Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 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 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
5c676f6d | 15 | #include <linux/gfs2_ondisk.h> |
b3b94faa DT |
16 | |
17 | #include "gfs2.h" | |
5c676f6d | 18 | #include "incore.h" |
2332c443 | 19 | #include "inode.h" |
b3b94faa DT |
20 | #include "glock.h" |
21 | #include "log.h" | |
22 | #include "lops.h" | |
23 | #include "meta_io.h" | |
24 | #include "recovery.h" | |
25 | #include "rgrp.h" | |
26 | #include "trans.h" | |
5c676f6d | 27 | #include "util.h" |
b3b94faa | 28 | |
9b9107a5 SW |
29 | /** |
30 | * gfs2_pin - Pin a buffer in memory | |
31 | * @sdp: The superblock | |
32 | * @bh: The buffer to be pinned | |
33 | * | |
34 | * The log lock must be held when calling this function | |
35 | */ | |
36 | static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh) | |
37 | { | |
38 | struct gfs2_bufdata *bd; | |
39 | ||
40 | gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)); | |
41 | ||
42 | clear_buffer_dirty(bh); | |
43 | if (test_set_buffer_pinned(bh)) | |
44 | gfs2_assert_withdraw(sdp, 0); | |
45 | if (!buffer_uptodate(bh)) | |
46 | gfs2_io_error_bh(sdp, bh); | |
47 | bd = bh->b_private; | |
48 | /* If this buffer is in the AIL and it has already been written | |
49 | * to in-place disk block, remove it from the AIL. | |
50 | */ | |
51 | if (bd->bd_ail) | |
52 | list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list); | |
53 | get_bh(bh); | |
54 | } | |
55 | ||
56 | /** | |
57 | * gfs2_unpin - Unpin a buffer | |
58 | * @sdp: the filesystem the buffer belongs to | |
59 | * @bh: The buffer to unpin | |
60 | * @ai: | |
61 | * | |
62 | */ | |
63 | ||
64 | static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh, | |
65 | struct gfs2_ail *ai) | |
66 | { | |
67 | struct gfs2_bufdata *bd = bh->b_private; | |
68 | ||
69 | gfs2_assert_withdraw(sdp, buffer_uptodate(bh)); | |
70 | ||
71 | if (!buffer_pinned(bh)) | |
72 | gfs2_assert_withdraw(sdp, 0); | |
73 | ||
74 | lock_buffer(bh); | |
75 | mark_buffer_dirty(bh); | |
76 | clear_buffer_pinned(bh); | |
77 | ||
78 | gfs2_log_lock(sdp); | |
79 | if (bd->bd_ail) { | |
80 | list_del(&bd->bd_ail_st_list); | |
81 | brelse(bh); | |
82 | } else { | |
83 | struct gfs2_glock *gl = bd->bd_gl; | |
84 | list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list); | |
85 | atomic_inc(&gl->gl_ail_count); | |
86 | } | |
87 | bd->bd_ail = ai; | |
88 | list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
2bcd610d | 89 | clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); |
9b9107a5 SW |
90 | gfs2_log_unlock(sdp); |
91 | unlock_buffer(bh); | |
92 | } | |
93 | ||
16615be1 SW |
94 | |
95 | static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh) | |
96 | { | |
97 | return (struct gfs2_log_descriptor *)bh->b_data; | |
98 | } | |
99 | ||
100 | static inline __be64 *bh_log_ptr(struct buffer_head *bh) | |
101 | { | |
102 | struct gfs2_log_descriptor *ld = bh_log_desc(bh); | |
103 | return (__force __be64 *)(ld + 1); | |
104 | } | |
105 | ||
106 | static inline __be64 *bh_ptr_end(struct buffer_head *bh) | |
107 | { | |
108 | return (__force __be64 *)(bh->b_data + bh->b_size); | |
109 | } | |
110 | ||
111 | ||
112 | static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type) | |
113 | { | |
114 | struct buffer_head *bh = gfs2_log_get_buf(sdp); | |
