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> |
7d308590 | 16 | #include <linux/lm_interface.h> |
b3b94faa DT |
17 | |
18 | #include "gfs2.h" | |
5c676f6d | 19 | #include "incore.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 DT |
28 | |
29 | static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
30 | { | |
31 | struct gfs2_glock *gl; | |
5c676f6d | 32 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa | 33 | |
5c676f6d | 34 | tr->tr_touched = 1; |
b3b94faa DT |
35 | |
36 | if (!list_empty(&le->le_list)) | |
37 | return; | |
38 | ||
39 | gl = container_of(le, struct gfs2_glock, gl_le); | |
40 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl))) | |
41 | return; | |
42 | gfs2_glock_hold(gl); | |
43 | set_bit(GLF_DIRTY, &gl->gl_flags); | |
44 | ||
45 | gfs2_log_lock(sdp); | |
46 | sdp->sd_log_num_gl++; | |
47 | list_add(&le->le_list, &sdp->sd_log_le_gl); | |
48 | gfs2_log_unlock(sdp); | |
49 | } | |
50 | ||
51 | static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
52 | { | |
53 | struct list_head *head = &sdp->sd_log_le_gl; | |
54 | struct gfs2_glock *gl; | |
55 | ||
56 | while (!list_empty(head)) { | |
57 | gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list); | |
58 | list_del_init(&gl->gl_le.le_list); | |
59 | sdp->sd_log_num_gl--; | |
60 | ||
61 | gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)); | |
62 | gfs2_glock_put(gl); | |
63 | } | |
64 | gfs2_assert_warn(sdp, !sdp->sd_log_num_gl); | |
65 | } | |
66 | ||
67 | static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
68 | { | |
69 | struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le); | |
70 | struct gfs2_trans *tr; | |
71 | ||
72 | if (!list_empty(&bd->bd_list_tr)) | |
73 | return; | |
74 | ||
5c676f6d | 75 | tr = current->journal_info; |
b3b94faa DT |
76 | tr->tr_touched = 1; |
77 | tr->tr_num_buf++; | |
78 | list_add(&bd->bd_list_tr, &tr->tr_list_buf); | |
79 | ||
80 | if (!list_empty(&le->le_list)) | |
81 | return; | |
82 | ||
83 | gfs2_trans_add_gl(bd->bd_gl); | |
84 | ||
85 | gfs2_meta_check(sdp, bd->bd_bh); | |
a98ab220 | 86 | gfs2_pin(sdp, bd->bd_bh); |
b3b94faa DT |
87 | |
88 | gfs2_log_lock(sdp); | |
89 | sdp->sd_log_num_buf++; | |
90 | list_add(&le->le_list, &sdp->sd_log_le_buf); | |
91 | gfs2_log_unlock(sdp); | |
92 | ||
93 | tr->tr_num_buf_new++; | |
94 | } | |
95 | ||
96 | static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
97 | { | |
98 | struct list_head *head = &tr->tr_list_buf; | |
99 | struct gfs2_bufdata *bd; | |
100 | ||
101 | while (!list_empty(head)) { | |
102 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr); | |
103 | list_del_init(&bd->bd_list_tr); | |
104 | tr->tr_num_buf--; | |
105 | } | |
106 | gfs2_assert_warn(sdp, !tr->tr_num_buf); | |
107 | } | |
108 | ||
109 | static void buf_lo_before_commit(struct gfs2_sbd *sdp) | |
110 | { | |
111 | struct buffer_head *bh; | |
112 | struct gfs2_log_descriptor *ld; | |
113 | struct gfs2_bufdata *bd1 = NULL, *bd2; | |
114 | unsigned int total = sdp->sd_log_num_buf; | |
115 | unsigned int offset = sizeof(struct gfs2_log_descriptor); | |
116 | unsigned int limit; | |
117 | unsigned int num; | |
118 | unsigned n; | |
119 | __be64 *ptr; | |
120 | ||
82ffa516 | 121 | offset += sizeof(__be64) - 1; |
b3b94faa DT |
122 | offset &= ~(sizeof(__be64) - 1); |
123 | limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64); | |
124 | /* for 4k blocks, limit = 503 */ | |
125 | ||
126 | bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list); | |
127 | while(total) { | |
128 | num = total; | |
129 | if (total > limit) | |
