Merge tag 'trace-v6.8' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux...
[linux-2.6-block.git] / fs / gfs2 / trans.c
CommitLineData
7336d0e6 1// SPDX-License-Identifier: GPL-2.0-only
b3b94faa
DT
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3a8a9a10 4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
b3b94faa
DT
5 */
6
d77d1b58
JP
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
b3b94faa
DT
9#include <linux/sched.h>
10#include <linux/slab.h>
11#include <linux/spinlock.h>
12#include <linux/completion.h>
13#include <linux/buffer_head.h>
d0dc80db 14#include <linux/kallsyms.h>
f057f6cd 15#include <linux/gfs2_ondisk.h>
b3b94faa
DT
16
17#include "gfs2.h"
5c676f6d 18#include "incore.h"
b3b94faa 19#include "glock.h"
767f433f 20#include "inode.h"
b3b94faa
DT
21#include "log.h"
22#include "lops.h"
23#include "meta_io.h"
24#include "trans.h"
5c676f6d 25#include "util.h"
5e687eac 26#include "trace_gfs2.h"
b3b94faa 27
b0be23b2
BP
28static void gfs2_print_trans(struct gfs2_sbd *sdp, const struct gfs2_trans *tr)
29{
30 fs_warn(sdp, "Transaction created at: %pSR\n", (void *)tr->tr_ip);
31 fs_warn(sdp, "blocks=%u revokes=%u reserved=%u touched=%u\n",
32 tr->tr_blocks, tr->tr_revokes, tr->tr_reserved,
33 test_bit(TR_TOUCHED, &tr->tr_flags));
2129b428 34 fs_warn(sdp, "Buf %u/%u Databuf %u/%u Revoke %u\n",
b0be23b2
BP
35 tr->tr_num_buf_new, tr->tr_num_buf_rm,
36 tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
2129b428 37 tr->tr_num_revoke);
b0be23b2
BP
38}
39
c968f578
AG
40int __gfs2_trans_begin(struct gfs2_trans *tr, struct gfs2_sbd *sdp,
41 unsigned int blocks, unsigned int revokes,
42 unsigned long ip)
b3b94faa 43{
2129b428
AG
44 unsigned int extra_revokes;
45
b0be23b2
BP
46 if (current->journal_info) {
47 gfs2_print_trans(sdp, current->journal_info);
48 BUG();
49 }
d0dc80db 50 BUG_ON(blocks == 0 && revokes == 0);
b3b94faa 51
a1c0643f
SW
52 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
53 return -EROFS;
54
c968f578 55 tr->tr_ip = ip;
b3b94faa
DT
56 tr->tr_blocks = blocks;
57 tr->tr_revokes = revokes;
fe3e3976 58 tr->tr_reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
297de318
AG
59 if (blocks) {
60 /*
61 * The reserved blocks are either used for data or metadata.
62 * We can have mixed data and metadata, each with its own log
63 * descriptor block; see calc_reserved().
64 */
65 tr->tr_reserved += blocks + 1 + DIV_ROUND_UP(blocks - 1, databuf_limit(sdp));
66 }
d69a3c65
SW
67 INIT_LIST_HEAD(&tr->tr_databuf);
68 INIT_LIST_HEAD(&tr->tr_buf);
462582b9 69 INIT_LIST_HEAD(&tr->tr_list);
cbcc89b6
BP
70 INIT_LIST_HEAD(&tr->tr_ail1_list);
71 INIT_LIST_HEAD(&tr->tr_ail2_list);
d69a3c65 72
c1eba1b0
AG
73 if (gfs2_assert_warn(sdp, tr->tr_reserved <= sdp->sd_jdesc->jd_blocks))
74 return -EINVAL;
75
39263d5e 76 sb_start_intwrite(sdp->sd_vfs);
b3b94faa 77
2129b428
AG
78 /*
79 * Try the reservations under sd_log_flush_lock to prevent log flushes
80 * from creating inconsistencies between the number of allocated and
81 * reserved revokes. If that fails, do a full-block allocation outside
82 * of the lock to avoid stalling log flushes. Then, allot the
83 * appropriate number of blocks to revokes, use as many revokes locally
84 * as needed, and "release" the surplus into the revokes pool.