115 | struct gfs2_log_descriptor *ld = bh_log_desc(bh); | |
116 | ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
117 | ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); | |
118 | ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); | |
119 | ld->ld_type = cpu_to_be32(ld_type); | |
120 | ld->ld_length = 0; | |
121 | ld->ld_data1 = 0; | |
122 | ld->ld_data2 = 0; | |
123 | memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); | |
124 | return bh; | |
125 | } | |
126 | ||
b3b94faa DT |
127 | static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) |
128 | { | |
129 | struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le); | |
130 | struct gfs2_trans *tr; | |
131 | ||
9b9107a5 | 132 | lock_buffer(bd->bd_bh); |
8bd95727 | 133 | gfs2_log_lock(sdp); |
9b9107a5 SW |
134 | if (!list_empty(&bd->bd_list_tr)) |
135 | goto out; | |
5c676f6d | 136 | tr = current->journal_info; |
b3b94faa DT |
137 | tr->tr_touched = 1; |
138 | tr->tr_num_buf++; | |
139 | list_add(&bd->bd_list_tr, &tr->tr_list_buf); | |
b3b94faa | 140 | if (!list_empty(&le->le_list)) |
9b9107a5 | 141 | goto out; |
2bcd610d SW |
142 | set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); |
143 | set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); | |
b3b94faa | 144 | gfs2_meta_check(sdp, bd->bd_bh); |
a98ab220 | 145 | gfs2_pin(sdp, bd->bd_bh); |
b3b94faa DT |
146 | sdp->sd_log_num_buf++; |
147 | list_add(&le->le_list, &sdp->sd_log_le_buf); | |
b3b94faa | 148 | tr->tr_num_buf_new++; |
9b9107a5 SW |
149 | out: |
150 | gfs2_log_unlock(sdp); | |
151 | unlock_buffer(bd->bd_bh); | |
b3b94faa DT |
152 | } |
153 | ||
b3b94faa DT |
154 | static void buf_lo_before_commit(struct gfs2_sbd *sdp) |
155 | { | |
156 | struct buffer_head *bh; | |
157 | struct gfs2_log_descriptor *ld; | |
158 | struct gfs2_bufdata *bd1 = NULL, *bd2; | |
905d2aef | 159 | unsigned int total; |
b3b94faa DT |
160 | unsigned int limit; |
161 | unsigned int num; | |
162 | unsigned n; | |
163 | __be64 *ptr; | |
164 | ||
2332c443 | 165 | limit = buf_limit(sdp); |
b3b94faa DT |
166 | /* for 4k blocks, limit = 503 */ |
167 | ||
905d2aef BP |
168 | gfs2_log_lock(sdp); |
169 | total = sdp->sd_log_num_buf; | |
b3b94faa DT |
170 | bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list); |
171 | while(total) { | |
172 | num = total; | |
173 | if (total > limit) | |
174 | num = limit; | |
905d2aef | 175 | gfs2_log_unlock(sdp); |
16615be1 | 176 | bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA); |
905d2aef | 177 | gfs2_log_lock(sdp); |
16615be1 SW |
178 | ld = bh_log_desc(bh); |
179 | ptr = bh_log_ptr(bh); | |
b3b94faa DT |
180 | ld->ld_length = cpu_to_be32(num + 1); |
181 | ld->ld_data1 = cpu_to_be32(num); | |
b3b94faa DT |
182 | |
183 | n = 0; | |
568f4c96 SW |
184 | list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf, |
185 | bd_le.le_list) { | |
b3b94faa DT |
186 | *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); |
187 | if (++n >= num) | |
188 | break; | |
189 | } | |
190 | ||
905d2aef | 191 | gfs2_log_unlock(sdp); |
16615be1 | 192 | submit_bh(WRITE, bh); |
905d2aef | 193 | gfs2_log_lock(sdp); |
b3b94faa DT |
194 | |
195 | n = 0; | |
568f4c96 SW |
196 | list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf, |
197 | bd_le.le_list) { | |
16615be1 | 198 | get_bh(bd2->bd_bh); |
905d2aef | 199 | gfs2_log_unlock(sdp); |
16615be1 | 200 | lock_buffer(bd2->bd_bh); |
