Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/fs/ext3/inode.c |
| 3 | * |
| 4 | * Copyright (C) 1992, 1993, 1994, 1995 |
| 5 | * Remy Card (card@masi.ibp.fr) |
| 6 | * Laboratoire MASI - Institut Blaise Pascal |
| 7 | * Universite Pierre et Marie Curie (Paris VI) |
| 8 | * |
| 9 | * from |
| 10 | * |
| 11 | * linux/fs/minix/inode.c |
| 12 | * |
| 13 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 14 | * |
| 15 | * Goal-directed block allocation by Stephen Tweedie |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 16 | * (sct@redhat.com), 1993, 1998 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | * Big-endian to little-endian byte-swapping/bitmaps by |
| 18 | * David S. Miller (davem@caip.rutgers.edu), 1995 |
| 19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 20 | * (jj@sunsite.ms.mff.cuni.cz) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | * |
| 22 | * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000 |
| 23 | */ |
| 24 | |
| 25 | #include <linux/module.h> |
| 26 | #include <linux/fs.h> |
| 27 | #include <linux/time.h> |
| 28 | #include <linux/ext3_jbd.h> |
| 29 | #include <linux/jbd.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/highuid.h> |
| 31 | #include <linux/pagemap.h> |
| 32 | #include <linux/quotaops.h> |
| 33 | #include <linux/string.h> |
| 34 | #include <linux/buffer_head.h> |
| 35 | #include <linux/writeback.h> |
| 36 | #include <linux/mpage.h> |
| 37 | #include <linux/uio.h> |
Jens Axboe | caa38fb | 2006-07-23 01:41:26 +0200 | [diff] [blame] | 38 | #include <linux/bio.h> |
Josef Bacik | 68c9d70 | 2008-10-03 17:32:43 -0400 | [diff] [blame] | 39 | #include <linux/fiemap.h> |
Duane Griffin | b5ed311 | 2008-12-19 20:47:14 +0000 | [diff] [blame] | 40 | #include <linux/namei.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | #include "xattr.h" |
| 42 | #include "acl.h" |
| 43 | |
| 44 | static int ext3_writepage_trans_blocks(struct inode *inode); |
| 45 | |
| 46 | /* |
| 47 | * Test whether an inode is a fast symlink. |
| 48 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 49 | static int ext3_inode_is_fast_symlink(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | { |
| 51 | int ea_blocks = EXT3_I(inode)->i_file_acl ? |
| 52 | (inode->i_sb->s_blocksize >> 9) : 0; |
| 53 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 54 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | } |
| 56 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 57 | /* |
| 58 | * The ext3 forget function must perform a revoke if we are freeing data |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | * which has been journaled. Metadata (eg. indirect blocks) must be |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 60 | * revoked in all cases. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | * |
| 62 | * "bh" may be NULL: a metadata block may have been freed from memory |
| 63 | * but there may still be a record of it in the journal, and that record |
| 64 | * still needs to be revoked. |
| 65 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 66 | int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode, |
Mingming Cao | 1c2bf37 | 2006-06-25 05:48:06 -0700 | [diff] [blame] | 67 | struct buffer_head *bh, ext3_fsblk_t blocknr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | { |
| 69 | int err; |
| 70 | |
| 71 | might_sleep(); |
| 72 | |
| 73 | BUFFER_TRACE(bh, "enter"); |
| 74 | |
| 75 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " |
| 76 | "data mode %lx\n", |
| 77 | bh, is_metadata, inode->i_mode, |
| 78 | test_opt(inode->i_sb, DATA_FLAGS)); |
| 79 | |
| 80 | /* Never use the revoke function if we are doing full data |
| 81 | * journaling: there is no need to, and a V1 superblock won't |
| 82 | * support it. Otherwise, only skip the revoke on un-journaled |
| 83 | * data blocks. */ |
| 84 | |
| 85 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA || |
| 86 | (!is_metadata && !ext3_should_journal_data(inode))) { |
| 87 | if (bh) { |
| 88 | BUFFER_TRACE(bh, "call journal_forget"); |
| 89 | return ext3_journal_forget(handle, bh); |
| 90 | } |
| 91 | return 0; |
| 92 | } |
| 93 | |
| 94 | /* |
| 95 | * data!=journal && (is_metadata || should_journal_data(inode)) |
| 96 | */ |
| 97 | BUFFER_TRACE(bh, "call ext3_journal_revoke"); |
| 98 | err = ext3_journal_revoke(handle, blocknr, bh); |
| 99 | if (err) |
Harvey Harrison | e05b6b5 | 2008-04-28 02:16:15 -0700 | [diff] [blame] | 100 | ext3_abort(inode->i_sb, __func__, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | "error %d when attempting revoke", err); |
| 102 | BUFFER_TRACE(bh, "exit"); |
| 103 | return err; |
| 104 | } |
| 105 | |
| 106 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 107 | * Work out how many blocks we need to proceed with the next chunk of a |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | * truncate transaction. |
| 109 | */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 110 | static unsigned long blocks_for_truncate(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | { |
| 112 | unsigned long needed; |
| 113 | |
| 114 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); |
| 115 | |
| 116 | /* Give ourselves just enough room to cope with inodes in which |
| 117 | * i_blocks is corrupt: we've seen disk corruptions in the past |
| 118 | * which resulted in random data in an inode which looked enough |
| 119 | * like a regular file for ext3 to try to delete it. Things |
| 120 | * will go a bit crazy if that happens, but at least we should |
| 121 | * try not to panic the whole kernel. */ |
| 122 | if (needed < 2) |
| 123 | needed = 2; |
| 124 | |
| 125 | /* But we need to bound the transaction so we don't overflow the |
| 126 | * journal. */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 127 | if (needed > EXT3_MAX_TRANS_DATA) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | needed = EXT3_MAX_TRANS_DATA; |
| 129 | |
Jan Kara | 1f54587 | 2005-06-23 22:01:04 -0700 | [diff] [blame] | 130 | return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | } |
| 132 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 133 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | * Truncate transactions can be complex and absolutely huge. So we need to |
| 135 | * be able to restart the transaction at a conventient checkpoint to make |
| 136 | * sure we don't overflow the journal. |
| 137 | * |
| 138 | * start_transaction gets us a new handle for a truncate transaction, |
| 139 | * and extend_transaction tries to extend the existing one a bit. If |
| 140 | * extend fails, we need to propagate the failure up and restart the |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 141 | * transaction in the top-level truncate loop. --sct |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 143 | static handle_t *start_transaction(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | { |
| 145 | handle_t *result; |
| 146 | |
| 147 | result = ext3_journal_start(inode, blocks_for_truncate(inode)); |
| 148 | if (!IS_ERR(result)) |
| 149 | return result; |
| 150 | |
| 151 | ext3_std_error(inode->i_sb, PTR_ERR(result)); |
| 152 | return result; |
| 153 | } |
| 154 | |
| 155 | /* |
| 156 | * Try to extend this transaction for the purposes of truncation. |
| 157 | * |
| 158 | * Returns 0 if we managed to create more room. If we can't create more |
| 159 | * room, and the transaction must be restarted we return 1. |
| 160 | */ |
| 161 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) |
| 162 | { |
| 163 | if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS) |
| 164 | return 0; |
| 165 | if (!ext3_journal_extend(handle, blocks_for_truncate(inode))) |
| 166 | return 0; |
| 167 | return 1; |
| 168 | } |
| 169 | |
| 170 | /* |
| 171 | * Restart the transaction associated with *handle. This does a commit, |
| 172 | * so before we call here everything must be consistently dirtied against |
| 173 | * this transaction. |
| 174 | */ |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 175 | static int truncate_restart_transaction(handle_t *handle, struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | { |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 177 | int ret; |
| 178 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | jbd_debug(2, "restarting handle %p\n", handle); |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 180 | /* |
| 181 | * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle |
| 182 | * At this moment, get_block can be called only for blocks inside |
| 183 | * i_size since page cache has been already dropped and writes are |
| 184 | * blocked by i_mutex. So we can safely drop the truncate_mutex. |
| 185 | */ |
| 186 | mutex_unlock(&EXT3_I(inode)->truncate_mutex); |
| 187 | ret = ext3_journal_restart(handle, blocks_for_truncate(inode)); |
| 188 | mutex_lock(&EXT3_I(inode)->truncate_mutex); |
| 189 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | } |
| 191 | |
| 192 | /* |
| 193 | * Called at the last iput() if i_nlink is zero. |
| 194 | */ |
| 195 | void ext3_delete_inode (struct inode * inode) |
| 196 | { |
| 197 | handle_t *handle; |
| 198 | |
Mark Fasheh | fef2665 | 2005-09-09 13:01:31 -0700 | [diff] [blame] | 199 | truncate_inode_pages(&inode->i_data, 0); |
| 200 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | if (is_bad_inode(inode)) |
| 202 | goto no_delete; |
| 203 | |
| 204 | handle = start_transaction(inode); |
| 205 | if (IS_ERR(handle)) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 206 | /* |
| 207 | * If we're going to skip the normal cleanup, we still need to |
| 208 | * make sure that the in-core orphan linked list is properly |
| 209 | * cleaned up. |
| 210 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | ext3_orphan_del(NULL, inode); |
| 212 | goto no_delete; |
| 213 | } |
| 214 | |
| 215 | if (IS_SYNC(inode)) |
| 216 | handle->h_sync = 1; |
| 217 | inode->i_size = 0; |
| 218 | if (inode->i_blocks) |
| 219 | ext3_truncate(inode); |
| 220 | /* |
| 221 | * Kill off the orphan record which ext3_truncate created. |
| 222 | * AKPM: I think this can be inside the above `if'. |
| 223 | * Note that ext3_orphan_del() has to be able to cope with the |
| 224 | * deletion of a non-existent orphan - this is because we don't |
| 225 | * know if ext3_truncate() actually created an orphan record. |
| 226 | * (Well, we could do this if we need to, but heck - it works) |
| 227 | */ |
| 228 | ext3_orphan_del(handle, inode); |
| 229 | EXT3_I(inode)->i_dtime = get_seconds(); |
| 230 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 231 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | * One subtle ordering requirement: if anything has gone wrong |
| 233 | * (transaction abort, IO errors, whatever), then we can still |
| 234 | * do these next steps (the fs will already have been marked as |
| 235 | * having errors), but we can't free the inode if the mark_dirty |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 236 | * fails. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 237 | */ |
| 238 | if (ext3_mark_inode_dirty(handle, inode)) |
| 239 | /* If that failed, just do the required in-core inode clear. */ |
| 240 | clear_inode(inode); |
| 241 | else |
| 242 | ext3_free_inode(handle, inode); |
| 243 | ext3_journal_stop(handle); |
| 244 | return; |
| 245 | no_delete: |
| 246 | clear_inode(inode); /* We must guarantee clearing of inode... */ |
| 247 | } |
| 248 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | typedef struct { |
| 250 | __le32 *p; |
| 251 | __le32 key; |
| 252 | struct buffer_head *bh; |
| 253 | } Indirect; |
| 254 | |
| 255 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) |
| 256 | { |
| 257 | p->key = *(p->p = v); |
| 258 | p->bh = bh; |
| 259 | } |
| 260 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 261 | static int verify_chain(Indirect *from, Indirect *to) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 262 | { |
| 263 | while (from <= to && from->key == *from->p) |
| 264 | from++; |
| 265 | return (from > to); |
| 266 | } |
| 267 | |
| 268 | /** |
| 269 | * ext3_block_to_path - parse the block number into array of offsets |
| 270 | * @inode: inode in question (we are only interested in its superblock) |
| 271 | * @i_block: block number to be parsed |
| 272 | * @offsets: array to store the offsets in |
| 273 | * @boundary: set this non-zero if the referred-to block is likely to be |
| 274 | * followed (on disk) by an indirect block. |
| 275 | * |
| 276 | * To store the locations of file's data ext3 uses a data structure common |
| 277 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
| 278 | * data blocks at leaves and indirect blocks in intermediate nodes. |
| 279 | * This function translates the block number into path in that tree - |
| 280 | * return value is the path length and @offsets[n] is the offset of |
| 281 | * pointer to (n+1)th node in the nth one. If @block is out of range |
| 282 | * (negative or too large) warning is printed and zero returned. |
| 283 | * |
| 284 | * Note: function doesn't find node addresses, so no IO is needed. All |
| 285 | * we need to know is the capacity of indirect blocks (taken from the |
| 286 | * inode->i_sb). |
| 287 | */ |
| 288 | |
| 289 | /* |
| 290 | * Portability note: the last comparison (check that we fit into triple |
| 291 | * indirect block) is spelled differently, because otherwise on an |
| 292 | * architecture with 32-bit longs and 8Kb pages we might get into trouble |
| 293 | * if our filesystem had 8Kb blocks. We might use long long, but that would |
| 294 | * kill us on x86. Oh, well, at least the sign propagation does not matter - |
| 295 | * i_block would have to be negative in the very beginning, so we would not |
| 296 | * get there at all. |
| 297 | */ |
| 298 | |
| 299 | static int ext3_block_to_path(struct inode *inode, |
| 300 | long i_block, int offsets[4], int *boundary) |
| 301 | { |
| 302 | int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb); |
| 303 | int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb); |
| 304 | const long direct_blocks = EXT3_NDIR_BLOCKS, |
| 305 | indirect_blocks = ptrs, |
| 306 | double_blocks = (1 << (ptrs_bits * 2)); |
| 307 | int n = 0; |
| 308 | int final = 0; |
| 309 | |
| 310 | if (i_block < 0) { |
| 311 | ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0"); |
| 312 | } else if (i_block < direct_blocks) { |
| 313 | offsets[n++] = i_block; |
| 314 | final = direct_blocks; |
| 315 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { |
| 316 | offsets[n++] = EXT3_IND_BLOCK; |
| 317 | offsets[n++] = i_block; |
| 318 | final = ptrs; |
| 319 | } else if ((i_block -= indirect_blocks) < double_blocks) { |
| 320 | offsets[n++] = EXT3_DIND_BLOCK; |
| 321 | offsets[n++] = i_block >> ptrs_bits; |
| 322 | offsets[n++] = i_block & (ptrs - 1); |
| 323 | final = ptrs; |
| 324 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { |
| 325 | offsets[n++] = EXT3_TIND_BLOCK; |
| 326 | offsets[n++] = i_block >> (ptrs_bits * 2); |
| 327 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); |
| 328 | offsets[n++] = i_block & (ptrs - 1); |
| 329 | final = ptrs; |
| 330 | } else { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 331 | ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | } |
| 333 | if (boundary) |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 334 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | return n; |
| 336 | } |
| 337 | |
| 338 | /** |
| 339 | * ext3_get_branch - read the chain of indirect blocks leading to data |
| 340 | * @inode: inode in question |
| 341 | * @depth: depth of the chain (1 - direct pointer, etc.) |
| 342 | * @offsets: offsets of pointers in inode/indirect blocks |
| 343 | * @chain: place to store the result |
| 344 | * @err: here we store the error value |
| 345 | * |
| 346 | * Function fills the array of triples <key, p, bh> and returns %NULL |
| 347 | * if everything went OK or the pointer to the last filled triple |
| 348 | * (incomplete one) otherwise. Upon the return chain[i].key contains |
| 349 | * the number of (i+1)-th block in the chain (as it is stored in memory, |
| 350 | * i.e. little-endian 32-bit), chain[i].p contains the address of that |
| 351 | * number (it points into struct inode for i==0 and into the bh->b_data |
| 352 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect |
| 353 | * block for i>0 and NULL for i==0. In other words, it holds the block |
| 354 | * numbers of the chain, addresses they were taken from (and where we can |
| 355 | * verify that chain did not change) and buffer_heads hosting these |
| 356 | * numbers. |
| 357 | * |
| 358 | * Function stops when it stumbles upon zero pointer (absent block) |
| 359 | * (pointer to last triple returned, *@err == 0) |
| 360 | * or when it gets an IO error reading an indirect block |
| 361 | * (ditto, *@err == -EIO) |
| 362 | * or when it notices that chain had been changed while it was reading |
| 363 | * (ditto, *@err == -EAGAIN) |
| 364 | * or when it reads all @depth-1 indirect blocks successfully and finds |
| 365 | * the whole chain, all way to the data (returns %NULL, *err == 0). |
| 366 | */ |
| 367 | static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets, |
| 368 | Indirect chain[4], int *err) |
| 369 | { |
| 370 | struct super_block *sb = inode->i_sb; |
| 371 | Indirect *p = chain; |
| 372 | struct buffer_head *bh; |
| 373 | |
| 374 | *err = 0; |
| 375 | /* i_data is not going away, no lock needed */ |
| 376 | add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets); |
| 377 | if (!p->key) |
| 378 | goto no_block; |
| 379 | while (--depth) { |
| 380 | bh = sb_bread(sb, le32_to_cpu(p->key)); |
| 381 | if (!bh) |
| 382 | goto failure; |
| 383 | /* Reader: pointers */ |
| 384 | if (!verify_chain(chain, p)) |
| 385 | goto changed; |
| 386 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); |
| 387 | /* Reader: end */ |
| 388 | if (!p->key) |
| 389 | goto no_block; |
| 390 | } |
| 391 | return NULL; |
| 392 | |
| 393 | changed: |
| 394 | brelse(bh); |
| 395 | *err = -EAGAIN; |
| 396 | goto no_block; |
| 397 | failure: |
| 398 | *err = -EIO; |
| 399 | no_block: |
| 400 | return p; |
| 401 | } |
| 402 | |
| 403 | /** |
| 404 | * ext3_find_near - find a place for allocation with sufficient locality |
| 405 | * @inode: owner |
| 406 | * @ind: descriptor of indirect block. |
| 407 | * |
Benoit Boissinot | 1cc8dcf | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 408 | * This function returns the preferred place for block allocation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | * It is used when heuristic for sequential allocation fails. |
| 410 | * Rules are: |
| 411 | * + if there is a block to the left of our position - allocate near it. |
| 412 | * + if pointer will live in indirect block - allocate near that block. |
| 413 | * + if pointer will live in inode - allocate in the same |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 414 | * cylinder group. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | * |
| 416 | * In the latter case we colour the starting block by the callers PID to |
| 417 | * prevent it from clashing with concurrent allocations for a different inode |
| 418 | * in the same block group. The PID is used here so that functionally related |
| 419 | * files will be close-by on-disk. |
