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>
32 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
35 * None of the buffers should be dirty, locked, or pinned.
38 static void __gfs2_ail_flush(struct gfs2_glock *gl)
40 struct gfs2_sbd *sdp = gl->gl_sbd;
41 struct list_head *head = &gl->gl_ail_list;
42 struct gfs2_bufdata *bd;
43 struct buffer_head *bh;
45 spin_lock(&sdp->sd_ail_lock);
46 while (!list_empty(head)) {
47 bd = list_entry(head->next, struct gfs2_bufdata,
50 gfs2_remove_from_ail(bd);
53 spin_unlock(&sdp->sd_ail_lock);
55 bd->bd_blkno = bh->b_blocknr;
57 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
58 gfs2_trans_add_revoke(sdp, bd);
61 spin_lock(&sdp->sd_ail_lock);
63 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
64 spin_unlock(&sdp->sd_ail_lock);
68 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
70 struct gfs2_sbd *sdp = gl->gl_sbd;
73 memset(&tr, 0, sizeof(tr));
74 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
79 /* A shortened, inline version of gfs2_trans_begin() */
80 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
81 tr.tr_ip = (unsigned long)__builtin_return_address(0);
82 INIT_LIST_HEAD(&tr.tr_list_buf);
83 gfs2_log_reserve(sdp, tr.tr_reserved);
84 BUG_ON(current->journal_info);
85 current->journal_info = &tr;
90 gfs2_log_flush(sdp, NULL);
93 void gfs2_ail_flush(struct gfs2_glock *gl)
95 struct gfs2_sbd *sdp = gl->gl_sbd;
96 unsigned int revokes = atomic_read(&gl->gl_ail_count);
102 ret = gfs2_trans_begin(sdp, 0, revokes);
105 __gfs2_ail_flush(gl);
107 gfs2_log_flush(sdp, NULL);
111 * rgrp_go_sync - sync out the metadata for this glock
114 * Called when demoting or unlocking an EX glock. We must flush
115 * to disk all dirty buffers/pages relating to this glock, and must not
116 * not return to caller to demote/unlock the glock until I/O is complete.
119 static void rgrp_go_sync(struct gfs2_glock *gl)
121 struct address_space *metamapping = gfs2_glock2aspace(gl);
124 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
126 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
128 gfs2_log_flush(gl->gl_sbd, gl);
129 filemap_fdatawrite(metamapping);
130 error = filemap_fdatawait(metamapping);
131 mapping_set_error(metamapping, error);
132 gfs2_ail_empty_gl(gl);
136 * rgrp_go_inval - invalidate the metadata for this glock
140 * We never used LM_ST_DEFERRED with resource groups, so that we
141 * should always see the metadata flag set here.
145 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
147 struct address_space *mapping = gfs2_glock2aspace(gl);
149 BUG_ON(!(flags & DIO_METADATA));
150 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
151 truncate_inode_pages(mapping, 0);
154 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
155 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
160 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
161 * @gl: the glock protecting the inode
165 static void inode_go_sync(struct gfs2_glock *gl)
167 struct gfs2_inode *ip = gl->gl_object;
168 struct address_space *metamapping = gfs2_glock2aspace(gl);
171 if (ip && !S_ISREG(ip->i_inode.i_mode))
173 if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
174 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
175 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
178 BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
180 gfs2_log_flush(gl->gl_sbd, gl);
181 filemap_fdatawrite(metamapping);
183 struct address_space *mapping = ip->i_inode.i_mapping;
184 filemap_fdatawrite(mapping);
185 error = filemap_fdatawait(mapping);
186 mapping_set_error(mapping, error);
188 error = filemap_fdatawait(metamapping);
189 mapping_set_error(metamapping, error);
190 gfs2_ail_empty_gl(gl);
192 * Writeback of the data mapping may cause the dirty flag to be set
193 * so we have to clear it again here.
195 smp_mb__before_clear_bit();
196 clear_bit(GLF_DIRTY, &gl->gl_flags);
200 * inode_go_inval - prepare a inode glock to be released
204 * Normally we invlidate everything, but if we are moving into
205 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
206 * can keep hold of the metadata, since it won't have changed.