85 */
c1eba1b0
AG
86
87 down_read(&sdp->sd_log_flush_lock);
2129b428
AG
88 if (gfs2_log_try_reserve(sdp, tr, &extra_revokes))
89 goto reserved;
90 up_read(&sdp->sd_log_flush_lock);
91 gfs2_log_reserve(sdp, tr, &extra_revokes);
92 down_read(&sdp->sd_log_flush_lock);
93
94reserved:
95 gfs2_log_release_revokes(sdp, extra_revokes);
c1eba1b0 96 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
2129b428 97 gfs2_log_release_revokes(sdp, tr->tr_revokes);
c1eba1b0 98 up_read(&sdp->sd_log_flush_lock);
5ae8fff8 99 gfs2_log_release(sdp, tr->tr_reserved);
c968f578 100 sb_end_intwrite(sdp->sd_vfs);
c1eba1b0 101 return -EROFS;
c968f578 102 }
b3b94faa 103
5c676f6d 104 current->journal_info = tr;
b3b94faa
DT
105
106 return 0;
c968f578 107}
b3b94faa 108
c968f578
AG
109int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
110 unsigned int revokes)
111{
112 struct gfs2_trans *tr;
113 int error;
b3b94faa 114
c968f578
AG
115 tr = kmem_cache_zalloc(gfs2_trans_cachep, GFP_NOFS);
116 if (!tr)
117 return -ENOMEM;
118 error = __gfs2_trans_begin(tr, sdp, blocks, revokes, _RET_IP_);
119 if (error)
120 kmem_cache_free(gfs2_trans_cachep, tr);
b3b94faa
DT
121 return error;
122}
123
124void gfs2_trans_end(struct gfs2_sbd *sdp)
125{
f4154ea0 126 struct gfs2_trans *tr = current->journal_info;
c50b91c4 127 s64 nbuf;
2e60d768 128
5c676f6d 129 current->journal_info = NULL;
b3b94faa 130
9862ca05 131 if (!test_bit(TR_TOUCHED, &tr->tr_flags)) {
2129b428 132 gfs2_log_release_revokes(sdp, tr->tr_revokes);
c1eba1b0 133 up_read(&sdp->sd_log_flush_lock);
2129b428 134 gfs2_log_release(sdp, tr->tr_reserved);
c968f578 135 if (!test_bit(TR_ONSTACK, &tr->tr_flags))
b839dada 136 gfs2_trans_free(sdp, tr);
15e20a30 137 sb_end_intwrite(sdp->sd_vfs);
b3b94faa
DT
138 return;
139 }
140
2129b428
AG
141 gfs2_log_release_revokes(sdp, tr->tr_revokes - tr->tr_num_revoke);
142
c50b91c4
SW
143 nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new;
144 nbuf -= tr->tr_num_buf_rm;
145 nbuf -= tr->tr_num_databuf_rm;
146
625a8edd
AG
147 if (gfs2_assert_withdraw(sdp, nbuf <= tr->tr_blocks) ||
148 gfs2_assert_withdraw(sdp, tr->tr_num_revoke <= tr->tr_revokes))
e54c78a2 149 gfs2_print_trans(sdp, tr);
b3b94faa
DT
150
151 gfs2_log_commit(sdp, tr);
c968f578
AG
152 if (!test_bit(TR_ONSTACK, &tr->tr_flags) &&
153 !test_bit(TR_ATTACHED, &tr->tr_flags))
b839dada 154 gfs2_trans_free(sdp, tr);
16ca9412 155 up_read(&sdp->sd_log_flush_lock);
b3b94faa 156
1751e8a6 157 if (sdp->sd_vfs->s_flags & SB_SYNCHRONOUS)
805c0907
BP
158 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
159 GFS2_LFC_TRANS_END);
15e20a30 160 sb_end_intwrite(sdp->sd_vfs);
b3b94faa
DT
161}
162
c76c4d96 163static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
cbbe76c8 164 struct buffer_head *bh)
c76c4d96
SW
165{
166 struct gfs2_bufdata *bd;
167
168 bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
169 bd->bd_bh = bh;
170 bd->bd_gl = gl;
c76c4d96 171 INIT_LIST_HEAD(&bd->bd_list);
1a5a2cfd
BP
172 INIT_LIST_HEAD(&bd->bd_ail_st_list);
173 INIT_LIST_HEAD(&bd->bd_ail_gl_list);
c76c4d96
SW
174 bh->b_private = bd;
175 return bd;
176}
177
b3b94faa 178/**
45138990
SW
179 * gfs2_trans_add_data - Add a databuf to the transaction.
