lops.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777
  1. /*
  2. * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
  3. * Copyright (C) 2004-2006 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 version 2.
  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/gfs2_ondisk.h>
  15. #include <linux/bio.h>
  16. #include <linux/fs.h>
  17. #include "gfs2.h"
  18. #include "incore.h"
  19. #include "inode.h"
  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"
  27. #include "util.h"
  28. #include "trace_gfs2.h"
  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. gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
  40. clear_buffer_dirty(bh);
  41. if (test_set_buffer_pinned(bh))
  42. gfs2_assert_withdraw(sdp, 0);
  43. if (!buffer_uptodate(bh))
  44. gfs2_io_error_bh(sdp, bh);
  45. bd = bh->b_private;
  46. /* If this buffer is in the AIL and it has already been written
  47. * to in-place disk block, remove it from the AIL.
  48. */
  49. if (bd->bd_ail)
  50. list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
  51. get_bh(bh);
  52. atomic_inc(&sdp->sd_log_pinned);
  53. trace_gfs2_pin(bd, 1);
  54. }
  55. /**
  56. * gfs2_unpin - Unpin a buffer
  57. * @sdp: the filesystem the buffer belongs to
  58. * @bh: The buffer to unpin
  59. * @ai:
  60. *
  61. */
  62. static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  63. struct gfs2_ail *ai)
  64. {
  65. struct gfs2_bufdata *bd = bh->b_private;
  66. gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
  67. if (!buffer_pinned(bh))
  68. gfs2_assert_withdraw(sdp, 0);
  69. lock_buffer(bh);
  70. mark_buffer_dirty(bh);
  71. clear_buffer_pinned(bh);
  72. gfs2_log_lock(sdp);
  73. if (bd->bd_ail) {
  74. list_del(&bd->bd_ail_st_list);
  75. brelse(bh);
  76. } else {
  77. struct gfs2_glock *gl = bd->bd_gl;
  78. list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
  79. atomic_inc(&gl->gl_ail_count);
  80. }
  81. bd->bd_ail = ai;
  82. list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
  83. if (test_and_clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags))
  84. gfs2_glock_schedule_for_reclaim(bd->bd_gl);
  85. trace_gfs2_pin(bd, 0);
  86. gfs2_log_unlock(sdp);
  87. unlock_buffer(bh);
  88. atomic_dec(&sdp->sd_log_pinned);
  89. }
  90. static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
  91. {
  92. return (struct gfs2_log_descriptor *)bh->b_data;
  93. }
  94. static inline __be64 *bh_log_ptr(struct buffer_head *bh)
  95. {
  96. struct gfs2_log_descriptor *ld = bh_log_desc(bh);
  97. return (__force __be64 *)(ld + 1);
  98. }
  99. static inline __be64 *bh_ptr_end(struct buffer_head *bh)
  100. {
  101. return (__force __be64 *)(bh->b_data + bh->b_size);
  102. }
  103. static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
  104. {
  105. struct buffer_head *bh = gfs2_log_get_buf(sdp);
  106. struct gfs2_log_descriptor *ld = bh_log_desc(bh);
  107. ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
  108. ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
  109. ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
  110. ld->ld_type = cpu_to_be32(ld_type);
  111. ld->ld_length = 0;
  112. ld->ld_data1 = 0;
  113. ld->ld_data2 = 0;
  114. memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
  115. return bh;
  116. }
  117. static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  118. {
  119. struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
  120. struct gfs2_meta_header *mh;
  121. struct gfs2_trans *tr;
  122. lock_buffer(bd->bd_bh);
  123. gfs2_log_lock(sdp);
  124. if (!list_empty(&bd->bd_list_tr))
  125. goto out;
  126. tr = current->journal_info;
  127. tr->tr_touched = 1;
  128. tr->tr_num_buf++;
  129. list_add(&bd->bd_list_tr, &tr->tr_list_buf);
  130. if (!list_empty(&le->le_list))
  131. goto out;
  132. set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
  133. set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
  134. gfs2_meta_check(sdp, bd->bd_bh);
  135. gfs2_pin(sdp, bd->bd_bh);
  136. mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
  137. mh->__pad0 = cpu_to_be64(0);
  138. mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
  139. sdp->sd_log_num_buf++;
  140. list_add(&le->le_list, &sdp->sd_log_le_buf);
  141. tr->tr_num_buf_new++;
  142. out:
  143. gfs2_log_unlock(sdp);
  144. unlock_buffer(bd->bd_bh);
  145. }
  146. static void buf_lo_before_commit(struct gfs2_sbd *sdp)
  147. {
  148. struct buffer_head *bh;
  149. struct gfs2_log_descriptor *ld;
  150. struct gfs2_bufdata *bd1 = NULL, *bd2;
