Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3 | * Copyright (C) 2004-2005 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 v.2. | |
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 | #include <asm/semaphore.h> |
17 | ||
18 | #include "gfs2.h" | |
5c676f6d SW |
19 | #include "lm_interface.h" |
20 | #include "incore.h" | |
b3b94faa DT |
21 | #include "glock.h" |
22 | #include "log.h" | |
23 | #include "lops.h" | |
24 | #include "meta_io.h" | |
25 | #include "recovery.h" | |
26 | #include "rgrp.h" | |
27 | #include "trans.h" | |
5c676f6d | 28 | #include "util.h" |
b3b94faa DT |
29 | |
30 | static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
31 | { | |
32 | struct gfs2_glock *gl; | |
5c676f6d | 33 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa | 34 | |
5c676f6d | 35 | tr->tr_touched = 1; |
b3b94faa DT |
36 | |
37 | if (!list_empty(&le->le_list)) | |
38 | return; | |
39 | ||
40 | gl = container_of(le, struct gfs2_glock, gl_le); | |
41 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl))) | |
42 | return; | |
43 | gfs2_glock_hold(gl); | |
44 | set_bit(GLF_DIRTY, &gl->gl_flags); | |
45 | ||
46 | gfs2_log_lock(sdp); | |
47 | sdp->sd_log_num_gl++; | |
48 | list_add(&le->le_list, &sdp->sd_log_le_gl); | |
49 | gfs2_log_unlock(sdp); | |
50 | } | |
51 | ||
52 | static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
53 | { | |
54 | struct list_head *head = &sdp->sd_log_le_gl; | |
55 | struct gfs2_glock *gl; | |
56 | ||
57 | while (!list_empty(head)) { | |
58 | gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list); | |
59 | list_del_init(&gl->gl_le.le_list); | |
60 | sdp->sd_log_num_gl--; | |
61 | ||
62 | gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)); | |
63 | gfs2_glock_put(gl); | |
64 | } | |
65 | gfs2_assert_warn(sdp, !sdp->sd_log_num_gl); | |
66 | } | |
67 | ||
68 | static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
69 | { | |
70 | struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le); | |
71 | struct gfs2_trans *tr; | |
72 | ||
73 | if (!list_empty(&bd->bd_list_tr)) | |
74 | return; | |
75 | ||
5c676f6d | 76 | tr = current->journal_info; |
b3b94faa DT |
77 | tr->tr_touched = 1; |
78 | tr->tr_num_buf++; | |
79 | list_add(&bd->bd_list_tr, &tr->tr_list_buf); | |
80 | ||
81 | if (!list_empty(&le->le_list)) | |
82 | return; | |
83 | ||
84 | gfs2_trans_add_gl(bd->bd_gl); | |
85 | ||
86 | gfs2_meta_check(sdp, bd->bd_bh); | |
a98ab220 | 87 | gfs2_pin(sdp, bd->bd_bh); |
b3b94faa DT |
88 | |
89 | gfs2_log_lock(sdp); | |
90 | sdp->sd_log_num_buf++; | |
91 | list_add(&le->le_list, &sdp->sd_log_le_buf); | |
92 | gfs2_log_unlock(sdp); | |
93 | ||
94 | tr->tr_num_buf_new++; | |
95 | } | |
96 | ||
97 | static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
98 | { | |
99 | struct list_head *head = &tr->tr_list_buf; | |
100 | struct gfs2_bufdata *bd; | |
101 | ||
102 | while (!list_empty(head)) { | |
103 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr); | |
104 | list_del_init(&bd->bd_list_tr); | |
105 | tr->tr_num_buf--; | |
106 | } | |
107 | gfs2_assert_warn(sdp, !tr->tr_num_buf); | |
108 | } | |
109 | ||
110 | static void buf_lo_before_commit(struct gfs2_sbd *sdp) | |
111 | { | |
112 | struct buffer_head *bh; | |
113 | struct gfs2_log_descriptor *ld; | |
