1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. |
4 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
5 | */ |
6 | |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
8 | |
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> |
14 | #include <linux/kallsyms.h> |
15 | #include <linux/gfs2_ondisk.h> |
16 | |
17 | #include "gfs2.h" |
18 | #include "incore.h" |
19 | #include "glock.h" |
20 | #include "inode.h" |
21 | #include "log.h" |
22 | #include "lops.h" |
23 | #include "meta_io.h" |
24 | #include "trans.h" |
25 | #include "util.h" |
26 | #include "trace_gfs2.h" |
27 | |
28 | static 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)); |
34 | fs_warn(sdp, "Buf %u/%u Databuf %u/%u Revoke %u\n" , |
35 | tr->tr_num_buf_new, tr->tr_num_buf_rm, |
36 | tr->tr_num_databuf_new, tr->tr_num_databuf_rm, |
37 | tr->tr_num_revoke); |
38 | } |
39 | |
40 | int __gfs2_trans_begin(struct gfs2_trans *tr, struct gfs2_sbd *sdp, |
41 | unsigned int blocks, unsigned int revokes, |
42 | unsigned long ip) |
43 | { |
44 | unsigned int ; |
45 | |
46 | if (current->journal_info) { |
47 | gfs2_print_trans(sdp, current->journal_info); |
48 | BUG(); |
49 | } |
50 | BUG_ON(blocks == 0 && revokes == 0); |
51 | |
52 | if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) |
53 | return -EROFS; |
54 | |
55 | tr->tr_ip = ip; |
56 | tr->tr_blocks = blocks; |
57 | tr->tr_revokes = revokes; |
58 | tr->tr_reserved = GFS2_LOG_FLUSH_MIN_BLOCKS; |
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 | } |
67 | INIT_LIST_HEAD(list: &tr->tr_databuf); |
68 | INIT_LIST_HEAD(list: &tr->tr_buf); |
69 | INIT_LIST_HEAD(list: &tr->tr_list); |
70 | INIT_LIST_HEAD(list: &tr->tr_ail1_list); |
71 | INIT_LIST_HEAD(list: &tr->tr_ail2_list); |
72 | |
73 | if (gfs2_assert_warn(sdp, tr->tr_reserved <= sdp->sd_jdesc->jd_blocks)) |
74 | return -EINVAL; |
75 | |
76 | sb_start_intwrite(sb: sdp->sd_vfs); |
77 | |
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 | */ |
86 | |
87 | down_read(sem: &sdp->sd_log_flush_lock); |
88 | if (gfs2_log_try_reserve(sdp, tr, extra_revokes: &extra_revokes)) |
89 | goto reserved; |
90 | up_read(sem: &sdp->sd_log_flush_lock); |
91 | gfs2_log_reserve(sdp, tr, extra_revokes: &extra_revokes); |
92 | down_read(sem: &sdp->sd_log_flush_lock); |
93 | |
94 | reserved: |
95 | gfs2_log_release_revokes(sdp, revokes: extra_revokes); |
96 | if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) { |
97 | gfs2_log_release_revokes(sdp, revokes: tr->tr_revokes); |
98 | up_read(sem: &sdp->sd_log_flush_lock); |
99 | gfs2_log_release(sdp, blks: tr->tr_reserved); |
100 | sb_end_intwrite(sb: sdp->sd_vfs); |
101 | return -EROFS; |
102 | } |
103 | |
104 | current->journal_info = tr; |
105 | |
106 | return 0; |
107 | } |
108 | |
109 | int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks, |
110 | unsigned int revokes) |
111 | { |
112 | struct gfs2_trans *tr; |
113 | int error; |
114 | |
115 | tr = kmem_cache_zalloc(k: 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(s: gfs2_trans_cachep, objp: tr); |
121 | return error; |
122 | } |
123 | |
124 | void gfs2_trans_end(struct gfs2_sbd *sdp) |