| 420 | * |
| 421 | * Caller must make sure that @ind is valid and will stay that way. |
| 422 | */ |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 423 | static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | { |
| 425 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 426 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; |
| 427 | __le32 *p; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 428 | ext3_fsblk_t bg_start; |
| 429 | ext3_grpblk_t colour; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | |
| 431 | /* Try to find previous block */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 432 | for (p = ind->p - 1; p >= start; p--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | if (*p) |
| 434 | return le32_to_cpu(*p); |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 435 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | |
| 437 | /* No such thing, so let's try location of indirect block */ |
| 438 | if (ind->bh) |
| 439 | return ind->bh->b_blocknr; |
| 440 | |
| 441 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 442 | * It is going to be referred to from the inode itself? OK, just put it |
| 443 | * into the same cylinder group then. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | */ |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 445 | bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | colour = (current->pid % 16) * |
| 447 | (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
| 448 | return bg_start + colour; |
| 449 | } |
| 450 | |
| 451 | /** |
Benoit Boissinot | 1cc8dcf | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 452 | * ext3_find_goal - find a preferred place for allocation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | * @inode: owner |
| 454 | * @block: block we want |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 455 | * @partial: pointer to the last triple within a chain |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | * |
Benoit Boissinot | 1cc8dcf | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 457 | * Normally this function find the preferred place for block allocation, |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 458 | * returns it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | */ |
| 460 | |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 461 | static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block, |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 462 | Indirect *partial) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 463 | { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 464 | struct ext3_block_alloc_info *block_i; |
| 465 | |
| 466 | block_i = EXT3_I(inode)->i_block_alloc_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | |
| 468 | /* |
| 469 | * try the heuristic for sequential allocation, |
| 470 | * failing that at least try to get decent locality. |
| 471 | */ |
| 472 | if (block_i && (block == block_i->last_alloc_logical_block + 1) |
| 473 | && (block_i->last_alloc_physical_block != 0)) { |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 474 | return block_i->last_alloc_physical_block + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | } |
| 476 | |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 477 | return ext3_find_near(inode, partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | } |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 479 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 480 | /** |
| 481 | * ext3_blks_to_allocate: Look up the block map and count the number |
| 482 | * of direct blocks need to be allocated for the given branch. |
| 483 | * |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 484 | * @branch: chain of indirect blocks |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 485 | * @k: number of blocks need for indirect blocks |
| 486 | * @blks: number of data blocks to be mapped. |
| 487 | * @blocks_to_boundary: the offset in the indirect block |
| 488 | * |
| 489 | * return the total number of blocks to be allocate, including the |
| 490 | * direct and indirect blocks. |
| 491 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 492 | static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 493 | int blocks_to_boundary) |
| 494 | { |
| 495 | unsigned long count = 0; |
| 496 | |
| 497 | /* |
| 498 | * Simple case, [t,d]Indirect block(s) has not allocated yet |
| 499 | * then it's clear blocks on that path have not allocated |
| 500 | */ |
| 501 | if (k > 0) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 502 | /* right now we don't handle cross boundary allocation */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 503 | if (blks < blocks_to_boundary + 1) |
| 504 | count += blks; |
| 505 | else |
| 506 | count += blocks_to_boundary + 1; |
| 507 | return count; |
| 508 | } |
| 509 | |
| 510 | count++; |
| 511 | while (count < blks && count <= blocks_to_boundary && |
| 512 | le32_to_cpu(*(branch[0].p + count)) == 0) { |
| 513 | count++; |
| 514 | } |
| 515 | return count; |
| 516 | } |
| 517 | |
| 518 | /** |
| 519 | * ext3_alloc_blocks: multiple allocate blocks needed for a branch |
| 520 | * @indirect_blks: the number of blocks need to allocate for indirect |
| 521 | * blocks |
| 522 | * |
| 523 | * @new_blocks: on return it will store the new block numbers for |
| 524 | * the indirect blocks(if needed) and the first direct block, |
| 525 | * @blks: on return it will store the total number of allocated |
| 526 | * direct blocks |
| 527 | */ |
| 528 | static int ext3_alloc_blocks(handle_t *handle, struct inode *inode, |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 529 | ext3_fsblk_t goal, int indirect_blks, int blks, |
| 530 | ext3_fsblk_t new_blocks[4], int *err) |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 531 | { |
| 532 | int target, i; |
| 533 | unsigned long count = 0; |
| 534 | int index = 0; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 535 | ext3_fsblk_t current_block = 0; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 536 | int ret = 0; |
| 537 | |
| 538 | /* |
| 539 | * Here we try to allocate the requested multiple blocks at once, |
| 540 | * on a best-effort basis. |
| 541 | * To build a branch, we should allocate blocks for |
| 542 | * the indirect blocks(if not allocated yet), and at least |
| 543 | * the first direct block of this branch. That's the |
| 544 | * minimum number of blocks need to allocate(required) |
| 545 | */ |
| 546 | target = blks + indirect_blks; |
| 547 | |
| 548 | while (1) { |
| 549 | count = target; |
| 550 | /* allocating blocks for indirect blocks and direct blocks */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 551 | current_block = ext3_new_blocks(handle,inode,goal,&count,err); |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 552 | if (*err) |
| 553 | goto failed_out; |
| 554 | |
| 555 | target -= count; |
| 556 | /* allocate blocks for indirect blocks */ |
| 557 | while (index < indirect_blks && count) { |
| 558 | new_blocks[index++] = current_block++; |
| 559 | count--; |
| 560 | } |
| 561 | |
| 562 | if (count > 0) |
| 563 | break; |
| 564 | } |
| 565 | |
| 566 | /* save the new block number for the first direct block */ |
| 567 | new_blocks[index] = current_block; |
| 568 | |
| 569 | /* total number of blocks allocated for direct blocks */ |
| 570 | ret = count; |
| 571 | *err = 0; |
| 572 | return ret; |
| 573 | failed_out: |
| 574 | for (i = 0; i <index; i++) |
| 575 | ext3_free_blocks(handle, inode, new_blocks[i], 1); |
| 576 | return ret; |
| 577 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | |
| 579 | /** |
| 580 | * ext3_alloc_branch - allocate and set up a chain of blocks. |
| 581 | * @inode: owner |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 582 | * @indirect_blks: number of allocated indirect blocks |
| 583 | * @blks: number of allocated direct blocks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 584 | * @offsets: offsets (in the blocks) to store the pointers to next. |
| 585 | * @branch: place to store the chain in. |
| 586 | * |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 587 | * This function allocates blocks, zeroes out all but the last one, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | * links them into chain and (if we are synchronous) writes them to disk. |
| 589 | * In other words, it prepares a branch that can be spliced onto the |
| 590 | * inode. It stores the information about that chain in the branch[], in |
| 591 | * the same format as ext3_get_branch() would do. We are calling it after |
| 592 | * we had read the existing part of chain and partial points to the last |
| 593 | * triple of that (one with zero ->key). Upon the exit we have the same |
Glauber de Oliveira Costa | 5b11687 | 2005-10-30 15:02:48 -0800 | [diff] [blame] | 594 | * picture as after the successful ext3_get_block(), except that in one |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | * place chain is disconnected - *branch->p is still zero (we did not |
| 596 | * set the last link), but branch->key contains the number that should |
| 597 | * be placed into *branch->p to fill that gap. |
| 598 | * |
| 599 | * If allocation fails we free all blocks we've allocated (and forget |
| 600 | * their buffer_heads) and return the error value the from failed |
| 601 | * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
| 602 | * as described above and return 0. |
| 603 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | static int ext3_alloc_branch(handle_t *handle, struct inode *inode, |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 605 | int indirect_blks, int *blks, ext3_fsblk_t goal, |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 606 | int *offsets, Indirect *branch) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | { |
| 608 | int blocksize = inode->i_sb->s_blocksize; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 609 | int i, n = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | int err = 0; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 611 | struct buffer_head *bh; |
| 612 | int num; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 613 | ext3_fsblk_t new_blocks[4]; |
| 614 | ext3_fsblk_t current_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 616 | num = ext3_alloc_blocks(handle, inode, goal, indirect_blks, |
| 617 | *blks, new_blocks, &err); |
| 618 | if (err) |
| 619 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 621 | branch[0].key = cpu_to_le32(new_blocks[0]); |
| 622 | /* |
| 623 | * metadata blocks and data blocks are allocated. |
| 624 | */ |
| 625 | for (n = 1; n <= indirect_blks; n++) { |
| 626 | /* |
| 627 | * Get buffer_head for parent block, zero it out |
| 628 | * and set the pointer to new one, then send |
| 629 | * parent to disk. |
| 630 | */ |
| 631 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); |
| 632 | branch[n].bh = bh; |
| 633 | lock_buffer(bh); |
| 634 | BUFFER_TRACE(bh, "call get_create_access"); |
| 635 | err = ext3_journal_get_create_access(handle, bh); |
| 636 | if (err) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | unlock_buffer(bh); |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 638 | brelse(bh); |
| 639 | goto failed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 642 | memset(bh->b_data, 0, blocksize); |
| 643 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; |
| 644 | branch[n].key = cpu_to_le32(new_blocks[n]); |
| 645 | *branch[n].p = branch[n].key; |
| 646 | if ( n == indirect_blks) { |
| 647 | current_block = new_blocks[n]; |
| 648 | /* |
| 649 | * End of chain, update the last new metablock of |
| 650 | * the chain to point to the new allocated |
| 651 | * data blocks numbers |
| 652 | */ |
| 653 | for (i=1; i < num; i++) |
| 654 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
| 655 | } |
| 656 | BUFFER_TRACE(bh, "marking uptodate"); |
| 657 | set_buffer_uptodate(bh); |
| 658 | unlock_buffer(bh); |
| 659 | |
| 660 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
| 661 | err = ext3_journal_dirty_metadata(handle, bh); |
| 662 | if (err) |
| 663 | goto failed; |
| 664 | } |
| 665 | *blks = num; |
| 666 | return err; |
| 667 | failed: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | /* Allocation failed, free what we already allocated */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 669 | for (i = 1; i <= n ; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | BUFFER_TRACE(branch[i].bh, "call journal_forget"); |
| 671 | ext3_journal_forget(handle, branch[i].bh); |
| 672 | } |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 673 | for (i = 0; i <indirect_blks; i++) |
| 674 | ext3_free_blocks(handle, inode, new_blocks[i], 1); |
| 675 | |
| 676 | ext3_free_blocks(handle, inode, new_blocks[i], num); |
| 677 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | return err; |
| 679 | } |
| 680 | |
| 681 | /** |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 682 | * ext3_splice_branch - splice the allocated branch onto inode. |
| 683 | * @inode: owner |
| 684 | * @block: (logical) number of block we are adding |
| 685 | * @chain: chain of indirect blocks (with a missing link - see |
| 686 | * ext3_alloc_branch) |
| 687 | * @where: location of missing link |
| 688 | * @num: number of indirect blocks we are adding |
| 689 | * @blks: number of direct blocks we are adding |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | * |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 691 | * This function fills the missing link and does all housekeeping needed in |
| 692 | * inode (->i_blocks, etc.). In case of success we end up with the full |
| 693 | * chain to new block and return 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 695 | static int ext3_splice_branch(handle_t *handle, struct inode *inode, |
| 696 | long block, Indirect *where, int num, int blks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | { |
| 698 | int i; |
| 699 | int err = 0; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 700 | struct ext3_block_alloc_info *block_i; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 701 | ext3_fsblk_t current_block; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 702 | |
| 703 | block_i = EXT3_I(inode)->i_block_alloc_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | /* |
| 705 | * If we're splicing into a [td]indirect block (as opposed to the |
| 706 | * inode) then we need to get write access to the [td]indirect block |
| 707 | * before the splice. |
| 708 | */ |
| 709 | if (where->bh) { |
| 710 | BUFFER_TRACE(where->bh, "get_write_access"); |
| 711 | err = ext3_journal_get_write_access(handle, where->bh); |
| 712 | if (err) |
| 713 | goto err_out; |
| 714 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 715 | /* That's it */ |
| 716 | |
| 717 | *where->p = where->key; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 718 | |
| 719 | /* |
| 720 | * Update the host buffer_head or inode to point to more just allocated |
| 721 | * direct blocks blocks |
| 722 | */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 723 | if (num == 0 && blks > 1) { |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 724 | current_block = le32_to_cpu(where->key) + 1; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 725 | for (i = 1; i < blks; i++) |
| 726 | *(where->p + i ) = cpu_to_le32(current_block++); |
| 727 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | |
| 729 | /* |
| 730 | * update the most recently allocated logical & physical block |
| 731 | * in i_block_alloc_info, to assist find the proper goal block for next |
| 732 | * allocation |
| 733 | */ |
| 734 | if (block_i) { |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 735 | block_i->last_alloc_logical_block = block + blks - 1; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 736 | block_i->last_alloc_physical_block = |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 737 | le32_to_cpu(where[num].key) + blks - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | } |
| 739 | |
| 740 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
| 741 | |
| 742 | inode->i_ctime = CURRENT_TIME_SEC; |
| 743 | ext3_mark_inode_dirty(handle, inode); |
| 744 | |
| 745 | /* had we spliced it onto indirect block? */ |
| 746 | if (where->bh) { |
| 747 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 748 | * If we spliced it onto an indirect block, we haven't |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | * altered the inode. Note however that if it is being spliced |
| 750 | * onto an indirect block at the very end of the file (the |
| 751 | * file is growing) then we *will* alter the inode to reflect |
| 752 | * the new i_size. But that is not done here - it is done in |
| 753 | * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode. |
| 754 | */ |
| 755 | jbd_debug(5, "splicing indirect only\n"); |
| 756 | BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata"); |
| 757 | err = ext3_journal_dirty_metadata(handle, where->bh); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 758 | if (err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | goto err_out; |
| 760 | } else { |
| 761 | /* |
| 762 | * OK, we spliced it into the inode itself on a direct block. |
| 763 | * Inode was dirtied above. |
| 764 | */ |
| 765 | jbd_debug(5, "splicing direct\n"); |
| 766 | } |
| 767 | return err; |
| 768 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | err_out: |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 770 | for (i = 1; i <= num; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | BUFFER_TRACE(where[i].bh, "call journal_forget"); |
| 772 | ext3_journal_forget(handle, where[i].bh); |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 773 | ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | } |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 775 | ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
| 776 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 777 | return err; |
| 778 | } |
| 779 | |
| 780 | /* |
| 781 | * Allocation strategy is simple: if we have to allocate something, we will |
| 782 | * have to go the whole way to leaf. So let's do it before attaching anything |
| 783 | * to tree, set linkage between the newborn blocks, write them if sync is |
| 784 | * required, recheck the path, free and repeat if check fails, otherwise |
| 785 | * set the last missing link (that will protect us from any truncate-generated |
| 786 | * removals - all blocks on the path are immune now) and possibly force the |
| 787 | * write on the parent block. |
| 788 | * That has a nice additional property: no special recovery from the failed |
| 789 | * allocations is needed - we simply release blocks and do not touch anything |
| 790 | * reachable from inode. |
| 791 | * |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 792 | * `handle' can be NULL if create == 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | * |
| 794 | * The BKL may not be held on entry here. Be sure to take it early. |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 795 | * return > 0, # of blocks mapped or allocated. |
| 796 | * return = 0, if plain lookup failed. |
| 797 | * return < 0, error case. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 799 | int ext3_get_blocks_handle(handle_t *handle, struct inode *inode, |