114 | struct gfs2_bufdata *bd1 = NULL, *bd2; | |
115 | unsigned int total = sdp->sd_log_num_buf; | |
116 | unsigned int offset = sizeof(struct gfs2_log_descriptor); | |
117 | unsigned int limit; | |
118 | unsigned int num; | |
119 | unsigned n; | |
120 | __be64 *ptr; | |
121 | ||
122 | offset += (sizeof(__be64) - 1); | |
123 | offset &= ~(sizeof(__be64) - 1); | |
124 | limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64); | |
125 | /* for 4k blocks, limit = 503 */ | |
126 | ||
127 | bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list); | |
128 | while(total) { | |
129 | num = total; | |
130 | if (total > limit) | |
131 | num = limit; | |
132 | bh = gfs2_log_get_buf(sdp); | |
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); | |
136 | ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD); | |
137 | ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD); | |
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, | |
185 | struct gfs2_log_header *head, int pass) | |
186 | { | |
5c676f6d SW |
187 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
188 | struct gfs2_sbd *sdp = ip->i_sbd; | |
b3b94faa DT |
189 | |
190 | if (pass != 0) | |
191 | return; | |
192 | ||
193 | sdp->sd_found_blocks = 0; | |
194 | sdp->sd_replayed_blocks = 0; | |
195 | } | |
196 | ||
197 | static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
198 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
199 | int pass) | |
200 | { | |
5c676f6d SW |
201 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
202 | struct gfs2_sbd *sdp = ip->i_sbd; | |
203 | struct gfs2_glock *gl = ip->i_gl; | |
b3b94faa DT |
204 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
205 | struct buffer_head *bh_log, *bh_ip; | |
206 | uint64_t blkno; | |
207 | int error = 0; | |
208 | ||
209 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) | |
210 | return 0; | |
211 | ||
212 | gfs2_replay_incr_blk(sdp, &start); | |
213 | ||
214 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
215 | blkno = be64_to_cpu(*ptr++); | |
216 | ||
217 | sdp->sd_found_blocks++; | |
218 | ||
219 | if (gfs2_revoke_check(sdp, blkno, start)) | |
220 | continue; | |
221 | ||
222 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
223 | if (error) | |
224 | return error; | |
225 | ||
226 | bh_ip = gfs2_meta_new(gl, blkno); | |
227 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
228 | ||
229 | if (gfs2_meta_check(sdp, bh_ip)) | |
230 | error = -EIO; | |
231 | else | |
232 | mark_buffer_dirty(bh_ip); | |
233 | ||
234 | brelse(bh_log); | |
235 | brelse(bh_ip); | |
236 | ||
237 | if (error) | |
238 | break; | |
239 | ||
240 | sdp->sd_replayed_blocks++; | |
241 | } | |
242 | ||
243 | return error; | |
244 | } | |
245 | ||
246 | static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
247 | { | |
5c676f6d SW |
248 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
249 | struct gfs2_sbd *sdp = ip->i_sbd; | |
b3b94faa DT |
250 | |
251 | if (error) { | |
5c676f6d | 252 | gfs2_meta_sync(ip->i_gl, |
568f4c96 | 253 | DIO_START | DIO_WAIT); |
b3b94faa DT |
254 | return; |
255 | } | |
256 | if (pass != 1) | |
257 | return; | |
258 | ||
5c676f6d | 259 | gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT); |
b3b94faa DT |
260 | |
