]> git.ipfire.org Git - people/ms/linux.git/blame - fs/gfs2/glops.c
GFS2: Fix locking bug in failed shared to exclusive conversion
[people/ms/linux.git] / fs / gfs2 / glops.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
cf45b752 3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
b3b94faa
DT
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT
8 */
9
b3b94faa
DT
10#include <linux/slab.h>
11#include <linux/spinlock.h>
12#include <linux/completion.h>
13#include <linux/buffer_head.h>
5c676f6d 14#include <linux/gfs2_ondisk.h>
6802e340 15#include <linux/bio.h>
b3b94faa
DT
16
17#include "gfs2.h"
5c676f6d 18#include "incore.h"
b3b94faa
DT
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"
b3b94faa
DT
25#include "recovery.h"
26#include "rgrp.h"
5c676f6d 27#include "util.h"
ddacfaf7 28#include "trans.h"
b3b94faa 29
ddacfaf7
SW
30/**
31 * ail_empty_gl - remove all buffers for a given lock from the AIL
32 * @gl: the glock
33 *
34 * None of the buffers should be dirty, locked, or pinned.
35 */
36
37static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
38{
39 struct gfs2_sbd *sdp = gl->gl_sbd;
ddacfaf7
SW
40 struct list_head *head = &gl->gl_ail_list;
41 struct gfs2_bufdata *bd;
42 struct buffer_head *bh;
d8348de0 43 struct gfs2_trans tr;
ddacfaf7 44
d8348de0
SW
45 memset(&tr, 0, sizeof(tr));
46 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
ddacfaf7 47
d8348de0 48 if (!tr.tr_revokes)
ddacfaf7
SW
49 return;
50
d8348de0
SW
51 /* A shortened, inline version of gfs2_trans_begin() */
52 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
53 tr.tr_ip = (unsigned long)__builtin_return_address(0);
54 INIT_LIST_HEAD(&tr.tr_list_buf);
55 gfs2_log_reserve(sdp, tr.tr_reserved);
56 BUG_ON(current->journal_info);
57 current->journal_info = &tr;
58
ddacfaf7
SW
59 gfs2_log_lock(sdp);
60 while (!list_empty(head)) {
61 bd = list_entry(head->next, struct gfs2_bufdata,
62 bd_ail_gl_list);
63 bh = bd->bd_bh;
f91a0d3e 64 gfs2_remove_from_ail(bd);
1ad38c43
SW
65 bd->bd_bh = NULL;
66 bh->b_private = NULL;
67 bd->bd_blkno = bh->b_blocknr;
68 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
69 gfs2_trans_add_revoke(sdp, bd);
ddacfaf7
SW
70 }
71 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
72 gfs2_log_unlock(sdp);
73
74 gfs2_trans_end(sdp);
75 gfs2_log_flush(sdp, NULL);
76}
ba7f7290
SW
77
78/**
79 * gfs2_pte_inval - Sync and invalidate all PTEs associated with a glock
80 * @gl: the glock
81 *
82 */
83
84static void gfs2_pte_inval(struct gfs2_glock *gl)
85{
86 struct gfs2_inode *ip;
87 struct inode *inode;
88
89 ip = gl->gl_object;
90 inode = &ip->i_inode;
b60623c2 91 if (!ip || !S_ISREG(inode->i_mode))
ba7f7290
SW
92 return;
93
ba7f7290 94 unmap_shared_mapping_range(inode->i_mapping, 0, 0);
ba7f7290
SW
95 if (test_bit(GIF_SW_PAGED, &ip->i_flags))
96 set_bit(GLF_DIRTY, &gl->gl_flags);
97
ba7f7290
SW
98}
99
b3b94faa
DT
100/**
101 * meta_go_sync - sync out the metadata for this glock
102 * @gl: the glock
b3b94faa
DT
103 *
104 * Called when demoting or unlocking an EX glock. We must flush
105 * to disk all dirty buffers/pages relating to this glock, and must not
106 * not return to caller to demote/unlock the glock until I/O is complete.