| 800 | sector_t iblock, unsigned long maxblocks, |
| 801 | struct buffer_head *bh_result, |
Jan Kara | 43237b5 | 2009-05-20 18:41:58 +0200 | [diff] [blame] | 802 | int create) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 803 | { |
| 804 | int err = -EIO; |
| 805 | int offsets[4]; |
| 806 | Indirect chain[4]; |
| 807 | Indirect *partial; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 808 | ext3_fsblk_t goal; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 809 | int indirect_blks; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 810 | int blocks_to_boundary = 0; |
| 811 | int depth; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | struct ext3_inode_info *ei = EXT3_I(inode); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 813 | int count = 0; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 814 | ext3_fsblk_t first_block = 0; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 815 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 816 | |
| 817 | J_ASSERT(handle != NULL || create == 0); |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 818 | depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | |
| 820 | if (depth == 0) |
| 821 | goto out; |
| 822 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 823 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
| 824 | |
| 825 | /* Simplest case - block found, no allocation needed */ |
| 826 | if (!partial) { |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 827 | first_block = le32_to_cpu(chain[depth - 1].key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | clear_buffer_new(bh_result); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 829 | count++; |
| 830 | /*map more blocks*/ |
| 831 | while (count < maxblocks && count <= blocks_to_boundary) { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 832 | ext3_fsblk_t blk; |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 833 | |
Jan Kara | e8ef7aa | 2009-06-17 16:26:23 -0700 | [diff] [blame] | 834 | if (!verify_chain(chain, chain + depth - 1)) { |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 835 | /* |
| 836 | * Indirect block might be removed by |
| 837 | * truncate while we were reading it. |
| 838 | * Handling of that case: forget what we've |
| 839 | * got now. Flag the err as EAGAIN, so it |
| 840 | * will reread. |
| 841 | */ |
| 842 | err = -EAGAIN; |
| 843 | count = 0; |
| 844 | break; |
| 845 | } |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 846 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
| 847 | |
| 848 | if (blk == first_block + count) |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 849 | count++; |
| 850 | else |
| 851 | break; |
| 852 | } |
| 853 | if (err != -EAGAIN) |
| 854 | goto got_it; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | } |
| 856 | |
| 857 | /* Next simple case - plain lookup or failed read of indirect block */ |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 858 | if (!create || err == -EIO) |
| 859 | goto cleanup; |
| 860 | |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 861 | mutex_lock(&ei->truncate_mutex); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 862 | |
| 863 | /* |
| 864 | * If the indirect block is missing while we are reading |
| 865 | * the chain(ext3_get_branch() returns -EAGAIN err), or |
| 866 | * if the chain has been changed after we grab the semaphore, |
| 867 | * (either because another process truncated this branch, or |
| 868 | * another get_block allocated this branch) re-grab the chain to see if |
| 869 | * the request block has been allocated or not. |
| 870 | * |
| 871 | * Since we already block the truncate/other get_block |
| 872 | * at this point, we will have the current copy of the chain when we |
| 873 | * splice the branch into the tree. |
| 874 | */ |
| 875 | if (err == -EAGAIN || !verify_chain(chain, partial)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | while (partial > chain) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | brelse(partial->bh); |
| 878 | partial--; |
| 879 | } |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 880 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
| 881 | if (!partial) { |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 882 | count++; |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 883 | mutex_unlock(&ei->truncate_mutex); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 884 | if (err) |
| 885 | goto cleanup; |
| 886 | clear_buffer_new(bh_result); |
| 887 | goto got_it; |
| 888 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | } |
| 890 | |
| 891 | /* |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 892 | * Okay, we need to do block allocation. Lazily initialize the block |
| 893 | * allocation info here if necessary |
| 894 | */ |
| 895 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | ext3_init_block_alloc_info(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 897 | |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 898 | goal = ext3_find_goal(inode, iblock, partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 899 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 900 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
| 901 | indirect_blks = (chain + depth) - partial - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 902 | |
| 903 | /* |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 904 | * Next look up the indirect map to count the totoal number of |
| 905 | * direct blocks to allocate for this branch. |
| 906 | */ |
| 907 | count = ext3_blks_to_allocate(partial, indirect_blks, |
| 908 | maxblocks, blocks_to_boundary); |
| 909 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 910 | * Block out ext3_truncate while we alter the tree |
| 911 | */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 912 | err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal, |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 913 | offsets + (partial - chain), partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 914 | |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 915 | /* |
| 916 | * The ext3_splice_branch call will free and forget any buffers |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 917 | * on the new chain if there is a failure, but that risks using |
| 918 | * up transaction credits, especially for bitmaps where the |
| 919 | * credits cannot be returned. Can we handle this somehow? We |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 920 | * may need to return -EAGAIN upwards in the worst case. --sct |
| 921 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | if (!err) |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 923 | err = ext3_splice_branch(handle, inode, iblock, |
| 924 | partial, indirect_blks, count); |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 925 | mutex_unlock(&ei->truncate_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 926 | if (err) |
| 927 | goto cleanup; |
| 928 | |
| 929 | set_buffer_new(bh_result); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 930 | got_it: |
| 931 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); |
Suparna Bhattacharya | 20acaa1 | 2006-09-16 12:15:58 -0700 | [diff] [blame] | 932 | if (count > blocks_to_boundary) |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 933 | set_buffer_boundary(bh_result); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 934 | err = count; |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 935 | /* Clean up and exit */ |
| 936 | partial = chain + depth - 1; /* the whole chain */ |
| 937 | cleanup: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | while (partial > chain) { |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 939 | BUFFER_TRACE(partial->bh, "call brelse"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 940 | brelse(partial->bh); |
| 941 | partial--; |
| 942 | } |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 943 | BUFFER_TRACE(bh_result, "returned"); |
| 944 | out: |
| 945 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | } |
| 947 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 948 | /* Maximum number of blocks we map for direct IO at once. */ |
| 949 | #define DIO_MAX_BLOCKS 4096 |
| 950 | /* |
| 951 | * Number of credits we need for writing DIO_MAX_BLOCKS: |
| 952 | * We need sb + group descriptor + bitmap + inode -> 4 |
| 953 | * For B blocks with A block pointers per block we need: |
| 954 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). |
| 955 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. |
| 956 | */ |
| 957 | #define DIO_CREDITS 25 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 958 | |
Badari Pulavarty | f91a2ad | 2006-03-26 01:38:04 -0800 | [diff] [blame] | 959 | static int ext3_get_block(struct inode *inode, sector_t iblock, |
| 960 | struct buffer_head *bh_result, int create) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | { |
Dmitriy Monakhov | 3e4fdaf | 2007-02-10 01:46:35 -0800 | [diff] [blame] | 962 | handle_t *handle = ext3_journal_current_handle(); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 963 | int ret = 0, started = 0; |
Badari Pulavarty | 1d8fa7a | 2006-03-26 01:38:02 -0800 | [diff] [blame] | 964 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 965 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 966 | if (create && !handle) { /* Direct IO write... */ |
| 967 | if (max_blocks > DIO_MAX_BLOCKS) |
| 968 | max_blocks = DIO_MAX_BLOCKS; |
| 969 | handle = ext3_journal_start(inode, DIO_CREDITS + |
| 970 | 2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb)); |
| 971 | if (IS_ERR(handle)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | ret = PTR_ERR(handle); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 973 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 974 | } |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 975 | started = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | } |
| 977 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 978 | ret = ext3_get_blocks_handle(handle, inode, iblock, |
Jan Kara | 43237b5 | 2009-05-20 18:41:58 +0200 | [diff] [blame] | 979 | max_blocks, bh_result, create); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 980 | if (ret > 0) { |
| 981 | bh_result->b_size = (ret << inode->i_blkbits); |
| 982 | ret = 0; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 983 | } |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 984 | if (started) |
| 985 | ext3_journal_stop(handle); |
| 986 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | return ret; |
| 988 | } |
| 989 | |
Josef Bacik | 68c9d70 | 2008-10-03 17:32:43 -0400 | [diff] [blame] | 990 | int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
| 991 | u64 start, u64 len) |
| 992 | { |
| 993 | return generic_block_fiemap(inode, fieinfo, start, len, |
| 994 | ext3_get_block); |
| 995 | } |
| 996 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 997 | /* |
| 998 | * `handle' can be NULL if create is zero |
| 999 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1000 | struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode, |
| 1001 | long block, int create, int *errp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | { |
| 1003 | struct buffer_head dummy; |
| 1004 | int fatal = 0, err; |
| 1005 | |
| 1006 | J_ASSERT(handle != NULL || create == 0); |
| 1007 | |
| 1008 | dummy.b_state = 0; |
| 1009 | dummy.b_blocknr = -1000; |
| 1010 | buffer_trace_init(&dummy.b_history); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1011 | err = ext3_get_blocks_handle(handle, inode, block, 1, |
Jan Kara | 43237b5 | 2009-05-20 18:41:58 +0200 | [diff] [blame] | 1012 | &dummy, create); |
Badari Pulavarty | 3665d0e | 2006-09-08 09:48:21 -0700 | [diff] [blame] | 1013 | /* |
| 1014 | * ext3_get_blocks_handle() returns number of blocks |
| 1015 | * mapped. 0 in case of a HOLE. |
| 1016 | */ |
| 1017 | if (err > 0) { |
| 1018 | if (err > 1) |
| 1019 | WARN_ON(1); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1020 | err = 0; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1021 | } |
| 1022 | *errp = err; |
| 1023 | if (!err && buffer_mapped(&dummy)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | struct buffer_head *bh; |
| 1025 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); |
Glauber de Oliveira Costa | 2973dfd | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1026 | if (!bh) { |
| 1027 | *errp = -EIO; |
| 1028 | goto err; |
| 1029 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | if (buffer_new(&dummy)) { |
| 1031 | J_ASSERT(create != 0); |
Stephen Hemminger | c80544d | 2007-10-18 03:07:05 -0700 | [diff] [blame] | 1032 | J_ASSERT(handle != NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1033 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1034 | /* |
| 1035 | * Now that we do not always journal data, we should |
| 1036 | * keep in mind whether this should always journal the |
| 1037 | * new buffer as metadata. For now, regular file |
| 1038 | * writes use ext3_get_block instead, so it's not a |
| 1039 | * problem. |
| 1040 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | lock_buffer(bh); |
| 1042 | BUFFER_TRACE(bh, "call get_create_access"); |
| 1043 | fatal = ext3_journal_get_create_access(handle, bh); |
| 1044 | if (!fatal && !buffer_uptodate(bh)) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1045 | memset(bh->b_data,0,inode->i_sb->s_blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | set_buffer_uptodate(bh); |
| 1047 | } |
| 1048 | unlock_buffer(bh); |
| 1049 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
| 1050 | err = ext3_journal_dirty_metadata(handle, bh); |
| 1051 | if (!fatal) |
| 1052 | fatal = err; |
| 1053 | } else { |
| 1054 | BUFFER_TRACE(bh, "not a new buffer"); |
| 1055 | } |
| 1056 | if (fatal) { |
| 1057 | *errp = fatal; |
| 1058 | brelse(bh); |
| 1059 | bh = NULL; |
| 1060 | } |
| 1061 | return bh; |
| 1062 | } |
Glauber de Oliveira Costa | 2973dfd | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1063 | err: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | return NULL; |
| 1065 | } |
| 1066 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1067 | struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | int block, int create, int *err) |
| 1069 | { |
| 1070 | struct buffer_head * bh; |
| 1071 | |
| 1072 | bh = ext3_getblk(handle, inode, block, create, err); |
| 1073 | if (!bh) |
| 1074 | return bh; |
| 1075 | if (buffer_uptodate(bh)) |
| 1076 | return bh; |
Jens Axboe | caa38fb | 2006-07-23 01:41:26 +0200 | [diff] [blame] | 1077 | ll_rw_block(READ_META, 1, &bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | wait_on_buffer(bh); |
| 1079 | if (buffer_uptodate(bh)) |
| 1080 | return bh; |
| 1081 | put_bh(bh); |
| 1082 | *err = -EIO; |
| 1083 | return NULL; |
| 1084 | } |
| 1085 | |
| 1086 | static int walk_page_buffers( handle_t *handle, |
| 1087 | struct buffer_head *head, |
| 1088 | unsigned from, |
| 1089 | unsigned to, |
| 1090 | int *partial, |
| 1091 | int (*fn)( handle_t *handle, |
| 1092 | struct buffer_head *bh)) |
| 1093 | { |
| 1094 | struct buffer_head *bh; |
| 1095 | unsigned block_start, block_end; |
| 1096 | unsigned blocksize = head->b_size; |
| 1097 | int err, ret = 0; |
| 1098 | struct buffer_head *next; |
| 1099 | |
| 1100 | for ( bh = head, block_start = 0; |
| 1101 | ret == 0 && (bh != head || !block_start); |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 1102 | block_start = block_end, bh = next) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | { |
| 1104 | next = bh->b_this_page; |
| 1105 | block_end = block_start + blocksize; |
| 1106 | if (block_end <= from || block_start >= to) { |
| 1107 | if (partial && !buffer_uptodate(bh)) |
| 1108 | *partial = 1; |
| 1109 | continue; |
| 1110 | } |
| 1111 | err = (*fn)(handle, bh); |
| 1112 | if (!ret) |
| 1113 | ret = err; |
| 1114 | } |
| 1115 | return ret; |
| 1116 | } |
| 1117 | |
| 1118 | /* |
| 1119 | * To preserve ordering, it is essential that the hole instantiation and |
| 1120 | * the data write be encapsulated in a single transaction. We cannot |
| 1121 | * close off a transaction and start a new one between the ext3_get_block() |
| 1122 | * and the commit_write(). So doing the journal_start at the start of |
| 1123 | * prepare_write() is the right place. |
| 1124 | * |
| 1125 | * Also, this function can nest inside ext3_writepage() -> |
| 1126 | * block_write_full_page(). In that case, we *know* that ext3_writepage() |
| 1127 | * has generated enough buffer credits to do the whole page. So we won't |
| 1128 | * block on the journal in that case, which is good, because the caller may |
| 1129 | * be PF_MEMALLOC. |
| 1130 | * |
| 1131 | * By accident, ext3 can be reentered when a transaction is open via |
| 1132 | * quota file writes. If we were to commit the transaction while thus |
| 1133 | * reentered, there can be a deadlock - we would be holding a quota |
| 1134 | * lock, and the commit would never complete if another thread had a |
| 1135 | * transaction open and was blocking on the quota lock - a ranking |
| 1136 | * violation. |
| 1137 | * |
| 1138 | * So what we do is to rely on the fact that journal_stop/journal_start |
| 1139 | * will _not_ run commit under these circumstances because handle->h_ref |
| 1140 | * is elevated. We'll still have enough credits for the tiny quotafile |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1141 | * write. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1142 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1143 | static int do_journal_get_write_access(handle_t *handle, |
| 1144 | struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | { |
| 1146 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1147 | return 0; |
| 1148 | return ext3_journal_get_write_access(handle, bh); |
| 1149 | } |
| 1150 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1151 | static int ext3_write_begin(struct file *file, struct address_space *mapping, |
| 1152 | loff_t pos, unsigned len, unsigned flags, |
| 1153 | struct page **pagep, void **fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | { |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1155 | struct inode *inode = mapping->host; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1156 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1157 | handle_t *handle; |
| 1158 | int retries = 0; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1159 | struct page *page; |
| 1160 | pgoff_t index; |
| 1161 | unsigned from, to; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1162 | /* Reserve one block more for addition to orphan list in case |
| 1163 | * we allocate blocks but write fails for some reason */ |
| 1164 | int needed_blocks = ext3_writepage_trans_blocks(inode) + 1; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1165 | |
| 1166 | index = pos >> PAGE_CACHE_SHIFT; |
| 1167 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1168 | to = from + len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | |
| 1170 | retry: |
Nick Piggin | 54566b2 | 2009-01-04 12:00:53 -0800 | [diff] [blame] | 1171 | page = grab_cache_page_write_begin(mapping, index, flags); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1172 | if (!page) |