261 | fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", | |
262 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
263 | } | |
264 | ||
265 | static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
266 | { | |
267 | struct gfs2_trans *tr; | |
268 | ||
5c676f6d | 269 | tr = current->journal_info; |
b3b94faa DT |
270 | tr->tr_touched = 1; |
271 | tr->tr_num_revoke++; | |
272 | ||
273 | gfs2_log_lock(sdp); | |
274 | sdp->sd_log_num_revoke++; | |
275 | list_add(&le->le_list, &sdp->sd_log_le_revoke); | |
276 | gfs2_log_unlock(sdp); | |
277 | } | |
278 | ||
279 | static void revoke_lo_before_commit(struct gfs2_sbd *sdp) | |
280 | { | |
281 | struct gfs2_log_descriptor *ld; | |
282 | struct gfs2_meta_header *mh; | |
283 | struct buffer_head *bh; | |
284 | unsigned int offset; | |
285 | struct list_head *head = &sdp->sd_log_le_revoke; | |
286 | struct gfs2_revoke *rv; | |
287 | ||
288 | if (!sdp->sd_log_num_revoke) | |
289 | return; | |
290 | ||
291 | bh = gfs2_log_get_buf(sdp); | |
292 | ld = (struct gfs2_log_descriptor *)bh->b_data; | |
293 | ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
294 | ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD); | |
295 | ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD); | |
296 | ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE); | |
568f4c96 SW |
297 | ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, |
298 | sizeof(uint64_t))); | |
b3b94faa DT |
299 | ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke); |
300 | ld->ld_data2 = cpu_to_be32(0); | |
301 | memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); | |
302 | offset = sizeof(struct gfs2_log_descriptor); | |
303 | ||
304 | while (!list_empty(head)) { | |
305 | rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list); | |
13538b8e | 306 | list_del_init(&rv->rv_le.le_list); |
b3b94faa DT |
307 | sdp->sd_log_num_revoke--; |
308 | ||
309 | if (offset + sizeof(uint64_t) > sdp->sd_sb.sb_bsize) { | |
310 | set_buffer_dirty(bh); | |
311 | ll_rw_block(WRITE, 1, &bh); | |
312 | ||
313 | bh = gfs2_log_get_buf(sdp); | |
314 | mh = (struct gfs2_meta_header *)bh->b_data; | |
315 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); | |
316 | mh->mh_type = cpu_to_be16(GFS2_METATYPE_LB); | |
317 | mh->mh_format = cpu_to_be16(GFS2_FORMAT_LB); | |
318 | offset = sizeof(struct gfs2_meta_header); | |
319 | } | |
320 | ||
321 | *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno); | |
322 | kfree(rv); | |
323 | ||
324 | offset += sizeof(uint64_t); | |
325 | } | |
326 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
327 | ||
328 | set_buffer_dirty(bh); | |
329 | ll_rw_block(WRITE, 1, &bh); | |
330 | } | |
331 | ||
332 | static void revoke_lo_before_scan(struct gfs2_jdesc *jd, | |
333 | struct gfs2_log_header *head, int pass) | |
334 | { | |
5c676f6d SW |
335 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
336 | struct gfs2_sbd *sdp = ip->i_sbd; | |
b3b94faa DT |
337 | |
338 | if (pass != 0) | |
339 | return; | |
340 | ||
341 | sdp->sd_found_revokes = 0; | |
342 | sdp->sd_replay_tail = head->lh_tail; | |
343 | } | |
344 | ||
345 | static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
346 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
347 | int pass) | |
348 | { | |
5c676f6d SW |
