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 26 #ifndef OCFS2_JOURNAL_H 27 #define OCFS2_JOURNAL_H 28 29 #include <linux/fs.h> 30 #ifndef CONFIG_OCFS2_COMPAT_JBD 31 # include <linux/jbd2.h> 32 #else 33 # include <linux/jbd.h> 34 # include "ocfs2_jbd_compat.h" 35 #endif 36 37 enum ocfs2_journal_state { 38 OCFS2_JOURNAL_FREE = 0, 39 OCFS2_JOURNAL_LOADED, 40 OCFS2_JOURNAL_IN_SHUTDOWN, 41 }; 42 43 struct ocfs2_super; 44 struct ocfs2_dinode; 45 46 struct ocfs2_journal { 47 enum ocfs2_journal_state j_state; /* Journals current state */ 48 49 journal_t *j_journal; /* The kernels journal type */ 50 struct inode *j_inode; /* Kernel inode pointing to 51 * this journal */ 52 struct ocfs2_super *j_osb; /* pointer to the super 53 * block for the node 54 * we're currently 55 * running on -- not 56 * necessarily the super 57 * block from the node 58 * which we usually run 59 * from (recovery, 60 * etc) */ 61 struct buffer_head *j_bh; /* Journal disk inode block */ 62 atomic_t j_num_trans; /* Number of transactions 63 * currently in the system. */ 64 unsigned long j_trans_id; 65 struct rw_semaphore j_trans_barrier; 66 wait_queue_head_t j_checkpointed; 67 68 spinlock_t j_lock; 69 struct list_head j_la_cleanups; 70 struct work_struct j_recovery_work; 71 }; 72 73 extern spinlock_t trans_inc_lock; 74 75 /* wrap j_trans_id so we never have it equal to zero. */ 76 static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j) 77 { 78 unsigned long old_id; 79 spin_lock(&trans_inc_lock); 80 old_id = j->j_trans_id++; 81 if (unlikely(!j->j_trans_id)) 82 j->j_trans_id = 1; 83 spin_unlock(&trans_inc_lock); 84 return old_id; 85 } 86 87 static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal *journal, 88 struct inode *inode) 89 { 90 spin_lock(&trans_inc_lock); 91 OCFS2_I(inode)->ip_last_trans = journal->j_trans_id; 92 spin_unlock(&trans_inc_lock); 93 } 94 95 /* Used to figure out whether it's safe to drop a metadata lock on an 96 * inode. Returns true if all the inodes changes have been 97 * checkpointed to disk. You should be holding the spinlock on the 98 * metadata lock while calling this to be sure that nobody can take 99 * the lock and put it on another transaction. */ 100 static inline int ocfs2_inode_fully_checkpointed(struct inode *inode) 101 { 102 int ret; 103 struct ocfs2_journal *journal = OCFS2_SB(inode->i_sb)->journal; 104 105 spin_lock(&trans_inc_lock); 106 ret = time_after(journal->j_trans_id, OCFS2_I(inode)->ip_last_trans); 107 spin_unlock(&trans_inc_lock); 108 return ret; 109 } 110 111 /* convenience function to check if an inode is still new (has never 112 * hit disk) Will do you a favor and set created_trans = 0 when you've 113 * been checkpointed. returns '1' if the inode is still new. */ 114 static inline int ocfs2_inode_is_new(struct inode *inode) 115 { 116 int ret; 117 118 /* System files are never "new" as they're written out by 119 * mkfs. This helps us early during mount, before we have the 120 * journal open and j_trans_id could be junk. */ 121 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) 122 return 0; 123 spin_lock(&trans_inc_lock); 124 ret = !(time_after(OCFS2_SB(inode->i_sb)->journal->j_trans_id, 125 OCFS2_I(inode)->ip_created_trans)); 126 if (!ret) 127 OCFS2_I(inode)->ip_created_trans = 0; 128 spin_unlock(&trans_inc_lock); 129 return ret; 130 } 131 132 static inline void ocfs2_inode_set_new(struct ocfs2_super *osb, 133 struct inode *inode) 134 { 135 spin_lock(&trans_inc_lock); 136 OCFS2_I(inode)->ip_created_trans = osb->journal->j_trans_id; 137 spin_unlock(&trans_inc_lock); 138 } 139 140 /* Exported only for the journal struct init code in super.c. Do not call. */ 141 void ocfs2_complete_recovery(struct work_struct *work); 142 void ocfs2_wait_for_recovery(struct ocfs2_super *osb); 143 144 int ocfs2_recovery_init(struct ocfs2_super *osb); 145 void ocfs2_recovery_exit(struct ocfs2_super *osb); 146 147 /* 148 * Journal Control: 149 * Initialize, Load, Shutdown, Wipe a journal. 