| 1173 | return -ENOMEM; |
| 1174 | *pagep = page; |
| 1175 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1176 | handle = ext3_journal_start(inode, needed_blocks); |
Andrew Morton | 1aa9b4b | 2007-04-01 23:49:43 -0700 | [diff] [blame] | 1177 | if (IS_ERR(handle)) { |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1178 | unlock_page(page); |
| 1179 | page_cache_release(page); |
Andrew Morton | 1aa9b4b | 2007-04-01 23:49:43 -0700 | [diff] [blame] | 1180 | ret = PTR_ERR(handle); |
| 1181 | goto out; |
| 1182 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1183 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
| 1184 | ext3_get_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1185 | if (ret) |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1186 | goto write_begin_failed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1187 | |
| 1188 | if (ext3_should_journal_data(inode)) { |
| 1189 | ret = walk_page_buffers(handle, page_buffers(page), |
| 1190 | from, to, NULL, do_journal_get_write_access); |
| 1191 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1192 | write_begin_failed: |
| 1193 | if (ret) { |
Aneesh Kumar K.V | 5ec8b75 | 2008-10-18 20:28:00 -0700 | [diff] [blame] | 1194 | /* |
| 1195 | * block_write_begin may have instantiated a few blocks |
| 1196 | * outside i_size. Trim these off again. Don't need |
| 1197 | * i_size_read because we hold i_mutex. |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1198 | * |
| 1199 | * Add inode to orphan list in case we crash before truncate |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1200 | * finishes. Do this only if ext3_can_truncate() agrees so |
| 1201 | * that orphan processing code is happy. |
Aneesh Kumar K.V | 5ec8b75 | 2008-10-18 20:28:00 -0700 | [diff] [blame] | 1202 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1203 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1204 | ext3_orphan_add(handle, inode); |
| 1205 | ext3_journal_stop(handle); |
| 1206 | unlock_page(page); |
| 1207 | page_cache_release(page); |
| 1208 | if (pos + len > inode->i_size) |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1209 | ext3_truncate(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1210 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
| 1212 | goto retry; |
Andrew Morton | 1aa9b4b | 2007-04-01 23:49:43 -0700 | [diff] [blame] | 1213 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1214 | return ret; |
| 1215 | } |
| 1216 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1217 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1218 | int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1219 | { |
| 1220 | int err = journal_dirty_data(handle, bh); |
| 1221 | if (err) |
Harvey Harrison | e05b6b5 | 2008-04-28 02:16:15 -0700 | [diff] [blame] | 1222 | ext3_journal_abort_handle(__func__, __func__, |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1223 | bh, handle, err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | return err; |
| 1225 | } |
| 1226 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1227 | /* For ordered writepage and write_end functions */ |
| 1228 | static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) |
| 1229 | { |
| 1230 | /* |
| 1231 | * Write could have mapped the buffer but it didn't copy the data in |
| 1232 | * yet. So avoid filing such buffer into a transaction. |
| 1233 | */ |
| 1234 | if (buffer_mapped(bh) && buffer_uptodate(bh)) |
| 1235 | return ext3_journal_dirty_data(handle, bh); |
| 1236 | return 0; |
| 1237 | } |
| 1238 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1239 | /* For write_end() in data=journal mode */ |
| 1240 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | { |
| 1242 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1243 | return 0; |
| 1244 | set_buffer_uptodate(bh); |
| 1245 | return ext3_journal_dirty_metadata(handle, bh); |
| 1246 | } |
| 1247 | |
| 1248 | /* |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1249 | * This is nasty and subtle: ext3_write_begin() could have allocated blocks |
| 1250 | * for the whole page but later we failed to copy the data in. Update inode |
| 1251 | * size according to what we managed to copy. The rest is going to be |
| 1252 | * truncated in write_end function. |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1253 | */ |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1254 | static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied) |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1255 | { |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1256 | /* What matters to us is i_disksize. We don't write i_size anywhere */ |
| 1257 | if (pos + copied > inode->i_size) |
| 1258 | i_size_write(inode, pos + copied); |
| 1259 | if (pos + copied > EXT3_I(inode)->i_disksize) { |
| 1260 | EXT3_I(inode)->i_disksize = pos + copied; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1261 | mark_inode_dirty(inode); |
| 1262 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1263 | } |
| 1264 | |
| 1265 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1266 | * We need to pick up the new inode size which generic_commit_write gave us |
| 1267 | * `file' can be NULL - eg, when called from page_symlink(). |
| 1268 | * |
| 1269 | * ext3 never places buffers on inode->i_mapping->private_list. metadata |
| 1270 | * buffers are managed internally. |
| 1271 | */ |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1272 | static int ext3_ordered_write_end(struct file *file, |
| 1273 | struct address_space *mapping, |
| 1274 | loff_t pos, unsigned len, unsigned copied, |
| 1275 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | { |
| 1277 | handle_t *handle = ext3_journal_current_handle(); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1278 | struct inode *inode = file->f_mapping->host; |
| 1279 | unsigned from, to; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | int ret = 0, ret2; |
| 1281 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1282 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 1283 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1284 | from = pos & (PAGE_CACHE_SIZE - 1); |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1285 | to = from + copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | ret = walk_page_buffers(handle, page_buffers(page), |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1287 | from, to, NULL, journal_dirty_data_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1289 | if (ret == 0) |
| 1290 | update_file_sizes(inode, pos, copied); |
| 1291 | /* |
| 1292 | * There may be allocated blocks outside of i_size because |
| 1293 | * we failed to copy some data. Prepare for truncate. |
| 1294 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1295 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1296 | ext3_orphan_add(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1297 | ret2 = ext3_journal_stop(handle); |
| 1298 | if (!ret) |
| 1299 | ret = ret2; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1300 | unlock_page(page); |
| 1301 | page_cache_release(page); |
| 1302 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1303 | if (pos + len > inode->i_size) |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1304 | ext3_truncate(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1305 | return ret ? ret : copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1306 | } |
| 1307 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1308 | static int ext3_writeback_write_end(struct file *file, |
| 1309 | struct address_space *mapping, |
| 1310 | loff_t pos, unsigned len, unsigned copied, |
| 1311 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1312 | { |
| 1313 | handle_t *handle = ext3_journal_current_handle(); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1314 | struct inode *inode = file->f_mapping->host; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1315 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1317 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 1318 | update_file_sizes(inode, pos, copied); |
| 1319 | /* |
| 1320 | * There may be allocated blocks outside of i_size because |
| 1321 | * we failed to copy some data. Prepare for truncate. |
| 1322 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1323 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1324 | ext3_orphan_add(handle, inode); |
| 1325 | ret = ext3_journal_stop(handle); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1326 | unlock_page(page); |
| 1327 | page_cache_release(page); |
| 1328 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1329 | if (pos + len > inode->i_size) |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1330 | ext3_truncate(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1331 | return ret ? ret : copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1332 | } |
| 1333 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1334 | static int ext3_journalled_write_end(struct file *file, |
| 1335 | struct address_space *mapping, |
| 1336 | loff_t pos, unsigned len, unsigned copied, |
| 1337 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | { |
| 1339 | handle_t *handle = ext3_journal_current_handle(); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1340 | struct inode *inode = mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | int ret = 0, ret2; |
| 1342 | int partial = 0; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1343 | unsigned from, to; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1345 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1346 | to = from + len; |
| 1347 | |
| 1348 | if (copied < len) { |
| 1349 | if (!PageUptodate(page)) |
| 1350 | copied = 0; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1351 | page_zero_new_buffers(page, from + copied, to); |
| 1352 | to = from + copied; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1353 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | |
| 1355 | ret = walk_page_buffers(handle, page_buffers(page), from, |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1356 | to, &partial, write_end_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | if (!partial) |
| 1358 | SetPageUptodate(page); |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1359 | |
| 1360 | if (pos + copied > inode->i_size) |
| 1361 | i_size_write(inode, pos + copied); |
| 1362 | /* |
| 1363 | * There may be allocated blocks outside of i_size because |
| 1364 | * we failed to copy some data. Prepare for truncate. |
| 1365 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1366 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1367 | ext3_orphan_add(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | EXT3_I(inode)->i_state |= EXT3_STATE_JDATA; |
| 1369 | if (inode->i_size > EXT3_I(inode)->i_disksize) { |
| 1370 | EXT3_I(inode)->i_disksize = inode->i_size; |
| 1371 | ret2 = ext3_mark_inode_dirty(handle, inode); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1372 | if (!ret) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | ret = ret2; |
| 1374 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1375 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | ret2 = ext3_journal_stop(handle); |
| 1377 | if (!ret) |
| 1378 | ret = ret2; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1379 | unlock_page(page); |
| 1380 | page_cache_release(page); |
| 1381 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1382 | if (pos + len > inode->i_size) |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1383 | ext3_truncate(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1384 | return ret ? ret : copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | } |
| 1386 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1387 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | * bmap() is special. It gets used by applications such as lilo and by |
| 1389 | * the swapper to find the on-disk block of a specific piece of data. |
| 1390 | * |
| 1391 | * Naturally, this is dangerous if the block concerned is still in the |
| 1392 | * journal. If somebody makes a swapfile on an ext3 data-journaling |
| 1393 | * filesystem and enables swap, then they may get a nasty shock when the |
| 1394 | * data getting swapped to that swapfile suddenly gets overwritten by |
| 1395 | * the original zero's written out previously to the journal and |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1396 | * awaiting writeback in the kernel's buffer cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | * |
| 1398 | * So, if we see any bmap calls here on a modified, data-journaled file, |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1399 | * take extra steps to flush any blocks which might be in the cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1400 | */ |
| 1401 | static sector_t ext3_bmap(struct address_space *mapping, sector_t block) |
| 1402 | { |
| 1403 | struct inode *inode = mapping->host; |
| 1404 | journal_t *journal; |
| 1405 | int err; |
| 1406 | |
| 1407 | if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) { |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1408 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | * This is a REALLY heavyweight approach, but the use of |
| 1410 | * bmap on dirty files is expected to be extremely rare: |
| 1411 | * only if we run lilo or swapon on a freshly made file |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1412 | * do we expect this to happen. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1413 | * |
| 1414 | * (bmap requires CAP_SYS_RAWIO so this does not |
| 1415 | * represent an unprivileged user DOS attack --- we'd be |
| 1416 | * in trouble if mortal users could trigger this path at |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1417 | * will.) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | * |
| 1419 | * NB. EXT3_STATE_JDATA is not set on files other than |
| 1420 | * regular files. If somebody wants to bmap a directory |
| 1421 | * or symlink and gets confused because the buffer |
| 1422 | * hasn't yet been flushed to disk, they deserve |
| 1423 | * everything they get. |
| 1424 | */ |
| 1425 | |
| 1426 | EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA; |
| 1427 | journal = EXT3_JOURNAL(inode); |
| 1428 | journal_lock_updates(journal); |
| 1429 | err = journal_flush(journal); |
| 1430 | journal_unlock_updates(journal); |
| 1431 | |
| 1432 | if (err) |
| 1433 | return 0; |
| 1434 | } |
| 1435 | |
| 1436 | return generic_block_bmap(mapping,block,ext3_get_block); |
| 1437 | } |
| 1438 | |
| 1439 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
| 1440 | { |
| 1441 | get_bh(bh); |
| 1442 | return 0; |
| 1443 | } |
| 1444 | |
| 1445 | static int bput_one(handle_t *handle, struct buffer_head *bh) |
| 1446 | { |
| 1447 | put_bh(bh); |
| 1448 | return 0; |
| 1449 | } |
| 1450 | |
Jan Kara | 9e80d40 | 2009-03-26 13:08:04 +0100 | [diff] [blame] | 1451 | static int buffer_unmapped(handle_t *handle, struct buffer_head *bh) |
| 1452 | { |
| 1453 | return !buffer_mapped(bh); |
| 1454 | } |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1455 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1456 | /* |
| 1457 | * Note that we always start a transaction even if we're not journalling |
| 1458 | * data. This is to preserve ordering: any hole instantiation within |
| 1459 | * __block_write_full_page -> ext3_get_block() should be journalled |
| 1460 | * along with the data so we don't crash and then get metadata which |
| 1461 | * refers to old data. |
| 1462 | * |
| 1463 | * In all journalling modes block_write_full_page() will start the I/O. |
| 1464 | * |
| 1465 | * Problem: |
| 1466 | * |
| 1467 | * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
| 1468 | * ext3_writepage() |
| 1469 | * |
| 1470 | * Similar for: |
| 1471 | * |
| 1472 | * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
| 1473 | * |
| 1474 | * Same applies to ext3_get_block(). We will deadlock on various things like |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 1475 | * lock_journal and i_truncate_mutex. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | * |
| 1477 | * Setting PF_MEMALLOC here doesn't work - too many internal memory |
| 1478 | * allocations fail. |
| 1479 | * |
| 1480 | * 16May01: If we're reentered then journal_current_handle() will be |
| 1481 | * non-zero. We simply *return*. |
| 1482 | * |
| 1483 | * 1 July 2001: @@@ FIXME: |
| 1484 | * In journalled data mode, a data buffer may be metadata against the |
| 1485 | * current transaction. But the same file is part of a shared mapping |
| 1486 | * and someone does a writepage() on it. |
| 1487 | * |
| 1488 | * We will move the buffer onto the async_data list, but *after* it has |
| 1489 | * been dirtied. So there's a small window where we have dirty data on |
| 1490 | * BJ_Metadata. |
| 1491 | * |
| 1492 | * Note that this only applies to the last partial page in the file. The |
| 1493 | * bit which block_write_full_page() uses prepare/commit for. (That's |
| 1494 | * broken code anyway: it's wrong for msync()). |
| 1495 | * |
| 1496 | * It's a rare case: affects the final partial page, for journalled data |
| 1497 | * where the file is subject to bith write() and writepage() in the same |
| 1498 | * transction. To fix it we'll need a custom block_write_full_page(). |
| 1499 | * We'll probably need that anyway for journalling writepage() output. |
| 1500 | * |
| 1501 | * We don't honour synchronous mounts for writepage(). That would be |
| 1502 | * disastrous. Any write() or metadata operation will sync the fs for |
| 1503 | * us. |
| 1504 | * |
| 1505 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, |
| 1506 | * we don't need to open a transaction here. |
| 1507 | */ |
| 1508 | static int ext3_ordered_writepage(struct page *page, |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1509 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | { |
| 1511 | struct inode *inode = page->mapping->host; |
| 1512 | struct buffer_head *page_bufs; |
| 1513 | handle_t *handle = NULL; |
| 1514 | int ret = 0; |
| 1515 | int err; |
| 1516 | |
| 1517 | J_ASSERT(PageLocked(page)); |
| 1518 | |
| 1519 | /* |
| 1520 | * We give up here if we're reentered, because it might be for a |
| 1521 | * different filesystem. |
| 1522 | */ |
| 1523 | if (ext3_journal_current_handle()) |
| 1524 | goto out_fail; |
| 1525 | |
Jan Kara | 9e80d40 | 2009-03-26 13:08:04 +0100 | [diff] [blame] | 1526 | if (!page_has_buffers(page)) { |
| 1527 | create_empty_buffers(page, inode->i_sb->s_blocksize, |
| 1528 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
Jan Kara | 430db32 | 2009-04-07 18:25:01 -0400 | [diff] [blame] | 1529 | page_bufs = page_buffers(page); |
| 1530 | } else { |
| 1531 | page_bufs = page_buffers(page); |
| 1532 | if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE, |
| 1533 | NULL, buffer_unmapped)) { |
| 1534 | /* Provide NULL get_block() to catch bugs if buffers |
| 1535 | * weren't really mapped */ |
| 1536 | return block_write_full_page(page, NULL, wbc); |
| 1537 | } |
Jan Kara | 9e80d40 | 2009-03-26 13:08:04 +0100 | [diff] [blame] | 1538 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1539 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
| 1540 | |
