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 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
33 fs_err(gl->gl_sbd, "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page state 0x%lx\n",
34 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
35 bh->b_page->mapping, bh->b_page->flags);
36 fs_err(gl->gl_sbd, "AIL glock %u:%llu mapping %p\n",
37 gl->gl_name.ln_type, gl->gl_name.ln_number,
38 gfs2_glock2aspace(gl));
39 gfs2_lm_withdraw(gl->gl_sbd, "AIL error\n");
43 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
45 * @fsync: set when called from fsync (not all buffers will be clean)
47 * None of the buffers should be dirty, locked, or pinned.
50 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
52 struct gfs2_sbd *sdp = gl->gl_sbd;
53 struct list_head *head = &gl->gl_ail_list;
54 struct gfs2_bufdata *bd, *tmp;
55 struct buffer_head *bh;
56 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
60 spin_lock(&sdp->sd_ail_lock);
61 list_for_each_entry_safe(bd, tmp, head, bd_ail_gl_list) {
63 if (bh->b_state & b_state) {
66 gfs2_ail_error(gl, bh);
68 blocknr = bh->b_blocknr;
70 gfs2_remove_from_ail(bd); /* drops ref on bh */
73 bd->bd_blkno = blocknr;
75 gfs2_trans_add_revoke(sdp, bd);
77 BUG_ON(!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_sbd;
88 memset(&tr, 0, sizeof(tr));
89 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
94 /* A shortened, inline version of gfs2_trans_begin() */
95 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
96 tr.tr_ip = (unsigned long)__builtin_return_address(0);
97 INIT_LIST_HEAD(&tr.tr_list_buf);
98 gfs2_log_reserve(sdp, tr.tr_reserved);
99 BUG_ON(current->journal_info);
100 current->journal_info = &tr;
102 __gfs2_ail_flush(gl, 0);
105 gfs2_log_flush(sdp, NULL);
108 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
110 struct gfs2_sbd *sdp = gl->gl_sbd;
111 unsigned int revokes = atomic_read(&gl->gl_ail_count);
117 ret = gfs2_trans_begin(sdp, 0, revokes);
120 __gfs2_ail_flush(gl, fsync);
122 gfs2_log_flush(sdp, NULL);
126 * rgrp_go_sync - sync out the metadata for this glock
129 * Called when demoting or unlocking an EX glock. We must flush
130 * to disk all dirty buffers/pages relating to this glock, and must not
131 * not return to caller to demote/unlock the glock until I/O is complete.
134 static void rgrp_go_sync(struct gfs2_glock *gl)
136 struct address_space *metamapping = gfs2_glock2aspace(gl);
137 struct gfs2_rgrpd *rgd;
140 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
142 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
144 gfs2_log_flush(gl->gl_sbd, gl);
145 filemap_fdatawrite(metamapping);
146 error = filemap_fdatawait(metamapping);
147 mapping_set_error(metamapping, error);
148 gfs2_ail_empty_gl(gl);
150 spin_lock(&gl->gl_spin);
153 gfs2_free_clones(rgd);
154 spin_unlock(&gl->gl_spin);
158 * rgrp_go_inval - invalidate the metadata for this glock
162 * We never used LM_ST_DEFERRED with resource groups, so that we
163 * should always see the metadata flag set here.
167 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
169 struct address_space *mapping = gfs2_glock2aspace(gl);
171 BUG_ON(!(flags & DIO_METADATA));
172 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
173 truncate_inode_pages(mapping, 0);
176 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
177 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
182 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
183 * @gl: the glock protecting the inode
187 static void inode_go_sync(struct gfs2_glock *gl)
189 struct gfs2_inode *ip = gl->gl_object;
190 struct address_space *metamapping = gfs2_glock2aspace(gl);
193 if (ip && !S_ISREG(ip->i_inode.i_mode))
195 if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
196 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
197 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
200 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
202 gfs2_log_flush(gl->gl_sbd, gl);
203 filemap_fdatawrite(metamapping);
205 struct address_space *mapping = ip->i_inode.i_mapping;
206 filemap_fdatawrite(mapping);
207 error = filemap_fdatawait(mapping);
208 mapping_set_error(mapping, error);
210 error = filemap_fdatawait(metamapping);
211 mapping_set_error(metamapping, error);
212 gfs2_ail_empty_gl(gl);
214 * Writeback of the data mapping may cause the dirty flag to be set
215 * so we have to clear it again here.