150 * 151 * ocfs2_journal_init - Initialize journal structures in the OSB. 152 * ocfs2_journal_load - Load the given journal off disk. Replay it if 153 * there's transactions still in there. 154 * ocfs2_journal_shutdown - Shutdown a journal, this will flush all 155 * uncommitted, uncheckpointed transactions. 156 * ocfs2_journal_wipe - Wipe transactions from a journal. Optionally 157 * zero out each block. 158 * ocfs2_recovery_thread - Perform recovery on a node. osb is our own osb. 159 * ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat 160 * event on. 161 * ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint. 162 */ 163 void ocfs2_set_journal_params(struct ocfs2_super *osb); 164 int ocfs2_journal_init(struct ocfs2_journal *journal, 165 int *dirty); 166 void ocfs2_journal_shutdown(struct ocfs2_super *osb); 167 int ocfs2_journal_wipe(struct ocfs2_journal *journal, 168 int full); 169 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, 170 int replayed); 171 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb); 172 void ocfs2_recovery_thread(struct ocfs2_super *osb, 173 int node_num); 174 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb); 175 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb); 176 177 static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb) 178 { 179 atomic_set(&osb->needs_checkpoint, 1); 180 wake_up(&osb->checkpoint_event); 181 } 182 183 static inline void ocfs2_checkpoint_inode(struct inode *inode) 184 { 185 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 186 187 if (ocfs2_mount_local(osb)) 188 return; 189 190 if (!ocfs2_inode_fully_checkpointed(inode)) { 191 /* WARNING: This only kicks off a single 192 * checkpoint. If someone races you and adds more 193 * metadata to the journal, you won't know, and will 194 * wind up waiting *alot* longer than necessary. Right 195 * now we only use this in clear_inode so that's 196 * OK. */ 197 ocfs2_start_checkpoint(osb); 198 199 wait_event(osb->journal->j_checkpointed, 200 ocfs2_inode_fully_checkpointed(inode)); 201 } 202 } 203 204 /* 205 * Transaction Handling: 206 * Manage the lifetime of a transaction handle. 207 * 208 * ocfs2_start_trans - Begin a transaction. Give it an upper estimate of 209 * the number of blocks that will be changed during 210 * this handle. 211 * ocfs2_commit_trans - Complete a handle. It might return -EIO if 212 * the journal was aborted. The majority of paths don't 213 * check the return value as an error there comes too 214 * late to do anything (and will be picked up in a 215 * later transaction). 216 * ocfs2_extend_trans - Extend a handle by nblocks credits. This may 217 * commit the handle to disk in the process, but will 218 * not release any locks taken during the transaction. 219 * ocfs2_journal_access - Notify the handle that we want to journal this 220 * buffer. Will have to call ocfs2_journal_dirty once 221 * we've actually dirtied it. Type is one of . or . 222 * ocfs2_journal_dirty - Mark a journalled buffer as having dirty data. 223 * ocfs2_jbd2_file_inode - Mark an inode so that its data goes out before 224 * the current handle commits. 225 */ 226 227 /* You must always start_trans with a number of buffs > 0, but it's 228 * perfectly legal to go through an entire transaction without having 229 * dirtied any buffers. */ 230 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, 231 int max_buffs); 232 int ocfs2_commit_trans(struct ocfs2_super *osb, 233 handle_t *handle); 234 int ocfs2_extend_trans(handle_t *handle, int nblocks); 235 236 /* 237 * Create access is for when we get a newly created buffer and we're 238 * not gonna read it off disk, but rather fill it ourselves. Right 239 * now, we don't do anything special with this (it turns into a write 240 * request), but this is a good placeholder in case we do... 241 * 242 * Write access is for when we read a block off disk and are going to 243 * modify it. This way the journalling layer knows