180 * @gl: The inode glock associated with the buffer
181 * @bh: The buffer to add
b3b94faa 182 *
845802b1
AG
183 * This is used in journaled data mode.
184 * We need to journal the data block in the same way as metadata in
185 * the functions above. The difference is that here we have a tag
186 * which is two __be64's being the block number (as per meta data)
187 * and a flag which says whether the data block needs escaping or
188 * not. This means we need a new log entry for each 251 or so data
189 * blocks, which isn't an enormous overhead but twice as much as
190 * for normal metadata blocks.
b3b94faa 191 */
45138990 192void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
767f433f
SW
193{
194 struct gfs2_trans *tr = current->journal_info;
15562c43 195 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
45138990 196 struct gfs2_bufdata *bd;
767f433f 197
96e5d1d3 198 lock_buffer(bh);
aacee720
BP
199 if (buffer_pinned(bh)) {
200 set_bit(TR_TOUCHED, &tr->tr_flags);
201 goto out;
202 }
96e5d1d3 203 gfs2_log_lock(sdp);
5c676f6d 204 bd = bh->b_private;
c76c4d96 205 if (bd == NULL) {
96e5d1d3
BM
206 gfs2_log_unlock(sdp);
207 unlock_buffer(bh);
c76c4d96 208 if (bh->b_private == NULL)
cbbe76c8 209 bd = gfs2_alloc_bufdata(gl, bh);
491e94f7
BP
210 else
211 bd = bh->b_private;
96e5d1d3
BM
212 lock_buffer(bh);
213 gfs2_log_lock(sdp);
b3b94faa 214 }
c76c4d96 215 gfs2_assert(sdp, bd->bd_gl == gl);
9862ca05 216 set_bit(TR_TOUCHED, &tr->tr_flags);
45138990
SW
217 if (list_empty(&bd->bd_list)) {
218 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
219 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
220 gfs2_pin(sdp, bd->bd_bh);
221 tr->tr_num_databuf_new++;
d69a3c65 222 list_add_tail(&bd->bd_list, &tr->tr_databuf);
45138990 223 }
96e5d1d3 224 gfs2_log_unlock(sdp);
aacee720 225out:
96e5d1d3 226 unlock_buffer(bh);
b3b94faa
DT
227}
228
192738b7 229void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
350a9b0a 230{
192738b7
BP
231
232 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2cbd8064 233 struct super_block *sb = sdp->sd_vfs;
192738b7 234 struct gfs2_bufdata *bd;
767f433f 235 struct gfs2_meta_header *mh;
aacee720 236 struct gfs2_trans *tr = current->journal_info;
2cbd8064 237 bool withdraw = false;
767f433f 238
192738b7 239 lock_buffer(bh);
aacee720
BP
240 if (buffer_pinned(bh)) {
241 set_bit(TR_TOUCHED, &tr->tr_flags);
242 goto out;
243 }
192738b7
BP
244 gfs2_log_lock(sdp);
245 bd = bh->b_private;
246 if (bd == NULL) {
247 gfs2_log_unlock(sdp);
248 unlock_buffer(bh);
249 lock_page(bh->b_page);
250 if (bh->b_private == NULL)
cbbe76c8 251 bd = gfs2_alloc_bufdata(gl, bh);
192738b7
BP
252 else
253 bd = bh->b_private;
254 unlock_page(bh->b_page);
255 lock_buffer(bh);
256 gfs2_log_lock(sdp);
257 }
258 gfs2_assert(sdp, bd->bd_gl == gl);
9862ca05 259 set_bit(TR_TOUCHED, &tr->tr_flags);