130 | num = limit; | |
131 | bh = gfs2_log_get_buf(sdp); | |
b09e593d | 132 | sdp->sd_log_num_hdrs++; |
b3b94faa DT |
133 | ld = (struct gfs2_log_descriptor *)bh->b_data; |
134 | ptr = (__be64 *)(bh->b_data + offset); | |
135 | ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
e3167ded SW |
136 | ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); |
137 | ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); | |
b3b94faa DT |
138 | ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA); |
139 | ld->ld_length = cpu_to_be32(num + 1); | |
140 | ld->ld_data1 = cpu_to_be32(num); | |
141 | ld->ld_data2 = cpu_to_be32(0); | |
142 | memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); | |
143 | ||
144 | n = 0; | |
568f4c96 SW |
145 | list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf, |
146 | bd_le.le_list) { | |
b3b94faa DT |
147 | *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); |
148 | if (++n >= num) | |
149 | break; | |
150 | } | |
151 | ||
152 | set_buffer_dirty(bh); | |
153 | ll_rw_block(WRITE, 1, &bh); | |
154 | ||
155 | n = 0; | |
568f4c96 SW |
156 | list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf, |
157 | bd_le.le_list) { | |
b3b94faa DT |
158 | bh = gfs2_log_fake_buf(sdp, bd2->bd_bh); |
159 | set_buffer_dirty(bh); | |
160 | ll_rw_block(WRITE, 1, &bh); | |
161 | if (++n >= num) | |
162 | break; | |
163 | } | |
164 | ||
165 | total -= num; | |
166 | } | |
167 | } | |
168 | ||
169 | static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
170 | { | |
171 | struct list_head *head = &sdp->sd_log_le_buf; | |
172 | struct gfs2_bufdata *bd; | |
173 | ||
174 | while (!list_empty(head)) { | |
175 | bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); | |
176 | list_del_init(&bd->bd_le.le_list); | |
177 | sdp->sd_log_num_buf--; | |
178 | ||
a98ab220 | 179 | gfs2_unpin(sdp, bd->bd_bh, ai); |
b3b94faa DT |
180 | } |
181 | gfs2_assert_warn(sdp, !sdp->sd_log_num_buf); | |
182 | } | |
183 | ||
184 | static void buf_lo_before_scan(struct gfs2_jdesc *jd, | |
55167622 | 185 | struct gfs2_log_header_host *head, int pass) |
b3b94faa | 186 | { |
feaa7bba | 187 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
188 | |
189 | if (pass != 0) | |
190 | return; | |
191 | ||
192 | sdp->sd_found_blocks = 0; | |
193 | sdp->sd_replayed_blocks = 0; | |
194 | } | |
195 | ||
196 | static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
197 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
198 | int pass) | |
199 | { | |
feaa7bba SW |
200 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
201 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
5c676f6d | 202 | struct gfs2_glock *gl = ip->i_gl; |
b3b94faa DT |
203 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
204 | struct buffer_head *bh_log, *bh_ip; | |
cd915493 | 205 | u64 blkno; |
b3b94faa DT |
206 | int error = 0; |
207 | ||
208 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) | |
209 | return 0; | |
210 | ||
211 | gfs2_replay_incr_blk(sdp, &start); | |
212 | ||
213 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
214 | blkno = be64_to_cpu(*ptr++); | |
215 | ||
216 | sdp->sd_found_blocks++; | |
217 | ||
218 | if (gfs2_revoke_check(sdp, blkno, start)) | |
219 | continue; | |
220 | ||
221 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
82ffa516 SW |
222 | if (error) |
223 | return error; | |
b3b94faa DT |
224 | |
225 | bh_ip = gfs2_meta_new(gl, blkno); | |
226 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
227 | ||
228 | if (gfs2_meta_check(sdp, bh_ip)) | |
229 | error = -EIO; | |