b3b94faa | 201 | bh = gfs2_log_fake_buf(sdp, bd2->bd_bh); |
16615be1 | 202 | submit_bh(WRITE, bh); |
905d2aef | 203 | gfs2_log_lock(sdp); |
b3b94faa DT |
204 | if (++n >= num) |
205 | break; | |
206 | } | |
207 | ||
905d2aef | 208 | BUG_ON(total < num); |
b3b94faa DT |
209 | total -= num; |
210 | } | |
905d2aef | 211 | gfs2_log_unlock(sdp); |
b3b94faa DT |
212 | } |
213 | ||
214 | static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
215 | { | |
216 | struct list_head *head = &sdp->sd_log_le_buf; | |
217 | struct gfs2_bufdata *bd; | |
218 | ||
219 | while (!list_empty(head)) { | |
220 | bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); | |
221 | list_del_init(&bd->bd_le.le_list); | |
222 | sdp->sd_log_num_buf--; | |
223 | ||
a98ab220 | 224 | gfs2_unpin(sdp, bd->bd_bh, ai); |
b3b94faa DT |
225 | } |
226 | gfs2_assert_warn(sdp, !sdp->sd_log_num_buf); | |
227 | } | |
228 | ||
229 | static void buf_lo_before_scan(struct gfs2_jdesc *jd, | |
55167622 | 230 | struct gfs2_log_header_host *head, int pass) |
b3b94faa | 231 | { |
feaa7bba | 232 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
233 | |
234 | if (pass != 0) | |
235 | return; | |
236 | ||
237 | sdp->sd_found_blocks = 0; | |
238 | sdp->sd_replayed_blocks = 0; | |
239 | } | |
240 | ||
241 | static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
242 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
243 | int pass) | |
244 | { | |
feaa7bba SW |
245 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
246 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
5c676f6d | 247 | struct gfs2_glock *gl = ip->i_gl; |
b3b94faa DT |
248 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
249 | struct buffer_head *bh_log, *bh_ip; | |
cd915493 | 250 | u64 blkno; |
b3b94faa DT |
251 | int error = 0; |
252 | ||
253 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) | |
254 | return 0; | |
255 | ||
256 | gfs2_replay_incr_blk(sdp, &start); | |
257 | ||
258 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
259 | blkno = be64_to_cpu(*ptr++); | |
260 | ||
261 | sdp->sd_found_blocks++; | |
262 | ||
263 | if (gfs2_revoke_check(sdp, blkno, start)) | |
264 | continue; | |
265 | ||
266 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
82ffa516 SW |
267 | if (error) |
268 | return error; | |
b3b94faa DT |
269 | |
270 | bh_ip = gfs2_meta_new(gl, blkno); | |
271 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
272 | ||
273 | if (gfs2_meta_check(sdp, bh_ip)) | |
274 | error = -EIO; | |
275 | else | |
276 | mark_buffer_dirty(bh_ip); | |
277 | ||
278 | brelse(bh_log); | |
279 | brelse(bh_ip); | |
280 | ||
281 | if (error) | |
282 | break; | |
283 | ||
284 | sdp->sd_replayed_blocks++; | |
285 | } | |
286 | ||
287 | return error; | |
288 | } | |
289 | ||
290 | static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
291 | { | |
feaa7bba SW |
292 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
293 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
b3b94faa DT |
294 | |
295 | if (error) { | |
7276b3b0 | 296 | gfs2_meta_sync(ip->i_gl); |
b3b94faa DT |
297 | return; |
298 | } | |
299 | if (pass != 1) | |
300 | return; | |
301 | ||
7276b3b0 | 302 | gfs2_meta_sync(ip->i_gl); |
b3b94faa DT |
303 | |
304 | fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", | |
305 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
306 | } | |
307 | ||