  151. unsigned int total;
  152. unsigned int limit;
  153. unsigned int num;
  154. unsigned n;
  155. __be64 *ptr;
  156. limit = buf_limit(sdp);
  157. /* for 4k blocks, limit = 503 */
  158. gfs2_log_lock(sdp);
  159. total = sdp->sd_log_num_buf;
  160. bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
  161. while(total) {
  162. num = total;
  163. if (total > limit)
  164. num = limit;
  165. gfs2_log_unlock(sdp);
  166. bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
  167. gfs2_log_lock(sdp);
  168. ld = bh_log_desc(bh);
  169. ptr = bh_log_ptr(bh);
  170. ld->ld_length = cpu_to_be32(num + 1);
  171. ld->ld_data1 = cpu_to_be32(num);
  172. n = 0;
  173. list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
  174. bd_le.le_list) {
  175. *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
  176. if (++n >= num)
  177. break;
  178. }
  179. gfs2_log_unlock(sdp);
  180. submit_bh(WRITE_SYNC_PLUG, bh);
  181. gfs2_log_lock(sdp);
  182. n = 0;
  183. list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
  184. bd_le.le_list) {
  185. get_bh(bd2->bd_bh);
  186. gfs2_log_unlock(sdp);
  187. lock_buffer(bd2->bd_bh);
  188. bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
  189. submit_bh(WRITE_SYNC_PLUG, bh);
  190. gfs2_log_lock(sdp);
  191. if (++n >= num)
  192. break;
  193. }
  194. BUG_ON(total < num);
  195. total -= num;
  196. }
  197. gfs2_log_unlock(sdp);
  198. }
  199. static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  200. {
  201. struct list_head *head = &sdp->sd_log_le_buf;
  202. struct gfs2_bufdata *bd;
  203. while (!list_empty(head)) {
  204. bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
  205. list_del_init(&bd->bd_le.le_list);
  206. sdp->sd_log_num_buf--;
  207. gfs2_unpin(sdp, bd->bd_bh, ai);
  208. }
  209. gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
  210. }
  211. static void buf_lo_before_scan(struct gfs2_jdesc *jd,
  212. struct gfs2_log_header_host *head, int pass)
  213. {
  214. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  215. if (pass != 0)
  216. return;
  217. sdp->sd_found_blocks = 0;
  218. sdp->sd_replayed_blocks = 0;
  219. }
  220. static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
  221. struct gfs2_log_descriptor *ld, __be64 *ptr,
  222. int pass)
  223. {
  224. struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
  225. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  226. struct gfs2_glock *gl = ip->i_gl;
  227. unsigned int blks = be32_to_cpu(ld->ld_data1);
  228. struct buffer_head *bh_log, *bh_ip;
  229. u64 blkno;
  230. int error = 0;
  231. if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
  232. return 0;
  233. gfs2_replay_incr_blk(sdp, &start);
  234. for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
  235. blkno = be64_to_cpu(*ptr++);
  236. sdp->sd_found_blocks++;
  237. if (gfs2_revoke_check(sdp, blkno, start))
  238. continue;
  239. error = gfs2_replay_read_block(jd, start, &bh_log);
  240. if (error)
  241. return error;
  242. bh_ip = gfs2_meta_new(gl, blkno);
  243. memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
  244. if (gfs2_meta_check(sdp, bh_ip))
  245. error = -EIO;
  246. else
  247. mark_buffer_dirty(bh_ip);
  248. brelse(bh_log);
  249. brelse(bh_ip);
  250. if (error)
  251. break;
  252. sdp->sd_replayed_blocks++;
  253. }
  254. return error;
  255. }
  256. static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
  257. {
  258. struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
  259. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  260. if (error) {
  261. gfs2_meta_sync(ip->i_gl);
  262. return;
  263. }
  264. if (pass != 1)
  265. return;
  266. gfs2_meta_sync(ip->i_gl);
  267. fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
  268. jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
  269. }
  270. static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  271. {
  272. struct gfs2_trans *tr;
  273. tr = current->journal_info;
  274. tr->tr_touched = 1;
  275. tr->tr_num_revoke++;
  276. sdp->sd_log_num_revoke++;
  277. list_add(&le->le_list, &sdp->sd_log_le_revoke);
  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_bufdata *bd;
  287. if (!sdp->sd_log_num_revoke)
  288. return;
  289. bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
  290. ld = bh_log_desc(bh);
  291. ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
  292. sizeof(u64)));
  293. ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
  294. offset = sizeof(struct gfs2_log_descriptor);
  295. while (!list_empty(head)) {