349 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
350 | struct gfs2_sbd *sdp = ip->i_sbd; | |
b3b94faa DT |
351 | unsigned int blks = be32_to_cpu(ld->ld_length); |
352 | unsigned int revokes = be32_to_cpu(ld->ld_data1); | |
353 | struct buffer_head *bh; | |
354 | unsigned int offset; | |
355 | uint64_t blkno; | |
356 | int first = 1; | |
357 | int error; | |
358 | ||
359 | if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) | |
360 | return 0; | |
361 | ||
362 | offset = sizeof(struct gfs2_log_descriptor); | |
363 | ||
364 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
365 | error = gfs2_replay_read_block(jd, start, &bh); | |
366 | if (error) | |
367 | return error; | |
368 | ||
369 | if (!first) | |
370 | gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); | |
371 | ||
372 | while (offset + sizeof(uint64_t) <= sdp->sd_sb.sb_bsize) { | |
373 | blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); | |
374 | ||
375 | error = gfs2_revoke_add(sdp, blkno, start); | |
376 | if (error < 0) | |
377 | return error; | |
378 | else if (error) | |
379 | sdp->sd_found_revokes++; | |
380 | ||
381 | if (!--revokes) | |
382 | break; | |
383 | offset += sizeof(uint64_t); | |
384 | } | |
385 | ||
386 | brelse(bh); | |
387 | offset = sizeof(struct gfs2_meta_header); | |
388 | first = 0; | |
389 | } | |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
394 | static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
395 | { | |
5c676f6d SW |
396 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
397 | struct gfs2_sbd *sdp = ip->i_sbd; | |
b3b94faa DT |
398 | |
399 | if (error) { | |
400 | gfs2_revoke_clean(sdp); | |
401 | return; | |
402 | } | |
403 | if (pass != 1) | |
404 | return; | |
405 | ||
406 | fs_info(sdp, "jid=%u: Found %u revoke tags\n", | |
407 | jd->jd_jid, sdp->sd_found_revokes); | |
408 | ||
409 | gfs2_revoke_clean(sdp); | |
410 | } | |
411 | ||
412 | static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) | |
413 | { | |
414 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 415 | struct gfs2_trans *tr = current->journal_info; |
b3b94faa | 416 | |
5c676f6d | 417 | tr->tr_touched = 1; |
b3b94faa DT |
418 | |
419 | if (!list_empty(&le->le_list)) | |
420 | return; | |
421 | ||
422 | rgd = container_of(le, struct gfs2_rgrpd, rd_le); | |
423 | gfs2_rgrp_bh_hold(rgd); | |
424 | ||
425 | gfs2_log_lock(sdp); | |
426 | sdp->sd_log_num_rg++; | |
427 | list_add(&le->le_list, &sdp->sd_log_le_rg); | |
428 | gfs2_log_unlock(sdp); | |
429 | } | |
430 | ||
431 | static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
432 | { | |
433 | struct list_head *head = &sdp->sd_log_le_rg; | |
434 | struct gfs2_rgrpd *rgd; | |
435 | ||
436 | while (!list_empty(head)) { | |
437 | rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list); | |
438 | list_del_init(&rgd->rd_le.le_list); | |
439 | sdp->sd_log_num_rg--; | |
440 | ||
441 | gfs2_rgrp_repolish_clones(rgd); | |
442 | gfs2_rgrp_bh_put(rgd); | |
443 | } | |
444 | gfs2_assert_warn(sdp, !sdp->sd_log_num_rg); | |
445 | } | |
446 | ||
18ec7d5c SW |
447 | /** |
448 | * databuf_lo_add - Add a databuf to the transaction. | |
449 | * | |
450 | * This is used in two distinct cases: | |
451 | * i) In ordered write mode | |