230 | else | |
231 | mark_buffer_dirty(bh_ip); | |
232 | ||
233 | brelse(bh_log); | |
234 | brelse(bh_ip); | |
235 | ||
236 | if (error) | |
237 | break; | |
238 | ||
239 | sdp->sd_replayed_blocks++; | |
240 | } | |
241 | ||
242 | return error; | |
243 | } | |
244 | ||
245 | static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
246 | { | |
feaa7bba SW |
247 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
248 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
b3b94faa DT |
249 | |
250 | if (error) { | |
7276b3b0 | 251 | gfs2_meta_sync(ip->i_gl); |
b3b94faa DT |
252 | return; |
253 | } | |
254 | if (pass != 1) | |
255 | return; | |
256 | ||
7276b3b0 | 257 | gfs2_meta_sync(ip->i_gl); |
b3b94faa DT |
258 | |
259 | fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", | |
260 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
261 | } | |
262 | ||
263 | static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
264 | { | |
265 | struct gfs2_trans *tr; | |
266 | ||
5c676f6d | 267 | tr = current->journal_info; |
b3b94faa DT |
268 | tr->tr_touched = 1; |
269 | tr->tr_num_revoke++; | |
270 | ||
271 | gfs2_log_lock(sdp); | |
272 | sdp->sd_log_num_revoke++; | |
273 | list_add(&le->le_list, &sdp->sd_log_le_revoke); | |
274 | gfs2_log_unlock(sdp); | |
275 | } | |
276 | ||
277 | static void revoke_lo_before_commit(struct gfs2_sbd *sdp) | |
278 | { | |
279 | struct gfs2_log_descriptor *ld; | |
280 | struct gfs2_meta_header *mh; | |
281 | struct buffer_head *bh; | |
282 | unsigned int offset; | |
283 | struct list_head *head = &sdp->sd_log_le_revoke; | |
284 | struct gfs2_revoke *rv; | |
285 | ||
286 | if (!sdp->sd_log_num_revoke) | |
287 | return; | |
288 | ||
289 | bh = gfs2_log_get_buf(sdp); | |
290 | ld = (struct gfs2_log_descriptor *)bh->b_data; | |
291 | ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
e3167ded SW |
292 | ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); |
293 | ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); | |
b3b94faa | 294 | ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE); |
568f4c96 | 295 | ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, |
cd915493 | 296 | sizeof(u64))); |
b3b94faa DT |
297 | ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke); |
298 | ld->ld_data2 = cpu_to_be32(0); | |
299 | memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); | |
300 | offset = sizeof(struct gfs2_log_descriptor); | |
301 | ||
302 | while (!list_empty(head)) { | |
303 | rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list); | |
13538b8e | 304 | list_del_init(&rv->rv_le.le_list); |
b3b94faa DT |
305 | sdp->sd_log_num_revoke--; |
306 | ||
cd915493 | 307 | if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) { |
b3b94faa DT |
308 | set_buffer_dirty(bh); |
309 | ll_rw_block(WRITE, 1, &bh); | |
310 | ||
311 | bh = gfs2_log_get_buf(sdp); | |
312 | mh = (struct gfs2_meta_header *)bh->b_data; | |
313 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
e3167ded SW |
314 | mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB); |
315 | mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB); | |
b3b94faa DT |
316 | offset = sizeof(struct gfs2_meta_header); |
317 | } | |
318 | ||
319 | *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno); | |
320 | kfree(rv); | |
321 | ||
cd915493 | 322 | offset += sizeof(u64); |
b3b94faa DT |
323 | } |
324 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
325 | ||
326 | set_buffer_dirty(bh); | |
327 | ll_rw_block(WRITE, 1, &bh); | |
328 | } | |
329 | ||
330 | static void revoke_lo_before_scan(struct gfs2_jdesc *jd, | |