308 | static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
309 | { | |
310 | struct gfs2_trans *tr; | |
311 | ||
5c676f6d | 312 | tr = current->journal_info; |
b3b94faa DT |
313 | tr->tr_touched = 1; |
314 | tr->tr_num_revoke++; | |
b3b94faa DT |
315 | sdp->sd_log_num_revoke++; |
316 | list_add(&le->le_list, &sdp->sd_log_le_revoke); | |
b3b94faa DT |
317 | } |
318 | ||
319 | static void revoke_lo_before_commit(struct gfs2_sbd *sdp) | |
320 | { | |
321 | struct gfs2_log_descriptor *ld; | |
322 | struct gfs2_meta_header *mh; | |
323 | struct buffer_head *bh; | |
324 | unsigned int offset; | |
325 | struct list_head *head = &sdp->sd_log_le_revoke; | |
82e86087 | 326 | struct gfs2_bufdata *bd; |
b3b94faa DT |
327 | |
328 | if (!sdp->sd_log_num_revoke) | |
329 | return; | |
330 | ||
16615be1 SW |
331 | bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE); |
332 | ld = bh_log_desc(bh); | |
568f4c96 | 333 | ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, |
cd915493 | 334 | sizeof(u64))); |
b3b94faa | 335 | ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke); |
b3b94faa DT |
336 | offset = sizeof(struct gfs2_log_descriptor); |
337 | ||
338 | while (!list_empty(head)) { | |
82e86087 SW |
339 | bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); |
340 | list_del_init(&bd->bd_le.le_list); | |
b3b94faa DT |
341 | sdp->sd_log_num_revoke--; |
342 | ||
cd915493 | 343 | if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) { |
16615be1 | 344 | submit_bh(WRITE, bh); |
b3b94faa DT |
345 | |
346 | bh = gfs2_log_get_buf(sdp); | |
347 | mh = (struct gfs2_meta_header *)bh->b_data; | |
348 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
e3167ded SW |
349 | mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB); |
350 | mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB); | |
b3b94faa DT |
351 | offset = sizeof(struct gfs2_meta_header); |
352 | } | |
353 | ||
82e86087 | 354 | *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno); |
0820ab51 | 355 | kmem_cache_free(gfs2_bufdata_cachep, bd); |
b3b94faa | 356 | |
cd915493 | 357 | offset += sizeof(u64); |
b3b94faa DT |
358 | } |
359 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
360 | ||
16615be1 | 361 | submit_bh(WRITE, bh); |
b3b94faa DT |
362 | } |
363 | ||
364 | static void revoke_lo_before_scan(struct gfs2_jdesc *jd, | |
55167622 | 365 | struct gfs2_log_header_host *head, int pass) |
b3b94faa | 366 | { |
feaa7bba | 367 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
368 | |
369 | if (pass != 0) | |
370 | return; | |
371 | ||
372 | sdp->sd_found_revokes = 0; | |
373 | sdp->sd_replay_tail = head->lh_tail; | |
374 | } | |
375 | ||
376 | static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
377 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
378 | int pass) | |
379 | { | |
feaa7bba | 380 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
381 | unsigned int blks = be32_to_cpu(ld->ld_length); |
382 | unsigned int revokes = be32_to_cpu(ld->ld_data1); | |
383 | struct buffer_head *bh; | |
384 | unsigned int offset; | |
cd915493 | 385 | u64 blkno; |
b3b94faa DT |
386 | int first = 1; |
387 | int error; | |
388 | ||
389 | if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) | |
390 | return 0; | |
391 | ||
392 | offset = sizeof(struct gfs2_log_descriptor); | |
393 | ||
394 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
