[XFS] Fix race in xfs_write() b/w dmapi callout and direct I/O checks.
[linux-2.6/verdex.git] / fs / gfs2 / glops.c
blob39c8ae23bd9c04a8b47b0cec2a371a597b5a2450
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 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.
8 */
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/lm_interface.h>
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "inode.h"
23 #include "log.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "rgrp.h"
27 #include "util.h"
28 #include "trans.h"
30 /**
31 * ail_empty_gl - remove all buffers for a given lock from the AIL
32 * @gl: the glock
34 * None of the buffers should be dirty, locked, or pinned.
37 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
39 struct gfs2_sbd *sdp = gl->gl_sbd;
40 unsigned int blocks;
41 struct list_head *head = &gl->gl_ail_list;
42 struct gfs2_bufdata *bd;
43 struct buffer_head *bh;
44 u64 blkno;
45 int error;
47 blocks = atomic_read(&gl->gl_ail_count);
48 if (!blocks)
49 return;
51 error = gfs2_trans_begin(sdp, 0, blocks);
52 if (gfs2_assert_withdraw(sdp, !error))
53 return;
55 gfs2_log_lock(sdp);
56 while (!list_empty(head)) {
57 bd = list_entry(head->next, struct gfs2_bufdata,
58 bd_ail_gl_list);
59 bh = bd->bd_bh;
60 blkno = bh->b_blocknr;
61 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
63 bd->bd_ail = NULL;
64 list_del(&bd->bd_ail_st_list);
65 list_del(&bd->bd_ail_gl_list);
66 atomic_dec(&gl->gl_ail_count);
67 brelse(bh);
68 gfs2_log_unlock(sdp);
70 gfs2_trans_add_revoke(sdp, blkno);
72 gfs2_log_lock(sdp);
74 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
75 gfs2_log_unlock(sdp);
77 gfs2_trans_end(sdp);
78 gfs2_log_flush(sdp, NULL);
81 /**
82 * gfs2_pte_inval - Sync and invalidate all PTEs associated with a glock
83 * @gl: the glock
87 static void gfs2_pte_inval(struct gfs2_glock *gl)
89 struct gfs2_inode *ip;
90 struct inode *inode;
92 ip = gl->gl_object;
93 inode = &ip->i_inode;
94 if (!ip || !S_ISREG(inode->i_mode))
95 return;
97 if (!test_bit(GIF_PAGED, &ip->i_flags))
98 return;
100 unmap_shared_mapping_range(inode->i_mapping, 0, 0);
102 if (test_bit(GIF_SW_PAGED, &ip->i_flags))
103 set_bit(GLF_DIRTY, &gl->gl_flags);
105 clear_bit(GIF_SW_PAGED, &ip->i_flags);
109 * meta_go_sync - sync out the metadata for this glock
110 * @gl: the glock
112 * Called when demoting or unlocking an EX glock. We must flush
113 * to disk all dirty buffers/pages relating to this glock, and must not
114 * not return to caller to demote/unlock the glock until I/O is complete.
117 static void meta_go_sync(struct gfs2_glock *gl)
119 if (gl->gl_state != LM_ST_EXCLUSIVE)
120 return;
122 if (test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) {
123 gfs2_log_flush(gl->gl_sbd, gl);
124 gfs2_meta_sync(gl);
125 gfs2_ail_empty_gl(gl);
130 * meta_go_inval - invalidate the metadata for this glock
131 * @gl: the glock
132 * @flags:
136 static void meta_go_inval(struct gfs2_glock *gl, int flags)
138 if (!(flags & DIO_METADATA))
139 return;
141 gfs2_meta_inval(gl);
142 gl->gl_vn++;
146 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
147 * @gl: the glock protecting the inode
151 static void inode_go_sync(struct gfs2_glock *gl)
153 struct gfs2_inode *ip = gl->gl_object;
155 if (ip && !S_ISREG(ip->i_inode.i_mode))
156 ip = NULL;
158 if (test_bit(GLF_DIRTY, &gl->gl_flags)) {
159 gfs2_log_flush(gl->gl_sbd, gl);
160 if (ip)
161 filemap_fdatawrite(ip->i_inode.i_mapping);
162 gfs2_meta_sync(gl);
163 if (ip) {
164 struct address_space *mapping = ip->i_inode.i_mapping;
165 int error = filemap_fdatawait(mapping);
166 if (error == -ENOSPC)
167 set_bit(AS_ENOSPC, &mapping->flags);
168 else if (error)
169 set_bit(AS_EIO, &mapping->flags);
171 clear_bit(GLF_DIRTY, &gl->gl_flags);
172 gfs2_ail_empty_gl(gl);
177 * inode_go_xmote_th - promote/demote a glock
178 * @gl: the glock
179 * @state: the requested state
180 * @flags:
184 static void inode_go_xmote_th(struct gfs2_glock *gl)
186 if (gl->gl_state != LM_ST_UNLOCKED)
187 gfs2_pte_inval(gl);
188 if (gl->gl_state == LM_ST_EXCLUSIVE)
189 inode_go_sync(gl);
193 * inode_go_xmote_bh - After promoting/demoting a glock
194 * @gl: the glock
198 static void inode_go_xmote_bh(struct gfs2_glock *gl)
200 struct gfs2_holder *gh = gl->gl_req_gh;
201 struct buffer_head *bh;
202 int error;
204 if (gl->gl_state != LM_ST_UNLOCKED &&
205 (!gh || !(gh->gh_flags & GL_SKIP))) {
206 error = gfs2_meta_read(gl, gl->gl_name.ln_number, 0, &bh);
207 if (!error)
208 brelse(bh);
213 * inode_go_drop_th - unlock a glock
214 * @gl: the glock
216 * Invoked from rq_demote().