55167622 | 331 | struct gfs2_log_header_host *head, int pass) |
b3b94faa | 332 | { |
feaa7bba | 333 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
334 | |
335 | if (pass != 0) | |
336 | return; | |
337 | ||
338 | sdp->sd_found_revokes = 0; | |
339 | sdp->sd_replay_tail = head->lh_tail; | |
340 | } | |
341 | ||
342 | static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
343 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
344 | int pass) | |
345 | { | |
feaa7bba | 346 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
347 | unsigned int blks = be32_to_cpu(ld->ld_length); |
348 | unsigned int revokes = be32_to_cpu(ld->ld_data1); | |
349 | struct buffer_head *bh; | |
350 | unsigned int offset; | |
cd915493 | 351 | u64 blkno; |
b3b94faa DT |
352 | int first = 1; |
353 | int error; | |
354 | ||
355 | if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) | |
356 | return 0; | |
357 | ||
358 | offset = sizeof(struct gfs2_log_descriptor); | |
359 | ||
360 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
361 | error = gfs2_replay_read_block(jd, start, &bh); | |
362 | if (error) | |
363 | return error; | |
364 | ||
365 | if (!first) | |
366 | gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); | |
367 | ||
cd915493 | 368 | while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) { |
b3b94faa DT |
369 | blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); |
370 | ||
371 | error = gfs2_revoke_add(sdp, blkno, start); | |
372 | if (error < 0) | |
373 | return error; | |
374 | else if (error) | |
375 | sdp->sd_found_revokes++; | |
376 | ||
377 | if (!--revokes) | |
378 | break; | |
cd915493 | 379 | offset += sizeof(u64); |
b3b94faa DT |
380 | } |
381 | ||
382 | brelse(bh); | |
383 | offset = sizeof(struct gfs2_meta_header); | |
384 | first = 0; | |
385 | } | |
386 | ||
387 | return 0; | |
388 | } | |
389 | ||
390 | static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
391 | { | |
feaa7bba | 392 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
393 | |
394 | if (error) { | |
395 | gfs2_revoke_clean(sdp); | |
396 | return; | |
397 | } | |
398 | if (pass != 1) | |
399 | return; | |
400 | ||
401 | fs_info(sdp, "jid=%u: Found %u revoke tags\n", | |
402 | jd->jd_jid, sdp->sd_found_revokes); | |
403 | ||
404 | gfs2_revoke_clean(sdp); | |
405 | } | |
406 | ||
407 | static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
408 | { | |
409 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 410 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa | 411 | |
5c676f6d | 412 | tr->tr_touched = 1; |
b3b94faa DT |
413 | |
414 | if (!list_empty(&le->le_list)) | |
415 | return; | |
416 | ||
417 | rgd = container_of(le, struct gfs2_rgrpd, rd_le); | |
418 | gfs2_rgrp_bh_hold(rgd); | |
419 | ||
420 | gfs2_log_lock(sdp); | |
421 | sdp->sd_log_num_rg++; | |
422 | list_add(&le->le_list, &sdp->sd_log_le_rg); | |
907b9bce | 423 | gfs2_log_unlock(sdp); |
b3b94faa DT |
424 | } |
425 | ||
426 | static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
427 | { | |
428 | struct list_head *head = &sdp->sd_log_le_rg; | |
429 | struct gfs2_rgrpd *rgd; | |
430 | ||
431 | while (!list_empty(head)) { | |
432 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list); | |
433 | list_del_init(&rgd->rd_le.le_list); | |
434 | sdp->sd_log_num_rg--; | |
435 | ||
436 | gfs2_rgrp_repolish_clones(rgd); | |
437 | gfs2_rgrp_bh_put(rgd); | |
438 | } | |
439 | gfs2_assert_warn(sdp, !sdp->sd_log_num_rg); | |
440 | } | |
441 | ||
18ec7d5c SW |
442 | /** |
443 | * databuf_lo_add - Add a databuf to the transaction. | |