395 | error = gfs2_replay_read_block(jd, start, &bh); | |
396 | if (error) | |
397 | return error; | |
398 | ||
399 | if (!first) | |
400 | gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); | |
401 | ||
cd915493 | 402 | while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) { |
b3b94faa DT |
403 | blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); |
404 | ||
405 | error = gfs2_revoke_add(sdp, blkno, start); | |
3ad62e87 BP |
406 | if (error < 0) { |
407 | brelse(bh); | |
b3b94faa | 408 | return error; |
3ad62e87 | 409 | } |
b3b94faa DT |
410 | else if (error) |
411 | sdp->sd_found_revokes++; | |
412 | ||
413 | if (!--revokes) | |
414 | break; | |
cd915493 | 415 | offset += sizeof(u64); |
b3b94faa DT |
416 | } |
417 | ||
418 | brelse(bh); | |
419 | offset = sizeof(struct gfs2_meta_header); | |
420 | first = 0; | |
421 | } | |
422 | ||
423 | return 0; | |
424 | } | |
425 | ||
426 | static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
427 | { | |
feaa7bba | 428 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
429 | |
430 | if (error) { | |
431 | gfs2_revoke_clean(sdp); | |
432 | return; | |
433 | } | |
434 | if (pass != 1) | |
435 | return; | |
436 | ||
437 | fs_info(sdp, "jid=%u: Found %u revoke tags\n", | |
438 | jd->jd_jid, sdp->sd_found_revokes); | |
439 | ||
440 | gfs2_revoke_clean(sdp); | |
441 | } | |
442 | ||
443 | static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
444 | { | |
445 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 446 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa | 447 | |
5c676f6d | 448 | tr->tr_touched = 1; |
b3b94faa | 449 | |
b3b94faa | 450 | rgd = container_of(le, struct gfs2_rgrpd, rd_le); |
b3b94faa DT |
451 | |
452 | gfs2_log_lock(sdp); | |
68835625 BM |
453 | if (!list_empty(&le->le_list)){ |
454 | gfs2_log_unlock(sdp); | |
455 | return; | |
456 | } | |
457 | gfs2_rgrp_bh_hold(rgd); | |
b3b94faa DT |
458 | sdp->sd_log_num_rg++; |
459 | list_add(&le->le_list, &sdp->sd_log_le_rg); | |
907b9bce | 460 | gfs2_log_unlock(sdp); |
b3b94faa DT |
461 | } |
462 | ||
463 | static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
464 | { | |
465 | struct list_head *head = &sdp->sd_log_le_rg; | |
466 | struct gfs2_rgrpd *rgd; | |
467 | ||
468 | while (!list_empty(head)) { | |
469 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list); | |
470 | list_del_init(&rgd->rd_le.le_list); | |
471 | sdp->sd_log_num_rg--; | |
472 | ||
473 | gfs2_rgrp_repolish_clones(rgd); | |
474 | gfs2_rgrp_bh_put(rgd); | |
475 | } | |
476 | gfs2_assert_warn(sdp, !sdp->sd_log_num_rg); | |
477 | } | |
478 | ||
18ec7d5c SW |
479 | /** |
480 | * databuf_lo_add - Add a databuf to the transaction. | |
481 | * | |
482 | * This is used in two distinct cases: | |
483 | * i) In ordered write mode | |
484 | * We put the data buffer on a list so that we can ensure that its | |
485 | * synced to disk at the right time | |
486 | * ii) In journaled data mode | |
487 | * We need to journal the data block in the same way as metadata in | |
488 | * the functions above. The difference is that here we have a tag | |
489 | * which is two __be64's being the block number (as per meta data) | |
490 | * and a flag which says whether the data block needs escaping or | |
491 | * not. This means we need a new log entry for each 251 or so data | |
492 | * blocks, which isn't an enormous overhead but twice as much as | |
