journal.h 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621
  1. /* -*- mode: c; c-basic-offset: 8; -*-
  2. * vim: noexpandtab sw=8 ts=8 sts=0:
  3. *
  4. * journal.h
  5. *
  6. * Defines journalling api and structures.
  7. *
  8. * Copyright (C) 2003, 2005 Oracle. All rights reserved.
  9. *
  10. * This program is free software; you can redistribute it and/or
  11. * modify it under the terms of the GNU General Public
  12. * License as published by the Free Software Foundation; either
  13. * version 2 of the License, or (at your option) any later version.
  14. *
  15. * This program is distributed in the hope that it will be useful,
  16. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  18. * General Public License for more details.
  19. *
  20. * You should have received a copy of the GNU General Public
  21. * License along with this program; if not, write to the
  22. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23. * Boston, MA 021110-1307, USA.
  24. */
  25. #ifndef OCFS2_JOURNAL_H
  26. #define OCFS2_JOURNAL_H
  27. #include <linux/fs.h>
  28. #include <linux/jbd2.h>
  29. enum ocfs2_journal_state {
  30. OCFS2_JOURNAL_FREE = 0,
  31. OCFS2_JOURNAL_LOADED,
  32. OCFS2_JOURNAL_IN_SHUTDOWN,
  33. };
  34. struct ocfs2_super;
  35. struct ocfs2_dinode;
  36. /*
  37. * The recovery_list is a simple linked list of node numbers to recover.
  38. * It is protected by the recovery_lock.
  39. */
  40. struct ocfs2_recovery_map {
  41. unsigned int rm_used;
  42. unsigned int *rm_entries;
  43. };
  44. struct ocfs2_journal {
  45. enum ocfs2_journal_state j_state; /* Journals current state */
  46. journal_t *j_journal; /* The kernels journal type */
  47. struct inode *j_inode; /* Kernel inode pointing to
  48. * this journal */
  49. struct ocfs2_super *j_osb; /* pointer to the super
  50. * block for the node
  51. * we're currently
  52. * running on -- not
  53. * necessarily the super
  54. * block from the node
  55. * which we usually run
  56. * from (recovery,
  57. * etc) */
  58. struct buffer_head *j_bh; /* Journal disk inode block */
  59. atomic_t j_num_trans; /* Number of transactions
  60. * currently in the system. */
  61. spinlock_t j_lock;
  62. unsigned long j_trans_id;
  63. struct rw_semaphore j_trans_barrier;
  64. wait_queue_head_t j_checkpointed;
  65. /* both fields protected by j_lock*/
  66. struct list_head j_la_cleanups;
  67. struct work_struct j_recovery_work;
  68. };
  69. extern spinlock_t trans_inc_lock;
  70. /* wrap j_trans_id so we never have it equal to zero. */
  71. static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
  72. {
  73. unsigned long old_id;
  74. spin_lock(&trans_inc_lock);
  75. old_id = j->j_trans_id++;
  76. if (unlikely(!j->j_trans_id))
  77. j->j_trans_id = 1;
  78. spin_unlock(&trans_inc_lock);
  79. return old_id;
  80. }
  81. static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
  82. struct ocfs2_caching_info *ci)
  83. {
  84. spin_lock(&trans_inc_lock);
  85. ci->ci_last_trans = journal->j_trans_id;
  86. spin_unlock(&trans_inc_lock);
  87. }
  88. /* Used to figure out whether it's safe to drop a metadata lock on an
  89. * cached object. Returns true if all the object's changes have been
  90. * checkpointed to disk. You should be holding the spinlock on the
  91. * metadata lock while calling this to be sure that nobody can take
  92. * the lock and put it on another transaction. */
  93. static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
  94. {
  95. int ret;
  96. struct ocfs2_journal *journal =
  97. OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
  98. spin_lock(&trans_inc_lock);
  99. ret = time_after(journal->j_trans_id, ci->ci_last_trans);
  100. spin_unlock(&trans_inc_lock);
  101. return ret;
  102. }
  103. /* convenience function to check if an object backed by struct
  104. * ocfs2_caching_info is still new (has never hit disk) Will do you a
  105. * favor and set created_trans = 0 when you've
  106. * been checkpointed. returns '1' if the ci is still new. */
  107. static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
  108. {
  109. int ret;
  110. struct ocfs2_journal *journal =
  111. OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
  112. spin_lock(&trans_inc_lock);
  113. ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