452 | * We put the data buffer on a list so that we can ensure that its | |
453 | * synced to disk at the right time | |
454 | * ii) In journaled data mode | |
455 | * We need to journal the data block in the same way as metadata in | |
456 | * the functions above. The difference is that here we have a tag | |
457 | * which is two __be64's being the block number (as per meta data) | |
458 | * and a flag which says whether the data block needs escaping or | |
459 | * not. This means we need a new log entry for each 251 or so data | |
460 | * blocks, which isn't an enormous overhead but twice as much as | |
461 | * for normal metadata blocks. | |
462 | */ | |
b3b94faa DT |
463 | static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le) |
464 | { | |
18ec7d5c | 465 | struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le); |
5c676f6d | 466 | struct gfs2_trans *tr = current->journal_info; |
18ec7d5c | 467 | struct address_space *mapping = bd->bd_bh->b_page->mapping; |
5c676f6d | 468 | struct gfs2_inode *ip = mapping->host->u.generic_ip; |
b3b94faa | 469 | |
18ec7d5c SW |
470 | tr->tr_touched = 1; |
471 | if (!list_empty(&bd->bd_list_tr) && | |
472 | (ip->i_di.di_flags & GFS2_DIF_JDATA)) { | |
473 | tr->tr_num_buf++; | |
474 | gfs2_trans_add_gl(bd->bd_gl); | |
475 | list_add(&bd->bd_list_tr, &tr->tr_list_buf); | |
476 | gfs2_pin(sdp, bd->bd_bh); | |
18ec7d5c | 477 | } |
b3b94faa | 478 | gfs2_log_lock(sdp); |
13538b8e SW |
479 | if (!list_empty(&le->le_list)) { |
480 | if (ip->i_di.di_flags & GFS2_DIF_JDATA) | |
481 | sdp->sd_log_num_jdata++; | |
482 | sdp->sd_log_num_databuf++; | |
483 | list_add(&le->le_list, &sdp->sd_log_le_databuf); | |
484 | } | |
b3b94faa DT |
485 | gfs2_log_unlock(sdp); |
486 | } | |
487 | ||
18ec7d5c SW |
488 | static int gfs2_check_magic(struct buffer_head *bh) |
489 | { | |
490 | struct page *page = bh->b_page; | |
491 | void *kaddr; | |
492 | __be32 *ptr; | |
493 | int rv = 0; | |
494 | ||
495 | kaddr = kmap_atomic(page, KM_USER0); | |
496 | ptr = kaddr + bh_offset(bh); | |
497 | if (*ptr == cpu_to_be32(GFS2_MAGIC)) | |
498 | rv = 1; | |
499 | kunmap_atomic(page, KM_USER0); | |
500 | ||
501 | return rv; | |
502 | } | |
503 | ||
504 | /** | |
505 | * databuf_lo_before_commit - Scan the data buffers, writing as we go | |
506 | * | |
507 | * Here we scan through the lists of buffers and make the assumption | |
508 | * that any buffer thats been pinned is being journaled, and that | |
509 | * any unpinned buffer is an ordered write data buffer and therefore | |
510 | * will be written back rather than journaled. | |
511 | */ | |
b3b94faa DT |
512 | static void databuf_lo_before_commit(struct gfs2_sbd *sdp) |
513 | { | |
b3b94faa | 514 | LIST_HEAD(started); |
18ec7d5c SW |
515 | struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt; |
516 | struct buffer_head *bh = NULL; | |
517 | unsigned int offset = sizeof(struct gfs2_log_descriptor); | |
518 | struct gfs2_log_descriptor *ld; | |
519 | unsigned int limit; | |
520 | unsigned int total_dbuf = sdp->sd_log_num_databuf; | |
521 | unsigned int total_jdata = sdp->sd_log_num_jdata; | |
522 | unsigned int num, n; | |
f55ab26a | 523 | __be64 *ptr = NULL; |
b3b94faa | 524 | |
18ec7d5c SW |
525 | offset += (2*sizeof(__be64) - 1); |
