quota: Do more fine-grained locking in dquot_acquire()
[sfrench/cifs-2.6.git] / fs / gfs2 / trans.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/buffer_head.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "glock.h"
23 #include "inode.h"
24 #include "log.h"
25 #include "lops.h"
26 #include "meta_io.h"
27 #include "trans.h"
28 #include "util.h"
29 #include "trace_gfs2.h"
30
31 int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
32                      unsigned int revokes)
33 {
34         struct gfs2_trans *tr;
35         int error;
36
37         BUG_ON(current->journal_info);
38         BUG_ON(blocks == 0 && revokes == 0);
39
40         if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
41                 return -EROFS;
42
43         tr = kzalloc(sizeof(struct gfs2_trans), GFP_NOFS);
44         if (!tr)
45                 return -ENOMEM;
46
47         tr->tr_ip = _RET_IP_;
48         tr->tr_blocks = blocks;
49         tr->tr_revokes = revokes;
50         tr->tr_reserved = 1;
51         set_bit(TR_ALLOCED, &tr->tr_flags);
52         if (blocks)
53                 tr->tr_reserved += 6 + blocks;
54         if (revokes)
55                 tr->tr_reserved += gfs2_struct2blk(sdp, revokes,
56                                                    sizeof(u64));
57         INIT_LIST_HEAD(&tr->tr_databuf);
58         INIT_LIST_HEAD(&tr->tr_buf);
59
60         sb_start_intwrite(sdp->sd_vfs);
61
62         error = gfs2_log_reserve(sdp, tr->tr_reserved);
63         if (error)
64                 goto fail;
65
66         current->journal_info = tr;
67
68         return 0;
69
70 fail:
71         sb_end_intwrite(sdp->sd_vfs);
72         kfree(tr);
73
74         return error;
75 }
76
77 static void gfs2_print_trans(const struct gfs2_trans *tr)
78 {
79         pr_warn("Transaction created at: %pSR\n", (void *)tr->tr_ip);
80         pr_warn("blocks=%u revokes=%u reserved=%u touched=%u\n",
81                 tr->tr_blocks, tr->tr_revokes, tr->tr_reserved,
82                 test_bit(TR_TOUCHED, &tr->tr_flags));
83         pr_warn("Buf %u/%u Databuf %u/%u Revoke %u/%u\n",
84                 tr->tr_num_buf_new, tr->tr_num_buf_rm,
85                 tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
86                 tr->tr_num_revoke, tr->tr_num_revoke_rm);
87 }
88
89 void gfs2_trans_end(struct gfs2_sbd *sdp)
90 {
91         struct gfs2_trans *tr = current->journal_info;
92         s64 nbuf;
93         int alloced = test_bit(TR_ALLOCED, &tr->tr_flags);
94
95         BUG_ON(!tr);
96         current->journal_info = NULL;
97
98         if (!test_bit(TR_TOUCHED, &tr->tr_flags)) {
99                 gfs2_log_release(sdp, tr->tr_reserved);
100                 if (alloced) {
101                         kfree(tr);
102                         sb_end_intwrite(sdp->sd_vfs);
103                 }
104                 return;
105         }
106
107         nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new;
108         nbuf -= tr->tr_num_buf_rm;
109         nbuf -= tr->tr_num_databuf_rm;
110
111         if (gfs2_assert_withdraw(sdp, (nbuf <= tr->tr_blocks) &&
112                                        (tr->tr_num_revoke <= tr->tr_revokes)))
113                 gfs2_print_trans(tr);
114
115         gfs2_log_commit(sdp, tr);
116         if (alloced && !test_bit(TR_ATTACHED, &tr->tr_flags))
117                 kfree(tr);
118         up_read(&sdp->sd_log_flush_lock);
119
120         if (sdp->sd_vfs->s_flags & MS_SYNCHRONOUS)
121                 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
122         if (alloced)
123                 sb_end_intwrite(sdp->sd_vfs);
124 }
125
126 static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
127                                                struct buffer_head *bh,
128                                                const struct gfs2_log_operations *lops)
129 {
130         struct gfs2_bufdata *bd;
131
132         bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
133         bd->bd_bh = bh;
134         bd->bd_gl = gl;
135         bd->bd_ops = lops;
136         INIT_LIST_HEAD(&bd->bd_list);
137         bh->b_private = bd;
138         return bd;
139 }
140
141 /**
142  * gfs2_trans_add_data - Add a databuf to the transaction.
143  * @gl: The inode glock associated with the buffer
144  * @bh: The buffer to add
145  *
146  * This is used in two distinct cases:
147  * i) In ordered write mode
148  *    We put the data buffer on a list so that we can ensure that its
149  *    synced to disk at the right time
150  * ii) In journaled data mode
151  *    We need to journal the data block in the same way as metadata in
152  *    the functions above. The difference is that here we have a tag
153  *    which is two __be64's being the block number (as per meta data)
154  *    and a flag which says whether the data block needs escaping or
155  *    not. This means we need a new log entry for each 251 or so data
156  *    blocks, which isn't an enormous overhead but twice as much as
157  *    for normal metadata blocks.