  114. if (!ret)
  115. ci->ci_created_trans = 0;
  116. spin_unlock(&trans_inc_lock);
  117. return ret;
  118. }
  119. /* Wrapper for inodes so we can check system files */
  120. static inline int ocfs2_inode_is_new(struct inode *inode)
  121. {
  122. /* System files are never "new" as they're written out by
  123. * mkfs. This helps us early during mount, before we have the
  124. * journal open and j_trans_id could be junk. */
  125. if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
  126. return 0;
  127. return ocfs2_ci_is_new(INODE_CACHE(inode));
  128. }
  129. static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
  130. struct ocfs2_caching_info *ci)
  131. {
  132. spin_lock(&trans_inc_lock);
  133. ci->ci_created_trans = osb->journal->j_trans_id;
  134. spin_unlock(&trans_inc_lock);
  135. }
  136. /* Exported only for the journal struct init code in super.c. Do not call. */
  137. void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
  138. void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
  139. void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
  140. void ocfs2_orphan_scan_exit(struct ocfs2_super *osb);
  141. void ocfs2_complete_recovery(struct work_struct *work);
  142. void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
  143. int ocfs2_recovery_init(struct ocfs2_super *osb);
  144. void ocfs2_recovery_exit(struct ocfs2_super *osb);
  145. int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
  146. /*
  147. * Journal Control:
  148. * Initialize, Load, Shutdown, Wipe a journal.
  149. *
  150. * ocfs2_journal_init - Initialize journal structures in the OSB.
  151. * ocfs2_journal_load - Load the given journal off disk. Replay it if
  152. * there's transactions still in there.
  153. * ocfs2_journal_shutdown - Shutdown a journal, this will flush all
  154. * uncommitted, uncheckpointed transactions.
  155. * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally
  156. * zero out each block.
  157. * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb.
  158. * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
  159. * event on.
  160. * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
  161. */
  162. void ocfs2_set_journal_params(struct ocfs2_super *osb);
  163. int ocfs2_journal_init(struct ocfs2_journal *journal,
  164. int *dirty);
  165. void ocfs2_journal_shutdown(struct ocfs2_super *osb);
  166. int ocfs2_journal_wipe(struct ocfs2_journal *journal,
  167. int full);
  168. int ocfs2_journal_load(struct ocfs2_journal *journal, int local,
  169. int replayed);
  170. int ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
  171. void ocfs2_recovery_thread(struct ocfs2_super *osb,
  172. int node_num);
  173. int ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
  174. void ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
  175. void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
  176. static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
  177. {
  178. atomic_set(&osb->needs_checkpoint, 1);
  179. wake_up(&osb->checkpoint_event);
  180. }
  181. static inline void ocfs2_checkpoint_inode(struct inode *inode)
  182. {
  183. struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
  184. if (ocfs2_mount_local(osb))
  185. return;
  186. if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
  187. /* WARNING: This only kicks off a single
  188. * checkpoint. If someone races you and adds more
  189. * metadata to the journal, you won't know, and will
  190. * wind up waiting *a lot* longer than necessary. Right
  191. * now we only use this in clear_inode so that's
  192. * OK. */
  193. ocfs2_start_checkpoint(osb);
  194. wait_event(osb->journal->j_checkpointed,
  195. ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
  196. }
  197. }
  198. /*
  199. * Transaction Handling:
  200. * Manage the lifetime of a transaction handle.
  201. *
  202. * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of
  203. * the number of blocks that will be changed during
  204. * this handle.
  205. * ocfs2_commit_trans - Complete a handle. It might return -EIO if
  206. * the journal was aborted. The majority of paths don't
  207. * check the return value as an error there comes too
  208. * late to do anything (and will be picked up in a
  209. * later transaction).
