1/*
2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
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
7 * of the GNU General Public License version 2.
8 */
9
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
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"
29
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;
40	struct list_head *head = &gl->gl_ail_list;
41	struct gfs2_bufdata *bd;
42	struct buffer_head *bh;
43	struct gfs2_trans tr;
44
45	memset(&tr, 0, sizeof(tr));
46	tr.tr_revokes = atomic_read(&gl->gl_ail_count);
47
48	if (!tr.tr_revokes)
49		return;
50
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
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;
64		gfs2_remove_from_ail(bd);
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);
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}
77
78/**
79 * rgrp_go_sync - sync out the metadata for this glock
80 * @gl: the glock
81 *
82 * Called when demoting or unlocking an EX glock.  We must flush
83 * to disk all dirty buffers/pages relating to this glock, and must not
84 * not return to caller to demote/unlock the glock until I/O is complete.
85 */
86
87static void rgrp_go_sync(struct gfs2_glock *gl)
88{
89	struct address_space *metamapping = gfs2_glock2aspace(gl);
90	int error;
91
92	if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
93		return;
94	BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
95
96	gfs2_log_flush(gl->gl_sbd, gl);
97	filemap_fdatawrite(metamapping);
98	error = filemap_fdatawait(metamapping);
99        mapping_set_error(metamapping, error);
100	gfs2_ail_empty_gl(gl);
101}
102
103/**
104 * rgrp_go_inval - invalidate the metadata for this glock
105 * @gl: the glock
106 * @flags:
107 *
108 * We never used LM_ST_DEFERRED with resource groups, so that we
109 * should always see the metadata flag set here.
110 *
111 */
112
113static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
114{
115	struct address_space *mapping = gfs2_glock2aspace(gl);
116
117	BUG_ON(!(flags & DIO_METADATA));
118	gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
119	truncate_inode_pages(mapping, 0);
120
121	if (gl->gl_object) {
122		struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
123		rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
124	}
125}
126
127/**
128 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
129 * @gl: the glock protecting the inode
130 *
131 */
132
133static void inode_go_sync(struct gfs2_glock *gl)
134{
135	struct gfs2_inode *ip = gl->gl_object;
136	struct address_space *metamapping = gfs2_glock2aspace(gl);
137	int error;
138
139	if (ip && !S_ISREG(ip->i_inode.i_mode))
140		ip = NULL;
141	if (ip && test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
142		unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
143	if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
144		return;
145
146	BUG_ON(gl->gl_state != LM_ST_EXCLUSIVE);
147
148	gfs2_log_flush(gl->gl_sbd, gl);
149	filemap_fdatawrite(metamapping);
150	if (ip) {
151		struct address_space *mapping = ip->i_inode.i_mapping;
152		filemap_fdatawrite(mapping);
153		error = filemap_fdatawait(mapping);
154		mapping_set_error(mapping, error);
155	}
156	error = filemap_fdatawait(metamapping);
157	mapping_set_error(metamapping, error);
158	gfs2_ail_empty_gl(gl);
159	/*
160	 * Writeback of the data mapping may cause the dirty flag to be set
161	 * so we have to clear it again here.
162	 */
163	smp_mb__before_clear_bit();
164	clear_bit(GLF_DIRTY, &gl->gl_flags);
165}
166
167/**
168 * inode_go_inval - prepare a inode glock to be released
169 * @gl: the glock
170 * @flags:
171 *
172 * Normally we invlidate everything, but if we are moving into
173 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
174 * can keep hold of the metadata, since it won't have changed.