| 1541 | if (IS_ERR(handle)) { |
| 1542 | ret = PTR_ERR(handle); |
| 1543 | goto out_fail; |
| 1544 | } |
| 1545 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | walk_page_buffers(handle, page_bufs, 0, |
| 1547 | PAGE_CACHE_SIZE, NULL, bget_one); |
| 1548 | |
| 1549 | ret = block_write_full_page(page, ext3_get_block, wbc); |
| 1550 | |
| 1551 | /* |
| 1552 | * The page can become unlocked at any point now, and |
| 1553 | * truncate can then come in and change things. So we |
| 1554 | * can't touch *page from now on. But *page_bufs is |
| 1555 | * safe due to elevated refcount. |
| 1556 | */ |
| 1557 | |
| 1558 | /* |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1559 | * And attach them to the current transaction. But only if |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1560 | * block_write_full_page() succeeded. Otherwise they are unmapped, |
| 1561 | * and generally junk. |
| 1562 | */ |
| 1563 | if (ret == 0) { |
| 1564 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, |
| 1565 | NULL, journal_dirty_data_fn); |
| 1566 | if (!ret) |
| 1567 | ret = err; |
| 1568 | } |
| 1569 | walk_page_buffers(handle, page_bufs, 0, |
| 1570 | PAGE_CACHE_SIZE, NULL, bput_one); |
| 1571 | err = ext3_journal_stop(handle); |
| 1572 | if (!ret) |
| 1573 | ret = err; |
| 1574 | return ret; |
| 1575 | |
| 1576 | out_fail: |
| 1577 | redirty_page_for_writepage(wbc, page); |
| 1578 | unlock_page(page); |
| 1579 | return ret; |
| 1580 | } |
| 1581 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1582 | static int ext3_writeback_writepage(struct page *page, |
| 1583 | struct writeback_control *wbc) |
| 1584 | { |
| 1585 | struct inode *inode = page->mapping->host; |
| 1586 | handle_t *handle = NULL; |
| 1587 | int ret = 0; |
| 1588 | int err; |
| 1589 | |
| 1590 | if (ext3_journal_current_handle()) |
| 1591 | goto out_fail; |
| 1592 | |
Jan Kara | 430db32 | 2009-04-07 18:25:01 -0400 | [diff] [blame] | 1593 | if (page_has_buffers(page)) { |
| 1594 | if (!walk_page_buffers(NULL, page_buffers(page), 0, |
| 1595 | PAGE_CACHE_SIZE, NULL, buffer_unmapped)) { |
| 1596 | /* Provide NULL get_block() to catch bugs if buffers |
| 1597 | * weren't really mapped */ |
| 1598 | return block_write_full_page(page, NULL, wbc); |
| 1599 | } |
| 1600 | } |
| 1601 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1602 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
| 1603 | if (IS_ERR(handle)) { |
| 1604 | ret = PTR_ERR(handle); |
| 1605 | goto out_fail; |
| 1606 | } |
| 1607 | |
Badari Pulavarty | 0e31f51 | 2006-07-30 03:04:14 -0700 | [diff] [blame] | 1608 | if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1609 | ret = nobh_writepage(page, ext3_get_block, wbc); |
| 1610 | else |
| 1611 | ret = block_write_full_page(page, ext3_get_block, wbc); |
| 1612 | |
| 1613 | err = ext3_journal_stop(handle); |
| 1614 | if (!ret) |
| 1615 | ret = err; |
| 1616 | return ret; |
| 1617 | |
| 1618 | out_fail: |
| 1619 | redirty_page_for_writepage(wbc, page); |
| 1620 | unlock_page(page); |
| 1621 | return ret; |
| 1622 | } |
| 1623 | |
| 1624 | static int ext3_journalled_writepage(struct page *page, |
| 1625 | struct writeback_control *wbc) |
| 1626 | { |
| 1627 | struct inode *inode = page->mapping->host; |
| 1628 | handle_t *handle = NULL; |
| 1629 | int ret = 0; |
| 1630 | int err; |
| 1631 | |
| 1632 | if (ext3_journal_current_handle()) |
| 1633 | goto no_write; |
| 1634 | |
| 1635 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
| 1636 | if (IS_ERR(handle)) { |
| 1637 | ret = PTR_ERR(handle); |
| 1638 | goto no_write; |
| 1639 | } |
| 1640 | |
| 1641 | if (!page_has_buffers(page) || PageChecked(page)) { |
| 1642 | /* |
| 1643 | * It's mmapped pagecache. Add buffers and journal it. There |
| 1644 | * doesn't seem much point in redirtying the page here. |
| 1645 | */ |
| 1646 | ClearPageChecked(page); |
| 1647 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, |
| 1648 | ext3_get_block); |
Denis Lunev | ab4eb43 | 2005-11-13 16:07:17 -0800 | [diff] [blame] | 1649 | if (ret != 0) { |
| 1650 | ext3_journal_stop(handle); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1651 | goto out_unlock; |
Denis Lunev | ab4eb43 | 2005-11-13 16:07:17 -0800 | [diff] [blame] | 1652 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | ret = walk_page_buffers(handle, page_buffers(page), 0, |
| 1654 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); |
| 1655 | |
| 1656 | err = walk_page_buffers(handle, page_buffers(page), 0, |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1657 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1658 | if (ret == 0) |
| 1659 | ret = err; |
| 1660 | EXT3_I(inode)->i_state |= EXT3_STATE_JDATA; |
| 1661 | unlock_page(page); |
| 1662 | } else { |
| 1663 | /* |
| 1664 | * It may be a page full of checkpoint-mode buffers. We don't |
| 1665 | * really know unless we go poke around in the buffer_heads. |
| 1666 | * But block_write_full_page will do the right thing. |
| 1667 | */ |
| 1668 | ret = block_write_full_page(page, ext3_get_block, wbc); |
| 1669 | } |
| 1670 | err = ext3_journal_stop(handle); |
| 1671 | if (!ret) |
| 1672 | ret = err; |
| 1673 | out: |
| 1674 | return ret; |
| 1675 | |
| 1676 | no_write: |
| 1677 | redirty_page_for_writepage(wbc, page); |
| 1678 | out_unlock: |
| 1679 | unlock_page(page); |
| 1680 | goto out; |
| 1681 | } |
| 1682 | |
| 1683 | static int ext3_readpage(struct file *file, struct page *page) |
| 1684 | { |
| 1685 | return mpage_readpage(page, ext3_get_block); |
| 1686 | } |
| 1687 | |
| 1688 | static int |
| 1689 | ext3_readpages(struct file *file, struct address_space *mapping, |
| 1690 | struct list_head *pages, unsigned nr_pages) |
| 1691 | { |
| 1692 | return mpage_readpages(mapping, pages, nr_pages, ext3_get_block); |
| 1693 | } |
| 1694 | |
NeilBrown | 2ff28e2 | 2006-03-26 01:37:18 -0800 | [diff] [blame] | 1695 | static void ext3_invalidatepage(struct page *page, unsigned long offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1696 | { |
| 1697 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); |
| 1698 | |
| 1699 | /* |
| 1700 | * If it's a full truncate we just forget about the pending dirtying |
| 1701 | */ |
| 1702 | if (offset == 0) |
| 1703 | ClearPageChecked(page); |
| 1704 | |
NeilBrown | 2ff28e2 | 2006-03-26 01:37:18 -0800 | [diff] [blame] | 1705 | journal_invalidatepage(journal, page, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1706 | } |
| 1707 | |
Al Viro | 27496a8 | 2005-10-21 03:20:48 -0400 | [diff] [blame] | 1708 | static int ext3_releasepage(struct page *page, gfp_t wait) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1709 | { |
| 1710 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); |
| 1711 | |
| 1712 | WARN_ON(PageChecked(page)); |
| 1713 | if (!page_has_buffers(page)) |
| 1714 | return 0; |
| 1715 | return journal_try_to_free_buffers(journal, page, wait); |
| 1716 | } |
| 1717 | |
| 1718 | /* |
| 1719 | * If the O_DIRECT write will extend the file then add this inode to the |
| 1720 | * orphan list. So recovery will truncate it back to the original size |
| 1721 | * if the machine crashes during the write. |
| 1722 | * |
| 1723 | * If the O_DIRECT write is intantiating holes inside i_size and the machine |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1724 | * crashes then stale disk data _may_ be exposed inside the file. But current |
| 1725 | * VFS code falls back into buffered path in that case so we are safe. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1726 | */ |
| 1727 | static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb, |
| 1728 | const struct iovec *iov, loff_t offset, |
| 1729 | unsigned long nr_segs) |
| 1730 | { |
| 1731 | struct file *file = iocb->ki_filp; |
| 1732 | struct inode *inode = file->f_mapping->host; |
| 1733 | struct ext3_inode_info *ei = EXT3_I(inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1734 | handle_t *handle; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1735 | ssize_t ret; |
| 1736 | int orphan = 0; |
| 1737 | size_t count = iov_length(iov, nr_segs); |
| 1738 | |
| 1739 | if (rw == WRITE) { |
| 1740 | loff_t final_size = offset + count; |
| 1741 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1742 | if (final_size > inode->i_size) { |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1743 | /* Credits for sb + inode write */ |
| 1744 | handle = ext3_journal_start(inode, 2); |
| 1745 | if (IS_ERR(handle)) { |
| 1746 | ret = PTR_ERR(handle); |
| 1747 | goto out; |
| 1748 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1749 | ret = ext3_orphan_add(handle, inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1750 | if (ret) { |
| 1751 | ext3_journal_stop(handle); |
| 1752 | goto out; |
| 1753 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1754 | orphan = 1; |
| 1755 | ei->i_disksize = inode->i_size; |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1756 | ext3_journal_stop(handle); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | } |
| 1758 | } |
| 1759 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1760 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | offset, nr_segs, |
Badari Pulavarty | f91a2ad | 2006-03-26 01:38:04 -0800 | [diff] [blame] | 1762 | ext3_get_block, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1763 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1764 | if (orphan) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1765 | int err; |
| 1766 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1767 | /* Credits for sb + inode write */ |
| 1768 | handle = ext3_journal_start(inode, 2); |
| 1769 | if (IS_ERR(handle)) { |
| 1770 | /* This is really bad luck. We've written the data |
| 1771 | * but cannot extend i_size. Bail out and pretend |
| 1772 | * the write failed... */ |
| 1773 | ret = PTR_ERR(handle); |
| 1774 | goto out; |
| 1775 | } |
| 1776 | if (inode->i_nlink) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1777 | ext3_orphan_del(handle, inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1778 | if (ret > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1779 | loff_t end = offset + ret; |
| 1780 | if (end > inode->i_size) { |
| 1781 | ei->i_disksize = end; |
| 1782 | i_size_write(inode, end); |
| 1783 | /* |
| 1784 | * We're going to return a positive `ret' |
| 1785 | * here due to non-zero-length I/O, so there's |
| 1786 | * no way of reporting error returns from |
| 1787 | * ext3_mark_inode_dirty() to userspace. So |
| 1788 | * ignore it. |
| 1789 | */ |
| 1790 | ext3_mark_inode_dirty(handle, inode); |
| 1791 | } |
| 1792 | } |
| 1793 | err = ext3_journal_stop(handle); |
| 1794 | if (ret == 0) |
| 1795 | ret = err; |
| 1796 | } |
| 1797 | out: |
| 1798 | return ret; |
| 1799 | } |
| 1800 | |
| 1801 | /* |
| 1802 | * Pages can be marked dirty completely asynchronously from ext3's journalling |
| 1803 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
| 1804 | * much here because ->set_page_dirty is called under VFS locks. The page is |
| 1805 | * not necessarily locked. |
| 1806 | * |
| 1807 | * We cannot just dirty the page and leave attached buffers clean, because the |
| 1808 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
| 1809 | * or jbddirty because all the journalling code will explode. |
| 1810 | * |
| 1811 | * So what we do is to mark the page "pending dirty" and next time writepage |
| 1812 | * is called, propagate that into the buffers appropriately. |
| 1813 | */ |
| 1814 | static int ext3_journalled_set_page_dirty(struct page *page) |
| 1815 | { |
| 1816 | SetPageChecked(page); |
| 1817 | return __set_page_dirty_nobuffers(page); |
| 1818 | } |
| 1819 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1820 | static const struct address_space_operations ext3_ordered_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1821 | .readpage = ext3_readpage, |
| 1822 | .readpages = ext3_readpages, |
| 1823 | .writepage = ext3_ordered_writepage, |
| 1824 | .sync_page = block_sync_page, |
| 1825 | .write_begin = ext3_write_begin, |
| 1826 | .write_end = ext3_ordered_write_end, |
| 1827 | .bmap = ext3_bmap, |
| 1828 | .invalidatepage = ext3_invalidatepage, |
| 1829 | .releasepage = ext3_releasepage, |
| 1830 | .direct_IO = ext3_direct_IO, |
| 1831 | .migratepage = buffer_migrate_page, |
| 1832 | .is_partially_uptodate = block_is_partially_uptodate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | }; |
| 1834 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1835 | static const struct address_space_operations ext3_writeback_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1836 | .readpage = ext3_readpage, |
| 1837 | .readpages = ext3_readpages, |
| 1838 | .writepage = ext3_writeback_writepage, |
| 1839 | .sync_page = block_sync_page, |
| 1840 | .write_begin = ext3_write_begin, |
| 1841 | .write_end = ext3_writeback_write_end, |
| 1842 | .bmap = ext3_bmap, |
| 1843 | .invalidatepage = ext3_invalidatepage, |
| 1844 | .releasepage = ext3_releasepage, |
| 1845 | .direct_IO = ext3_direct_IO, |
| 1846 | .migratepage = buffer_migrate_page, |
| 1847 | .is_partially_uptodate = block_is_partially_uptodate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1848 | }; |
| 1849 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1850 | static const struct address_space_operations ext3_journalled_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1851 | .readpage = ext3_readpage, |
| 1852 | .readpages = ext3_readpages, |
| 1853 | .writepage = ext3_journalled_writepage, |
| 1854 | .sync_page = block_sync_page, |
| 1855 | .write_begin = ext3_write_begin, |
| 1856 | .write_end = ext3_journalled_write_end, |
| 1857 | .set_page_dirty = ext3_journalled_set_page_dirty, |
| 1858 | .bmap = ext3_bmap, |
| 1859 | .invalidatepage = ext3_invalidatepage, |
| 1860 | .releasepage = ext3_releasepage, |
| 1861 | .is_partially_uptodate = block_is_partially_uptodate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1862 | }; |
| 1863 | |
| 1864 | void ext3_set_aops(struct inode *inode) |
| 1865 | { |
| 1866 | if (ext3_should_order_data(inode)) |
| 1867 | inode->i_mapping->a_ops = &ext3_ordered_aops; |
| 1868 | else if (ext3_should_writeback_data(inode)) |
| 1869 | inode->i_mapping->a_ops = &ext3_writeback_aops; |
| 1870 | else |
| 1871 | inode->i_mapping->a_ops = &ext3_journalled_aops; |
| 1872 | } |
| 1873 | |
| 1874 | /* |
| 1875 | * ext3_block_truncate_page() zeroes out a mapping from file offset `from' |
| 1876 | * up to the end of the block which corresponds to `from'. |
| 1877 | * This required during truncate. We need to physically zero the tail end |
| 1878 | * of that block so it doesn't yield old data if the file is later grown. |
| 1879 | */ |
| 1880 | static int ext3_block_truncate_page(handle_t *handle, struct page *page, |
| 1881 | struct address_space *mapping, loff_t from) |
| 1882 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 1883 | ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
| 1885 | unsigned blocksize, iblock, length, pos; |
| 1886 | struct inode *inode = mapping->host; |
| 1887 | struct buffer_head *bh; |
| 1888 | int err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | |
| 1890 | blocksize = inode->i_sb->s_blocksize; |
| 1891 | length = blocksize - (offset & (blocksize - 1)); |
| 1892 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
| 1893 | |
| 1894 | /* |
| 1895 | * For "nobh" option, we can only work if we don't need to |
| 1896 | * read-in the page - otherwise we create buffers to do the IO. |
| 1897 | */ |
Badari Pulavarty | cd6ef84 | 2006-03-11 03:27:14 -0800 | [diff] [blame] | 1898 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && |
| 1899 | ext3_should_writeback_data(inode) && PageUptodate(page)) { |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 1900 | zero_user(page, offset, length); |
Badari Pulavarty | cd6ef84 | 2006-03-11 03:27:14 -0800 | [diff] [blame] | 1901 | set_page_dirty(page); |
| 1902 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1903 | } |
| 1904 | |
| 1905 | if (!page_has_buffers(page)) |
| 1906 | create_empty_buffers(page, blocksize, 0); |
| 1907 | |
| 1908 | /* Find the buffer that contains "offset" */ |
| 1909 | bh = page_buffers(page); |
| 1910 | pos = blocksize; |
| 1911 | while (offset >= pos) { |
| 1912 | bh = bh->b_this_page; |
| 1913 | iblock++; |
| 1914 | pos += blocksize; |
| 1915 | } |
| 1916 | |
| 1917 | err = 0; |
| 1918 | if (buffer_freed(bh)) { |
| 1919 | BUFFER_TRACE(bh, "freed: skip"); |
| 1920 | goto unlock; |
| 1921 | } |
| 1922 | |
| 1923 | if (!buffer_mapped(bh)) { |
| 1924 | BUFFER_TRACE(bh, "unmapped"); |
| 1925 | ext3_get_block(inode, iblock, bh, 0); |
| 1926 | /* unmapped? It's a hole - nothing to do */ |
| 1927 | if (!buffer_mapped(bh)) { |
| 1928 | BUFFER_TRACE(bh, "still unmapped"); |
| 1929 | goto unlock; |
| 1930 | } |
| 1931 | } |
| 1932 | |
| 1933 | /* Ok, it's mapped. Make sure it's up-to-date */ |
| 1934 | if (PageUptodate(page)) |
| 1935 | set_buffer_uptodate(bh); |
| 1936 | |
| 1937 | if (!buffer_uptodate(bh)) { |
| 1938 | err = -EIO; |
| 1939 | ll_rw_block(READ, 1, &bh); |
| 1940 | wait_on_buffer(bh); |
| 1941 | /* Uhhuh. Read error. Complain and punt. */ |
| 1942 | if (!buffer_uptodate(bh)) |
| 1943 | goto unlock; |
| 1944 | } |
| 1945 | |
| 1946 | if (ext3_should_journal_data(inode)) { |
| 1947 | BUFFER_TRACE(bh, "get write access"); |
| 1948 | err = ext3_journal_get_write_access(handle, bh); |
| 1949 | if (err) |
| 1950 | goto unlock; |
| 1951 | } |
| 1952 | |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 1953 | zero_user(page, offset, length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1954 | BUFFER_TRACE(bh, "zeroed end of block"); |
| 1955 | |
| 1956 | err = 0; |
| 1957 | if (ext3_should_journal_data(inode)) { |
| 1958 | err = ext3_journal_dirty_metadata(handle, bh); |
| 1959 | } else { |
| 1960 | if (ext3_should_order_data(inode)) |
| 1961 | err = ext3_journal_dirty_data(handle, bh); |
| 1962 | mark_buffer_dirty(bh); |
| 1963 | } |
| 1964 | |
| 1965 | unlock: |
| 1966 | unlock_page(page); |
| 1967 | page_cache_release(page); |
| 1968 | return err; |
| 1969 | } |
| 1970 | |
| 1971 | /* |
| 1972 | * Probably it should be a library function... search for first non-zero word |
| 1973 | * or memcmp with zero_page, whatever is better for particular architecture. |
| 1974 | * Linus? |
| 1975 | */ |
| 1976 | static inline int all_zeroes(__le32 *p, __le32 *q) |
| 1977 | { |
| 1978 | while (p < q) |
| 1979 | if (*p++) |
| 1980 | return 0; |
| 1981 | return 1; |
| 1982 | } |
| 1983 | |
| 1984 | /** |
| 1985 | * ext3_find_shared - find the indirect blocks for partial truncation. |
| 1986 | * @inode: inode in question |
| 1987 | * @depth: depth of the affected branch |
| 1988 | * @offsets: offsets of pointers in that branch (see ext3_block_to_path) |
| 1989 | * @chain: place to store the pointers to partial indirect blocks |
| 1990 | * @top: place to the (detached) top of branch |
| 1991 | * |
| 1992 | * This is a helper function used by ext3_truncate(). |
| 1993 | * |
| 1994 | * When we do truncate() we may have to clean the ends of several |
| 1995 | * indirect blocks but leave the blocks themselves alive. Block is |
| 1996 | * partially truncated if some data below the new i_size is refered |
| 1997 | * from it (and it is on the path to the first completely truncated |