  210. * ocfs2_extend_trans - Extend a handle by nblocks credits. This may
  211. * commit the handle to disk in the process, but will
  212. * not release any locks taken during the transaction.
  213. * ocfs2_journal_access* - Notify the handle that we want to journal this
  214. * buffer. Will have to call ocfs2_journal_dirty once
  215. * we've actually dirtied it. Type is one of . or .
  216. * Always call the specific flavor of
  217. * ocfs2_journal_access_*() unless you intend to
  218. * manage the checksum by hand.
  219. * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data.
  220. * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before
  221. * the current handle commits.
  222. */
  223. /* You must always start_trans with a number of buffs > 0, but it's
  224. * perfectly legal to go through an entire transaction without having
  225. * dirtied any buffers. */
  226. handle_t *ocfs2_start_trans(struct ocfs2_super *osb,
  227. int max_buffs);
  228. int ocfs2_commit_trans(struct ocfs2_super *osb,
  229. handle_t *handle);
  230. int ocfs2_extend_trans(handle_t *handle, int nblocks);
  231. /*
  232. * Create access is for when we get a newly created buffer and we're
  233. * not gonna read it off disk, but rather fill it ourselves. Right
  234. * now, we don't do anything special with this (it turns into a write
  235. * request), but this is a good placeholder in case we do...
  236. *
  237. * Write access is for when we read a block off disk and are going to
  238. * modify it. This way the journalling layer knows it may need to make
  239. * a copy of that block (if it's part of another, uncommitted
  240. * transaction) before we do so.
  241. */
  242. #define OCFS2_JOURNAL_ACCESS_CREATE 0
  243. #define OCFS2_JOURNAL_ACCESS_WRITE 1
  244. #define OCFS2_JOURNAL_ACCESS_UNDO 2
  245. /* ocfs2_inode */
  246. int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
  247. struct buffer_head *bh, int type);
  248. /* ocfs2_extent_block */
  249. int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
  250. struct buffer_head *bh, int type);
  251. /* ocfs2_refcount_block */
  252. int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
  253. struct buffer_head *bh, int type);
  254. /* ocfs2_group_desc */
  255. int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
  256. struct buffer_head *bh, int type);
  257. /* ocfs2_xattr_block */
  258. int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
  259. struct buffer_head *bh, int type);
  260. /* quota blocks */
  261. int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
  262. struct buffer_head *bh, int type);
  263. /* dirblock */
  264. int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
  265. struct buffer_head *bh, int type);
  266. /* ocfs2_dx_root_block */
  267. int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
  268. struct buffer_head *bh, int type);
  269. /* ocfs2_dx_leaf */
  270. int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
  271. struct buffer_head *bh, int type);
  272. /* Anything that has no ecc */
  273. int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
  274. struct buffer_head *bh, int type);
  275. /*
  276. * A word about the journal_access/journal_dirty "dance". It is
  277. * entirely legal to journal_access a buffer more than once (as long
  278. * as the access type is the same -- I'm not sure what will happen if
  279. * access type is different but this should never happen anyway) It is
  280. * also legal to journal_dirty a buffer more than once. In fact, you
  281. * can even journal_access a buffer after you've done a
  282. * journal_access/journal_dirty pair. The only thing you cannot do
  283. * however, is journal_dirty a buffer which you haven't yet passed to
  284. * journal_access at least once.
  285. *
  286. * That said, 99% of the time this doesn't matter and this is what the
  287. * path looks like:
  288. *
  289. * <read a bh>
  290. * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE);
  291. * <modify the bh>
  292. * ocfs2_journal_dirty(handle, bh);
  293. */
  294. void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
  295. /*
  296. * Credit Macros:
  297. * Convenience macros to calculate number of credits needed.
  298. *
  299. * For convenience sake, I have a set of macros here which calculate
  300. * the *maximum* number of sectors which will be changed for various
  301. * metadata updates.