175 *
176 */
177
178static void inode_go_inval(struct gfs2_glock *gl, int flags)
179{
180	struct gfs2_inode *ip = gl->gl_object;
181
182	gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
183
184	if (flags & DIO_METADATA) {
185		struct address_space *mapping = gfs2_glock2aspace(gl);
186		truncate_inode_pages(mapping, 0);
187		if (ip) {
188			set_bit(GIF_INVALID, &ip->i_flags);
189			forget_all_cached_acls(&ip->i_inode);
190		}
191	}
192
193	if (ip == GFS2_I(gl->gl_sbd->sd_rindex))
194		gl->gl_sbd->sd_rindex_uptodate = 0;
195	if (ip && S_ISREG(ip->i_inode.i_mode))
196		truncate_inode_pages(ip->i_inode.i_mapping, 0);
197}
198
199/**
200 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
201 * @gl: the glock
202 *
203 * Returns: 1 if it's ok
204 */
205
206static int inode_go_demote_ok(const struct gfs2_glock *gl)
207{
208	struct gfs2_sbd *sdp = gl->gl_sbd;
209	if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
210		return 0;
211	return 1;
212}
213
214/**
215 * inode_go_lock - operation done after an inode lock is locked by a process
216 * @gl: the glock
217 * @flags:
218 *
219 * Returns: errno
220 */
221
222static int inode_go_lock(struct gfs2_holder *gh)
223{
224	struct gfs2_glock *gl = gh->gh_gl;
225	struct gfs2_sbd *sdp = gl->gl_sbd;
226	struct gfs2_inode *ip = gl->gl_object;
227	int error = 0;
228
229	if (!ip || (gh->gh_flags & GL_SKIP))
230		return 0;
231
232	if (test_bit(GIF_INVALID, &ip->i_flags)) {
233		error = gfs2_inode_refresh(ip);
234		if (error)
235			return error;
236	}
237
238	if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
239	    (gl->gl_state == LM_ST_EXCLUSIVE) &&
240	    (gh->gh_state == LM_ST_EXCLUSIVE)) {
241		spin_lock(&sdp->sd_trunc_lock);
242		if (list_empty(&ip->i_trunc_list))
243			list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
244		spin_unlock(&sdp->sd_trunc_lock);
245		wake_up(&sdp->sd_quota_wait);
246		return 1;
247	}
248
249	return error;
250}
251
252/**
253 * inode_go_dump - print information about an inode
254 * @seq: The iterator
255 * @ip: the inode
256 *
257 * Returns: 0 on success, -ENOBUFS when we run out of space
258 */
259
260static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
261{
262	const struct gfs2_inode *ip = gl->gl_object;
263	if (ip == NULL)
264		return 0;
265	gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu/%llu\n",
266		  (unsigned long long)ip->i_no_formal_ino,
267		  (unsigned long long)ip->i_no_addr,
268		  IF2DT(ip->i_inode.i_mode), ip->i_flags,
269		  (unsigned int)ip->i_diskflags,
270		  (unsigned long long)ip->i_inode.i_size,
271		  (unsigned long long)ip->i_disksize);
272	return 0;
273}
274
275/**
276 * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock
277 * @gl: the glock
278 *
279 * Returns: 1 if it's ok
280 */
281
282static int rgrp_go_demote_ok(const struct gfs2_glock *gl)
283{
284	const struct address_space *mapping = (const struct address_space *)(gl + 1);
285	return !mapping->nrpages;
286}
287
288/**
289 * rgrp_go_lock - operation done after an rgrp lock is locked by
290 *    a first holder on this node.
291 * @gl: the glock
292 * @flags:
293 *
294 * Returns: errno
295 */
296
297static int rgrp_go_lock(struct gfs2_holder *gh)
298{
299	return gfs2_rgrp_bh_get(gh->gh_gl->gl_object);
300}
301
302/**
303 * rgrp_go_unlock - operation done before an rgrp lock is unlocked by
304 *    a last holder on this node.