217 smp_mb__before_clear_bit();
218 clear_bit(GLF_DIRTY, &gl->gl_flags);
222 * inode_go_inval - prepare a inode glock to be released
226 * Normally we invlidate everything, but if we are moving into
227 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
228 * can keep hold of the metadata, since it won't have changed.
232 static void inode_go_inval(struct gfs2_glock *gl, int flags)
234 struct gfs2_inode *ip = gl->gl_object;
236 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
238 if (flags & DIO_METADATA) {
239 struct address_space *mapping = gfs2_glock2aspace(gl);
240 truncate_inode_pages(mapping, 0);
242 set_bit(GIF_INVALID, &ip->i_flags);
243 forget_all_cached_acls(&ip->i_inode);
244 gfs2_dir_hash_inval(ip);
248 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
249 gfs2_log_flush(gl->gl_sbd, NULL);
250 gl->gl_sbd->sd_rindex_uptodate = 0;
252 if (ip && S_ISREG(ip->i_inode.i_mode))
253 truncate_inode_pages(ip->i_inode.i_mapping, 0);
257 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
260 * Returns: 1 if it's ok
263 static int inode_go_demote_ok(const struct gfs2_glock *gl)
265 struct gfs2_sbd *sdp = gl->gl_sbd;
266 struct gfs2_holder *gh;
268 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
271 if (!list_empty(&gl->gl_holders)) {
272 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
273 if (gh->gh_list.next != &gl->gl_holders)
281 * gfs2_set_nlink - Set the inode's link count based on on-disk info
282 * @inode: The inode in question
283 * @nlink: The link count
285 * If the link count has hit zero, it must never be raised, whatever the
286 * on-disk inode might say. When new struct inodes are created the link
287 * count is set to 1, so that we can safely use this test even when reading
288 * in on disk information for the first time.
291 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
294 * We will need to review setting the nlink count here in the
295 * light of the forthcoming ro bind mount work. This is a reminder
298 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
302 set_nlink(inode, nlink);
306 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
308 const struct gfs2_dinode *str = buf;
309 struct timespec atime;
312 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
314 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
315 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
316 ip->i_inode.i_rdev = 0;
317 switch (ip->i_inode.i_mode & S_IFMT) {
320 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
321 be32_to_cpu(str->di_minor));
325 ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
326 ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
327 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
328 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
329 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
330 atime.tv_sec = be64_to_cpu(str->di_atime);
331 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
332 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
333 ip->i_inode.i_atime = atime;
334 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
335 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
336 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
337 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
339 ip->i_goal = be64_to_cpu(str->di_goal_meta);
340 ip->i_generation = be64_to_cpu(str->di_generation);
342 ip->i_diskflags = be32_to_cpu(str->di_flags);
343 ip->i_eattr = be64_to_cpu(str->di_eattr);
344 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
345 gfs2_set_inode_flags(&ip->i_inode);
346 height = be16_to_cpu(str->di_height);
347 if (unlikely(height > GFS2_MAX_META_HEIGHT))
349 ip->i_height = (u8)height;
351 depth = be16_to_cpu(str->di_depth);
352 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
354 ip->i_depth = (u8)depth;
355 ip->i_entries = be32_to_cpu(str->di_entries);
357 if (S_ISREG(ip->i_inode.i_mode))
358 gfs2_set_aops(&ip->i_inode);
362 gfs2_consist_inode(ip);
367 * gfs2_inode_refresh - Refresh the incore copy of the dinode
368 * @ip: The GFS2 inode
373 int gfs2_inode_refresh(struct gfs2_inode *ip)
375 struct buffer_head *dibh;
378 error = gfs2_meta_inode_buffer(ip, &dibh);
382 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
387 error = gfs2_dinode_in(ip, dibh->b_data);
389 clear_bit(GIF_INVALID, &ip->i_flags);
395 * inode_go_lock - operation done after an inode lock is locked by a process
402 static int inode_go_lock(struct gfs2_holder *gh)
404 struct gfs2_glock *gl = gh->gh_gl;
405 struct gfs2_sbd *sdp = gl->gl_sbd;