  296. bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
  297. list_del_init(&bd->bd_le.le_list);
  298. sdp->sd_log_num_revoke--;
  299. if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
  300. submit_bh(WRITE_SYNC_PLUG, bh);
  301. bh = gfs2_log_get_buf(sdp);
  302. mh = (struct gfs2_meta_header *)bh->b_data;
  303. mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
  304. mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
  305. mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
  306. offset = sizeof(struct gfs2_meta_header);
  307. }
  308. *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
  309. kmem_cache_free(gfs2_bufdata_cachep, bd);
  310. offset += sizeof(u64);
  311. }
  312. gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
  313. submit_bh(WRITE_SYNC_PLUG, bh);
  314. }
  315. static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
  316. struct gfs2_log_header_host *head, int pass)
  317. {
  318. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  319. if (pass != 0)
  320. return;
  321. sdp->sd_found_revokes = 0;
  322. sdp->sd_replay_tail = head->lh_tail;
  323. }
  324. static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
  325. struct gfs2_log_descriptor *ld, __be64 *ptr,
  326. int pass)
  327. {
  328. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  329. unsigned int blks = be32_to_cpu(ld->ld_length);
  330. unsigned int revokes = be32_to_cpu(ld->ld_data1);
  331. struct buffer_head *bh;
  332. unsigned int offset;
  333. u64 blkno;
  334. int first = 1;
  335. int error;
  336. if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
  337. return 0;
  338. offset = sizeof(struct gfs2_log_descriptor);
  339. for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
  340. error = gfs2_replay_read_block(jd, start, &bh);
  341. if (error)
  342. return error;
  343. if (!first)
  344. gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
  345. while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
  346. blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
  347. error = gfs2_revoke_add(sdp, blkno, start);
  348. if (error < 0) {
  349. brelse(bh);
  350. return error;
  351. }
  352. else if (error)
  353. sdp->sd_found_revokes++;
  354. if (!--revokes)
  355. break;
  356. offset += sizeof(u64);
  357. }
  358. brelse(bh);
  359. offset = sizeof(struct gfs2_meta_header);
  360. first = 0;
  361. }
  362. return 0;
  363. }
  364. static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
  365. {
  366. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  367. if (error) {
  368. gfs2_revoke_clean(sdp);
  369. return;
  370. }
  371. if (pass != 1)
  372. return;
  373. fs_info(sdp, "jid=%u: Found %u revoke tags\n",
  374. jd->jd_jid, sdp->sd_found_revokes);
  375. gfs2_revoke_clean(sdp);
  376. }
  377. static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  378. {
  379. struct gfs2_rgrpd *rgd;
  380. struct gfs2_trans *tr = current->journal_info;
  381. tr->tr_touched = 1;
  382. rgd = container_of(le, struct gfs2_rgrpd, rd_le);
  383. gfs2_log_lock(sdp);
  384. if (!list_empty(&le->le_list)){
  385. gfs2_log_unlock(sdp);
  386. return;
  387. }
  388. gfs2_rgrp_bh_hold(rgd);
  389. sdp->sd_log_num_rg++;
  390. list_add(&le->le_list, &sdp->sd_log_le_rg);
  391. gfs2_log_unlock(sdp);
  392. }
  393. static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  394. {
  395. struct list_head *head = &sdp->sd_log_le_rg;
  396. struct gfs2_rgrpd *rgd;
  397. while (!list_empty(head)) {
  398. rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
  399. list_del_init(&rgd->rd_le.le_list);
  400. sdp->sd_log_num_rg--;
  401. gfs2_rgrp_repolish_clones(rgd);
  402. gfs2_rgrp_bh_put(rgd);
  403. }
  404. gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
  405. }
  406. /**
  407. * databuf_lo_add - Add a databuf to the transaction.
  408. *
  409. * This is used in two distinct cases:
  410. * i) In ordered write mode
  411. * We put the data buffer on a list so that we can ensure that its
  412. * synced to disk at the right time
  413. * ii) In journaled data mode
  414. * We need to journal the data block in the same way as metadata in
  415. * the functions above. The difference is that here we have a tag
  416. * which is two __be64's being the block number (as per meta data)
  417. * and a flag which says whether the data block needs escaping or
  418. * not. This means we need a new log entry for each 251 or so data
  419. * blocks, which isn't an enormous overhead but twice as much as
  420. * for normal metadata blocks.