305 * @gl: the glock
306 * @flags:
307 *
308 */
309
310static void rgrp_go_unlock(struct gfs2_holder *gh)
311{
312	gfs2_rgrp_bh_put(gh->gh_gl->gl_object);
313}
314
315/**
316 * trans_go_sync - promote/demote the transaction glock
317 * @gl: the glock
318 * @state: the requested state
319 * @flags:
320 *
321 */
322
323static void trans_go_sync(struct gfs2_glock *gl)
324{
325	struct gfs2_sbd *sdp = gl->gl_sbd;
326
327	if (gl->gl_state != LM_ST_UNLOCKED &&
328	    test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
329		flush_workqueue(gfs2_delete_workqueue);
330		gfs2_meta_syncfs(sdp);
331		gfs2_log_shutdown(sdp);
332	}
333}
334
335/**
336 * trans_go_xmote_bh - After promoting/demoting the transaction glock
337 * @gl: the glock
338 *
339 */
340
341static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
342{
343	struct gfs2_sbd *sdp = gl->gl_sbd;
344	struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
345	struct gfs2_glock *j_gl = ip->i_gl;
346	struct gfs2_log_header_host head;
347	int error;
348
349	if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
350		j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
351
352		error = gfs2_find_jhead(sdp->sd_jdesc, &head);
353		if (error)
354			gfs2_consist(sdp);
355		if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
356			gfs2_consist(sdp);
357
358		/*  Initialize some head of the log stuff  */
359		if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
360			sdp->sd_log_sequence = head.lh_sequence + 1;
361			gfs2_log_pointers_init(sdp, head.lh_blkno);
362		}
363	}
364	return 0;
365}
366
367/**
368 * trans_go_demote_ok
369 * @gl: the glock
370 *
371 * Always returns 0
372 */
373
374static int trans_go_demote_ok(const struct gfs2_glock *gl)
375{
376	return 0;
377}
378
379/**
380 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
381 * @gl: the glock
382 *
383 * gl_spin lock is held while calling this
384 */
385static void iopen_go_callback(struct gfs2_glock *gl)
386{
387	struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
388
389	if (gl->gl_demote_state == LM_ST_UNLOCKED &&
390	    gl->gl_state == LM_ST_SHARED && ip) {
391		gfs2_glock_hold(gl);
392		if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
393			gfs2_glock_put_nolock(gl);
394	}
395}
396
397const struct gfs2_glock_operations gfs2_meta_glops = {
398	.go_type = LM_TYPE_META,
399};
400
401const struct gfs2_glock_operations gfs2_inode_glops = {
402	.go_xmote_th = inode_go_sync,
403	.go_inval = inode_go_inval,
404	.go_demote_ok = inode_go_demote_ok,
405	.go_lock = inode_go_lock,
406	.go_dump = inode_go_dump,
407	.go_type = LM_TYPE_INODE,
408	.go_min_hold_time = HZ / 5,
409	.go_flags = GLOF_ASPACE,
410};
411
412const struct gfs2_glock_operations gfs2_rgrp_glops = {
413	.go_xmote_th = rgrp_go_sync,
414	.go_inval = rgrp_go_inval,
415	.go_demote_ok = rgrp_go_demote_ok,
416	.go_lock = rgrp_go_lock,
417	.go_unlock = rgrp_go_unlock,
418	.go_dump = gfs2_rgrp_dump,
419	.go_type = LM_TYPE_RGRP,
420	.go_min_hold_time = HZ / 5,
421	.go_flags = GLOF_ASPACE,
422};
423
424const struct gfs2_glock_operations gfs2_trans_glops = {
425	.go_xmote_th = trans_go_sync,
426	.go_xmote_bh = trans_go_xmote_bh,
427	.go_demote_ok = trans_go_demote_ok,
428	.go_type = LM_TYPE_NONDISK,
429};
430
431const struct gfs2_glock_operations gfs2_iopen_glops = {
432	.go_type = LM_TYPE_IOPEN,
433	.go_callback = iopen_go_callback,
434};
435
436const struct gfs2_glock_operations gfs2_flock_glops = {
437	.go_type = LM_TYPE_FLOCK,
438};
439
440const struct gfs2_glock_operations gfs2_nondisk_glops = {
441	.go_type = LM_TYPE_NONDISK,
442};
443
444const struct gfs2_glock_operations gfs2_quota_glops = {
445	.go_type = LM_TYPE_QUOTA,
446};
447
448const struct gfs2_glock_operations gfs2_journal_glops = {
449	.go_type = LM_TYPE_JOURNAL,
450};
451
452const struct gfs2_glock_operations *gfs2_glops_list[] = {
453	[LM_TYPE_META] = &gfs2_meta_glops,
454	[LM_TYPE_INODE] = &gfs2_inode_glops,
455	[LM_TYPE_RGRP] = &gfs2_rgrp_glops,
456	[LM_TYPE_NONDISK] = &gfs2_trans_glops,
457	[LM_TYPE_IOPEN] = &gfs2_iopen_glops,
458	[LM_TYPE_FLOCK] = &gfs2_flock_glops,
459	[LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
460	[LM_TYPE_QUOTA] = &gfs2_quota_glops,
461	[LM_TYPE_JOURNAL] = &gfs2_journal_glops,
462};
463