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 INIT_LIST_HEAD(&tr.tr_ail1_list);
93 INIT_LIST_HEAD(&tr.tr_ail2_list);
94 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
99 /* A shortened, inline version of gfs2_trans_begin()
100 * tr->alloced is not set since the transaction structure is
102 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
104 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
106 WARN_ON_ONCE(current->journal_info);
107 current->journal_info = &tr;
109 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
112 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
113 GFS2_LFC_AIL_EMPTY_GL);
116 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
118 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
119 unsigned int revokes = atomic_read(&gl->gl_ail_count);
120 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
126 while (revokes > max_revokes)
127 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
129 ret = gfs2_trans_begin(sdp, 0, max_revokes);
132 __gfs2_ail_flush(gl, fsync, max_revokes);
134 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
139 * rgrp_go_sync - sync out the metadata for this glock
142 * Called when demoting or unlocking an EX glock. We must flush
143 * to disk all dirty buffers/pages relating to this glock, and must not
144 * return to caller to demote/unlock the glock until I/O is complete.
147 static void rgrp_go_sync(struct gfs2_glock *gl)
149 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
150 struct address_space *mapping = &sdp->sd_aspace;
151 struct gfs2_rgrpd *rgd;
154 spin_lock(&gl->gl_lockref.lock);
157 gfs2_rgrp_brelse(rgd);
158 spin_unlock(&gl->gl_lockref.lock);
160 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
162 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
164 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
165 GFS2_LFC_RGRP_GO_SYNC);
166 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
167 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
168 mapping_set_error(mapping, error);
169 gfs2_ail_empty_gl(gl);
171 spin_lock(&gl->gl_lockref.lock);
174 gfs2_free_clones(rgd);
175 spin_unlock(&gl->gl_lockref.lock);
179 * rgrp_go_inval - invalidate the metadata for this glock
183 * We never used LM_ST_DEFERRED with resource groups, so that we
184 * should always see the metadata flag set here.
188 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
190 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
191 struct address_space *mapping = &sdp->sd_aspace;
192 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
195 gfs2_rgrp_brelse(rgd);
197 WARN_ON_ONCE(!(flags & DIO_METADATA));
198 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
199 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
202 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
205 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
207 struct gfs2_inode *ip;
209 spin_lock(&gl->gl_lockref.lock);
212 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
213 spin_unlock(&gl->gl_lockref.lock);
217 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
219 struct gfs2_rgrpd *rgd;
221 spin_lock(&gl->gl_lockref.lock);
223 spin_unlock(&gl->gl_lockref.lock);
228 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
233 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
234 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
238 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
239 * @gl: the glock protecting the inode
243 static void inode_go_sync(struct gfs2_glock *gl)
245 struct gfs2_inode *ip = gfs2_glock2inode(gl);
246 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
247 struct address_space *metamapping = gfs2_glock2aspace(gl);
251 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
252 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
253 inode_dio_wait(&ip->i_inode);
255 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
258 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
260 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
261 GFS2_LFC_INODE_GO_SYNC);
262 filemap_fdatawrite(metamapping);
264 struct address_space *mapping = ip->i_inode.i_mapping;
265 filemap_fdatawrite(mapping);
266 error = filemap_fdatawait(mapping);
267 mapping_set_error(mapping, error);
269 error = filemap_fdatawait(metamapping);
270 mapping_set_error(metamapping, error);
271 gfs2_ail_empty_gl(gl);
273 * Writeback of the data mapping may cause the dirty flag to be set
274 * so we have to clear it again here.
276 smp_mb__before_atomic();
277 clear_bit(GLF_DIRTY, &gl->gl_flags);
280 gfs2_clear_glop_pending(ip);
284 * inode_go_inval - prepare a inode glock to be released
288 * Normally we invalidate everything, but if we are moving into
289 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
290 * can keep hold of the metadata, since it won't have changed.