444 | * | |
445 | * This is used in two distinct cases: | |
446 | * i) In ordered write mode | |
447 | * We put the data buffer on a list so that we can ensure that its | |
448 | * synced to disk at the right time | |
449 | * ii) In journaled data mode | |
450 | * We need to journal the data block in the same way as metadata in | |
451 | * the functions above. The difference is that here we have a tag | |
452 | * which is two __be64's being the block number (as per meta data) | |
453 | * and a flag which says whether the data block needs escaping or | |
454 | * not. This means we need a new log entry for each 251 or so data | |
455 | * blocks, which isn't an enormous overhead but twice as much as | |
456 | * for normal metadata blocks. | |
457 | */ | |
b3b94faa DT |
458 | static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) |
459 | { | |
18ec7d5c | 460 | struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le); |
5c676f6d | 461 | struct gfs2_trans *tr = current->journal_info; |
18ec7d5c | 462 | struct address_space *mapping = bd->bd_bh->b_page->mapping; |
feaa7bba | 463 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa | 464 | |
18ec7d5c | 465 | tr->tr_touched = 1; |
15d00c0b | 466 | if (list_empty(&bd->bd_list_tr) && |
18ec7d5c SW |
467 | (ip->i_di.di_flags & GFS2_DIF_JDATA)) { |
468 | tr->tr_num_buf++; | |
18ec7d5c SW |
469 | list_add(&bd->bd_list_tr, &tr->tr_list_buf); |
470 | gfs2_pin(sdp, bd->bd_bh); | |
b4dc7291 | 471 | tr->tr_num_buf_new++; |
18ec7d5c | 472 | } |
b61dde79 | 473 | gfs2_trans_add_gl(bd->bd_gl); |
b3b94faa | 474 | gfs2_log_lock(sdp); |
15d00c0b | 475 | if (list_empty(&le->le_list)) { |
13538b8e SW |
476 | if (ip->i_di.di_flags & GFS2_DIF_JDATA) |
477 | sdp->sd_log_num_jdata++; | |
478 | sdp->sd_log_num_databuf++; | |
479 | list_add(&le->le_list, &sdp->sd_log_le_databuf); | |
480 | } | |
b3b94faa DT |
481 | gfs2_log_unlock(sdp); |
482 | } | |
483 | ||
18ec7d5c SW |
484 | static int gfs2_check_magic(struct buffer_head *bh) |
485 | { | |
486 | struct page *page = bh->b_page; | |
487 | void *kaddr; | |
488 | __be32 *ptr; | |
489 | int rv = 0; | |
490 | ||
491 | kaddr = kmap_atomic(page, KM_USER0); | |
492 | ptr = kaddr + bh_offset(bh); | |
493 | if (*ptr == cpu_to_be32(GFS2_MAGIC)) | |
494 | rv = 1; | |
c312c4fd | 495 | kunmap_atomic(kaddr, KM_USER0); |
18ec7d5c SW |
496 | |
497 | return rv; | |
498 | } | |
499 | ||
500 | /** | |
501 | * databuf_lo_before_commit - Scan the data buffers, writing as we go | |
502 | * | |
503 | * Here we scan through the lists of buffers and make the assumption | |
504 | * that any buffer thats been pinned is being journaled, and that | |
505 | * any unpinned buffer is an ordered write data buffer and therefore | |
506 | * will be written back rather than journaled. | |
507 | */ | |
b3b94faa DT |
508 | static void databuf_lo_before_commit(struct gfs2_sbd *sdp) |
509 | { | |
b3b94faa | 510 | LIST_HEAD(started); |
18ec7d5c | 511 | struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt; |
70209331 | 512 | struct buffer_head *bh = NULL,*bh1 = NULL; |
18ec7d5c SW |
513 | unsigned int offset = sizeof(struct gfs2_log_descriptor); |
514 | struct gfs2_log_descriptor *ld; | |
515 | unsigned int limit; | |
516 | unsigned int total_dbuf = sdp->sd_log_num_databuf; | |
517 | unsigned int total_jdata = sdp->sd_log_num_jdata; | |
518 | unsigned int num, n; | |
f55ab26a | 519 | __be64 *ptr = NULL; |
b3b94faa | 520 | |
a67cdbd4 | 521 | offset += 2*sizeof(__be64) - 1; |
18ec7d5c SW |
522 | offset &= ~(2*sizeof(__be64) - 1); |