107 */
108
1a14d3a6 109static void meta_go_sync(struct gfs2_glock *gl)
b3b94faa 110{
b5d32bea
SW
111 if (gl->gl_state != LM_ST_EXCLUSIVE)
112 return;
113
b3b94faa 114 if (test_and_clear_bit(GLF_DIRTY, &gl->gl_flags)) {
b09e593d 115 gfs2_log_flush(gl->gl_sbd, gl);
7276b3b0 116 gfs2_meta_sync(gl);
1a14d3a6 117 gfs2_ail_empty_gl(gl);
b3b94faa 118 }
b3b94faa
DT
119}
120
121/**
122 * meta_go_inval - invalidate the metadata for this glock
123 * @gl: the glock
124 * @flags:
125 *
126 */
127
128static void meta_go_inval(struct gfs2_glock *gl, int flags)
129{
130 if (!(flags & DIO_METADATA))
131 return;
132
133 gfs2_meta_inval(gl);
cf45b752
BP
134 if (gl->gl_object == GFS2_I(gl->gl_sbd->sd_rindex))
135 gl->gl_sbd->sd_rindex_uptodate = 0;
136 else if (gl->gl_ops == &gfs2_rgrp_glops && gl->gl_object) {
137 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
138
139 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
140 }
b3b94faa
DT
141}
142
b5d32bea
SW
143/**
144 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
145 * @gl: the glock protecting the inode
146 *
147 */
148
149static void inode_go_sync(struct gfs2_glock *gl)
150{
151 struct gfs2_inode *ip = gl->gl_object;
3042a2cc
SW
152 struct address_space *metamapping = gl->gl_aspace->i_mapping;
153 int error;
154
155 if (gl->gl_state != LM_ST_UNLOCKED)
156 gfs2_pte_inval(gl);
157 if (gl->gl_state != LM_ST_EXCLUSIVE)
158 return;
b5d32bea
SW
159
160 if (ip && !S_ISREG(ip->i_inode.i_mode))
161 ip = NULL;
162
163 if (test_bit(GLF_DIRTY, &gl->gl_flags)) {
b524fe64 164 gfs2_log_flush(gl->gl_sbd, gl);
3042a2cc 165 filemap_fdatawrite(metamapping);
b5d32bea
SW
166 if (ip) {
167 struct address_space *mapping = ip->i_inode.i_mapping;
3042a2cc
SW
168 filemap_fdatawrite(mapping);
169 error = filemap_fdatawait(mapping);
3e9f45bd 170 mapping_set_error(mapping, error);
b5d32bea 171 }
3042a2cc
SW
172 error = filemap_fdatawait(metamapping);
173 mapping_set_error(metamapping, error);
b5d32bea
SW
174 clear_bit(GLF_DIRTY, &gl->gl_flags);
175 gfs2_ail_empty_gl(gl);
176 }
177}
178
b3b94faa
DT
179/**
180 * inode_go_inval - prepare a inode glock to be released
181 * @gl: the glock
182 * @flags:
183 *
184 */
185
186static void inode_go_inval(struct gfs2_glock *gl, int flags)
187{
b004157a 188 struct gfs2_inode *ip = gl->gl_object;
b3b94faa 189 int meta = (flags & DIO_METADATA);
b3b94faa
DT
190
191 if (meta) {
192 gfs2_meta_inval(gl);
b004157a
SW
193 if (ip)
194 set_bit(GIF_INVALID, &ip->i_flags);
195 }
196
3cc3f710 197 if (ip && S_ISREG(ip->i_inode.i_mode))
b004157a 198 truncate_inode_pages(ip->i_inode.i_mapping, 0);
b3b94faa
DT
199}
200
201/**
202 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
203 * @gl: the glock
204 *
205 * Returns: 1 if it's ok
206 */
207
97cc1025 208static int inode_go_demote_ok(const struct gfs2_glock *gl)
b3b94faa
DT
209{
210 struct gfs2_sbd *sdp = gl->gl_sbd;
97cc1025
SW
211 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
212 return 0;
213 return 1;
b3b94faa
DT
214}
215
216/**
217 * inode_go_lock - operation done after an inode lock is locked by a process
218 * @gl: the glock
219 * @flags:
220 *
221 * Returns: errno
222 */
223
224static int inode_go_lock(struct gfs2_holder *gh)
225{
226 struct gfs2_glock *gl = gh->gh_gl;
813e0c46 227 struct gfs2_sbd *sdp = gl->gl_sbd;
5c676f6d 228 struct gfs2_inode *ip = gl->gl_object;
b3b94faa
DT
229 int error = 0;
230
091806ed 231 if (!ip || (gh->gh_flags & GL_SKIP))
b3b94faa
DT