294 static void inode_go_inval(struct gfs2_glock *gl, int flags)
296 struct gfs2_inode *ip = gfs2_glock2inode(gl);
298 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
300 if (flags & DIO_METADATA) {
301 struct address_space *mapping = gfs2_glock2aspace(gl);
302 truncate_inode_pages(mapping, 0);
304 set_bit(GIF_INVALID, &ip->i_flags);
305 forget_all_cached_acls(&ip->i_inode);
306 security_inode_invalidate_secctx(&ip->i_inode);
307 gfs2_dir_hash_inval(ip);
311 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
312 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
313 GFS2_LOG_HEAD_FLUSH_NORMAL |
314 GFS2_LFC_INODE_GO_INVAL);
315 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
317 if (ip && S_ISREG(ip->i_inode.i_mode))
318 truncate_inode_pages(ip->i_inode.i_mapping, 0);
320 gfs2_clear_glop_pending(ip);
324 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
327 * Returns: 1 if it's ok
330 static int inode_go_demote_ok(const struct gfs2_glock *gl)
332 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
334 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
340 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
342 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
343 const struct gfs2_dinode *str = buf;
344 struct timespec64 atime;
347 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
349 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
350 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
351 ip->i_inode.i_rdev = 0;
352 switch (ip->i_inode.i_mode & S_IFMT) {
355 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
356 be32_to_cpu(str->di_minor));
360 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
361 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
362 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
363 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
364 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
365 atime.tv_sec = be64_to_cpu(str->di_atime);
366 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
367 if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
368 ip->i_inode.i_atime = atime;
369 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
370 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
371 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
372 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
374 ip->i_goal = be64_to_cpu(str->di_goal_meta);
375 ip->i_generation = be64_to_cpu(str->di_generation);
377 ip->i_diskflags = be32_to_cpu(str->di_flags);
378 ip->i_eattr = be64_to_cpu(str->di_eattr);
379 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
380 gfs2_set_inode_flags(&ip->i_inode);
381 height = be16_to_cpu(str->di_height);
382 if (unlikely(height > sdp->sd_max_height))
384 ip->i_height = (u8)height;
386 depth = be16_to_cpu(str->di_depth);
387 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
389 ip->i_depth = (u8)depth;
390 ip->i_entries = be32_to_cpu(str->di_entries);
392 if (gfs2_is_stuffed(ip) && ip->i_inode.i_size > gfs2_max_stuffed_size(ip))
395 if (S_ISREG(ip->i_inode.i_mode))
396 gfs2_set_aops(&ip->i_inode);
400 gfs2_consist_inode(ip);
405 * gfs2_inode_refresh - Refresh the incore copy of the dinode
406 * @ip: The GFS2 inode
411 int gfs2_inode_refresh(struct gfs2_inode *ip)
413 struct buffer_head *dibh;
416 error = gfs2_meta_inode_buffer(ip, &dibh);
420 error = gfs2_dinode_in(ip, dibh->b_data);
422 clear_bit(GIF_INVALID, &ip->i_flags);
428 * inode_go_lock - operation done after an inode lock is locked by a process
435 static int inode_go_lock(struct gfs2_holder *gh)
437 struct gfs2_glock *gl = gh->gh_gl;
438 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
439 struct gfs2_inode *ip = gl->gl_object;
442 if (!ip || (gh->gh_flags & GL_SKIP))
445 if (test_bit(GIF_INVALID, &ip->i_flags)) {
446 error = gfs2_inode_refresh(ip);
451 if (gh->gh_state != LM_ST_DEFERRED)
452 inode_dio_wait(&ip->i_inode);
454 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
455 (gl->gl_state == LM_ST_EXCLUSIVE) &&
456 (gh->gh_state == LM_ST_EXCLUSIVE)) {
457 spin_lock(&sdp->sd_trunc_lock);
458 if (list_empty(&ip->i_trunc_list))
459 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
460 spin_unlock(&sdp->sd_trunc_lock);
461 wake_up(&sdp->sd_quota_wait);
469 * inode_go_dump - print information about an inode