493 | * for normal metadata blocks. | |
494 | */ | |
b3b94faa DT |
495 | static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) |
496 | { | |
18ec7d5c | 497 | struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le); |
5c676f6d | 498 | struct gfs2_trans *tr = current->journal_info; |
18ec7d5c | 499 | struct address_space *mapping = bd->bd_bh->b_page->mapping; |
feaa7bba | 500 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa | 501 | |
9b9107a5 | 502 | lock_buffer(bd->bd_bh); |
8bd95727 | 503 | gfs2_log_lock(sdp); |
9ff8ec32 SW |
504 | if (tr) { |
505 | if (!list_empty(&bd->bd_list_tr)) | |
506 | goto out; | |
507 | tr->tr_touched = 1; | |
508 | if (gfs2_is_jdata(ip)) { | |
509 | tr->tr_num_buf++; | |
510 | list_add(&bd->bd_list_tr, &tr->tr_list_buf); | |
511 | } | |
773ed1a0 | 512 | } |
2332c443 | 513 | if (!list_empty(&le->le_list)) |
9b9107a5 | 514 | goto out; |
2332c443 | 515 | |
2bcd610d SW |
516 | set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); |
517 | set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags); | |
2332c443 | 518 | if (gfs2_is_jdata(ip)) { |
2332c443 RP |
519 | gfs2_pin(sdp, bd->bd_bh); |
520 | tr->tr_num_databuf_new++; | |
d7b616e2 SW |
521 | sdp->sd_log_num_databuf++; |
522 | list_add(&le->le_list, &sdp->sd_log_le_databuf); | |
523 | } else { | |
524 | list_add(&le->le_list, &sdp->sd_log_le_ordered); | |
9b9107a5 | 525 | } |
9b9107a5 | 526 | out: |
b3b94faa | 527 | gfs2_log_unlock(sdp); |
9b9107a5 | 528 | unlock_buffer(bd->bd_bh); |
b3b94faa DT |
529 | } |
530 | ||
16615be1 | 531 | static void gfs2_check_magic(struct buffer_head *bh) |
18ec7d5c | 532 | { |
18ec7d5c SW |
533 | void *kaddr; |
534 | __be32 *ptr; | |
18ec7d5c | 535 | |
16615be1 SW |
536 | clear_buffer_escaped(bh); |
537 | kaddr = kmap_atomic(bh->b_page, KM_USER0); | |
18ec7d5c SW |
538 | ptr = kaddr + bh_offset(bh); |
539 | if (*ptr == cpu_to_be32(GFS2_MAGIC)) | |
16615be1 | 540 | set_buffer_escaped(bh); |
c312c4fd | 541 | kunmap_atomic(kaddr, KM_USER0); |
18ec7d5c SW |
542 | } |
543 | ||
16615be1 SW |
544 | static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh, |
545 | struct list_head *list, struct list_head *done, | |
546 | unsigned int n) | |
b3b94faa | 547 | { |
16615be1 | 548 | struct buffer_head *bh1; |
18ec7d5c | 549 | struct gfs2_log_descriptor *ld; |
16615be1 SW |
550 | struct gfs2_bufdata *bd; |
551 | __be64 *ptr; | |
d7b616e2 | 552 | |
16615be1 SW |
553 | if (!bh) |
554 | return; | |
b3b94faa | 555 | |
16615be1 SW |
556 | ld = bh_log_desc(bh); |
557 | ld->ld_length = cpu_to_be32(n + 1); | |
558 | ld->ld_data1 = cpu_to_be32(n); | |
b3b94faa | 559 | |
16615be1 SW |
560 | ptr = bh_log_ptr(bh); |
561 | ||
562 | get_bh(bh); | |
563 | submit_bh(WRITE, bh); | |
f55ab26a | 564 | gfs2_log_lock(sdp); |
16615be1 SW |
565 | while(!list_empty(list)) { |
566 | bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list); | |
567 | list_move_tail(&bd->bd_le.le_list, done); | |
568 | get_bh(bd->bd_bh); | |
569 | while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) { | |
570 | gfs2_log_incr_head(sdp); | |
571 | ptr += 2; | |
18ec7d5c | 572 | } |
f55ab26a | 573 | gfs2_log_unlock(sdp); |
16615be1 SW |
574 | lock_buffer(bd->bd_bh); |
575 | if (buffer_escaped(bd->bd_bh)) { | |
576 | void *kaddr; | |
577 | bh1 = gfs2_log_get_buf(sdp); | |
578 | kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0); | |