526 | offset &= ~(2*sizeof(__be64) - 1); | |
527 | limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64); | |
b3b94faa | 528 | |
18ec7d5c SW |
529 | /* |
530 | * Start writing ordered buffers, write journaled buffers | |
531 | * into the log along with a header | |
532 | */ | |
f55ab26a | 533 | gfs2_log_lock(sdp); |
568f4c96 SW |
534 | bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf, |
535 | bd_le.le_list); | |
18ec7d5c SW |
536 | while(total_dbuf) { |
537 | num = total_jdata; | |
538 | if (num > limit) | |
539 | num = limit; | |
540 | n = 0; | |
568f4c96 SW |
541 | list_for_each_entry_safe_continue(bd1, bdt, |
542 | &sdp->sd_log_le_databuf, | |
543 | bd_le.le_list) { | |
18ec7d5c SW |
544 | /* An ordered write buffer */ |
545 | if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) { | |
546 | list_move(&bd1->bd_le.le_list, &started); | |
547 | if (bd1 == bd2) { | |
548 | bd2 = NULL; | |
568f4c96 SW |
549 | bd2 = list_prepare_entry(bd2, |
550 | &sdp->sd_log_le_databuf, | |
551 | bd_le.le_list); | |
18ec7d5c SW |
552 | } |
553 | total_dbuf--; | |
554 | if (bd1->bd_bh) { | |
555 | get_bh(bd1->bd_bh); | |
18ec7d5c | 556 | if (buffer_dirty(bd1->bd_bh)) { |
f55ab26a | 557 | gfs2_log_unlock(sdp); |
18ec7d5c | 558 | wait_on_buffer(bd1->bd_bh); |
568f4c96 SW |
559 | ll_rw_block(WRITE, 1, |
560 | &bd1->bd_bh); | |
f55ab26a | 561 | gfs2_log_lock(sdp); |
18ec7d5c SW |
562 | } |
563 | brelse(bd1->bd_bh); | |
564 | continue; | |
565 | } | |
18ec7d5c SW |
566 | continue; |
567 | } else if (bd1->bd_bh) { /* A journaled buffer */ | |
568 | int magic; | |
569 | gfs2_log_unlock(sdp); | |
18ec7d5c SW |
570 | if (!bh) { |
571 | bh = gfs2_log_get_buf(sdp); | |
568f4c96 SW |
572 | ld = (struct gfs2_log_descriptor *) |
573 | bh->b_data; | |
18ec7d5c | 574 | ptr = (__be64 *)(bh->b_data + offset); |
568f4c96 SW |
575 | ld->ld_header.mh_magic = |
576 | cpu_to_be32(GFS2_MAGIC); | |
577 | ld->ld_header.mh_type = | |
578 | cpu_to_be16(GFS2_METATYPE_LD); | |
579 | ld->ld_header.mh_format = | |
580 | cpu_to_be16(GFS2_FORMAT_LD); | |
581 | ld->ld_type = | |
582 | cpu_to_be32(GFS2_LOG_DESC_JDATA); | |
18ec7d5c SW |
583 | ld->ld_length = cpu_to_be32(num + 1); |
584 | ld->ld_data1 = cpu_to_be32(num); | |
585 | ld->ld_data2 = cpu_to_be32(0); | |
586 | memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); | |
587 | } | |
588 | magic = gfs2_check_magic(bd1->bd_bh); | |
589 | *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); | |
590 | *ptr++ = cpu_to_be64((__u64)magic); | |
591 | clear_buffer_escaped(bd1->bd_bh); | |
592 | if (unlikely(magic != 0)) | |
593 | set_buffer_escaped(bd1->bd_bh); | |
f55ab26a | 594 | gfs2_log_lock(sdp); |
18ec7d5c SW |
595 | if (n++ > num) |
596 | break; | |
597 | } | |
598 | } | |
f55ab26a | 599 | gfs2_log_unlock(sdp); |
b3b94faa | 600 | if (bh) { |
18ec7d5c SW |
601 | set_buffer_dirty(bh); |
602 | ll_rw_block(WRITE, 1, &bh); | |
603 | bh = NULL; | |
604 | } | |
605 | n = 0; | |
f55ab26a | 606 | gfs2_log_lock(sdp); |
568f4c96 SW |
607 | list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf, |
608 | bd_le.le_list) { | |
18ec7d5c SW |
609 | if (!bd2->bd_bh) |
610 | continue; | |
611 | /* copy buffer if it needs escaping */ | |
f55ab26a | 612 | gfs2_log_unlock(sdp); |
18ec7d5c SW |
613 | if (unlikely(buffer_escaped(bd2->bd_bh))) { |