232 return 0;
233
bfded27b 234 if (test_bit(GIF_INVALID, &ip->i_flags)) {
b3b94faa
DT
235 error = gfs2_inode_refresh(ip);
236 if (error)
237 return error;
b3b94faa
DT
238 }
239
383f01fb 240 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
b3b94faa 241 (gl->gl_state == LM_ST_EXCLUSIVE) &&
813e0c46
SW
242 (gh->gh_state == LM_ST_EXCLUSIVE)) {
243 spin_lock(&sdp->sd_trunc_lock);
244 if (list_empty(&ip->i_trunc_list))
245 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
246 spin_unlock(&sdp->sd_trunc_lock);
247 wake_up(&sdp->sd_quota_wait);
248 return 1;
249 }
b3b94faa
DT
250
251 return error;
252}
253
6802e340
SW
254/**
255 * inode_go_dump - print information about an inode
256 * @seq: The iterator
257 * @ip: the inode
258 *
259 * Returns: 0 on success, -ENOBUFS when we run out of space
260 */
261
262static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
263{
264 const struct gfs2_inode *ip = gl->gl_object;
265 if (ip == NULL)
266 return 0;
fa75cedc 267 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu/%llu\n",
6802e340
SW
268 (unsigned long long)ip->i_no_formal_ino,
269 (unsigned long long)ip->i_no_addr,
fa75cedc
SW
270 IF2DT(ip->i_inode.i_mode), ip->i_flags,
271 (unsigned int)ip->i_diskflags,
272 (unsigned long long)ip->i_inode.i_size,
273 (unsigned long long)ip->i_disksize);
6802e340
SW
274 return 0;
275}
276
b3b94faa
DT
277/**
278 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock
279 * @gl: the glock
280 *
281 * Returns: 1 if it's ok
282 */
283
97cc1025 284static int rgrp_go_demote_ok(const struct gfs2_glock *gl)
b3b94faa
DT
285{
286 return !gl->gl_aspace->i_mapping->nrpages;
287}
288
289/**
290 * rgrp_go_lock - operation done after an rgrp lock is locked by
291 * a first holder on this node.
292 * @gl: the glock
293 * @flags:
294 *
295 * Returns: errno
296 */
297
298static int rgrp_go_lock(struct gfs2_holder *gh)
299{
5c676f6d 300 return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
b3b94faa
DT
301}
302
303/**
304 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
305 * a last holder on this node.
306 * @gl: the glock
307 * @flags:
308 *
309 */
310
311static void rgrp_go_unlock(struct gfs2_holder *gh)
312{
5c676f6d 313 gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
b3b94faa
DT
314}
315
6802e340
SW
316/**
317 * rgrp_go_dump - print out an rgrp
318 * @seq: The iterator
319 * @gl: The glock in question
320 *
321 */
322
323static int rgrp_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
324{
325 const struct gfs2_rgrpd *rgd = gl->gl_object;
326 if (rgd == NULL)
327 return 0;
fa75cedc
SW
328 gfs2_print_dbg(seq, " R: n:%llu f:%02x b:%u/%u i:%u\n",
329 (unsigned long long)rgd->rd_addr, rgd->rd_flags,
330 rgd->rd_free, rgd->rd_free_clone, rgd->rd_dinodes);
6802e340
SW
331 return 0;
332}
333
b3b94faa 334/**
3042a2cc 335 * trans_go_sync - promote/demote the transaction glock
b3b94faa
DT
336 * @gl: the glock
337 * @state: the requested state
338 * @flags:
339 *
340 */
341
3042a2cc 342static void trans_go_sync(struct gfs2_glock *gl)
b3b94faa
DT
343{
344 struct gfs2_sbd *sdp = gl->gl_sbd;
345
346 if (gl->gl_state != LM_ST_UNLOCKED &&
347 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
348 gfs2_meta_syncfs(sdp);
349 gfs2_log_shutdown(sdp);
350 }
b3b94faa
DT
351}
352
353/**
354 * trans_go_xmote_bh - After promoting/demoting the transaction glock
355 * @gl: the glock
356 *
357 */
358
6802e340 359static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