406 struct gfs2_inode *ip = gl->gl_object;
409 if (!ip || (gh->gh_flags & GL_SKIP))
412 if (test_bit(GIF_INVALID, &ip->i_flags)) {
413 error = gfs2_inode_refresh(ip);
418 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
419 (gl->gl_state == LM_ST_EXCLUSIVE) &&
420 (gh->gh_state == LM_ST_EXCLUSIVE)) {
421 spin_lock(&sdp->sd_trunc_lock);
422 if (list_empty(&ip->i_trunc_list))
423 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
424 spin_unlock(&sdp->sd_trunc_lock);
425 wake_up(&sdp->sd_quota_wait);
433 * inode_go_dump - print information about an inode
437 * Returns: 0 on success, -ENOBUFS when we run out of space
440 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
442 const struct gfs2_inode *ip = gl->gl_object;
445 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
446 (unsigned long long)ip->i_no_formal_ino,
447 (unsigned long long)ip->i_no_addr,
448 IF2DT(ip->i_inode.i_mode), ip->i_flags,
449 (unsigned int)ip->i_diskflags,
450 (unsigned long long)i_size_read(&ip->i_inode));
455 * trans_go_sync - promote/demote the transaction glock
457 * @state: the requested state
462 static void trans_go_sync(struct gfs2_glock *gl)
464 struct gfs2_sbd *sdp = gl->gl_sbd;
466 if (gl->gl_state != LM_ST_UNLOCKED &&
467 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
468 gfs2_meta_syncfs(sdp);
469 gfs2_log_shutdown(sdp);
474 * trans_go_xmote_bh - After promoting/demoting the transaction glock
479 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
481 struct gfs2_sbd *sdp = gl->gl_sbd;
482 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
483 struct gfs2_glock *j_gl = ip->i_gl;
484 struct gfs2_log_header_host head;
487 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
488 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
490 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
493 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
496 /* Initialize some head of the log stuff */
497 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
498 sdp->sd_log_sequence = head.lh_sequence + 1;
499 gfs2_log_pointers_init(sdp, head.lh_blkno);
512 static int trans_go_demote_ok(const struct gfs2_glock *gl)
518 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
521 * gl_spin lock is held while calling this
523 static void iopen_go_callback(struct gfs2_glock *gl)
525 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
526 struct gfs2_sbd *sdp = gl->gl_sbd;
528 if (sdp->sd_vfs->s_flags & MS_RDONLY)
531 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
532 gl->gl_state == LM_ST_SHARED && ip) {
534 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
535 gfs2_glock_put_nolock(gl);
539 const struct gfs2_glock_operations gfs2_meta_glops = {
540 .go_type = LM_TYPE_META,
543 const struct gfs2_glock_operations gfs2_inode_glops = {
544 .go_xmote_th = inode_go_sync,
545 .go_inval = inode_go_inval,
546 .go_demote_ok = inode_go_demote_ok,
547 .go_lock = inode_go_lock,
548 .go_dump = inode_go_dump,
549 .go_type = LM_TYPE_INODE,
550 .go_flags = GLOF_ASPACE,
553 const struct gfs2_glock_operations gfs2_rgrp_glops = {
554 .go_xmote_th = rgrp_go_sync,
555 .go_inval = rgrp_go_inval,
556 .go_lock = gfs2_rgrp_go_lock,
557 .go_unlock = gfs2_rgrp_go_unlock,
558 .go_dump = gfs2_rgrp_dump,
559 .go_type = LM_TYPE_RGRP,
560 .go_flags = GLOF_ASPACE,
563 const struct gfs2_glock_operations gfs2_trans_glops = {
564 .go_xmote_th = trans_go_sync,
565 .go_xmote_bh = trans_go_xmote_bh,
566 .go_demote_ok = trans_go_demote_ok,
567 .go_type = LM_TYPE_NONDISK,
570 const struct gfs2_glock_operations gfs2_iopen_glops = {
571 .go_type = LM_TYPE_IOPEN,
572 .go_callback = iopen_go_callback,
575 const struct gfs2_glock_operations gfs2_flock_glops = {
576 .go_type = LM_TYPE_FLOCK,
579 const struct gfs2_glock_operations gfs2_nondisk_glops = {
580 .go_type = LM_TYPE_NONDISK,
583 const struct gfs2_glock_operations gfs2_quota_glops = {
584 .go_type = LM_TYPE_QUOTA,
587 const struct gfs2_glock_operations gfs2_journal_glops = {
588 .go_type = LM_TYPE_JOURNAL,
591 const struct gfs2_glock_operations *gfs2_glops_list[] = {
592 [LM_TYPE_META] = &gfs2_meta_glops,
593 [LM_TYPE_INODE] = &gfs2_inode_glops,
594 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
595 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
596 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
597 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
598 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
599 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,