614 | void *kaddr; | |
615 | struct page *page = bd2->bd_bh->b_page; | |
616 | bh = gfs2_log_get_buf(sdp); | |
617 | kaddr = kmap_atomic(page, KM_USER0); | |
568f4c96 SW |
618 | memcpy(bh->b_data, |
619 | kaddr + bh_offset(bd2->bd_bh), | |
620 | sdp->sd_sb.sb_bsize); | |
18ec7d5c SW |
621 | kunmap_atomic(page, KM_USER0); |
622 | *(__be32 *)bh->b_data = 0; | |
623 | } else { | |
624 | bh = gfs2_log_fake_buf(sdp, bd2->bd_bh); | |
b3b94faa | 625 | } |
18ec7d5c SW |
626 | set_buffer_dirty(bh); |
627 | ll_rw_block(WRITE, 1, &bh); | |
f55ab26a | 628 | gfs2_log_lock(sdp); |
18ec7d5c SW |
629 | if (++n >= num) |
630 | break; | |
631 | } | |
632 | bh = NULL; | |
633 | total_dbuf -= num; | |
634 | total_jdata -= num; | |
b3b94faa | 635 | } |
f55ab26a SW |
636 | gfs2_log_unlock(sdp); |
637 | ||
18ec7d5c | 638 | /* Wait on all ordered buffers */ |
b3b94faa | 639 | while (!list_empty(&started)) { |
13538b8e | 640 | gfs2_log_lock(sdp); |
568f4c96 SW |
641 | bd1 = list_entry(started.next, struct gfs2_bufdata, |
642 | bd_le.le_list); | |
18ec7d5c | 643 | list_del(&bd1->bd_le.le_list); |
b3b94faa DT |
644 | sdp->sd_log_num_databuf--; |
645 | ||
18ec7d5c | 646 | bh = bd1->bd_bh; |
b3b94faa | 647 | if (bh) { |
5c676f6d | 648 | bh->b_private = NULL; |
b3b94faa DT |
649 | gfs2_log_unlock(sdp); |
650 | wait_on_buffer(bh); | |
651 | brelse(bh); | |
652 | } else | |
653 | gfs2_log_unlock(sdp); | |
654 | ||
18ec7d5c | 655 | kfree(bd1); |
b3b94faa DT |
656 | } |
657 | ||
18ec7d5c SW |
658 | /* We've removed all the ordered write bufs here, so only jdata left */ |
659 | gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata); | |
660 | } | |
661 | ||
662 | static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
663 | struct gfs2_log_descriptor *ld, | |
664 | __be64 *ptr, int pass) | |
665 | { | |
5c676f6d SW |
666 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
667 | struct gfs2_sbd *sdp = ip->i_sbd; | |
668 | struct gfs2_glock *gl = ip->i_gl; | |
18ec7d5c SW |
669 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
670 | struct buffer_head *bh_log, *bh_ip; | |
671 | uint64_t blkno; | |
672 | uint64_t esc; | |
673 | int error = 0; | |
674 | ||
675 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) | |
676 | return 0; | |
677 | ||
678 | gfs2_replay_incr_blk(sdp, &start); | |
679 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
680 | blkno = be64_to_cpu(*ptr++); | |
681 | esc = be64_to_cpu(*ptr++); | |
682 | ||
683 | sdp->sd_found_blocks++; | |
684 | ||
685 | if (gfs2_revoke_check(sdp, blkno, start)) | |
686 | continue; | |
687 | ||
688 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
689 | if (error) | |
690 | return error; | |
691 | ||
692 | bh_ip = gfs2_meta_new(gl, blkno); | |
693 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
694 | ||
695 | /* Unescape */ | |
696 | if (esc) { | |
697 | __be32 *eptr = (__be32 *)bh_ip->b_data; | |
698 | *eptr = cpu_to_be32(GFS2_MAGIC); | |
699 | } | |
700 | mark_buffer_dirty(bh_ip); | |
701 | ||
702 | brelse(bh_log); | |
703 | brelse(bh_ip); | |
704 | if (error) | |
705 | break; | |
706 | ||
707 | sdp->sd_replayed_blocks++; | |
708 | } | |
709 | ||
710 | return error; | |
711 | } | |
712 | ||
713 | /* FIXME: sort out accounting for log blocks etc. */ | |