475 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
477 const struct gfs2_inode *ip = gl->gl_object;
480 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
481 (unsigned long long)ip->i_no_formal_ino,
482 (unsigned long long)ip->i_no_addr,
483 IF2DT(ip->i_inode.i_mode), ip->i_flags,
484 (unsigned int)ip->i_diskflags,
485 (unsigned long long)i_size_read(&ip->i_inode));
489 * freeze_go_sync - promote/demote the freeze glock
491 * @state: the requested state
496 static void freeze_go_sync(struct gfs2_glock *gl)
499 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
501 if (gl->gl_state == LM_ST_SHARED &&
502 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
503 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
504 error = freeze_super(sdp->sd_vfs);
506 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
507 gfs2_assert_withdraw(sdp, 0);
509 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
510 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
511 GFS2_LFC_FREEZE_GO_SYNC);
516 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
521 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
523 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
524 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
525 struct gfs2_glock *j_gl = ip->i_gl;
526 struct gfs2_log_header_host head;
529 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
530 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
532 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
535 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
538 /* Initialize some head of the log stuff */
539 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
540 sdp->sd_log_sequence = head.lh_sequence + 1;
541 gfs2_log_pointers_init(sdp, head.lh_blkno);
554 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
560 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
563 * gl_lockref.lock lock is held while calling this
565 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
567 struct gfs2_inode *ip = gl->gl_object;
568 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
570 if (!remote || sb_rdonly(sdp->sd_vfs))
573 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
574 gl->gl_state == LM_ST_SHARED && ip) {
575 gl->gl_lockref.count++;
576 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
577 gl->gl_lockref.count--;
581 const struct gfs2_glock_operations gfs2_meta_glops = {
582 .go_type = LM_TYPE_META,
585 const struct gfs2_glock_operations gfs2_inode_glops = {
586 .go_sync = inode_go_sync,
587 .go_inval = inode_go_inval,
588 .go_demote_ok = inode_go_demote_ok,
589 .go_lock = inode_go_lock,
590 .go_dump = inode_go_dump,
591 .go_type = LM_TYPE_INODE,
592 .go_flags = GLOF_ASPACE | GLOF_LRU,
595 const struct gfs2_glock_operations gfs2_rgrp_glops = {
596 .go_sync = rgrp_go_sync,
597 .go_inval = rgrp_go_inval,
598 .go_lock = gfs2_rgrp_go_lock,
599 .go_unlock = gfs2_rgrp_go_unlock,
600 .go_dump = gfs2_rgrp_dump,
601 .go_type = LM_TYPE_RGRP,
602 .go_flags = GLOF_LVB,
605 const struct gfs2_glock_operations gfs2_freeze_glops = {
606 .go_sync = freeze_go_sync,
607 .go_xmote_bh = freeze_go_xmote_bh,
608 .go_demote_ok = freeze_go_demote_ok,
609 .go_type = LM_TYPE_NONDISK,
612 const struct gfs2_glock_operations gfs2_iopen_glops = {
613 .go_type = LM_TYPE_IOPEN,
614 .go_callback = iopen_go_callback,
615 .go_flags = GLOF_LRU,
618 const struct gfs2_glock_operations gfs2_flock_glops = {
619 .go_type = LM_TYPE_FLOCK,
620 .go_flags = GLOF_LRU,
623 const struct gfs2_glock_operations gfs2_nondisk_glops = {
624 .go_type = LM_TYPE_NONDISK,
627 const struct gfs2_glock_operations gfs2_quota_glops = {
628 .go_type = LM_TYPE_QUOTA,
629 .go_flags = GLOF_LVB | GLOF_LRU,
632 const struct gfs2_glock_operations gfs2_journal_glops = {
633 .go_type = LM_TYPE_JOURNAL,
636 const struct gfs2_glock_operations *gfs2_glops_list[] = {
637 [LM_TYPE_META] = &gfs2_meta_glops,
638 [LM_TYPE_INODE] = &gfs2_inode_glops,
639 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
640 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
641 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
642 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
643 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
644 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,