523 | limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64); | |
b3b94faa | 524 | |
18ec7d5c SW |
525 | /* |
526 | * Start writing ordered buffers, write journaled buffers | |
527 | * into the log along with a header | |
528 | */ | |
f55ab26a | 529 | gfs2_log_lock(sdp); |
568f4c96 SW |
530 | bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf, |
531 | bd_le.le_list); | |
18ec7d5c SW |
532 | while(total_dbuf) { |
533 | num = total_jdata; | |
534 | if (num > limit) | |
535 | num = limit; | |
536 | n = 0; | |
568f4c96 SW |
537 | list_for_each_entry_safe_continue(bd1, bdt, |
538 | &sdp->sd_log_le_databuf, | |
539 | bd_le.le_list) { | |
70209331 RC |
540 | /* store off the buffer head in a local ptr since |
541 | * gfs2_bufdata might change when we drop the log lock | |
542 | */ | |
543 | bh1 = bd1->bd_bh; | |
544 | ||
18ec7d5c | 545 | /* An ordered write buffer */ |
70209331 | 546 | if (bh1 && !buffer_pinned(bh1)) { |
18ec7d5c SW |
547 | list_move(&bd1->bd_le.le_list, &started); |
548 | if (bd1 == bd2) { | |
549 | bd2 = NULL; | |
568f4c96 SW |
550 | bd2 = list_prepare_entry(bd2, |
551 | &sdp->sd_log_le_databuf, | |
552 | bd_le.le_list); | |
18ec7d5c SW |
553 | } |
554 | total_dbuf--; | |
70209331 RC |
555 | if (bh1) { |
556 | if (buffer_dirty(bh1)) { | |
557 | get_bh(bh1); | |
558 | ||
f55ab26a | 559 | gfs2_log_unlock(sdp); |
70209331 RC |
560 | |
561 | ll_rw_block(SWRITE, 1, &bh1); | |
562 | brelse(bh1); | |
563 | ||
f55ab26a | 564 | gfs2_log_lock(sdp); |
18ec7d5c | 565 | } |
18ec7d5c SW |
566 | continue; |
567 | } | |
18ec7d5c | 568 | continue; |
70209331 | 569 | } else if (bh1) { /* A journaled buffer */ |
18ec7d5c SW |
570 | int magic; |
571 | gfs2_log_unlock(sdp); | |
18ec7d5c SW |
572 | if (!bh) { |
573 | bh = gfs2_log_get_buf(sdp); | |
b09e593d | 574 | sdp->sd_log_num_hdrs++; |
568f4c96 SW |
575 | ld = (struct gfs2_log_descriptor *) |
576 | bh->b_data; | |
18ec7d5c | 577 | ptr = (__be64 *)(bh->b_data + offset); |
568f4c96 SW |
578 | ld->ld_header.mh_magic = |
579 | cpu_to_be32(GFS2_MAGIC); | |
580 | ld->ld_header.mh_type = | |
e3167ded | 581 | cpu_to_be32(GFS2_METATYPE_LD); |
568f4c96 | 582 | ld->ld_header.mh_format = |
e3167ded | 583 | cpu_to_be32(GFS2_FORMAT_LD); |
568f4c96 SW |
584 | ld->ld_type = |
585 | cpu_to_be32(GFS2_LOG_DESC_JDATA); | |
18ec7d5c SW |
586 | ld->ld_length = cpu_to_be32(num + 1); |
587 | ld->ld_data1 = cpu_to_be32(num); | |
588 | ld->ld_data2 = cpu_to_be32(0); | |
589 | memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); | |
590 | } | |
70209331 RC |
591 | magic = gfs2_check_magic(bh1); |
592 | *ptr++ = cpu_to_be64(bh1->b_blocknr); | |
18ec7d5c | 593 | *ptr++ = cpu_to_be64((__u64)magic); |
70209331 | 594 | clear_buffer_escaped(bh1); |
18ec7d5c | 595 | if (unlikely(magic != 0)) |
70209331 | 596 | set_buffer_escaped(bh1); |
f55ab26a | 597 | gfs2_log_lock(sdp); |
18ec7d5c SW |
598 | if (n++ > num) |
599 | break; | |
70209331 | 600 | } else if (!bh1) { |
623d9355 SW |
601 | total_dbuf--; |
602 | sdp->sd_log_num_databuf--; | |
603 | list_del_init(&bd1->bd_le.le_list); | |
604 | if (bd1 == bd2) { | |
605 | bd2 = NULL; | |
606 | bd2 = list_prepare_entry(bd2, | |
607 | &sdp->sd_log_le_databuf, | |
608 | bd_le.le_list); | |
609 | } | |
610 | kmem_cache_free(gfs2_bufdata_cachep, bd1); | |
18ec7d5c SW |
611 | } |
612 | } | |
f55ab26a | 613 | gfs2_log_unlock(sdp); |
b3b94faa | 614 | if (bh) { |
18ec7d5c SW |
615 | set_buffer_dirty(bh); |
616 | ll_rw_block(WRITE, 1, &bh); | |
617 | bh = NULL; | |
618 | } | |
619 | n = 0; | |
f55ab26a | 620 | gfs2_log_lock(sdp); |
568f4c96 SW |