  302. */
  303. /* simple file updates like chmod, etc. */
  304. #define OCFS2_INODE_UPDATE_CREDITS 1
  305. /* extended attribute block update */
  306. #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
  307. /* Update of a single quota block */
  308. #define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
  309. /* global quotafile inode update, data block */
  310. #define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
  311. OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
  312. #define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
  313. /*
  314. * The two writes below can accidentally see global info dirty due
  315. * to set_info() quotactl so make them prepared for the writes.
  316. */
  317. /* quota data block, global info */
  318. /* Write to local quota file */
  319. #define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
  320. OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
  321. /* global quota data block, local quota data block, global quota inode,
  322. * global quota info */
  323. #define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
  324. 2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
  325. static inline int ocfs2_quota_trans_credits(struct super_block *sb)
  326. {
  327. int credits = 0;
  328. if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
  329. credits += OCFS2_QWRITE_CREDITS;
  330. if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
  331. credits += OCFS2_QWRITE_CREDITS;
  332. return credits;
  333. }
  334. /* group extend. inode update and last group update. */
  335. #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
  336. /* group add. inode update and the new group update. */
  337. #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
  338. /* get one bit out of a suballocator: dinode + group descriptor +
  339. * prev. group desc. if we relink. */
  340. #define OCFS2_SUBALLOC_ALLOC (3)
  341. static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
  342. {
  343. return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
  344. ocfs2_quota_trans_credits(sb);
  345. }
  346. /* dinode + group descriptor update. We don't relink on free yet. */
  347. #define OCFS2_SUBALLOC_FREE (2)
  348. #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
  349. #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \
  350. + OCFS2_TRUNCATE_LOG_UPDATE)
  351. static inline int ocfs2_remove_extent_credits(struct super_block *sb)
  352. {
  353. return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
  354. ocfs2_quota_trans_credits(sb);
  355. }
  356. /* data block for new dir/symlink, allocation of directory block, dx_root
  357. * update for free list */
  358. #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
  359. static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
  360. {
  361. /* 1 block for index, 2 allocs (data, metadata), 1 clusters
  362. * worth of blocks for initial extent. */
  363. return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
  364. ocfs2_clusters_to_blocks(sb, 1);
  365. }
  366. /* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
  367. * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
  368. * blocks + quota update */
  369. static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
  370. int xattr_credits)
  371. {
  372. int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
  373. if (is_dir)
  374. dir_credits += ocfs2_add_dir_index_credits(sb);
  375. return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
  376. ocfs2_quota_trans_credits(sb);
  377. }
  378. /* local alloc metadata change + main bitmap updates */
  379. #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \
  380. + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
  381. /* used when we don't need an allocation change for a dir extend. One
  382. * for the dinode, one for the new block. */
  383. #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
  384. /* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
  385. * update on dir + index leaf + dx root update for free list +
  386. * previous dirblock update in the free list */
  387. static inline int ocfs2_link_credits(struct super_block *sb)
  388. {
  389. return 2*OCFS2_INODE_UPDATE_CREDITS + 4 +
  390. ocfs2_quota_trans_credits(sb);
  391. }
  392. /* inode + dir inode (if we unlink a dir), + dir entry block + orphan
  393. * dir inode link + dir inode index leaf + dir index root */
  394. static inline int ocfs2_unlink_credits(struct super_block *sb)
  395. {
  396. /* The quota update from ocfs2_link_credits is unused here... */
  397. return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
  398. }
  399. /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
  400. * inode alloc group descriptor + orphan dir index root +
  401. * orphan dir index leaf */
  402. #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
  403. /* dinode update, old dir dinode update, new dir dinode update, old
  404. * dir dir entry, new dir dir entry, dir entry update for renaming
  405. * directory + target unlink + 3 x dir index leaves */
  406. static inline int ocfs2_rename_credits(struct super_block *sb)
  407. {
  408. return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
  409. }
  410. /* global bitmap dinode, group desc., relinked group,
  411. * suballocator dinode, group desc., relinked group,
  412. * dinode, xattr block */
  413. #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
  414. + OCFS2_INODE_UPDATE_CREDITS \
  415. + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
  416. /* inode update, removal of dx root block from allocator */
  417. #define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
  418. OCFS2_SUBALLOC_FREE)
  419. static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
  420. {
  421. int credits = 1 + OCFS2_SUBALLOC_ALLOC;
  422. credits += ocfs2_clusters_to_blocks(sb, 1);
  423. credits += ocfs2_quota_trans_credits(sb);
  424. return credits;
  425. }
  426. /* inode update, new refcount block and its allocation credits. */
  427. #define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
  428. + OCFS2_SUBALLOC_ALLOC)
  429. /* inode and the refcount block update. */
  430. #define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
  431. /*
  432. * inode and the refcount block update.