| 1998 | * data block, indeed). We have to free the top of that path along |
| 1999 | * with everything to the right of the path. Since no allocation |
| 2000 | * past the truncation point is possible until ext3_truncate() |
| 2001 | * finishes, we may safely do the latter, but top of branch may |
| 2002 | * require special attention - pageout below the truncation point |
| 2003 | * might try to populate it. |
| 2004 | * |
| 2005 | * We atomically detach the top of branch from the tree, store the |
| 2006 | * block number of its root in *@top, pointers to buffer_heads of |
| 2007 | * partially truncated blocks - in @chain[].bh and pointers to |
| 2008 | * their last elements that should not be removed - in |
| 2009 | * @chain[].p. Return value is the pointer to last filled element |
| 2010 | * of @chain. |
| 2011 | * |
| 2012 | * The work left to caller to do the actual freeing of subtrees: |
| 2013 | * a) free the subtree starting from *@top |
| 2014 | * b) free the subtrees whose roots are stored in |
| 2015 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) |
| 2016 | * c) free the subtrees growing from the inode past the @chain[0]. |
| 2017 | * (no partially truncated stuff there). */ |
| 2018 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2019 | static Indirect *ext3_find_shared(struct inode *inode, int depth, |
| 2020 | int offsets[4], Indirect chain[4], __le32 *top) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2021 | { |
| 2022 | Indirect *partial, *p; |
| 2023 | int k, err; |
| 2024 | |
| 2025 | *top = 0; |
| 2026 | /* Make k index the deepest non-null offest + 1 */ |
| 2027 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
| 2028 | ; |
| 2029 | partial = ext3_get_branch(inode, k, offsets, chain, &err); |
| 2030 | /* Writer: pointers */ |
| 2031 | if (!partial) |
| 2032 | partial = chain + k-1; |
| 2033 | /* |
| 2034 | * If the branch acquired continuation since we've looked at it - |
| 2035 | * fine, it should all survive and (new) top doesn't belong to us. |
| 2036 | */ |
| 2037 | if (!partial->key && *partial->p) |
| 2038 | /* Writer: end */ |
| 2039 | goto no_top; |
| 2040 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) |
| 2041 | ; |
| 2042 | /* |
| 2043 | * OK, we've found the last block that must survive. The rest of our |
| 2044 | * branch should be detached before unlocking. However, if that rest |
| 2045 | * of branch is all ours and does not grow immediately from the inode |
| 2046 | * it's easier to cheat and just decrement partial->p. |
| 2047 | */ |
| 2048 | if (p == chain + k - 1 && p > chain) { |
| 2049 | p->p--; |
| 2050 | } else { |
| 2051 | *top = *p->p; |
| 2052 | /* Nope, don't do this in ext3. Must leave the tree intact */ |
| 2053 | #if 0 |
| 2054 | *p->p = 0; |
| 2055 | #endif |
| 2056 | } |
| 2057 | /* Writer: end */ |
| 2058 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2059 | while(partial > p) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2060 | brelse(partial->bh); |
| 2061 | partial--; |
| 2062 | } |
| 2063 | no_top: |
| 2064 | return partial; |
| 2065 | } |
| 2066 | |
| 2067 | /* |
| 2068 | * Zero a number of block pointers in either an inode or an indirect block. |
| 2069 | * If we restart the transaction we must again get write access to the |
| 2070 | * indirect block for further modification. |
| 2071 | * |
| 2072 | * We release `count' blocks on disk, but (last - first) may be greater |
| 2073 | * than `count' because there can be holes in there. |
| 2074 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2075 | static void ext3_clear_blocks(handle_t *handle, struct inode *inode, |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2076 | struct buffer_head *bh, ext3_fsblk_t block_to_free, |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2077 | unsigned long count, __le32 *first, __le32 *last) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2078 | { |
| 2079 | __le32 *p; |
| 2080 | if (try_to_extend_transaction(handle, inode)) { |
| 2081 | if (bh) { |
| 2082 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
| 2083 | ext3_journal_dirty_metadata(handle, bh); |
| 2084 | } |
| 2085 | ext3_mark_inode_dirty(handle, inode); |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 2086 | truncate_restart_transaction(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2087 | if (bh) { |
| 2088 | BUFFER_TRACE(bh, "retaking write access"); |
| 2089 | ext3_journal_get_write_access(handle, bh); |
| 2090 | } |
| 2091 | } |
| 2092 | |
| 2093 | /* |
| 2094 | * Any buffers which are on the journal will be in memory. We find |
| 2095 | * them on the hash table so journal_revoke() will run journal_forget() |
| 2096 | * on them. We've already detached each block from the file, so |
| 2097 | * bforget() in journal_forget() should be safe. |
| 2098 | * |
| 2099 | * AKPM: turn on bforget in journal_forget()!!! |
| 2100 | */ |
| 2101 | for (p = first; p < last; p++) { |
| 2102 | u32 nr = le32_to_cpu(*p); |
| 2103 | if (nr) { |
| 2104 | struct buffer_head *bh; |
| 2105 | |
| 2106 | *p = 0; |
| 2107 | bh = sb_find_get_block(inode->i_sb, nr); |
| 2108 | ext3_forget(handle, 0, inode, bh, nr); |
| 2109 | } |
| 2110 | } |
| 2111 | |
| 2112 | ext3_free_blocks(handle, inode, block_to_free, count); |
| 2113 | } |
| 2114 | |
| 2115 | /** |
| 2116 | * ext3_free_data - free a list of data blocks |
| 2117 | * @handle: handle for this transaction |
| 2118 | * @inode: inode we are dealing with |
| 2119 | * @this_bh: indirect buffer_head which contains *@first and *@last |
| 2120 | * @first: array of block numbers |
| 2121 | * @last: points immediately past the end of array |
| 2122 | * |
| 2123 | * We are freeing all blocks refered from that array (numbers are stored as |
| 2124 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. |
| 2125 | * |
| 2126 | * We accumulate contiguous runs of blocks to free. Conveniently, if these |
| 2127 | * blocks are contiguous then releasing them at one time will only affect one |
| 2128 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't |
| 2129 | * actually use a lot of journal space. |
| 2130 | * |
| 2131 | * @this_bh will be %NULL if @first and @last point into the inode's direct |
| 2132 | * block pointers. |
| 2133 | */ |
| 2134 | static void ext3_free_data(handle_t *handle, struct inode *inode, |
| 2135 | struct buffer_head *this_bh, |
| 2136 | __le32 *first, __le32 *last) |
| 2137 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2138 | ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2139 | unsigned long count = 0; /* Number of blocks in the run */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
| 2141 | corresponding to |
| 2142 | block_to_free */ |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2143 | ext3_fsblk_t nr; /* Current block # */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | __le32 *p; /* Pointer into inode/ind |
| 2145 | for current block */ |
| 2146 | int err; |
| 2147 | |
| 2148 | if (this_bh) { /* For indirect block */ |
| 2149 | BUFFER_TRACE(this_bh, "get_write_access"); |
| 2150 | err = ext3_journal_get_write_access(handle, this_bh); |
| 2151 | /* Important: if we can't update the indirect pointers |
| 2152 | * to the blocks, we can't free them. */ |
| 2153 | if (err) |
| 2154 | return; |
| 2155 | } |
| 2156 | |
| 2157 | for (p = first; p < last; p++) { |
| 2158 | nr = le32_to_cpu(*p); |
| 2159 | if (nr) { |
| 2160 | /* accumulate blocks to free if they're contiguous */ |
| 2161 | if (count == 0) { |
| 2162 | block_to_free = nr; |
| 2163 | block_to_free_p = p; |
| 2164 | count = 1; |
| 2165 | } else if (nr == block_to_free + count) { |
| 2166 | count++; |
| 2167 | } else { |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2168 | ext3_clear_blocks(handle, inode, this_bh, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2169 | block_to_free, |
| 2170 | count, block_to_free_p, p); |
| 2171 | block_to_free = nr; |
| 2172 | block_to_free_p = p; |
| 2173 | count = 1; |
| 2174 | } |
| 2175 | } |
| 2176 | } |
| 2177 | |
| 2178 | if (count > 0) |
| 2179 | ext3_clear_blocks(handle, inode, this_bh, block_to_free, |
| 2180 | count, block_to_free_p, p); |
| 2181 | |
| 2182 | if (this_bh) { |
| 2183 | BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata"); |
Duane Griffin | 3ccc316 | 2008-07-25 01:46:26 -0700 | [diff] [blame] | 2184 | |
| 2185 | /* |
| 2186 | * The buffer head should have an attached journal head at this |
| 2187 | * point. However, if the data is corrupted and an indirect |
| 2188 | * block pointed to itself, it would have been detached when |
| 2189 | * the block was cleared. Check for this instead of OOPSing. |
| 2190 | */ |
| 2191 | if (bh2jh(this_bh)) |
| 2192 | ext3_journal_dirty_metadata(handle, this_bh); |
| 2193 | else |
| 2194 | ext3_error(inode->i_sb, "ext3_free_data", |
| 2195 | "circular indirect block detected, " |
| 2196 | "inode=%lu, block=%llu", |
| 2197 | inode->i_ino, |
| 2198 | (unsigned long long)this_bh->b_blocknr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | } |
| 2200 | } |
| 2201 | |
| 2202 | /** |
| 2203 | * ext3_free_branches - free an array of branches |
| 2204 | * @handle: JBD handle for this transaction |
| 2205 | * @inode: inode we are dealing with |
| 2206 | * @parent_bh: the buffer_head which contains *@first and *@last |
| 2207 | * @first: array of block numbers |
| 2208 | * @last: pointer immediately past the end of array |
| 2209 | * @depth: depth of the branches to free |
| 2210 | * |
| 2211 | * We are freeing all blocks refered from these branches (numbers are |
| 2212 | * stored as little-endian 32-bit) and updating @inode->i_blocks |
| 2213 | * appropriately. |
| 2214 | */ |
| 2215 | static void ext3_free_branches(handle_t *handle, struct inode *inode, |
| 2216 | struct buffer_head *parent_bh, |
| 2217 | __le32 *first, __le32 *last, int depth) |
| 2218 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2219 | ext3_fsblk_t nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2220 | __le32 *p; |
| 2221 | |
| 2222 | if (is_handle_aborted(handle)) |
| 2223 | return; |
| 2224 | |
| 2225 | if (depth--) { |
| 2226 | struct buffer_head *bh; |
| 2227 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); |
| 2228 | p = last; |
| 2229 | while (--p >= first) { |
| 2230 | nr = le32_to_cpu(*p); |
| 2231 | if (!nr) |
| 2232 | continue; /* A hole */ |
| 2233 | |
| 2234 | /* Go read the buffer for the next level down */ |
| 2235 | bh = sb_bread(inode->i_sb, nr); |
| 2236 | |
| 2237 | /* |
| 2238 | * A read failure? Report error and clear slot |
| 2239 | * (should be rare). |
| 2240 | */ |
| 2241 | if (!bh) { |
| 2242 | ext3_error(inode->i_sb, "ext3_free_branches", |
Eric Sandeen | eee194e | 2006-09-27 01:49:30 -0700 | [diff] [blame] | 2243 | "Read failure, inode=%lu, block="E3FSBLK, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2244 | inode->i_ino, nr); |
| 2245 | continue; |
| 2246 | } |
| 2247 | |
| 2248 | /* This zaps the entire block. Bottom up. */ |
| 2249 | BUFFER_TRACE(bh, "free child branches"); |
| 2250 | ext3_free_branches(handle, inode, bh, |
| 2251 | (__le32*)bh->b_data, |
| 2252 | (__le32*)bh->b_data + addr_per_block, |
| 2253 | depth); |
| 2254 | |
| 2255 | /* |
| 2256 | * We've probably journalled the indirect block several |
| 2257 | * times during the truncate. But it's no longer |
| 2258 | * needed and we now drop it from the transaction via |
| 2259 | * journal_revoke(). |
| 2260 | * |
| 2261 | * That's easy if it's exclusively part of this |
| 2262 | * transaction. But if it's part of the committing |
| 2263 | * transaction then journal_forget() will simply |
| 2264 | * brelse() it. That means that if the underlying |
| 2265 | * block is reallocated in ext3_get_block(), |
| 2266 | * unmap_underlying_metadata() will find this block |
| 2267 | * and will try to get rid of it. damn, damn. |
| 2268 | * |
| 2269 | * If this block has already been committed to the |
| 2270 | * journal, a revoke record will be written. And |
| 2271 | * revoke records must be emitted *before* clearing |
| 2272 | * this block's bit in the bitmaps. |
| 2273 | */ |
| 2274 | ext3_forget(handle, 1, inode, bh, bh->b_blocknr); |
| 2275 | |
| 2276 | /* |
| 2277 | * Everything below this this pointer has been |
| 2278 | * released. Now let this top-of-subtree go. |
| 2279 | * |
| 2280 | * We want the freeing of this indirect block to be |
| 2281 | * atomic in the journal with the updating of the |
| 2282 | * bitmap block which owns it. So make some room in |
| 2283 | * the journal. |
| 2284 | * |
| 2285 | * We zero the parent pointer *after* freeing its |
| 2286 | * pointee in the bitmaps, so if extend_transaction() |
| 2287 | * for some reason fails to put the bitmap changes and |
| 2288 | * the release into the same transaction, recovery |
| 2289 | * will merely complain about releasing a free block, |
| 2290 | * rather than leaking blocks. |
| 2291 | */ |
| 2292 | if (is_handle_aborted(handle)) |
| 2293 | return; |
| 2294 | if (try_to_extend_transaction(handle, inode)) { |
| 2295 | ext3_mark_inode_dirty(handle, inode); |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 2296 | truncate_restart_transaction(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2297 | } |
| 2298 | |
| 2299 | ext3_free_blocks(handle, inode, nr, 1); |
| 2300 | |
| 2301 | if (parent_bh) { |
| 2302 | /* |
| 2303 | * The block which we have just freed is |
| 2304 | * pointed to by an indirect block: journal it |
| 2305 | */ |
| 2306 | BUFFER_TRACE(parent_bh, "get_write_access"); |
| 2307 | if (!ext3_journal_get_write_access(handle, |
| 2308 | parent_bh)){ |
| 2309 | *p = 0; |
| 2310 | BUFFER_TRACE(parent_bh, |
| 2311 | "call ext3_journal_dirty_metadata"); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2312 | ext3_journal_dirty_metadata(handle, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2313 | parent_bh); |
| 2314 | } |
| 2315 | } |
| 2316 | } |
| 2317 | } else { |
| 2318 | /* We have reached the bottom of the tree. */ |
| 2319 | BUFFER_TRACE(parent_bh, "free data blocks"); |
| 2320 | ext3_free_data(handle, inode, parent_bh, first, last); |
| 2321 | } |
| 2322 | } |
| 2323 | |
Duane Griffin | ae76dd9 | 2008-07-25 01:46:23 -0700 | [diff] [blame] | 2324 | int ext3_can_truncate(struct inode *inode) |
| 2325 | { |
| 2326 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) |
| 2327 | return 0; |
| 2328 | if (S_ISREG(inode->i_mode)) |
| 2329 | return 1; |
| 2330 | if (S_ISDIR(inode->i_mode)) |
| 2331 | return 1; |
| 2332 | if (S_ISLNK(inode->i_mode)) |
| 2333 | return !ext3_inode_is_fast_symlink(inode); |
| 2334 | return 0; |
| 2335 | } |
| 2336 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | /* |
| 2338 | * ext3_truncate() |
| 2339 | * |
| 2340 | * We block out ext3_get_block() block instantiations across the entire |
| 2341 | * transaction, and VFS/VM ensures that ext3_truncate() cannot run |
| 2342 | * simultaneously on behalf of the same inode. |
| 2343 | * |
| 2344 | * As we work through the truncate and commmit bits of it to the journal there |
| 2345 | * is one core, guiding principle: the file's tree must always be consistent on |
| 2346 | * disk. We must be able to restart the truncate after a crash. |
| 2347 | * |
| 2348 | * The file's tree may be transiently inconsistent in memory (although it |
| 2349 | * probably isn't), but whenever we close off and commit a journal transaction, |
| 2350 | * the contents of (the filesystem + the journal) must be consistent and |
| 2351 | * restartable. It's pretty simple, really: bottom up, right to left (although |
| 2352 | * left-to-right works OK too). |
| 2353 | * |
| 2354 | * Note that at recovery time, journal replay occurs *before* the restart of |
| 2355 | * truncate against the orphan inode list. |
| 2356 | * |
| 2357 | * The committed inode has the new, desired i_size (which is the same as |
| 2358 | * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see |
| 2359 | * that this inode's truncate did not complete and it will again call |
| 2360 | * ext3_truncate() to have another go. So there will be instantiated blocks |
| 2361 | * to the right of the truncation point in a crashed ext3 filesystem. But |
| 2362 | * that's fine - as long as they are linked from the inode, the post-crash |
| 2363 | * ext3_truncate() run will find them and release them. |
| 2364 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2365 | void ext3_truncate(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2366 | { |
| 2367 | handle_t *handle; |
| 2368 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 2369 | __le32 *i_data = ei->i_data; |
| 2370 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); |
| 2371 | struct address_space *mapping = inode->i_mapping; |
| 2372 | int offsets[4]; |
| 2373 | Indirect chain[4]; |
| 2374 | Indirect *partial; |
| 2375 | __le32 nr = 0; |
| 2376 | int n; |
| 2377 | long last_block; |
| 2378 | unsigned blocksize = inode->i_sb->s_blocksize; |
| 2379 | struct page *page; |
| 2380 | |
Duane Griffin | ae76dd9 | 2008-07-25 01:46:23 -0700 | [diff] [blame] | 2381 | if (!ext3_can_truncate(inode)) |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2382 | goto out_notrans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2383 | |
Theodore Ts'o | f7ab34e | 2009-04-03 01:34:35 -0400 | [diff] [blame] | 2384 | if (inode->i_size == 0 && ext3_should_writeback_data(inode)) |
| 2385 | ei->i_state |= EXT3_STATE_FLUSH_ON_CLOSE; |
| 2386 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2387 | /* |
| 2388 | * We have to lock the EOF page here, because lock_page() nests |
| 2389 | * outside journal_start(). |
| 2390 | */ |
| 2391 | if ((inode->i_size & (blocksize - 1)) == 0) { |
| 2392 | /* Block boundary? Nothing to do */ |
| 2393 | page = NULL; |
| 2394 | } else { |
| 2395 | page = grab_cache_page(mapping, |
| 2396 | inode->i_size >> PAGE_CACHE_SHIFT); |
| 2397 | if (!page) |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2398 | goto out_notrans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2399 | } |
| 2400 | |
| 2401 | handle = start_transaction(inode); |
| 2402 | if (IS_ERR(handle)) { |
| 2403 | if (page) { |
| 2404 | clear_highpage(page); |
| 2405 | flush_dcache_page(page); |
| 2406 | unlock_page(page); |
| 2407 | page_cache_release(page); |
| 2408 | } |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2409 | goto out_notrans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2410 | } |
| 2411 | |
| 2412 | last_block = (inode->i_size + blocksize-1) |
| 2413 | >> EXT3_BLOCK_SIZE_BITS(inode->i_sb); |
| 2414 | |
| 2415 | if (page) |
| 2416 | ext3_block_truncate_page(handle, page, mapping, inode->i_size); |
| 2417 | |
| 2418 | n = ext3_block_to_path(inode, last_block, offsets, NULL); |
| 2419 | if (n == 0) |
| 2420 | goto out_stop; /* error */ |
| 2421 | |
| 2422 | /* |
| 2423 | * OK. This truncate is going to happen. We add the inode to the |
| 2424 | * orphan list, so that if this truncate spans multiple transactions, |
| 2425 | * and we crash, we will resume the truncate when the filesystem |
| 2426 | * recovers. It also marks the inode dirty, to catch the new size. |
| 2427 | * |
| 2428 | * Implication: the file must always be in a sane, consistent |
| 2429 | * truncatable state while each transaction commits. |
| 2430 | */ |
| 2431 | if (ext3_orphan_add(handle, inode)) |