621 | list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf, |
622 | bd_le.le_list) { | |
18ec7d5c SW |
623 | if (!bd2->bd_bh) |
624 | continue; | |
625 | /* copy buffer if it needs escaping */ | |
f55ab26a | 626 | gfs2_log_unlock(sdp); |
18ec7d5c SW |
627 | if (unlikely(buffer_escaped(bd2->bd_bh))) { |
628 | void *kaddr; | |
629 | struct page *page = bd2->bd_bh->b_page; | |
630 | bh = gfs2_log_get_buf(sdp); | |
631 | kaddr = kmap_atomic(page, KM_USER0); | |
568f4c96 SW |
632 | memcpy(bh->b_data, |
633 | kaddr + bh_offset(bd2->bd_bh), | |
634 | sdp->sd_sb.sb_bsize); | |
c312c4fd | 635 | kunmap_atomic(kaddr, KM_USER0); |
18ec7d5c SW |
636 | *(__be32 *)bh->b_data = 0; |
637 | } else { | |
638 | bh = gfs2_log_fake_buf(sdp, bd2->bd_bh); | |
b3b94faa | 639 | } |
18ec7d5c SW |
640 | set_buffer_dirty(bh); |
641 | ll_rw_block(WRITE, 1, &bh); | |
f55ab26a | 642 | gfs2_log_lock(sdp); |
18ec7d5c SW |
643 | if (++n >= num) |
644 | break; | |
645 | } | |
646 | bh = NULL; | |
647 | total_dbuf -= num; | |
648 | total_jdata -= num; | |
b3b94faa | 649 | } |
f55ab26a SW |
650 | gfs2_log_unlock(sdp); |
651 | ||
18ec7d5c | 652 | /* Wait on all ordered buffers */ |
b3b94faa | 653 | while (!list_empty(&started)) { |
13538b8e | 654 | gfs2_log_lock(sdp); |
568f4c96 SW |
655 | bd1 = list_entry(started.next, struct gfs2_bufdata, |
656 | bd_le.le_list); | |
08867605 | 657 | list_del_init(&bd1->bd_le.le_list); |
b3b94faa | 658 | sdp->sd_log_num_databuf--; |
18ec7d5c | 659 | bh = bd1->bd_bh; |
b3b94faa | 660 | if (bh) { |
5c676f6d | 661 | bh->b_private = NULL; |
15d00c0b | 662 | get_bh(bh); |
b3b94faa DT |
663 | gfs2_log_unlock(sdp); |
664 | wait_on_buffer(bh); | |
665 | brelse(bh); | |
666 | } else | |
667 | gfs2_log_unlock(sdp); | |
668 | ||
3a8476dd | 669 | kmem_cache_free(gfs2_bufdata_cachep, bd1); |
b3b94faa DT |
670 | } |
671 | ||
18ec7d5c SW |
672 | /* We've removed all the ordered write bufs here, so only jdata left */ |
673 | gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata); | |
674 | } | |
675 | ||
676 | static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
677 | struct gfs2_log_descriptor *ld, | |
678 | __be64 *ptr, int pass) | |
679 | { | |
feaa7bba SW |
680 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
681 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
5c676f6d | 682 | struct gfs2_glock *gl = ip->i_gl; |
18ec7d5c SW |
683 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
684 | struct buffer_head *bh_log, *bh_ip; | |
cd915493 SW |
685 | u64 blkno; |
686 | u64 esc; | |
18ec7d5c SW |
687 | int error = 0; |
688 | ||
689 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) | |
690 | return 0; | |
691 | ||
692 | gfs2_replay_incr_blk(sdp, &start); | |
693 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
694 | blkno = be64_to_cpu(*ptr++); | |
695 | esc = be64_to_cpu(*ptr++); | |
696 | ||
697 | sdp->sd_found_blocks++; | |
698 | ||
699 | if (gfs2_revoke_check(sdp, blkno, start)) | |
700 | continue; | |
701 | ||
702 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
703 | if (error) | |
704 | return error; | |
705 | ||
706 | bh_ip = gfs2_meta_new(gl, blkno); | |
707 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
708 | ||
709 | /* Unescape */ | |
710 | if (esc) { | |
711 | __be32 *eptr = (__be32 *)bh_ip->b_data; | |
712 | *eptr = cpu_to_be32(GFS2_MAGIC); | |
713 | } | |
714 | mark_buffer_dirty(bh_ip); | |
715 | ||
716 | brelse(bh_log); | |
717 | brelse(bh_ip); | |
718 | if (error) | |
719 | break; | |
720 | ||
721 | sdp->sd_replayed_blocks++; | |
722 | } | |
723 | ||