579 | memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh), | |
580 | bh1->b_size); | |
581 | kunmap_atomic(kaddr, KM_USER0); | |
582 | *(__be32 *)bh1->b_data = 0; | |
583 | clear_buffer_escaped(bd->bd_bh); | |
584 | unlock_buffer(bd->bd_bh); | |
585 | brelse(bd->bd_bh); | |
586 | } else { | |
587 | bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh); | |
18ec7d5c | 588 | } |
16615be1 | 589 | submit_bh(WRITE, bh1); |
f55ab26a | 590 | gfs2_log_lock(sdp); |
16615be1 SW |
591 | ptr += 2; |
592 | } | |
593 | gfs2_log_unlock(sdp); | |
594 | brelse(bh); | |
595 | } | |
596 | ||
597 | /** | |
598 | * databuf_lo_before_commit - Scan the data buffers, writing as we go | |
599 | * | |
600 | */ | |
601 | ||
602 | static void databuf_lo_before_commit(struct gfs2_sbd *sdp) | |
603 | { | |
604 | struct gfs2_bufdata *bd = NULL; | |
605 | struct buffer_head *bh = NULL; | |
606 | unsigned int n = 0; | |
607 | __be64 *ptr = NULL, *end = NULL; | |
608 | LIST_HEAD(processed); | |
609 | LIST_HEAD(in_progress); | |
610 | ||
611 | gfs2_log_lock(sdp); | |
612 | while (!list_empty(&sdp->sd_log_le_databuf)) { | |
613 | if (ptr == end) { | |
f55ab26a | 614 | gfs2_log_unlock(sdp); |
16615be1 SW |
615 | gfs2_write_blocks(sdp, bh, &in_progress, &processed, n); |
616 | n = 0; | |
617 | bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA); | |
618 | ptr = bh_log_ptr(bh); | |
619 | end = bh_ptr_end(bh) - 1; | |
f55ab26a | 620 | gfs2_log_lock(sdp); |
16615be1 | 621 | continue; |
18ec7d5c | 622 | } |
16615be1 SW |
623 | bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list); |
624 | list_move_tail(&bd->bd_le.le_list, &in_progress); | |
625 | gfs2_check_magic(bd->bd_bh); | |
626 | *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr); | |
627 | *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0); | |
628 | n++; | |
b3b94faa | 629 | } |
f55ab26a | 630 | gfs2_log_unlock(sdp); |
16615be1 SW |
631 | gfs2_write_blocks(sdp, bh, &in_progress, &processed, n); |
632 | gfs2_log_lock(sdp); | |
633 | list_splice(&processed, &sdp->sd_log_le_databuf); | |
634 | gfs2_log_unlock(sdp); | |
18ec7d5c SW |
635 | } |
636 | ||
637 | static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
638 | struct gfs2_log_descriptor *ld, | |
639 | __be64 *ptr, int pass) | |
640 | { | |
feaa7bba SW |
641 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
642 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
5c676f6d | 643 | struct gfs2_glock *gl = ip->i_gl; |
18ec7d5c SW |
644 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
645 | struct buffer_head *bh_log, *bh_ip; | |
cd915493 SW |
646 | u64 blkno; |
647 | u64 esc; | |
18ec7d5c SW |
648 | int error = 0; |
649 | ||
650 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) | |
651 | return 0; | |
652 | ||
653 | gfs2_replay_incr_blk(sdp, &start); | |
654 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
655 | blkno = be64_to_cpu(*ptr++); | |
656 | esc = be64_to_cpu(*ptr++); | |
657 | ||
658 | sdp->sd_found_blocks++; | |
659 | ||
660 | if (gfs2_revoke_check(sdp, blkno, start)) | |
661 | continue; | |
662 | ||
663 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
664 | if (error) | |
665 | return error; | |
666 | ||
667 | bh_ip = gfs2_meta_new(gl, blkno); | |
668 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
669 | ||
670 | /* Unescape */ | |
671 | if (esc) { | |