210 static void inode_go_inval(struct gfs2_glock *gl, int flags)
212 struct gfs2_inode *ip = gl->gl_object;
214 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
216 if (flags & DIO_METADATA) {
217 struct address_space *mapping = gfs2_glock2aspace(gl);
218 truncate_inode_pages(mapping, 0);
220 set_bit(GIF_INVALID, &ip->i_flags);
221 forget_all_cached_acls(&ip->i_inode);
222 gfs2_dir_hash_inval(ip);
226 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
227 gfs2_log_flush(gl->gl_sbd, NULL);
228 gl->gl_sbd->sd_rindex_uptodate = 0;
230 if (ip && S_ISREG(ip->i_inode.i_mode))
231 truncate_inode_pages(ip->i_inode.i_mapping, 0);
235 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
238 * Returns: 1 if it's ok
241 static int inode_go_demote_ok(const struct gfs2_glock *gl)
243 struct gfs2_sbd *sdp = gl->gl_sbd;
244 struct gfs2_holder *gh;
246 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
249 if (!list_empty(&gl->gl_holders)) {
250 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
251 if (gh->gh_list.next != &gl->gl_holders)
259 * gfs2_set_nlink - Set the inode's link count based on on-disk info
260 * @inode: The inode in question
261 * @nlink: The link count
263 * If the link count has hit zero, it must never be raised, whatever the
264 * on-disk inode might say. When new struct inodes are created the link
265 * count is set to 1, so that we can safely use this test even when reading
266 * in on disk information for the first time.
269 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
272 * We will need to review setting the nlink count here in the
273 * light of the forthcoming ro bind mount work. This is a reminder
276 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
280 inode->i_nlink = nlink;
284 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
286 const struct gfs2_dinode *str = buf;
287 struct timespec atime;
290 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
292 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
293 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
294 ip->i_inode.i_rdev = 0;
295 switch (ip->i_inode.i_mode & S_IFMT) {
298 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
299 be32_to_cpu(str->di_minor));
303 ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
304 ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
305 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
306 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
307 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
308 atime.tv_sec = be64_to_cpu(str->di_atime);
309 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
310 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
311 ip->i_inode.i_atime = atime;
312 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
313 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
314 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
315 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
317 ip->i_goal = be64_to_cpu(str->di_goal_meta);
318 ip->i_generation = be64_to_cpu(str->di_generation);
320 ip->i_diskflags = be32_to_cpu(str->di_flags);
321 ip->i_eattr = be64_to_cpu(str->di_eattr);
322 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
323 gfs2_set_inode_flags(&ip->i_inode);
324 height = be16_to_cpu(str->di_height);
325 if (unlikely(height > GFS2_MAX_META_HEIGHT))
327 ip->i_height = (u8)height;
329 depth = be16_to_cpu(str->di_depth);
330 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
332 ip->i_depth = (u8)depth;
333 ip->i_entries = be32_to_cpu(str->di_entries);
335 if (S_ISREG(ip->i_inode.i_mode))
336 gfs2_set_aops(&ip->i_inode);
340 gfs2_consist_inode(ip);
345 * gfs2_inode_refresh - Refresh the incore copy of the dinode
346 * @ip: The GFS2 inode
351 int gfs2_inode_refresh(struct gfs2_inode *ip)
353 struct buffer_head *dibh;
356 error = gfs2_meta_inode_buffer(ip, &dibh);
360 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
365 error = gfs2_dinode_in(ip, dibh->b_data);
367 clear_bit(GIF_INVALID, &ip->i_flags);
373 * inode_go_lock - operation done after an inode lock is locked by a process
380 static int inode_go_lock(struct gfs2_holder *gh)
382 struct gfs2_glock *gl = gh->gh_gl;
383 struct gfs2_sbd *sdp = gl->gl_sbd;
384 struct gfs2_inode *ip = gl->gl_object;
387 if (!ip || (gh->gh_flags & GL_SKIP))
390 if (test_bit(GIF_INVALID, &ip->i_flags)) {
391 error = gfs2_inode_refresh(ip);
396 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
397 (gl->gl_state == LM_ST_EXCLUSIVE) &&
398 (gh->gh_state == LM_ST_EXCLUSIVE)) {
399 spin_lock(&sdp->sd_trunc_lock);
400 if (list_empty(&ip->i_trunc_list))
401 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
402 spin_unlock(&sdp->sd_trunc_lock);
403 wake_up(&sdp->sd_quota_wait);
411 * inode_go_dump - print information about an inode
415 * Returns: 0 on success, -ENOBUFS when we run out of space
418 static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
420 const struct gfs2_inode *ip = gl->gl_object;
423 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
424 (unsigned long long)ip->i_no_formal_ino,
425 (unsigned long long)ip->i_no_addr,
426 IF2DT(ip->i_inode.i_mode), ip->i_flags,
427 (unsigned int)ip->i_diskflags,
428 (unsigned long long)i_size_read(&ip->i_inode));
433 * rgrp_go_lock - operation done after an rgrp lock is locked by
434 * a first holder on this node.