| 2432 | goto out_stop; |
| 2433 | |
| 2434 | /* |
| 2435 | * The orphan list entry will now protect us from any crash which |
| 2436 | * occurs before the truncate completes, so it is now safe to propagate |
| 2437 | * the new, shorter inode size (held for now in i_size) into the |
| 2438 | * on-disk inode. We do this via i_disksize, which is the value which |
| 2439 | * ext3 *really* writes onto the disk inode. |
| 2440 | */ |
| 2441 | ei->i_disksize = inode->i_size; |
| 2442 | |
| 2443 | /* |
| 2444 | * From here we block out all ext3_get_block() callers who want to |
| 2445 | * modify the block allocation tree. |
| 2446 | */ |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 2447 | mutex_lock(&ei->truncate_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2448 | |
| 2449 | if (n == 1) { /* direct blocks */ |
| 2450 | ext3_free_data(handle, inode, NULL, i_data+offsets[0], |
| 2451 | i_data + EXT3_NDIR_BLOCKS); |
| 2452 | goto do_indirects; |
| 2453 | } |
| 2454 | |
| 2455 | partial = ext3_find_shared(inode, n, offsets, chain, &nr); |
| 2456 | /* Kill the top of shared branch (not detached) */ |
| 2457 | if (nr) { |
| 2458 | if (partial == chain) { |
| 2459 | /* Shared branch grows from the inode */ |
| 2460 | ext3_free_branches(handle, inode, NULL, |
| 2461 | &nr, &nr+1, (chain+n-1) - partial); |
| 2462 | *partial->p = 0; |
| 2463 | /* |
| 2464 | * We mark the inode dirty prior to restart, |
| 2465 | * and prior to stop. No need for it here. |
| 2466 | */ |
| 2467 | } else { |
| 2468 | /* Shared branch grows from an indirect block */ |
| 2469 | BUFFER_TRACE(partial->bh, "get_write_access"); |
| 2470 | ext3_free_branches(handle, inode, partial->bh, |
| 2471 | partial->p, |
| 2472 | partial->p+1, (chain+n-1) - partial); |
| 2473 | } |
| 2474 | } |
| 2475 | /* Clear the ends of indirect blocks on the shared branch */ |
| 2476 | while (partial > chain) { |
| 2477 | ext3_free_branches(handle, inode, partial->bh, partial->p + 1, |
| 2478 | (__le32*)partial->bh->b_data+addr_per_block, |
| 2479 | (chain+n-1) - partial); |
| 2480 | BUFFER_TRACE(partial->bh, "call brelse"); |
| 2481 | brelse (partial->bh); |
| 2482 | partial--; |
| 2483 | } |
| 2484 | do_indirects: |
| 2485 | /* Kill the remaining (whole) subtrees */ |
| 2486 | switch (offsets[0]) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2487 | default: |
| 2488 | nr = i_data[EXT3_IND_BLOCK]; |
| 2489 | if (nr) { |
| 2490 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
| 2491 | i_data[EXT3_IND_BLOCK] = 0; |
| 2492 | } |
| 2493 | case EXT3_IND_BLOCK: |
| 2494 | nr = i_data[EXT3_DIND_BLOCK]; |
| 2495 | if (nr) { |
| 2496 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
| 2497 | i_data[EXT3_DIND_BLOCK] = 0; |
| 2498 | } |
| 2499 | case EXT3_DIND_BLOCK: |
| 2500 | nr = i_data[EXT3_TIND_BLOCK]; |
| 2501 | if (nr) { |
| 2502 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
| 2503 | i_data[EXT3_TIND_BLOCK] = 0; |
| 2504 | } |
| 2505 | case EXT3_TIND_BLOCK: |
| 2506 | ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2507 | } |
| 2508 | |
| 2509 | ext3_discard_reservation(inode); |
| 2510 | |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 2511 | mutex_unlock(&ei->truncate_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2512 | inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC; |
| 2513 | ext3_mark_inode_dirty(handle, inode); |
| 2514 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2515 | /* |
| 2516 | * In a multi-transaction truncate, we only make the final transaction |
| 2517 | * synchronous |
| 2518 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2519 | if (IS_SYNC(inode)) |
| 2520 | handle->h_sync = 1; |
| 2521 | out_stop: |
| 2522 | /* |
| 2523 | * If this was a simple ftruncate(), and the file will remain alive |
| 2524 | * then we need to clear up the orphan record which we created above. |
| 2525 | * However, if this was a real unlink then we were called by |
| 2526 | * ext3_delete_inode(), and we allow that function to clean up the |
| 2527 | * orphan info for us. |
| 2528 | */ |
| 2529 | if (inode->i_nlink) |
| 2530 | ext3_orphan_del(handle, inode); |
| 2531 | |
| 2532 | ext3_journal_stop(handle); |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2533 | return; |
| 2534 | out_notrans: |
| 2535 | /* |
| 2536 | * Delete the inode from orphan list so that it doesn't stay there |
| 2537 | * forever and trigger assertion on umount. |
| 2538 | */ |
| 2539 | if (inode->i_nlink) |
| 2540 | ext3_orphan_del(NULL, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2541 | } |
| 2542 | |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2543 | static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2544 | unsigned long ino, struct ext3_iloc *iloc) |
| 2545 | { |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2546 | unsigned long block_group; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2547 | unsigned long offset; |
| 2548 | ext3_fsblk_t block; |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2549 | struct ext3_group_desc *gdp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | |
Neil Brown | 2ccb48e | 2006-07-30 03:03:01 -0700 | [diff] [blame] | 2551 | if (!ext3_valid_inum(sb, ino)) { |
| 2552 | /* |
| 2553 | * This error is already checked for in namei.c unless we are |
| 2554 | * looking at an NFS filehandle, in which case no error |
| 2555 | * report is needed |
| 2556 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2557 | return 0; |
| 2558 | } |
Neil Brown | 2ccb48e | 2006-07-30 03:03:01 -0700 | [diff] [blame] | 2559 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2560 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2561 | gdp = ext3_get_group_desc(sb, block_group, NULL); |
| 2562 | if (!gdp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2563 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2564 | /* |
| 2565 | * Figure out the offset within the block group inode table |
| 2566 | */ |
| 2567 | offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) * |
| 2568 | EXT3_INODE_SIZE(sb); |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2569 | block = le32_to_cpu(gdp->bg_inode_table) + |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2570 | (offset >> EXT3_BLOCK_SIZE_BITS(sb)); |
| 2571 | |
| 2572 | iloc->block_group = block_group; |
| 2573 | iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1); |
| 2574 | return block; |
| 2575 | } |
| 2576 | |
| 2577 | /* |
| 2578 | * ext3_get_inode_loc returns with an extra refcount against the inode's |
| 2579 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
| 2580 | * data in memory that is needed to recreate the on-disk version of this |
| 2581 | * inode. |
| 2582 | */ |
| 2583 | static int __ext3_get_inode_loc(struct inode *inode, |
| 2584 | struct ext3_iloc *iloc, int in_mem) |
| 2585 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2586 | ext3_fsblk_t block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2587 | struct buffer_head *bh; |
| 2588 | |
| 2589 | block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc); |
| 2590 | if (!block) |
| 2591 | return -EIO; |
| 2592 | |
| 2593 | bh = sb_getblk(inode->i_sb, block); |
| 2594 | if (!bh) { |
| 2595 | ext3_error (inode->i_sb, "ext3_get_inode_loc", |
| 2596 | "unable to read inode block - " |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2597 | "inode=%lu, block="E3FSBLK, |
| 2598 | inode->i_ino, block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2599 | return -EIO; |
| 2600 | } |
| 2601 | if (!buffer_uptodate(bh)) { |
| 2602 | lock_buffer(bh); |
Hidehiro Kawai | 95450f5 | 2008-07-25 01:46:24 -0700 | [diff] [blame] | 2603 | |
| 2604 | /* |
| 2605 | * If the buffer has the write error flag, we have failed |
| 2606 | * to write out another inode in the same block. In this |
| 2607 | * case, we don't have to read the block because we may |
| 2608 | * read the old inode data successfully. |
| 2609 | */ |
| 2610 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) |
| 2611 | set_buffer_uptodate(bh); |
| 2612 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2613 | if (buffer_uptodate(bh)) { |
| 2614 | /* someone brought it uptodate while we waited */ |
| 2615 | unlock_buffer(bh); |
| 2616 | goto has_buffer; |
| 2617 | } |
| 2618 | |
| 2619 | /* |
| 2620 | * If we have all information of the inode in memory and this |
| 2621 | * is the only valid inode in the block, we need not read the |
| 2622 | * block. |
| 2623 | */ |
| 2624 | if (in_mem) { |
| 2625 | struct buffer_head *bitmap_bh; |
| 2626 | struct ext3_group_desc *desc; |
| 2627 | int inodes_per_buffer; |
| 2628 | int inode_offset, i; |
| 2629 | int block_group; |
| 2630 | int start; |
| 2631 | |
| 2632 | block_group = (inode->i_ino - 1) / |
| 2633 | EXT3_INODES_PER_GROUP(inode->i_sb); |
| 2634 | inodes_per_buffer = bh->b_size / |
| 2635 | EXT3_INODE_SIZE(inode->i_sb); |
| 2636 | inode_offset = ((inode->i_ino - 1) % |
| 2637 | EXT3_INODES_PER_GROUP(inode->i_sb)); |
| 2638 | start = inode_offset & ~(inodes_per_buffer - 1); |
| 2639 | |
| 2640 | /* Is the inode bitmap in cache? */ |
| 2641 | desc = ext3_get_group_desc(inode->i_sb, |
| 2642 | block_group, NULL); |
| 2643 | if (!desc) |
| 2644 | goto make_io; |
| 2645 | |
| 2646 | bitmap_bh = sb_getblk(inode->i_sb, |
| 2647 | le32_to_cpu(desc->bg_inode_bitmap)); |
| 2648 | if (!bitmap_bh) |
| 2649 | goto make_io; |
| 2650 | |
| 2651 | /* |
| 2652 | * If the inode bitmap isn't in cache then the |
| 2653 | * optimisation may end up performing two reads instead |
| 2654 | * of one, so skip it. |
| 2655 | */ |
| 2656 | if (!buffer_uptodate(bitmap_bh)) { |
| 2657 | brelse(bitmap_bh); |
| 2658 | goto make_io; |
| 2659 | } |
| 2660 | for (i = start; i < start + inodes_per_buffer; i++) { |
| 2661 | if (i == inode_offset) |
| 2662 | continue; |
| 2663 | if (ext3_test_bit(i, bitmap_bh->b_data)) |
| 2664 | break; |
| 2665 | } |
| 2666 | brelse(bitmap_bh); |
| 2667 | if (i == start + inodes_per_buffer) { |
| 2668 | /* all other inodes are free, so skip I/O */ |
| 2669 | memset(bh->b_data, 0, bh->b_size); |
| 2670 | set_buffer_uptodate(bh); |
| 2671 | unlock_buffer(bh); |
| 2672 | goto has_buffer; |
| 2673 | } |
| 2674 | } |
| 2675 | |
| 2676 | make_io: |
| 2677 | /* |
| 2678 | * There are other valid inodes in the buffer, this inode |
| 2679 | * has in-inode xattrs, or we don't have this inode in memory. |
| 2680 | * Read the block from disk. |
| 2681 | */ |
| 2682 | get_bh(bh); |
| 2683 | bh->b_end_io = end_buffer_read_sync; |
Jens Axboe | caa38fb | 2006-07-23 01:41:26 +0200 | [diff] [blame] | 2684 | submit_bh(READ_META, bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2685 | wait_on_buffer(bh); |
| 2686 | if (!buffer_uptodate(bh)) { |
| 2687 | ext3_error(inode->i_sb, "ext3_get_inode_loc", |
| 2688 | "unable to read inode block - " |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2689 | "inode=%lu, block="E3FSBLK, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2690 | inode->i_ino, block); |
| 2691 | brelse(bh); |
| 2692 | return -EIO; |
| 2693 | } |
| 2694 | } |
| 2695 | has_buffer: |
| 2696 | iloc->bh = bh; |
| 2697 | return 0; |
| 2698 | } |
| 2699 | |
| 2700 | int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc) |
| 2701 | { |
| 2702 | /* We have all inode data except xattrs in memory here. */ |
| 2703 | return __ext3_get_inode_loc(inode, iloc, |
| 2704 | !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR)); |
| 2705 | } |
| 2706 | |
| 2707 | void ext3_set_inode_flags(struct inode *inode) |
| 2708 | { |
| 2709 | unsigned int flags = EXT3_I(inode)->i_flags; |
| 2710 | |
| 2711 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); |
| 2712 | if (flags & EXT3_SYNC_FL) |
| 2713 | inode->i_flags |= S_SYNC; |
| 2714 | if (flags & EXT3_APPEND_FL) |
| 2715 | inode->i_flags |= S_APPEND; |
| 2716 | if (flags & EXT3_IMMUTABLE_FL) |
| 2717 | inode->i_flags |= S_IMMUTABLE; |
| 2718 | if (flags & EXT3_NOATIME_FL) |
| 2719 | inode->i_flags |= S_NOATIME; |
| 2720 | if (flags & EXT3_DIRSYNC_FL) |
| 2721 | inode->i_flags |= S_DIRSYNC; |
| 2722 | } |
| 2723 | |
Jan Kara | 28be5ab | 2007-05-08 00:30:33 -0700 | [diff] [blame] | 2724 | /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */ |
| 2725 | void ext3_get_inode_flags(struct ext3_inode_info *ei) |
| 2726 | { |
| 2727 | unsigned int flags = ei->vfs_inode.i_flags; |
| 2728 | |
| 2729 | ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL| |
| 2730 | EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL); |
| 2731 | if (flags & S_SYNC) |
| 2732 | ei->i_flags |= EXT3_SYNC_FL; |
| 2733 | if (flags & S_APPEND) |
| 2734 | ei->i_flags |= EXT3_APPEND_FL; |
| 2735 | if (flags & S_IMMUTABLE) |
| 2736 | ei->i_flags |= EXT3_IMMUTABLE_FL; |
| 2737 | if (flags & S_NOATIME) |
| 2738 | ei->i_flags |= EXT3_NOATIME_FL; |
| 2739 | if (flags & S_DIRSYNC) |
| 2740 | ei->i_flags |= EXT3_DIRSYNC_FL; |
| 2741 | } |
| 2742 | |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2743 | struct inode *ext3_iget(struct super_block *sb, unsigned long ino) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2744 | { |
| 2745 | struct ext3_iloc iloc; |
| 2746 | struct ext3_inode *raw_inode; |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2747 | struct ext3_inode_info *ei; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2748 | struct buffer_head *bh; |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2749 | struct inode *inode; |
| 2750 | long ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2751 | int block; |
| 2752 | |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2753 | inode = iget_locked(sb, ino); |
| 2754 | if (!inode) |
| 2755 | return ERR_PTR(-ENOMEM); |
| 2756 | if (!(inode->i_state & I_NEW)) |
| 2757 | return inode; |
| 2758 | |
| 2759 | ei = EXT3_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2760 | ei->i_block_alloc_info = NULL; |
| 2761 | |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2762 | ret = __ext3_get_inode_loc(inode, &iloc, 0); |
| 2763 | if (ret < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2764 | goto bad_inode; |
| 2765 | bh = iloc.bh; |
| 2766 | raw_inode = ext3_raw_inode(&iloc); |
| 2767 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
| 2768 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 2769 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
| 2770 | if(!(test_opt (inode->i_sb, NO_UID32))) { |
| 2771 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 2772 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
| 2773 | } |
| 2774 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); |
| 2775 | inode->i_size = le32_to_cpu(raw_inode->i_size); |
Markus Rechberger | 4d7bf11 | 2007-05-08 00:23:39 -0700 | [diff] [blame] | 2776 | inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime); |
| 2777 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime); |
| 2778 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2779 | inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0; |
| 2780 | |
| 2781 | ei->i_state = 0; |
| 2782 | ei->i_dir_start_lookup = 0; |
| 2783 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 2784 | /* We now have enough fields to check if the inode was active or not. |
| 2785 | * This is needed because nfsd might try to access dead inodes |
| 2786 | * the test is that same one that e2fsck uses |
| 2787 | * NeilBrown 1999oct15 |
| 2788 | */ |
| 2789 | if (inode->i_nlink == 0) { |
| 2790 | if (inode->i_mode == 0 || |
| 2791 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) { |
| 2792 | /* this inode is deleted */ |
| 2793 | brelse (bh); |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2794 | ret = -ESTALE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2795 | goto bad_inode; |
| 2796 | } |
| 2797 | /* The only unlinked inodes we let through here have |
| 2798 | * valid i_mode and are being read by the orphan |
| 2799 | * recovery code: that's fine, we're about to complete |
| 2800 | * the process of deleting those. */ |
| 2801 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2802 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); |
| 2803 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
| 2804 | #ifdef EXT3_FRAGMENTS |
| 2805 | ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); |
| 2806 | ei->i_frag_no = raw_inode->i_frag; |
| 2807 | ei->i_frag_size = raw_inode->i_fsize; |
| 2808 | #endif |
| 2809 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); |
| 2810 | if (!S_ISREG(inode->i_mode)) { |
| 2811 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); |
| 2812 | } else { |
| 2813 | inode->i_size |= |
| 2814 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; |
| 2815 | } |
| 2816 | ei->i_disksize = inode->i_size; |
| 2817 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 2818 | ei->i_block_group = iloc.block_group; |
| 2819 | /* |
| 2820 | * NOTE! The in-memory inode i_data array is in little-endian order |
| 2821 | * even on big-endian machines: we do NOT byteswap the block numbers! |
| 2822 | */ |
| 2823 | for (block = 0; block < EXT3_N_BLOCKS; block++) |
| 2824 | ei->i_data[block] = raw_inode->i_block[block]; |
| 2825 | INIT_LIST_HEAD(&ei->i_orphan); |
| 2826 | |
| 2827 | if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 && |
| 2828 | EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) { |
| 2829 | /* |
| 2830 | * When mke2fs creates big inodes it does not zero out |
| 2831 | * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE, |
| 2832 | * so ignore those first few inodes. |
| 2833 | */ |
| 2834 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
| 2835 | if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
Kirill Korotaev | e4a10a3 | 2007-06-23 17:16:48 -0700 | [diff] [blame] | 2836 | EXT3_INODE_SIZE(inode->i_sb)) { |
| 2837 | brelse (bh); |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2838 | ret = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2839 | goto bad_inode; |
Kirill Korotaev | e4a10a3 | 2007-06-23 17:16:48 -0700 | [diff] [blame] | 2840 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2841 | if (ei->i_extra_isize == 0) { |
| 2842 | /* The extra space is currently unused. Use it. */ |
| 2843 | ei->i_extra_isize = sizeof(struct ext3_inode) - |
| 2844 | EXT3_GOOD_OLD_INODE_SIZE; |
| 2845 | } else { |
| 2846 | __le32 *magic = (void *)raw_inode + |
| 2847 | EXT3_GOOD_OLD_INODE_SIZE + |
| 2848 | ei->i_extra_isize; |
| 2849 | if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC)) |
| 2850 | ei->i_state |= EXT3_STATE_XATTR; |
| 2851 | } |
| 2852 | } else |
| 2853 | ei->i_extra_isize = 0; |
| 2854 | |
| 2855 | if (S_ISREG(inode->i_mode)) { |
| 2856 | inode->i_op = &ext3_file_inode_operations; |
| 2857 | inode->i_fop = &ext3_file_operations; |
| 2858 | ext3_set_aops(inode); |
| 2859 | } else if (S_ISDIR(inode->i_mode)) { |
| 2860 | inode->i_op = &ext3_dir_inode_operations; |
| 2861 | inode->i_fop = &ext3_dir_operations; |
| 2862 | } else if (S_ISLNK(inode->i_mode)) { |
Duane Griffin | b5ed311 | 2008-12-19 20:47:14 +0000 | [diff] [blame] | 2863 | if (ext3_inode_is_fast_symlink(inode)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2864 | inode->i_op = &ext3_fast_symlink_inode_operations; |
Duane Griffin | b5ed311 | 2008-12-19 20:47:14 +0000 | [diff] [blame] | 2865 | nd_terminate_link(ei->i_data, inode->i_size, |
| 2866 | sizeof(ei->i_data) - 1); |