767f433f 260 if (!list_empty(&bd->bd_list))
192738b7 261 goto out_unlock;
767f433f
SW
262 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
263 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
264 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
265 if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) {
e54c78a2
BP
266 fs_err(sdp, "Attempting to add uninitialised block to "
267 "journal (inplace block=%lld)\n",
767f433f
SW
268 (unsigned long long)bd->bd_bh->b_blocknr);
269 BUG();
270 }
4d927b03 271 if (gfs2_withdrawing_or_withdrawn(sdp)) {
2ca0c2fb
BP
272 fs_info(sdp, "GFS2:adding buf while withdrawn! 0x%llx\n",
273 (unsigned long long)bd->bd_bh->b_blocknr);
2cbd8064
AG
274 goto out_unlock;
275 }
276 if (unlikely(sb->s_writers.frozen == SB_FREEZE_COMPLETE)) {
277 fs_info(sdp, "GFS2:adding buf while frozen\n");
278 withdraw = true;
279 goto out_unlock;
2ca0c2fb 280 }
767f433f
SW
281 gfs2_pin(sdp, bd->bd_bh);
282 mh->__pad0 = cpu_to_be64(0);
283 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
d69a3c65 284 list_add(&bd->bd_list, &tr->tr_buf);
767f433f 285 tr->tr_num_buf_new++;
192738b7 286out_unlock:
767f433f 287 gfs2_log_unlock(sdp);
2cbd8064
AG
288 if (withdraw)
289 gfs2_assert_withdraw(sdp, 0);
aacee720 290out:
767f433f 291 unlock_buffer(bh);
350a9b0a
SW
292}
293
1ad38c43 294void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
b3b94faa 295{
75f2b879
SW
296 struct gfs2_trans *tr = current->journal_info;
297
c0752aa7 298 BUG_ON(!list_empty(&bd->bd_list));
5d054964 299 gfs2_add_revoke(sdp, bd);
9862ca05 300 set_bit(TR_TOUCHED, &tr->tr_flags);
75f2b879 301 tr->tr_num_revoke++;
b3b94faa
DT
302}
303
fbb27873 304void gfs2_trans_remove_revoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len)
b3b94faa 305{
5731be53 306 struct gfs2_bufdata *bd, *tmp;
5731be53 307 unsigned int n = len;
b3b94faa
DT
308
309 gfs2_log_lock(sdp);
a5b1d3fc 310 list_for_each_entry_safe(bd, tmp, &sdp->sd_log_revokes, bd_list) {
5731be53 311 if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) {
c0752aa7 312 list_del_init(&bd->bd_list);
b3b94faa
DT
313 gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke);
314 sdp->sd_log_num_revoke--;
fe5e7ba1
BP
315 if (bd->bd_gl)
316 gfs2_glock_remove_revoke(bd->bd_gl);
5731be53 317 kmem_cache_free(gfs2_bufdata_cachep, bd);
2129b428 318 gfs2_log_release_revokes(sdp, 1);
5731be53
SW
319 if (--n == 0)
320 break;
b3b94faa
DT
321 }
322 }
b3b94faa 323 gfs2_log_unlock(sdp);
b3b94faa
DT
324}
325
b839dada
BP
326void gfs2_trans_free(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
327{
328 if (tr == NULL)
329 return;
330
331 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
332 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
333 gfs2_assert_warn(sdp, list_empty(&tr->tr_databuf));
334 gfs2_assert_warn(sdp, list_empty(&tr->tr_buf));
335 kmem_cache_free(gfs2_trans_cachep, tr);
336}