  421. */
  422. static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  423. {
  424. struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
  425. struct gfs2_trans *tr = current->journal_info;
  426. struct address_space *mapping = bd->bd_bh->b_page->mapping;
  427. struct gfs2_inode *ip = GFS2_I(mapping->host);
  428. lock_buffer(bd->bd_bh);
  429. gfs2_log_lock(sdp);
  430. if (tr) {
  431. if (!list_empty(&bd->bd_list_tr))
  432. goto out;
  433. tr->tr_touched = 1;
  434. if (gfs2_is_jdata(ip)) {
  435. tr->tr_num_buf++;
  436. list_add(&bd->bd_list_tr, &tr->tr_list_buf);
  437. }
  438. }
  439. if (!list_empty(&le->le_list))
  440. goto out;
  441. set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
  442. set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
  443. if (gfs2_is_jdata(ip)) {
  444. gfs2_pin(sdp, bd->bd_bh);
  445. tr->tr_num_databuf_new++;
  446. sdp->sd_log_num_databuf++;
  447. list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
  448. } else {
  449. list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
  450. }
  451. out:
  452. gfs2_log_unlock(sdp);
  453. unlock_buffer(bd->bd_bh);
  454. }
  455. static void gfs2_check_magic(struct buffer_head *bh)
  456. {
  457. void *kaddr;
  458. __be32 *ptr;
  459. clear_buffer_escaped(bh);
  460. kaddr = kmap_atomic(bh->b_page, KM_USER0);
  461. ptr = kaddr + bh_offset(bh);
  462. if (*ptr == cpu_to_be32(GFS2_MAGIC))
  463. set_buffer_escaped(bh);
  464. kunmap_atomic(kaddr, KM_USER0);
  465. }
  466. static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
  467. struct list_head *list, struct list_head *done,
  468. unsigned int n)
  469. {
  470. struct buffer_head *bh1;
  471. struct gfs2_log_descriptor *ld;
  472. struct gfs2_bufdata *bd;
  473. __be64 *ptr;
  474. if (!bh)
  475. return;
  476. ld = bh_log_desc(bh);
  477. ld->ld_length = cpu_to_be32(n + 1);
  478. ld->ld_data1 = cpu_to_be32(n);
  479. ptr = bh_log_ptr(bh);
  480. get_bh(bh);
  481. submit_bh(WRITE_SYNC_PLUG, bh);
  482. gfs2_log_lock(sdp);
  483. while(!list_empty(list)) {
  484. bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
  485. list_move_tail(&bd->bd_le.le_list, done);
  486. get_bh(bd->bd_bh);
  487. while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
  488. gfs2_log_incr_head(sdp);
  489. ptr += 2;
  490. }
  491. gfs2_log_unlock(sdp);
  492. lock_buffer(bd->bd_bh);
  493. if (buffer_escaped(bd->bd_bh)) {
  494. void *kaddr;
  495. bh1 = gfs2_log_get_buf(sdp);
  496. kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
  497. memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
  498. bh1->b_size);
  499. kunmap_atomic(kaddr, KM_USER0);
  500. *(__be32 *)bh1->b_data = 0;
  501. clear_buffer_escaped(bd->bd_bh);
  502. unlock_buffer(bd->bd_bh);
  503. brelse(bd->bd_bh);
  504. } else {
  505. bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
  506. }
  507. submit_bh(WRITE_SYNC_PLUG, bh1);
  508. gfs2_log_lock(sdp);
  509. ptr += 2;
  510. }
  511. gfs2_log_unlock(sdp);
  512. brelse(bh);
  513. }
  514. /**
  515. * databuf_lo_before_commit - Scan the data buffers, writing as we go
  516. *
  517. */
  518. static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
  519. {
  520. struct gfs2_bufdata *bd = NULL;
  521. struct buffer_head *bh = NULL;
  522. unsigned int n = 0;
  523. __be64 *ptr = NULL, *end = NULL;
  524. LIST_HEAD(processed);
  525. LIST_HEAD(in_progress);
  526. gfs2_log_lock(sdp);
  527. while (!list_empty(&sdp->sd_log_le_databuf)) {
  528. if (ptr == end) {
  529. gfs2_log_unlock(sdp);
  530. gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
  531. n = 0;
  532. bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
  533. ptr = bh_log_ptr(bh);
  534. end = bh_ptr_end(bh) - 1;
  535. gfs2_log_lock(sdp);
  536. continue;
  537. }
  538. bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
  539. list_move_tail(&bd->bd_le.le_list, &in_progress);
  540. gfs2_check_magic(bd->bd_bh);
  541. *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