724 | return error; | |
725 | } | |
726 | ||
727 | /* FIXME: sort out accounting for log blocks etc. */ | |
728 | ||
729 | static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
730 | { | |
feaa7bba SW |
731 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
732 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
18ec7d5c SW |
733 | |
734 | if (error) { | |
7276b3b0 | 735 | gfs2_meta_sync(ip->i_gl); |
18ec7d5c SW |
736 | return; |
737 | } | |
738 | if (pass != 1) | |
739 | return; | |
740 | ||
741 | /* data sync? */ | |
7276b3b0 | 742 | gfs2_meta_sync(ip->i_gl); |
18ec7d5c SW |
743 | |
744 | fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", | |
745 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
746 | } | |
747 | ||
748 | static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
749 | { | |
750 | struct list_head *head = &sdp->sd_log_le_databuf; | |
751 | struct gfs2_bufdata *bd; | |
752 | ||
753 | while (!list_empty(head)) { | |
754 | bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); | |
b8e1aabf | 755 | list_del_init(&bd->bd_le.le_list); |
18ec7d5c SW |
756 | sdp->sd_log_num_databuf--; |
757 | sdp->sd_log_num_jdata--; | |
758 | gfs2_unpin(sdp, bd->bd_bh, ai); | |
18ec7d5c | 759 | } |
b3b94faa | 760 | gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf); |
18ec7d5c | 761 | gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata); |
b3b94faa DT |
762 | } |
763 | ||
18ec7d5c | 764 | |
b09e593d | 765 | const struct gfs2_log_operations gfs2_glock_lops = { |
b3b94faa DT |
766 | .lo_add = glock_lo_add, |
767 | .lo_after_commit = glock_lo_after_commit, | |
ea67eedb | 768 | .lo_name = "glock", |
b3b94faa DT |
769 | }; |
770 | ||
b09e593d | 771 | const struct gfs2_log_operations gfs2_buf_lops = { |
b3b94faa DT |
772 | .lo_add = buf_lo_add, |
773 | .lo_incore_commit = buf_lo_incore_commit, | |
774 | .lo_before_commit = buf_lo_before_commit, | |
775 | .lo_after_commit = buf_lo_after_commit, | |
776 | .lo_before_scan = buf_lo_before_scan, | |
777 | .lo_scan_elements = buf_lo_scan_elements, | |
778 | .lo_after_scan = buf_lo_after_scan, | |
ea67eedb | 779 | .lo_name = "buf", |
b3b94faa DT |
780 | }; |
781 | ||
b09e593d | 782 | const struct gfs2_log_operations gfs2_revoke_lops = { |
b3b94faa DT |
783 | .lo_add = revoke_lo_add, |
784 | .lo_before_commit = revoke_lo_before_commit, | |
785 | .lo_before_scan = revoke_lo_before_scan, | |
786 | .lo_scan_elements = revoke_lo_scan_elements, | |
787 | .lo_after_scan = revoke_lo_after_scan, | |
ea67eedb | 788 | .lo_name = "revoke", |
b3b94faa DT |
789 | }; |
790 | ||
b09e593d | 791 | const struct gfs2_log_operations gfs2_rg_lops = { |
b3b94faa DT |
792 | .lo_add = rg_lo_add, |
793 | .lo_after_commit = rg_lo_after_commit, | |
ea67eedb | 794 | .lo_name = "rg", |
b3b94faa DT |
795 | }; |
796 | ||
b09e593d | 797 | const struct gfs2_log_operations gfs2_databuf_lops = { |
b3b94faa | 798 | .lo_add = databuf_lo_add, |
18ec7d5c | 799 | .lo_incore_commit = buf_lo_incore_commit, |
b3b94faa | 800 | .lo_before_commit = databuf_lo_before_commit, |
18ec7d5c SW |
801 | .lo_after_commit = databuf_lo_after_commit, |
802 | .lo_scan_elements = databuf_lo_scan_elements, | |
803 | .lo_after_scan = databuf_lo_after_scan, | |
ea67eedb | 804 | .lo_name = "databuf", |
b3b94faa DT |
805 | }; |
806 | ||
b09e593d | 807 | const struct gfs2_log_operations *gfs2_log_ops[] = { |
b3b94faa DT |
808 | &gfs2_glock_lops, |
809 | &gfs2_buf_lops, | |
810 | &gfs2_revoke_lops, | |
811 | &gfs2_rg_lops, | |
812 | &gfs2_databuf_lops, | |
ea67eedb | 813 | NULL, |
b3b94faa DT |
814 | }; |
815 |