  433. * It doesn't include the credits for sub alloc change.
  434. * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
  435. */
  436. #define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
  437. /* 2 metadata alloc, 2 new blocks and root refcount block */
  438. #define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
  439. /*
  440. * Please note that the caller must make sure that root_el is the root
  441. * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
  442. * the result may be wrong.
  443. */
  444. static inline int ocfs2_calc_extend_credits(struct super_block *sb,
  445. struct ocfs2_extent_list *root_el,
  446. u32 bits_wanted)
  447. {
  448. int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
  449. /* bitmap dinode, group desc. + relinked group. */
  450. bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
  451. /* we might need to shift tree depth so lets assume an
  452. * absolute worst case of complete fragmentation. Even with
  453. * that, we only need one update for the dinode, and then
  454. * however many metadata chunks needed * a remaining suballoc
  455. * alloc. */
  456. sysfile_bitmap_blocks = 1 +
  457. (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
  458. /* this does not include *new* metadata blocks, which are
  459. * accounted for in sysfile_bitmap_blocks. root_el +
  460. * prev. last_eb_blk + blocks along edge of tree.
  461. * calc_symlink_credits passes because we just need 1
  462. * credit for the dinode there. */
  463. extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
  464. return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
  465. ocfs2_quota_trans_credits(sb);
  466. }
  467. static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
  468. {
  469. int blocks = ocfs2_mknod_credits(sb, 0, 0);
  470. /* links can be longer than one block so we may update many
  471. * within our single allocated extent. */
  472. blocks += ocfs2_clusters_to_blocks(sb, 1);
  473. return blocks + ocfs2_quota_trans_credits(sb);
  474. }
  475. static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
  476. unsigned int cpg)
  477. {
  478. int blocks;
  479. int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
  480. /* parent inode update + new block group header + bitmap inode update
  481. + bitmap blocks affected */
  482. blocks = 1 + 1 + 1 + bitmap_blocks;
  483. return blocks;
  484. }
  485. /*
  486. * Allocating a discontiguous block group requires the credits from
  487. * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
  488. * the group descriptor's extent list. The caller already has started
  489. * the transaction with ocfs2_calc_group_alloc_credits(). They extend
  490. * it with these credits.
  491. */
  492. static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
  493. {
  494. return ocfs2_extent_recs_per_gd(sb);
  495. }
  496. static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
  497. unsigned int clusters_to_del,
  498. struct ocfs2_dinode *fe,
  499. struct ocfs2_extent_list *last_el)
  500. {
  501. /* for dinode + all headers in this pass + update to next leaf */
  502. u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
  503. u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
  504. int credits = 1 + tree_depth + 1;
  505. int i;
  506. i = next_free - 1;
  507. BUG_ON(i < 0);
  508. /* We may be deleting metadata blocks, so metadata alloc dinode +
  509. one desc. block for each possible delete. */
  510. if (tree_depth && next_free == 1 &&
  511. ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
  512. credits += 1 + tree_depth;
  513. /* update to the truncate log. */
  514. credits += OCFS2_TRUNCATE_LOG_UPDATE;
  515. credits += ocfs2_quota_trans_credits(sb);
  516. return credits;
  517. }
  518. static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
  519. {
  520. return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
  521. }
  522. static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
  523. loff_t new_size)
  524. {
  525. return jbd2_journal_begin_ordered_truncate(
  526. OCFS2_SB(inode->i_sb)->journal->j_journal,
  527. &OCFS2_I(inode)->ip_jinode,
  528. new_size);
  529. }
  530. #endif /* OCFS2_JOURNAL_H */