b3b94faa
DT
360{
361 struct gfs2_sbd *sdp = gl->gl_sbd;
feaa7bba 362 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
5c676f6d 363 struct gfs2_glock *j_gl = ip->i_gl;
55167622 364 struct gfs2_log_header_host head;
b3b94faa
DT
365 int error;
366
6802e340 367 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
1a14d3a6 368 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
b3b94faa
DT
369
370 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
371 if (error)
372 gfs2_consist(sdp);
373 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
374 gfs2_consist(sdp);
375
376 /* Initialize some head of the log stuff */
377 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
378 sdp->sd_log_sequence = head.lh_sequence + 1;
379 gfs2_log_pointers_init(sdp, head.lh_blkno);
380 }
381 }
6802e340 382 return 0;
b3b94faa
DT
383}
384
97cc1025
SW
385/**
386 * trans_go_demote_ok
387 * @gl: the glock
388 *
389 * Always returns 0
390 */
391
392static int trans_go_demote_ok(const struct gfs2_glock *gl)
393{
394 return 0;
395}
396
8fb4b536 397const struct gfs2_glock_operations gfs2_meta_glops = {
b5d32bea 398 .go_xmote_th = meta_go_sync,
ea67eedb 399 .go_type = LM_TYPE_META,
b3b94faa
DT
400};
401
8fb4b536 402const struct gfs2_glock_operations gfs2_inode_glops = {
3042a2cc 403 .go_xmote_th = inode_go_sync,
b3b94faa
DT
404 .go_inval = inode_go_inval,
405 .go_demote_ok = inode_go_demote_ok,
406 .go_lock = inode_go_lock,
6802e340 407 .go_dump = inode_go_dump,
ea67eedb 408 .go_type = LM_TYPE_INODE,
6802e340 409 .go_min_hold_time = HZ / 5,
b3b94faa
DT
410};
411
8fb4b536 412const struct gfs2_glock_operations gfs2_rgrp_glops = {
cad5b939 413 .go_xmote_th = meta_go_sync,
b3b94faa
DT
414 .go_inval = meta_go_inval,
415 .go_demote_ok = rgrp_go_demote_ok,
416 .go_lock = rgrp_go_lock,
417 .go_unlock = rgrp_go_unlock,
6802e340 418 .go_dump = rgrp_go_dump,
ea67eedb 419 .go_type = LM_TYPE_RGRP,
6802e340 420 .go_min_hold_time = HZ / 5,
b3b94faa
DT
421};
422
8fb4b536 423const struct gfs2_glock_operations gfs2_trans_glops = {
3042a2cc 424 .go_xmote_th = trans_go_sync,
b3b94faa 425 .go_xmote_bh = trans_go_xmote_bh,
97cc1025 426 .go_demote_ok = trans_go_demote_ok,
ea67eedb 427 .go_type = LM_TYPE_NONDISK,
b3b94faa
DT
428};
429
8fb4b536 430const struct gfs2_glock_operations gfs2_iopen_glops = {
ea67eedb 431 .go_type = LM_TYPE_IOPEN,
b3b94faa
DT
432};
433
8fb4b536 434const struct gfs2_glock_operations gfs2_flock_glops = {
ea67eedb 435 .go_type = LM_TYPE_FLOCK,
b3b94faa
DT
436};
437
8fb4b536 438const struct gfs2_glock_operations gfs2_nondisk_glops = {
ea67eedb 439 .go_type = LM_TYPE_NONDISK,
b3b94faa
DT
440};
441
8fb4b536 442const struct gfs2_glock_operations gfs2_quota_glops = {
ea67eedb 443 .go_type = LM_TYPE_QUOTA,
b3b94faa
DT
444};
445
8fb4b536 446const struct gfs2_glock_operations gfs2_journal_glops = {
ea67eedb 447 .go_type = LM_TYPE_JOURNAL,
b3b94faa
DT
448};
449
64d576ba
SW
450const struct gfs2_glock_operations *gfs2_glops_list[] = {
451 [LM_TYPE_META] = &gfs2_meta_glops,
452 [LM_TYPE_INODE] = &gfs2_inode_glops,
453 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
454 [LM_TYPE_NONDISK] = &gfs2_trans_glops,
455 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
456 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
457 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
458 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
459 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
460};
461