it may need to make 244 * a copy of that block (if it's part of another, uncommitted 245 * transaction) before we do so. 246 */ 247 #define OCFS2_JOURNAL_ACCESS_CREATE 0 248 #define OCFS2_JOURNAL_ACCESS_WRITE 1 249 #define OCFS2_JOURNAL_ACCESS_UNDO 2 250 251 int ocfs2_journal_access(handle_t *handle, 252 struct inode *inode, 253 struct buffer_head *bh, 254 int type); 255 /* 256 * A word about the journal_access/journal_dirty "dance". It is 257 * entirely legal to journal_access a buffer more than once (as long 258 * as the access type is the same -- I'm not sure what will happen if 259 * access type is different but this should never happen anyway) It is 260 * also legal to journal_dirty a buffer more than once. In fact, you 261 * can even journal_access a buffer after you've done a 262 * journal_access/journal_dirty pair. The only thing you cannot do 263 * however, is journal_dirty a buffer which you haven't yet passed to 264 * journal_access at least once. 265 * 266 * That said, 99% of the time this doesn't matter and this is what the 267 * path looks like: 268 * 269 * <read a bh> 270 * ocfs2_journal_access(handle, bh, OCFS2_JOURNAL_ACCESS_WRITE); 271 * <modify the bh> 272 * ocfs2_journal_dirty(handle, bh); 273 */ 274 int ocfs2_journal_dirty(handle_t *handle, 275 struct buffer_head *bh); 276 #ifdef CONFIG_OCFS2_COMPAT_JBD 277 int ocfs2_journal_dirty_data(handle_t *handle, 278 struct buffer_head *bh); 279 #endif 280 281 /* 282 * Credit Macros: 283 * Convenience macros to calculate number of credits needed. 284 * 285 * For convenience sake, I have a set of macros here which calculate 286 * the *maximum* number of sectors which will be changed for various 287 * metadata updates. 288 */ 289 290 /* simple file updates like chmod, etc. */ 291 #define OCFS2_INODE_UPDATE_CREDITS 1 292 293 /* extended attribute block update */ 294 #define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1 295 296 /* group extend. inode update and last group update. */ 297 #define OCFS2_GROUP_EXTEND_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1) 298 299 /* group add. inode update and the new group update. */ 300 #define OCFS2_GROUP_ADD_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1) 301 302 /* get one bit out of a suballocator: dinode + group descriptor + 303 * prev. group desc. if we relink. */ 304 #define OCFS2_SUBALLOC_ALLOC (3) 305 306 #define OCFS2_INLINE_TO_EXTENTS_CREDITS (OCFS2_SUBALLOC_ALLOC \ 307 + OCFS2_INODE_UPDATE_CREDITS) 308 309 /* dinode + group descriptor update. We don't relink on free yet. */ 310 #define OCFS2_SUBALLOC_FREE (2) 311 312 #define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS 313 #define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE \ 314 + OCFS2_TRUNCATE_LOG_UPDATE) 315 316 #define OCFS2_REMOVE_EXTENT_CREDITS (OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS) 317 318 /* data block for new dir/symlink, 2 for bitmap updates (bitmap fe + 319 * bitmap block for the new bit) */ 320 #define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + 2) 321 322 /* parent fe, parent block, new file entry, inode alloc fe, inode alloc 323 * group descriptor + mkdir/symlink blocks */ 324 #define OCFS2_MKNOD_CREDITS (3 + OCFS2_SUBALLOC_ALLOC \ 325 + OCFS2_DIR_LINK_ADDITIONAL_CREDITS) 326 327 /* local alloc metadata change + main bitmap updates */ 328 #define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS \ 329 + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE) 330 331 /* used when we don't need an allocation change for a dir extend. One 332 * for the dinode, one for the new block. */ 333 #define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2) 334 335 /* file update (nlink, etc) + directory mtime/ctime + dir entry block */ 336 #define OCFS2_LINK_CREDITS (2*OCFS2_INODE_UPDATE_CREDITS + 1) 337 338 /* inode + dir inode (if we unlink a dir), + dir entry block + orphan 339 * dir inode link */ 340 #define OCFS2_UNLINK_CREDITS (2 * OCFS2_INODE_UPDATE_CREDITS + 1 \ 341 + OCFS2_LINK_CREDITS) 342 343 /* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry + 344 * inode alloc group descriptor */ 345 #define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 1 + 1) 346 347 /* dinode update, old dir dinode update, new dir dinode update, old 348 * dir dir entry, new dir dir entry, dir entry update for renaming 349 * directory + target unlink */ 350 #define OCFS2_RENAME_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 3 \ 351 + OCFS2_UNLINK_CREDITS) 352 353 /* global bitmap dinode, group desc., relinked group, 354 * suballocator dinode, group desc., relinked group, 355 * dinode, xattr block */ 356 #define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \ 357 + OCFS2_INODE_UPDATE_CREDITS \ 358 + OCFS2_XATTR_BLOCK_UPDATE_CREDITS) 359 360 /* 361 * Please note that the caller must make sure that root_el is the root 362 * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise 363 * the result may be wrong. 364 */ 365 static inline int ocfs2_calc_extend_credits(struct super_block *sb, 366 struct ocfs2_extent_list *root_el, 367 u32 bits_wanted) 368 { 369 int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks; 370 371 /* bitmap dinode, group desc. + relinked group. */ 372 bitmap_blocks = OCFS2_SUBALLOC_ALLOC; 373 374 /* we might need to shift tree depth so lets assume an 375 * absolute worst case of complete fragmentation. Even with 376 * that, we only need one update for the dinode, and then 377 * however many metadata chunks needed * a remaining suballoc 378 * alloc. */ 379 sysfile_bitmap_blocks = 1 + 380 (OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el); 381 382 /* this does not include *new* metadata blocks, which are 383 * accounted for in sysfile_bitmap_blocks. root_el + 384 * prev. last_eb_blk + blocks along edge of tree. 385 * calc_symlink_credits passes because we just need 1 386 * credit for the dinode there. */ 387 extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth); 388 389 return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks; 390 } 391 392 static inline int ocfs2_calc_symlink_credits(struct super_block *sb) 393 { 394 int blocks = OCFS2_MKNOD_CREDITS; 395 396 /* links can be longer than one block so we may update many 397 * within our single allocated extent. */ 398 blocks += ocfs2_clusters_to_blocks(sb, 1); 399 400 return blocks; 401 } 402 403 static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb, 404 unsigned int cpg) 405 { 406 int blocks; 407 int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1; 408 /* parent inode update + new block group header + bitmap inode update 409 + bitmap blocks affected */ 410 blocks = 1 + 1 + 1 + bitmap_blocks; 411 return blocks; 412 } 413 414 static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb, 415 unsigned int clusters_to_del, 416 struct ocfs2_dinode *fe, 417 struct ocfs2_extent_list *last_el) 418 { 419 /* for dinode + all headers in this pass + update to next leaf */ 420 u16 next_free = le16_to_cpu(last_el->l_next_free_rec); 421 u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth); 422 int credits = 1 + tree_depth + 1; 423 int i; 424 425 i = next_free - 1; 426 BUG_ON(i < 0); 427 428 /* We may be deleting metadata blocks, so metadata alloc dinode + 429 one desc. block for each possible delete. */ 430 if (tree_depth && next_free == 1 && 431 ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del) 432 credits += 1 + tree_depth; 433 434 /* update to the truncate log. */ 435 credits += OCFS2_TRUNCATE_LOG_UPDATE; 436 437 return credits; 438 } 439 440 static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode) 441 { 442 return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode); 443 } 444 445 static inline int ocfs2_begin_ordered_truncate(struct inode *inode, 446 loff_t new_size) 447 { 448 return jbd2_journal_begin_ordered_truncate(&OCFS2_I(inode)->ip_jinode, 449 new_size); 450 } 451 452 #endif /* OCFS2_JOURNAL_H */ 453