeze_go_sync - promote/demote the freeze glock
472 * @state: the requested state
477 static void freeze_go_sync(struct gfs2_glock *gl)
480 struct gfs2_sbd *sdp = gl->gl_sbd;
482 if (gl->gl_state == LM_ST_SHARED &&
483 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
484 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
485 error = freeze_super(sdp->sd_vfs);
487 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
488 gfs2_assert_withdraw(sdp, 0);
490 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
491 gfs2_log_flush(sdp, NULL, FREEZE_FLUSH);
496 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
501 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
503 struct gfs2_sbd *sdp = gl->gl_sbd;
504 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
505 struct gfs2_glock *j_gl = ip->i_gl;
506 struct gfs2_log_header_host head;
509 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
510 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
512 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
515 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
518 /* Initialize some head of the log stuff */
519 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
520 sdp->sd_log_sequence = head.lh_sequence + 1;
521 gfs2_log_pointers_init(sdp, head.lh_blkno);
534 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
540 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
543 * gl_spin lock is held while calling this
545 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
547 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
548 struct gfs2_sbd *sdp = gl->gl_sbd;
550 if (!remote || (sdp->sd_vfs->s_flags & MS_RDONLY))
553 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
554 gl->gl_state == LM_ST_SHARED && ip) {
555 gl->gl_lockref.count++;
556 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
557 gl->gl_lockref.count--;
561 const struct gfs2_glock_operations gfs2_meta_glops = {
562 .go_type = LM_TYPE_META,
565 const struct gfs2_glock_operations gfs2_inode_glops = {
566 .go_sync = inode_go_sync,
567 .go_inval = inode_go_inval,
568 .go_demote_ok = inode_go_demote_ok,
569 .go_lock = inode_go_lock,
570 .go_dump = inode_go_dump,
571 .go_type = LM_TYPE_INODE,
572 .go_flags = GLOF_ASPACE | GLOF_LRU,
575 const struct gfs2_glock_operations gfs2_rgrp_glops = {
576 .go_sync = rgrp_go_sync,
577 .go_inval = rgrp_go_inval,
578 .go_lock = gfs2_rgrp_go_lock,
579 .go_unlock = gfs2_rgrp_go_unlock,
580 .go_dump = gfs2_rgrp_dump,
581 .go_type = LM_TYPE_RGRP,
582 .go_flags = GLOF_LVB,
585 const struct gfs2_glock_operations gfs2_freeze_glops = {
586 .go_sync = freeze_go_sync,
587 .go_xmote_bh = freeze_go_xmote_bh,
588 .go_demote_ok = freeze_go_demote_ok,
589 .go_type = LM_TYPE_NONDISK,
592 const struct gfs2_glock_operations gfs2_iopen_glops = {
593 .go_type = LM_TYPE_IOPEN,
594 .go_callback = iopen_go_callback,
595 .go_flags = GLOF_LRU,
598 const struct gfs2_glock_operations gfs2_flock_glops = {
599 .go_type = LM_TYPE_FLOCK,
600 .go_flags = GLOF_LRU,
603 const struct gfs2_glock_operations gfs2_nondisk_glops = {
604 .go_type = LM_TYPE_NONDISK,
607 const struct gfs2_glock_operations gfs2_quota_glops = {
608 .go_type = LM_TYPE_QUOTA,
609 .go_flags = GLOF_LVB | GLOF_LRU,
612 const struct gfs2_glock_operations gfs2_journal_glops = {
613 .go_type = LM_TYPE_JOURNAL,
616 const struct gfs2_glock_operations *gfs2_glops_list[] = {
617 [LM_TYPE_META] = &gfs2_meta_glops,
618 [LM_TYPE_INODE] = &gfs2_inode_glops,
619 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
620 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
621 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
622 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
623 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
624 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,