217 * Another node needs the lock in EXCLUSIVE mode, or lock (unused for too long)
218 * is being purged from our node's glock cache; we're dropping lock.
221 static void inode_go_drop_th(struct gfs2_glock *gl)
223 gfs2_pte_inval(gl);
224 if (gl->gl_state == LM_ST_EXCLUSIVE)
225 inode_go_sync(gl);
229 * inode_go_inval - prepare a inode glock to be released
230 * @gl: the glock
231 * @flags:
235 static void inode_go_inval(struct gfs2_glock *gl, int flags)
237 struct gfs2_inode *ip = gl->gl_object;
238 int meta = (flags & DIO_METADATA);
240 if (meta) {
241 gfs2_meta_inval(gl);
242 if (ip)
243 set_bit(GIF_INVALID, &ip->i_flags);
246 if (ip && S_ISREG(ip->i_inode.i_mode)) {
247 truncate_inode_pages(ip->i_inode.i_mapping, 0);
248 clear_bit(GIF_PAGED, &ip->i_flags);
253 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
254 * @gl: the glock
256 * Returns: 1 if it's ok
259 static int inode_go_demote_ok(struct gfs2_glock *gl)
261 struct gfs2_sbd *sdp = gl->gl_sbd;
262 int demote = 0;
264 if (!gl->gl_object && !gl->gl_aspace->i_mapping->nrpages)
265 demote = 1;
266 else if (!sdp->sd_args.ar_localcaching &&
267 time_after_eq(jiffies, gl->gl_stamp +
268 gfs2_tune_get(sdp, gt_demote_secs) * HZ))
269 demote = 1;
271 return demote;
275 * inode_go_lock - operation done after an inode lock is locked by a process
276 * @gl: the glock
277 * @flags:
279 * Returns: errno
282 static int inode_go_lock(struct gfs2_holder *gh)
284 struct gfs2_glock *gl = gh->gh_gl;
285 struct gfs2_inode *ip = gl->gl_object;
286 int error = 0;
288 if (!ip)
289 return 0;
291 if (test_bit(GIF_INVALID, &ip->i_flags)) {
292 error = gfs2_inode_refresh(ip);
293 if (error)
294 return error;
297 if ((ip->i_di.di_flags & GFS2_DIF_TRUNC_IN_PROG) &&
298 (gl->gl_state == LM_ST_EXCLUSIVE) &&
299 (gh->gh_state == LM_ST_EXCLUSIVE))
300 error = gfs2_truncatei_resume(ip);
302 return error;
306 * inode_go_unlock - operation done before an inode lock is unlocked by a
307 * process
308 * @gl: the glock
309 * @flags:
313 static void inode_go_unlock(struct gfs2_holder *gh)
315 struct gfs2_glock *gl = gh->gh_gl;
316 struct gfs2_inode *ip = gl->gl_object;
318 if (ip)
319 gfs2_meta_cache_flush(ip);
323 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock
324 * @gl: the glock
326 * Returns: 1 if it's ok
329 static int rgrp_go_demote_ok(struct gfs2_glock *gl)
331 return !gl->gl_aspace->i_mapping->nrpages;
335 * rgrp_go_lock - operation done after an rgrp lock is locked by
336 * a first holder on this node.