125 | { |
126 | struct gfs2_trans *tr = current->journal_info; |
127 | s64 nbuf; |
128 | |
129 | current->journal_info = NULL; |
130 | |
131 | if (!test_bit(TR_TOUCHED, &tr->tr_flags)) { |
132 | gfs2_log_release_revokes(sdp, revokes: tr->tr_revokes); |
133 | up_read(sem: &sdp->sd_log_flush_lock); |
134 | gfs2_log_release(sdp, blks: tr->tr_reserved); |
135 | if (!test_bit(TR_ONSTACK, &tr->tr_flags)) |
136 | gfs2_trans_free(sdp, tr); |
137 | sb_end_intwrite(sb: sdp->sd_vfs); |
138 | return; |
139 | } |
140 | |
141 | gfs2_log_release_revokes(sdp, revokes: tr->tr_revokes - tr->tr_num_revoke); |
142 | |
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 | |
147 | if (gfs2_assert_withdraw(sdp, nbuf <= tr->tr_blocks) || |
148 | gfs2_assert_withdraw(sdp, tr->tr_num_revoke <= tr->tr_revokes)) |
149 | gfs2_print_trans(sdp, tr); |
150 | |
151 | gfs2_log_commit(sdp, trans: tr); |
152 | if (!test_bit(TR_ONSTACK, &tr->tr_flags) && |
153 | !test_bit(TR_ATTACHED, &tr->tr_flags)) |
154 | gfs2_trans_free(sdp, tr); |
155 | up_read(sem: &sdp->sd_log_flush_lock); |
156 | |
157 | if (sdp->sd_vfs->s_flags & SB_SYNCHRONOUS) |
158 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
159 | GFS2_LFC_TRANS_END); |
160 | sb_end_intwrite(sb: sdp->sd_vfs); |
161 | } |
162 | |
163 | static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl, |
164 | struct buffer_head *bh) |
165 | { |
166 | struct gfs2_bufdata *bd; |
167 | |
168 | bd = kmem_cache_zalloc(k: gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); |
169 | bd->bd_bh = bh; |
170 | bd->bd_gl = gl; |
171 | INIT_LIST_HEAD(list: &bd->bd_list); |
172 | INIT_LIST_HEAD(list: &bd->bd_ail_st_list); |
173 | INIT_LIST_HEAD(list: &bd->bd_ail_gl_list); |
174 | bh->b_private = bd; |
175 | return bd; |
176 | } |
177 | |
178 | /** |
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 |
182 | * |
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. |
191 | */ |
192 | void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh) |
193 | { |
194 | struct gfs2_trans *tr = current->journal_info; |
195 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
196 | struct gfs2_bufdata *bd; |
197 | |
198 | lock_buffer(bh); |
199 | if (buffer_pinned(bh)) { |
200 | set_bit(nr: TR_TOUCHED, addr: &tr->tr_flags); |
201 | goto out; |
202 | } |
203 | gfs2_log_lock(sdp); |
204 | bd = bh->b_private; |
205 | if (bd == NULL) { |
206 | gfs2_log_unlock(sdp); |
207 | unlock_buffer(bh); |
208 | if (bh->b_private == NULL) |
209 | bd = gfs2_alloc_bufdata(gl, bh); |
210 | else |
211 | bd = bh->b_private; |
212 | lock_buffer(bh); |
213 | gfs2_log_lock(sdp); |
214 | } |
215 | gfs2_assert(sdp, bd->bd_gl == gl); |
216 | set_bit(nr: TR_TOUCHED, addr: &tr->tr_flags); |
217 | if (list_empty(head: &bd->bd_list)) { |
218 | set_bit(nr: GLF_LFLUSH, addr: &bd->bd_gl->gl_flags); |
219 | set_bit(nr: GLF_DIRTY, addr: &bd->bd_gl->gl_flags); |
220 | gfs2_pin(sdp, bh: bd->bd_bh); |
221 | tr->tr_num_databuf_new++; |
222 | list_add_tail(new: &bd->bd_list, head: &tr->tr_databuf); |
223 | } |
224 | gfs2_log_unlock(sdp); |
225 | out: |
226 | unlock_buffer(bh); |
227 | } |
228 | |