158  */
159 void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
160 {
161         struct gfs2_trans *tr = current->journal_info;
162         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
163         struct address_space *mapping = bh->b_page->mapping;
164         struct gfs2_inode *ip = GFS2_I(mapping->host);
165         struct gfs2_bufdata *bd;
166
167         if (!gfs2_is_jdata(ip)) {
168                 gfs2_ordered_add_inode(ip);
169                 return;
170         }
171
172         lock_buffer(bh);
173         if (buffer_pinned(bh)) {
174                 set_bit(TR_TOUCHED, &tr->tr_flags);
175                 goto out;
176         }
177         gfs2_log_lock(sdp);
178         bd = bh->b_private;
179         if (bd == NULL) {
180                 gfs2_log_unlock(sdp);
181                 unlock_buffer(bh);
182                 if (bh->b_private == NULL)
183                         bd = gfs2_alloc_bufdata(gl, bh, &gfs2_databuf_lops);
184                 else
185                         bd = bh->b_private;
186                 lock_buffer(bh);
187                 gfs2_log_lock(sdp);
188         }
189         gfs2_assert(sdp, bd->bd_gl == gl);
190         set_bit(TR_TOUCHED, &tr->tr_flags);
191         if (list_empty(&bd->bd_list)) {
192                 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
193                 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
194                 gfs2_pin(sdp, bd->bd_bh);
195                 tr->tr_num_databuf_new++;
196                 list_add_tail(&bd->bd_list, &tr->tr_databuf);
197         }
198         gfs2_log_unlock(sdp);
199 out:
200         unlock_buffer(bh);
201 }
202
203 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
204 {
205
206         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
207         struct gfs2_bufdata *bd;
208         struct gfs2_meta_header *mh;
209         struct gfs2_trans *tr = current->journal_info;
210         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
211
212         lock_buffer(bh);
213         if (buffer_pinned(bh)) {
214                 set_bit(TR_TOUCHED, &tr->tr_flags);
215                 goto out;
216         }
217         gfs2_log_lock(sdp);
218         bd = bh->b_private;
219         if (bd == NULL) {
220                 gfs2_log_unlock(sdp);
221                 unlock_buffer(bh);
222                 lock_page(bh->b_page);
223                 if (bh->b_private == NULL)
224                         bd = gfs2_alloc_bufdata(gl, bh, &gfs2_buf_lops);
225                 else
226                         bd = bh->b_private;
227                 unlock_page(bh->b_page);
228                 lock_buffer(bh);
229                 gfs2_log_lock(sdp);
230         }
231         gfs2_assert(sdp, bd->bd_gl == gl);
232         set_bit(TR_TOUCHED, &tr->tr_flags);
233         if (!list_empty(&bd->bd_list))
234                 goto out_unlock;
235         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
236         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
237         mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
238         if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) {
239                 pr_err("Attempting to add uninitialised block to journal (inplace block=%lld)\n",
240                        (unsigned long long)bd->bd_bh->b_blocknr);
241                 BUG();
242         }
243         if (unlikely(state == SFS_FROZEN)) {
244                 printk(KERN_INFO "GFS2:adding buf while frozen\n");
245                 gfs2_assert_withdraw(sdp, 0);
246         }
247         gfs2_pin(sdp, bd->bd_bh);
248         mh->__pad0 = cpu_to_be64(0);
249         mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
250         list_add(&bd->bd_list, &tr->tr_buf);
251         tr->tr_num_buf_new++;
252 out_unlock:
253         gfs2_log_unlock(sdp);
254 out:
255         unlock_buffer(bh);
256 }
257
258 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
259 {
260         struct gfs2_trans *tr = current->journal_info;
261
262         BUG_ON(!list_empty(&bd->bd_list));
263         gfs2_add_revoke(sdp, bd);
264         set_bit(TR_TOUCHED, &tr->tr_flags);
265         tr->tr_num_revoke++;
266 }
267
268 void gfs2_trans_add_unrevoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len)
269 {
270         struct gfs2_bufdata *bd, *tmp;
271         struct gfs2_trans *tr = current->journal_info;
272         unsigned int n = len;
273
274         gfs2_log_lock(sdp);
275         list_for_each_entry_safe(bd, tmp, &sdp->sd_log_le_revoke, bd_list) {
276                 if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) {
277                         list_del_init(&bd->bd_list);
278                         gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke);
279                         sdp->sd_log_num_revoke--;
280                         kmem_cache_free(gfs2_bufdata_cachep, bd);
281                         tr->tr_num_revoke_rm++;
282                         if (--n == 0)
283                                 break;
284                 }
285         }
286         gfs2_log_unlock(sdp);
287 }
288