337 * @gl: the glock
338 * @flags:
340 * Returns: errno
343 static int rgrp_go_lock(struct gfs2_holder *gh)
345 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
349 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
350 * a last holder on this node.
351 * @gl: the glock
352 * @flags:
356 static void rgrp_go_unlock(struct gfs2_holder *gh)
358 gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
362 * trans_go_xmote_th - promote/demote the transaction glock
363 * @gl: the glock
364 * @state: the requested state
365 * @flags:
369 static void trans_go_xmote_th(struct gfs2_glock *gl)
371 struct gfs2_sbd *sdp = gl->gl_sbd;
373 if (gl->gl_state != LM_ST_UNLOCKED &&
374 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
375 gfs2_meta_syncfs(sdp);
376 gfs2_log_shutdown(sdp);
381 * trans_go_xmote_bh - After promoting/demoting the transaction glock
382 * @gl: the glock
386 static void trans_go_xmote_bh(struct gfs2_glock *gl)
388 struct gfs2_sbd *sdp = gl->gl_sbd;
389 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
390 struct gfs2_glock *j_gl = ip->i_gl;
391 struct gfs2_log_header_host head;
392 int error;
394 if (gl->gl_state != LM_ST_UNLOCKED &&
395 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
396 gfs2_meta_cache_flush(GFS2_I(sdp->sd_jdesc->jd_inode));
397 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
399 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
400 if (error)
401 gfs2_consist(sdp);
402 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
403 gfs2_consist(sdp);
405 /* Initialize some head of the log stuff */
406 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
407 sdp->sd_log_sequence = head.lh_sequence + 1;
408 gfs2_log_pointers_init(sdp, head.lh_blkno);
414 * trans_go_drop_th - unlock the transaction glock
415 * @gl: the glock
417 * We want to sync the device even with localcaching. Remember
418 * that localcaching journal replay only marks buffers dirty.
421 static void trans_go_drop_th(struct gfs2_glock *gl)
423 struct gfs2_sbd *sdp = gl->gl_sbd;
425 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
426 gfs2_meta_syncfs(sdp);
427 gfs2_log_shutdown(sdp);
432 * quota_go_demote_ok - Check to see if it's ok to unlock a quota glock
433 * @gl: the glock
435 * Returns: 1 if it's ok
438 static int quota_go_demote_ok(struct gfs2_glock *gl)
440 return !atomic_read(&gl->gl_lvb_count);
443 const struct gfs2_glock_operations gfs2_meta_glops = {
444 .go_xmote_th = meta_go_sync,
445 .go_drop_th = meta_go_sync,
446 .go_type = LM_TYPE_META,
449 const struct gfs2_glock_operations gfs2_inode_glops = {
450 .go_xmote_th = inode_go_xmote_th,
451 .go_xmote_bh = inode_go_xmote_bh,
452 .go_drop_th = inode_go_drop_th,
453 .go_inval = inode_go_inval,
454 .go_demote_ok = inode_go_demote_ok,
455 .go_lock = inode_go_lock,
456 .go_unlock = inode_go_unlock,
457 .go_type = LM_TYPE_INODE,
460 const struct gfs2_glock_operations gfs2_rgrp_glops = {
461 .go_xmote_th = meta_go_sync,
462 .go_drop_th = meta_go_sync,
463 .go_inval = meta_go_inval,
464 .go_demote_ok = rgrp_go_demote_ok,
465 .go_lock = rgrp_go_lock,
466 .go_unlock = rgrp_go_unlock,
467 .go_type = LM_TYPE_RGRP,
470 const struct gfs2_glock_operations gfs2_trans_glops = {
471 .go_xmote_th = trans_go_xmote_th,
472 .go_xmote_bh = trans_go_xmote_bh,
473 .go_drop_th = trans_go_drop_th,
474 .go_type = LM_TYPE_NONDISK,
477 const struct gfs2_glock_operations gfs2_iopen_glops = {
478 .go_type = LM_TYPE_IOPEN,
481 const struct gfs2_glock_operations gfs2_flock_glops = {
482 .go_type = LM_TYPE_FLOCK,
485 const struct gfs2_glock_operations gfs2_nondisk_glops = {
486 .go_type = LM_TYPE_NONDISK,
489 const struct gfs2_glock_operations gfs2_quota_glops = {
490 .go_demote_ok = quota_go_demote_ok,
491 .go_type = LM_TYPE_QUOTA,
494 const struct gfs2_glock_operations gfs2_journal_glops = {
495 .go_type = LM_TYPE_JOURNAL,