  542. *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
  543. n++;
  544. }
  545. gfs2_log_unlock(sdp);
  546. gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
  547. gfs2_log_lock(sdp);
  548. list_splice(&processed, &sdp->sd_log_le_databuf);
  549. gfs2_log_unlock(sdp);
  550. }
  551. static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
  552. struct gfs2_log_descriptor *ld,
  553. __be64 *ptr, int pass)
  554. {
  555. struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
  556. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  557. struct gfs2_glock *gl = ip->i_gl;
  558. unsigned int blks = be32_to_cpu(ld->ld_data1);
  559. struct buffer_head *bh_log, *bh_ip;
  560. u64 blkno;
  561. u64 esc;
  562. int error = 0;
  563. if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
  564. return 0;
  565. gfs2_replay_incr_blk(sdp, &start);
  566. for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
  567. blkno = be64_to_cpu(*ptr++);
  568. esc = be64_to_cpu(*ptr++);
  569. sdp->sd_found_blocks++;
  570. if (gfs2_revoke_check(sdp, blkno, start))
  571. continue;
  572. error = gfs2_replay_read_block(jd, start, &bh_log);
  573. if (error)
  574. return error;
  575. bh_ip = gfs2_meta_new(gl, blkno);
  576. memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
  577. /* Unescape */
  578. if (esc) {
  579. __be32 *eptr = (__be32 *)bh_ip->b_data;
  580. *eptr = cpu_to_be32(GFS2_MAGIC);
  581. }
  582. mark_buffer_dirty(bh_ip);
  583. brelse(bh_log);
  584. brelse(bh_ip);
  585. if (error)
  586. break;
  587. sdp->sd_replayed_blocks++;
  588. }
  589. return error;
  590. }
  591. /* FIXME: sort out accounting for log blocks etc. */
  592. static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
  593. {
  594. struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
  595. struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
  596. if (error) {
  597. gfs2_meta_sync(ip->i_gl);
  598. return;
  599. }
  600. if (pass != 1)
  601. return;
  602. /* data sync? */
  603. gfs2_meta_sync(ip->i_gl);
  604. fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
  605. jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
  606. }
  607. static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  608. {
  609. struct list_head *head = &sdp->sd_log_le_databuf;
  610. struct gfs2_bufdata *bd;
  611. while (!list_empty(head)) {
  612. bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
  613. list_del_init(&bd->bd_le.le_list);
  614. sdp->sd_log_num_databuf--;
  615. gfs2_unpin(sdp, bd->bd_bh, ai);
  616. }
  617. gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
  618. }
  619. const struct gfs2_log_operations gfs2_buf_lops = {
  620. .lo_add = buf_lo_add,
  621. .lo_before_commit = buf_lo_before_commit,
  622. .lo_after_commit = buf_lo_after_commit,
  623. .lo_before_scan = buf_lo_before_scan,
  624. .lo_scan_elements = buf_lo_scan_elements,
  625. .lo_after_scan = buf_lo_after_scan,
  626. .lo_name = "buf",
  627. };
  628. const struct gfs2_log_operations gfs2_revoke_lops = {
  629. .lo_add = revoke_lo_add,
  630. .lo_before_commit = revoke_lo_before_commit,
  631. .lo_before_scan = revoke_lo_before_scan,
  632. .lo_scan_elements = revoke_lo_scan_elements,
  633. .lo_after_scan = revoke_lo_after_scan,
  634. .lo_name = "revoke",
  635. };
  636. const struct gfs2_log_operations gfs2_rg_lops = {
  637. .lo_add = rg_lo_add,
  638. .lo_after_commit = rg_lo_after_commit,
  639. .lo_name = "rg",
  640. };
  641. const struct gfs2_log_operations gfs2_databuf_lops = {
  642. .lo_add = databuf_lo_add,
  643. .lo_before_commit = databuf_lo_before_commit,
  644. .lo_after_commit = databuf_lo_after_commit,
  645. .lo_scan_elements = databuf_lo_scan_elements,
  646. .lo_after_scan = databuf_lo_after_scan,
  647. .lo_name = "databuf",
  648. };
  649. const struct gfs2_log_operations *gfs2_log_ops[] = {
  650. &gfs2_databuf_lops,
  651. &gfs2_buf_lops,
  652. &gfs2_rg_lops,
  653. &gfs2_revoke_lops,
  654. NULL,
  655. };