441 static int rgrp_go_lock(struct gfs2_holder *gh)
443 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
447 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
448 * a last holder on this node.
454 static void rgrp_go_unlock(struct gfs2_holder *gh)
456 gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
460 * trans_go_sync - promote/demote the transaction glock
462 * @state: the requested state
467 static void trans_go_sync(struct gfs2_glock *gl)
469 struct gfs2_sbd *sdp = gl->gl_sbd;
471 if (gl->gl_state != LM_ST_UNLOCKED &&
472 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
473 gfs2_meta_syncfs(sdp);
474 gfs2_log_shutdown(sdp);
479 * trans_go_xmote_bh - After promoting/demoting the transaction glock
484 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
486 struct gfs2_sbd *sdp = gl->gl_sbd;
487 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
488 struct gfs2_glock *j_gl = ip->i_gl;
489 struct gfs2_log_header_host head;
492 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
493 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
495 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
498 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
501 /* Initialize some head of the log stuff */
502 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
503 sdp->sd_log_sequence = head.lh_sequence + 1;
504 gfs2_log_pointers_init(sdp, head.lh_blkno);
517 static int trans_go_demote_ok(const struct gfs2_glock *gl)
523 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
526 * gl_spin lock is held while calling this
528 static void iopen_go_callback(struct gfs2_glock *gl)
530 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
531 struct gfs2_sbd *sdp = gl->gl_sbd;
533 if (sdp->sd_vfs->s_flags & MS_RDONLY)
536 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
537 gl->gl_state == LM_ST_SHARED && ip) {
539 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
540 gfs2_glock_put_nolock(gl);
544 const struct gfs2_glock_operations gfs2_meta_glops = {
545 .go_type = LM_TYPE_META,
548 const struct gfs2_glock_operations gfs2_inode_glops = {
549 .go_xmote_th = inode_go_sync,
550 .go_inval = inode_go_inval,
551 .go_demote_ok = inode_go_demote_ok,
552 .go_lock = inode_go_lock,
553 .go_dump = inode_go_dump,
554 .go_type = LM_TYPE_INODE,
555 .go_flags = GLOF_ASPACE,
558 const struct gfs2_glock_operations gfs2_rgrp_glops = {
559 .go_xmote_th = rgrp_go_sync,
560 .go_inval = rgrp_go_inval,
561 .go_lock = rgrp_go_lock,
562 .go_unlock = rgrp_go_unlock,
563 .go_dump = gfs2_rgrp_dump,
564 .go_type = LM_TYPE_RGRP,
565 .go_flags = GLOF_ASPACE,
568 const struct gfs2_glock_operations gfs2_trans_glops = {
569 .go_xmote_th = trans_go_sync,
570 .go_xmote_bh = trans_go_xmote_bh,
571 .go_demote_ok = trans_go_demote_ok,
572 .go_type = LM_TYPE_NONDISK,
575 const struct gfs2_glock_operations gfs2_iopen_glops = {
576 .go_type = LM_TYPE_IOPEN,
577 .go_callback = iopen_go_callback,
580 const struct gfs2_glock_operations gfs2_flock_glops = {
581 .go_type = LM_TYPE_FLOCK,
584 const struct gfs2_glock_operations gfs2_nondisk_glops = {
585 .go_type = LM_TYPE_NONDISK,
588 const struct gfs2_glock_operations gfs2_quota_glops = {
589 .go_type = LM_TYPE_QUOTA,
592 const struct gfs2_glock_operations gfs2_journal_glops = {
593 .go_type = LM_TYPE_JOURNAL,
596 const struct gfs2_glock_operations *gfs2_glops_list[] = {
597 [LM_TYPE_META] = &gfs2_meta_glops,
598 [LM_TYPE_INODE] = &gfs2_inode_glops,
599 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
600 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
601 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
602 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
603 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
604 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,