672 | __be32 *eptr = (__be32 *)bh_ip->b_data; | |
673 | *eptr = cpu_to_be32(GFS2_MAGIC); | |
674 | } | |
675 | mark_buffer_dirty(bh_ip); | |
676 | ||
677 | brelse(bh_log); | |
678 | brelse(bh_ip); | |
679 | if (error) | |
680 | break; | |
681 | ||
682 | sdp->sd_replayed_blocks++; | |
683 | } | |
684 | ||
685 | return error; | |
686 | } | |
687 | ||
688 | /* FIXME: sort out accounting for log blocks etc. */ | |
689 | ||
690 | static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
691 | { | |
feaa7bba SW |
692 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
693 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
18ec7d5c SW |
694 | |
695 | if (error) { | |
7276b3b0 | 696 | gfs2_meta_sync(ip->i_gl); |
18ec7d5c SW |
697 | return; |
698 | } | |
699 | if (pass != 1) | |
700 | return; | |
701 | ||
702 | /* data sync? */ | |
7276b3b0 | 703 | gfs2_meta_sync(ip->i_gl); |
18ec7d5c SW |
704 | |
705 | fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", | |
706 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
707 | } | |
708 | ||
709 | static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
710 | { | |
711 | struct list_head *head = &sdp->sd_log_le_databuf; | |
712 | struct gfs2_bufdata *bd; | |
713 | ||
714 | while (!list_empty(head)) { | |
715 | bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); | |
b8e1aabf | 716 | list_del_init(&bd->bd_le.le_list); |
18ec7d5c | 717 | sdp->sd_log_num_databuf--; |
18ec7d5c | 718 | gfs2_unpin(sdp, bd->bd_bh, ai); |
18ec7d5c | 719 | } |
b3b94faa DT |
720 | gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf); |
721 | } | |
722 | ||
18ec7d5c | 723 | |
b09e593d | 724 | const struct gfs2_log_operations gfs2_buf_lops = { |
b3b94faa | 725 | .lo_add = buf_lo_add, |
b3b94faa DT |
726 | .lo_before_commit = buf_lo_before_commit, |
727 | .lo_after_commit = buf_lo_after_commit, | |
728 | .lo_before_scan = buf_lo_before_scan, | |
729 | .lo_scan_elements = buf_lo_scan_elements, | |
730 | .lo_after_scan = buf_lo_after_scan, | |
ea67eedb | 731 | .lo_name = "buf", |
b3b94faa DT |
732 | }; |
733 | ||
b09e593d | 734 | const struct gfs2_log_operations gfs2_revoke_lops = { |
b3b94faa DT |
735 | .lo_add = revoke_lo_add, |
736 | .lo_before_commit = revoke_lo_before_commit, | |
737 | .lo_before_scan = revoke_lo_before_scan, | |
738 | .lo_scan_elements = revoke_lo_scan_elements, | |
739 | .lo_after_scan = revoke_lo_after_scan, | |
ea67eedb | 740 | .lo_name = "revoke", |
b3b94faa DT |
741 | }; |
742 | ||
b09e593d | 743 | const struct gfs2_log_operations gfs2_rg_lops = { |
b3b94faa DT |
744 | .lo_add = rg_lo_add, |
745 | .lo_after_commit = rg_lo_after_commit, | |
ea67eedb | 746 | .lo_name = "rg", |
b3b94faa DT |
747 | }; |
748 | ||
b09e593d | 749 | const struct gfs2_log_operations gfs2_databuf_lops = { |
b3b94faa DT |
750 | .lo_add = databuf_lo_add, |
751 | .lo_before_commit = databuf_lo_before_commit, | |
18ec7d5c SW |
752 | .lo_after_commit = databuf_lo_after_commit, |
753 | .lo_scan_elements = databuf_lo_scan_elements, | |
754 | .lo_after_scan = databuf_lo_after_scan, | |
ea67eedb | 755 | .lo_name = "databuf", |
b3b94faa DT |
756 | }; |
757 | ||
b09e593d | 758 | const struct gfs2_log_operations *gfs2_log_ops[] = { |
16615be1 | 759 | &gfs2_databuf_lops, |
b3b94faa | 760 | &gfs2_buf_lops, |
b3b94faa | 761 | &gfs2_rg_lops, |
16615be1 | 762 | &gfs2_revoke_lops, |
ea67eedb | 763 | NULL, |
b3b94faa DT |
764 | }; |
765 |