229 | void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh) |
230 | { |
231 | |
232 | struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; |
233 | struct super_block *sb = sdp->sd_vfs; |
234 | struct gfs2_bufdata *bd; |
235 | struct gfs2_meta_header *mh; |
236 | struct gfs2_trans *tr = current->journal_info; |
237 | bool withdraw = false; |
238 | |
239 | lock_buffer(bh); |
240 | if (buffer_pinned(bh)) { |
241 | set_bit(nr: TR_TOUCHED, addr: &tr->tr_flags); |
242 | goto out; |
243 | } |
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(page: bh->b_page); |
250 | if (bh->b_private == NULL) |
251 | bd = gfs2_alloc_bufdata(gl, bh); |
252 | else |
253 | bd = bh->b_private; |
254 | unlock_page(page: bh->b_page); |
255 | lock_buffer(bh); |
256 | gfs2_log_lock(sdp); |
257 | } |
258 | gfs2_assert(sdp, bd->bd_gl == gl); |
259 | set_bit(nr: TR_TOUCHED, addr: &tr->tr_flags); |
260 | if (!list_empty(head: &bd->bd_list)) |
261 | goto out_unlock; |
262 | set_bit(nr: GLF_LFLUSH, addr: &bd->bd_gl->gl_flags); |
263 | set_bit(nr: GLF_DIRTY, addr: &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))) { |
266 | fs_err(sdp, "Attempting to add uninitialised block to " |
267 | "journal (inplace block=%lld)\n" , |
268 | (unsigned long long)bd->bd_bh->b_blocknr); |
269 | BUG(); |
270 | } |
271 | if (gfs2_withdrawing_or_withdrawn(sdp)) { |
272 | fs_info(sdp, "GFS2:adding buf while withdrawn! 0x%llx\n" , |
273 | (unsigned long long)bd->bd_bh->b_blocknr); |
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; |
280 | } |
281 | gfs2_pin(sdp, bh: bd->bd_bh); |
282 | mh->__pad0 = cpu_to_be64(0); |
283 | mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); |
284 | list_add(new: &bd->bd_list, head: &tr->tr_buf); |
285 | tr->tr_num_buf_new++; |
286 | out_unlock: |
287 | gfs2_log_unlock(sdp); |
288 | if (withdraw) |
289 | gfs2_assert_withdraw(sdp, 0); |
290 | out: |
291 | unlock_buffer(bh); |
292 | } |
293 | |
294 | void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) |
295 | { |
296 | struct gfs2_trans *tr = current->journal_info; |
297 | |
298 | BUG_ON(!list_empty(&bd->bd_list)); |
299 | gfs2_add_revoke(sdp, bd); |
300 | set_bit(nr: TR_TOUCHED, addr: &tr->tr_flags); |
301 | tr->tr_num_revoke++; |
302 | } |
303 | |
304 | void gfs2_trans_remove_revoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len) |
305 | { |
306 | struct gfs2_bufdata *bd, *tmp; |
307 | unsigned int n = len; |
308 | |
309 | gfs2_log_lock(sdp); |
310 | list_for_each_entry_safe(bd, tmp, &sdp->sd_log_revokes, bd_list) { |
311 | if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) { |
312 | list_del_init(entry: &bd->bd_list); |
313 | gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke); |
314 | sdp->sd_log_num_revoke--; |
315 | if (bd->bd_gl) |
316 | gfs2_glock_remove_revoke(gl: bd->bd_gl); |
317 | kmem_cache_free(s: gfs2_bufdata_cachep, objp: bd); |
318 | gfs2_log_release_revokes(sdp, revokes: 1); |
319 | if (--n == 0) |
320 | break; |
321 | } |
322 | } |
323 | gfs2_log_unlock(sdp); |
324 | } |
325 | |
326 | void 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(s: gfs2_trans_cachep, objp: tr); |
336 | } |
337 | |