714 | ||
715 | static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
716 | { | |
5c676f6d SW |
717 | struct gfs2_inode *ip = jd->jd_inode->u.generic_ip; |
718 | struct gfs2_sbd *sdp = ip->i_sbd; | |
18ec7d5c SW |
719 | |
720 | if (error) { | |
5c676f6d | 721 | gfs2_meta_sync(ip->i_gl, |
568f4c96 | 722 | DIO_START | DIO_WAIT); |
18ec7d5c SW |
723 | return; |
724 | } | |
725 | if (pass != 1) | |
726 | return; | |
727 | ||
728 | /* data sync? */ | |
5c676f6d | 729 | gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT); |
18ec7d5c SW |
730 | |
731 | fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", | |
732 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
733 | } | |
734 | ||
735 | static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
736 | { | |
737 | struct list_head *head = &sdp->sd_log_le_databuf; | |
738 | struct gfs2_bufdata *bd; | |
739 | ||
740 | while (!list_empty(head)) { | |
741 | bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); | |
f55ab26a | 742 | list_del(&bd->bd_le.le_list); |
18ec7d5c SW |
743 | sdp->sd_log_num_databuf--; |
744 | sdp->sd_log_num_jdata--; | |
745 | gfs2_unpin(sdp, bd->bd_bh, ai); | |
18ec7d5c | 746 | } |
b3b94faa | 747 | gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf); |
18ec7d5c | 748 | gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata); |
b3b94faa DT |
749 | } |
750 | ||
18ec7d5c | 751 | |
b3b94faa DT |
752 | struct gfs2_log_operations gfs2_glock_lops = { |
753 | .lo_add = glock_lo_add, | |
754 | .lo_after_commit = glock_lo_after_commit, | |
755 | .lo_name = "glock" | |
756 | }; | |
757 | ||
758 | struct gfs2_log_operations gfs2_buf_lops = { | |
759 | .lo_add = buf_lo_add, | |
760 | .lo_incore_commit = buf_lo_incore_commit, | |
761 | .lo_before_commit = buf_lo_before_commit, | |
762 | .lo_after_commit = buf_lo_after_commit, | |
763 | .lo_before_scan = buf_lo_before_scan, | |
764 | .lo_scan_elements = buf_lo_scan_elements, | |
765 | .lo_after_scan = buf_lo_after_scan, | |
766 | .lo_name = "buf" | |
767 | }; | |
768 | ||
769 | struct gfs2_log_operations gfs2_revoke_lops = { | |
770 | .lo_add = revoke_lo_add, | |
771 | .lo_before_commit = revoke_lo_before_commit, | |
772 | .lo_before_scan = revoke_lo_before_scan, | |
773 | .lo_scan_elements = revoke_lo_scan_elements, | |
774 | .lo_after_scan = revoke_lo_after_scan, | |
775 | .lo_name = "revoke" | |
776 | }; | |
777 | ||
778 | struct gfs2_log_operations gfs2_rg_lops = { | |
779 | .lo_add = rg_lo_add, | |
780 | .lo_after_commit = rg_lo_after_commit, | |
781 | .lo_name = "rg" | |
782 | }; | |
783 | ||
784 | struct gfs2_log_operations gfs2_databuf_lops = { | |
785 | .lo_add = databuf_lo_add, | |
18ec7d5c | 786 | .lo_incore_commit = buf_lo_incore_commit, |
b3b94faa | 787 | .lo_before_commit = databuf_lo_before_commit, |
18ec7d5c SW |
788 | .lo_after_commit = databuf_lo_after_commit, |
789 | .lo_scan_elements = databuf_lo_scan_elements, | |
790 | .lo_after_scan = databuf_lo_after_scan, | |
b3b94faa DT |
791 | .lo_name = "databuf" |
792 | }; | |
793 | ||
794 | struct gfs2_log_operations *gfs2_log_ops[] = { | |
795 | &gfs2_glock_lops, | |
796 | &gfs2_buf_lops, | |
797 | &gfs2_revoke_lops, | |
798 | &gfs2_rg_lops, | |
799 | &gfs2_databuf_lops, | |
800 | NULL | |
801 | }; | |
802 |