| 2867 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2868 | inode->i_op = &ext3_symlink_inode_operations; |
| 2869 | ext3_set_aops(inode); |
| 2870 | } |
| 2871 | } else { |
| 2872 | inode->i_op = &ext3_special_inode_operations; |
| 2873 | if (raw_inode->i_block[0]) |
| 2874 | init_special_inode(inode, inode->i_mode, |
| 2875 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2876 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2877 | init_special_inode(inode, inode->i_mode, |
| 2878 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
| 2879 | } |
| 2880 | brelse (iloc.bh); |
| 2881 | ext3_set_inode_flags(inode); |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2882 | unlock_new_inode(inode); |
| 2883 | return inode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2884 | |
| 2885 | bad_inode: |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2886 | iget_failed(inode); |
| 2887 | return ERR_PTR(ret); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2888 | } |
| 2889 | |
| 2890 | /* |
| 2891 | * Post the struct inode info into an on-disk inode location in the |
| 2892 | * buffer-cache. This gobbles the caller's reference to the |
| 2893 | * buffer_head in the inode location struct. |
| 2894 | * |
| 2895 | * The caller must have write access to iloc->bh. |
| 2896 | */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2897 | static int ext3_do_update_inode(handle_t *handle, |
| 2898 | struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2899 | struct ext3_iloc *iloc) |
| 2900 | { |
| 2901 | struct ext3_inode *raw_inode = ext3_raw_inode(iloc); |
| 2902 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 2903 | struct buffer_head *bh = iloc->bh; |
| 2904 | int err = 0, rc, block; |
| 2905 | |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 2906 | again: |
| 2907 | /* we can't allow multiple procs in here at once, its a bit racey */ |
| 2908 | lock_buffer(bh); |
| 2909 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2910 | /* For fields not not tracking in the in-memory inode, |
| 2911 | * initialise them to zero for new inodes. */ |
| 2912 | if (ei->i_state & EXT3_STATE_NEW) |
| 2913 | memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size); |
| 2914 | |
Jan Kara | 28be5ab | 2007-05-08 00:30:33 -0700 | [diff] [blame] | 2915 | ext3_get_inode_flags(ei); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2916 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
| 2917 | if(!(test_opt(inode->i_sb, NO_UID32))) { |
| 2918 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
| 2919 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); |
| 2920 | /* |
| 2921 | * Fix up interoperability with old kernels. Otherwise, old inodes get |
| 2922 | * re-used with the upper 16 bits of the uid/gid intact |
| 2923 | */ |
| 2924 | if(!ei->i_dtime) { |
| 2925 | raw_inode->i_uid_high = |
| 2926 | cpu_to_le16(high_16_bits(inode->i_uid)); |
| 2927 | raw_inode->i_gid_high = |
| 2928 | cpu_to_le16(high_16_bits(inode->i_gid)); |
| 2929 | } else { |
| 2930 | raw_inode->i_uid_high = 0; |
| 2931 | raw_inode->i_gid_high = 0; |
| 2932 | } |
| 2933 | } else { |
| 2934 | raw_inode->i_uid_low = |
| 2935 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); |
| 2936 | raw_inode->i_gid_low = |
| 2937 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); |
| 2938 | raw_inode->i_uid_high = 0; |
| 2939 | raw_inode->i_gid_high = 0; |
| 2940 | } |
| 2941 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
| 2942 | raw_inode->i_size = cpu_to_le32(ei->i_disksize); |
| 2943 | raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec); |
| 2944 | raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); |
| 2945 | raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); |
| 2946 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); |
| 2947 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
| 2948 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); |
| 2949 | #ifdef EXT3_FRAGMENTS |
| 2950 | raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); |
| 2951 | raw_inode->i_frag = ei->i_frag_no; |
| 2952 | raw_inode->i_fsize = ei->i_frag_size; |
| 2953 | #endif |
| 2954 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); |
| 2955 | if (!S_ISREG(inode->i_mode)) { |
| 2956 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); |
| 2957 | } else { |
| 2958 | raw_inode->i_size_high = |
| 2959 | cpu_to_le32(ei->i_disksize >> 32); |
| 2960 | if (ei->i_disksize > 0x7fffffffULL) { |
| 2961 | struct super_block *sb = inode->i_sb; |
| 2962 | if (!EXT3_HAS_RO_COMPAT_FEATURE(sb, |
| 2963 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE) || |
| 2964 | EXT3_SB(sb)->s_es->s_rev_level == |
| 2965 | cpu_to_le32(EXT3_GOOD_OLD_REV)) { |
| 2966 | /* If this is the first large file |
| 2967 | * created, add a flag to the superblock. |
| 2968 | */ |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 2969 | unlock_buffer(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2970 | err = ext3_journal_get_write_access(handle, |
| 2971 | EXT3_SB(sb)->s_sbh); |
| 2972 | if (err) |
| 2973 | goto out_brelse; |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 2974 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2975 | ext3_update_dynamic_rev(sb); |
| 2976 | EXT3_SET_RO_COMPAT_FEATURE(sb, |
| 2977 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2978 | handle->h_sync = 1; |
| 2979 | err = ext3_journal_dirty_metadata(handle, |
| 2980 | EXT3_SB(sb)->s_sbh); |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 2981 | /* get our lock and start over */ |
| 2982 | goto again; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2983 | } |
| 2984 | } |
| 2985 | } |
| 2986 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 2987 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 2988 | if (old_valid_dev(inode->i_rdev)) { |
| 2989 | raw_inode->i_block[0] = |
| 2990 | cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 2991 | raw_inode->i_block[1] = 0; |
| 2992 | } else { |
| 2993 | raw_inode->i_block[0] = 0; |
| 2994 | raw_inode->i_block[1] = |
| 2995 | cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 2996 | raw_inode->i_block[2] = 0; |
| 2997 | } |
| 2998 | } else for (block = 0; block < EXT3_N_BLOCKS; block++) |
| 2999 | raw_inode->i_block[block] = ei->i_data[block]; |
| 3000 | |
Andreas Gruenbacher | ff87b37 | 2005-07-07 17:57:00 -0700 | [diff] [blame] | 3001 | if (ei->i_extra_isize) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3002 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
| 3003 | |
| 3004 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 3005 | unlock_buffer(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3006 | rc = ext3_journal_dirty_metadata(handle, bh); |
| 3007 | if (!err) |
| 3008 | err = rc; |
| 3009 | ei->i_state &= ~EXT3_STATE_NEW; |
| 3010 | |
| 3011 | out_brelse: |
| 3012 | brelse (bh); |
| 3013 | ext3_std_error(inode->i_sb, err); |
| 3014 | return err; |
| 3015 | } |
| 3016 | |
| 3017 | /* |
| 3018 | * ext3_write_inode() |
| 3019 | * |
| 3020 | * We are called from a few places: |
| 3021 | * |
| 3022 | * - Within generic_file_write() for O_SYNC files. |
| 3023 | * Here, there will be no transaction running. We wait for any running |
| 3024 | * trasnaction to commit. |
| 3025 | * |
| 3026 | * - Within sys_sync(), kupdate and such. |
| 3027 | * We wait on commit, if tol to. |
| 3028 | * |
| 3029 | * - Within prune_icache() (PF_MEMALLOC == true) |
| 3030 | * Here we simply return. We can't afford to block kswapd on the |
| 3031 | * journal commit. |
| 3032 | * |
| 3033 | * In all cases it is actually safe for us to return without doing anything, |
| 3034 | * because the inode has been copied into a raw inode buffer in |
| 3035 | * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
| 3036 | * knfsd. |
| 3037 | * |
| 3038 | * Note that we are absolutely dependent upon all inode dirtiers doing the |
| 3039 | * right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 3040 | * which we are interested. |
| 3041 | * |
| 3042 | * It would be a bug for them to not do this. The code: |
| 3043 | * |
| 3044 | * mark_inode_dirty(inode) |
| 3045 | * stuff(); |
| 3046 | * inode->i_size = expr; |
| 3047 | * |
| 3048 | * is in error because a kswapd-driven write_inode() could occur while |
| 3049 | * `stuff()' is running, and the new i_size will be lost. Plus the inode |
| 3050 | * will no longer be on the superblock's dirty inode list. |
| 3051 | */ |
| 3052 | int ext3_write_inode(struct inode *inode, int wait) |
| 3053 | { |
| 3054 | if (current->flags & PF_MEMALLOC) |
| 3055 | return 0; |
| 3056 | |
| 3057 | if (ext3_journal_current_handle()) { |
Jose R. Santos | 9ad163a | 2007-10-18 23:39:23 -0700 | [diff] [blame] | 3058 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3059 | dump_stack(); |
| 3060 | return -EIO; |
| 3061 | } |
| 3062 | |
| 3063 | if (!wait) |
| 3064 | return 0; |
| 3065 | |
| 3066 | return ext3_force_commit(inode->i_sb); |
| 3067 | } |
| 3068 | |
| 3069 | /* |
| 3070 | * ext3_setattr() |
| 3071 | * |
| 3072 | * Called from notify_change. |
| 3073 | * |
| 3074 | * We want to trap VFS attempts to truncate the file as soon as |
| 3075 | * possible. In particular, we want to make sure that when the VFS |
| 3076 | * shrinks i_size, we put the inode on the orphan list and modify |
| 3077 | * i_disksize immediately, so that during the subsequent flushing of |
| 3078 | * dirty pages and freeing of disk blocks, we can guarantee that any |
| 3079 | * commit will leave the blocks being flushed in an unused state on |
| 3080 | * disk. (On recovery, the inode will get truncated and the blocks will |
| 3081 | * be freed, so we have a strong guarantee that no future commit will |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3082 | * leave these blocks visible to the user.) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3083 | * |
| 3084 | * Called with inode->sem down. |
| 3085 | */ |
| 3086 | int ext3_setattr(struct dentry *dentry, struct iattr *attr) |
| 3087 | { |
| 3088 | struct inode *inode = dentry->d_inode; |
| 3089 | int error, rc = 0; |
| 3090 | const unsigned int ia_valid = attr->ia_valid; |
| 3091 | |
| 3092 | error = inode_change_ok(inode, attr); |
| 3093 | if (error) |
| 3094 | return error; |
| 3095 | |
| 3096 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
| 3097 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { |
| 3098 | handle_t *handle; |
| 3099 | |
| 3100 | /* (user+group)*(old+new) structure, inode write (sb, |
| 3101 | * inode block, ? - but truncate inode update has it) */ |
Jan Kara | 1f54587 | 2005-06-23 22:01:04 -0700 | [diff] [blame] | 3102 | handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
| 3103 | EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3104 | if (IS_ERR(handle)) { |
| 3105 | error = PTR_ERR(handle); |
| 3106 | goto err_out; |
| 3107 | } |
Jan Kara | 81a0522 | 2009-01-26 16:58:01 +0100 | [diff] [blame] | 3108 | error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3109 | if (error) { |
| 3110 | ext3_journal_stop(handle); |
| 3111 | return error; |
| 3112 | } |
| 3113 | /* Update corresponding info in inode so that everything is in |
| 3114 | * one transaction */ |
| 3115 | if (attr->ia_valid & ATTR_UID) |
| 3116 | inode->i_uid = attr->ia_uid; |
| 3117 | if (attr->ia_valid & ATTR_GID) |
| 3118 | inode->i_gid = attr->ia_gid; |
| 3119 | error = ext3_mark_inode_dirty(handle, inode); |
| 3120 | ext3_journal_stop(handle); |
| 3121 | } |
| 3122 | |
| 3123 | if (S_ISREG(inode->i_mode) && |
| 3124 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { |
| 3125 | handle_t *handle; |
| 3126 | |
| 3127 | handle = ext3_journal_start(inode, 3); |
| 3128 | if (IS_ERR(handle)) { |
| 3129 | error = PTR_ERR(handle); |
| 3130 | goto err_out; |
| 3131 | } |
| 3132 | |
| 3133 | error = ext3_orphan_add(handle, inode); |
| 3134 | EXT3_I(inode)->i_disksize = attr->ia_size; |
| 3135 | rc = ext3_mark_inode_dirty(handle, inode); |
| 3136 | if (!error) |
| 3137 | error = rc; |
| 3138 | ext3_journal_stop(handle); |
| 3139 | } |
| 3140 | |
| 3141 | rc = inode_setattr(inode, attr); |
| 3142 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3143 | if (!rc && (ia_valid & ATTR_MODE)) |
| 3144 | rc = ext3_acl_chmod(inode); |
| 3145 | |
| 3146 | err_out: |
| 3147 | ext3_std_error(inode->i_sb, error); |
| 3148 | if (!error) |
| 3149 | error = rc; |
| 3150 | return error; |
| 3151 | } |
| 3152 | |
| 3153 | |
| 3154 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3155 | * How many blocks doth make a writepage()? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3156 | * |
| 3157 | * With N blocks per page, it may be: |
| 3158 | * N data blocks |
| 3159 | * 2 indirect block |
| 3160 | * 2 dindirect |
| 3161 | * 1 tindirect |
| 3162 | * N+5 bitmap blocks (from the above) |
| 3163 | * N+5 group descriptor summary blocks |
| 3164 | * 1 inode block |
| 3165 | * 1 superblock. |
| 3166 | * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files |
| 3167 | * |
| 3168 | * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS |
| 3169 | * |
| 3170 | * With ordered or writeback data it's the same, less the N data blocks. |
| 3171 | * |
| 3172 | * If the inode's direct blocks can hold an integral number of pages then a |
| 3173 | * page cannot straddle two indirect blocks, and we can only touch one indirect |
| 3174 | * and dindirect block, and the "5" above becomes "3". |
| 3175 | * |
| 3176 | * This still overestimates under most circumstances. If we were to pass the |
| 3177 | * start and end offsets in here as well we could do block_to_path() on each |
| 3178 | * block and work out the exact number of indirects which are touched. Pah. |
| 3179 | */ |
| 3180 | |
| 3181 | static int ext3_writepage_trans_blocks(struct inode *inode) |
| 3182 | { |
| 3183 | int bpp = ext3_journal_blocks_per_page(inode); |
| 3184 | int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3; |
| 3185 | int ret; |
| 3186 | |
| 3187 | if (ext3_should_journal_data(inode)) |
| 3188 | ret = 3 * (bpp + indirects) + 2; |
| 3189 | else |
| 3190 | ret = 2 * (bpp + indirects) + 2; |
| 3191 | |
| 3192 | #ifdef CONFIG_QUOTA |
Jan Kara | 81a0522 | 2009-01-26 16:58:01 +0100 | [diff] [blame] | 3193 | /* We know that structure was already allocated during vfs_dq_init so |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3194 | * we will be updating only the data blocks + inodes */ |
Jan Kara | 1f54587 | 2005-06-23 22:01:04 -0700 | [diff] [blame] | 3195 | ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3196 | #endif |
| 3197 | |
| 3198 | return ret; |
| 3199 | } |
| 3200 | |
| 3201 | /* |
| 3202 | * The caller must have previously called ext3_reserve_inode_write(). |
| 3203 | * Give this, we know that the caller already has write access to iloc->bh. |
| 3204 | */ |
| 3205 | int ext3_mark_iloc_dirty(handle_t *handle, |
| 3206 | struct inode *inode, struct ext3_iloc *iloc) |
| 3207 | { |
| 3208 | int err = 0; |
| 3209 | |
| 3210 | /* the do_update_inode consumes one bh->b_count */ |
| 3211 | get_bh(iloc->bh); |
| 3212 | |
| 3213 | /* ext3_do_update_inode() does journal_dirty_metadata */ |
| 3214 | err = ext3_do_update_inode(handle, inode, iloc); |
| 3215 | put_bh(iloc->bh); |
| 3216 | return err; |
| 3217 | } |
| 3218 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3219 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3220 | * On success, We end up with an outstanding reference count against |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3221 | * iloc->bh. This _must_ be cleaned up later. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3222 | */ |
| 3223 | |
| 3224 | int |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3225 | ext3_reserve_inode_write(handle_t *handle, struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3226 | struct ext3_iloc *iloc) |
| 3227 | { |
| 3228 | int err = 0; |
| 3229 | if (handle) { |
| 3230 | err = ext3_get_inode_loc(inode, iloc); |
| 3231 | if (!err) { |
| 3232 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 3233 | err = ext3_journal_get_write_access(handle, iloc->bh); |
| 3234 | if (err) { |
| 3235 | brelse(iloc->bh); |
| 3236 | iloc->bh = NULL; |
| 3237 | } |
| 3238 | } |
| 3239 | } |
| 3240 | ext3_std_error(inode->i_sb, err); |
| 3241 | return err; |
| 3242 | } |
| 3243 | |
| 3244 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3245 | * What we do here is to mark the in-core inode as clean with respect to inode |
| 3246 | * dirtiness (it may still be data-dirty). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3247 | * This means that the in-core inode may be reaped by prune_icache |
| 3248 | * without having to perform any I/O. This is a very good thing, |
| 3249 | * because *any* task may call prune_icache - even ones which |
| 3250 | * have a transaction open against a different journal. |
| 3251 | * |
| 3252 | * Is this cheating? Not really. Sure, we haven't written the |
| 3253 | * inode out, but prune_icache isn't a user-visible syncing function. |
| 3254 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 3255 | * we start and wait on commits. |
| 3256 | * |
| 3257 | * Is this efficient/effective? Well, we're being nice to the system |
| 3258 | * by cleaning up our inodes proactively so they can be reaped |
| 3259 | * without I/O. But we are potentially leaving up to five seconds' |
| 3260 | * worth of inodes floating about which prune_icache wants us to |
| 3261 | * write out. One way to fix that would be to get prune_icache() |
| 3262 | * to do a write_super() to free up some memory. It has the desired |
| 3263 | * effect. |
| 3264 | */ |
| 3265 | int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode) |
| 3266 | { |
| 3267 | struct ext3_iloc iloc; |
| 3268 | int err; |
| 3269 | |
| 3270 | might_sleep(); |
| 3271 | err = ext3_reserve_inode_write(handle, inode, &iloc); |
| 3272 | if (!err) |
| 3273 | err = ext3_mark_iloc_dirty(handle, inode, &iloc); |
| 3274 | return err; |
| 3275 | } |
| 3276 | |
| 3277 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3278 | * ext3_dirty_inode() is called from __mark_inode_dirty() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3279 | * |
| 3280 | * We're really interested in the case where a file is being extended. |
| 3281 | * i_size has been changed by generic_commit_write() and we thus need |
| 3282 | * to include the updated inode in the current transaction. |
| 3283 | * |
Jan Kara | 81a0522 | 2009-01-26 16:58:01 +0100 | [diff] [blame] | 3284 | * Also, vfs_dq_alloc_space() will always dirty the inode when blocks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3285 | * are allocated to the file. |
| 3286 | * |
| 3287 | * If the inode is marked synchronous, we don't honour that here - doing |
| 3288 | * so would cause a commit on atime updates, which we don't bother doing. |
| 3289 | * We handle synchronous inodes at the highest possible level. |
| 3290 | */ |
| 3291 | void ext3_dirty_inode(struct inode *inode) |
| 3292 | { |
| 3293 | handle_t *current_handle = ext3_journal_current_handle(); |
| 3294 | handle_t *handle; |
| 3295 | |
| 3296 | handle = ext3_journal_start(inode, 2); |
| 3297 | if (IS_ERR(handle)) |
| 3298 | goto out; |
| 3299 | if (current_handle && |
| 3300 | current_handle->h_transaction != handle->h_transaction) { |
| 3301 | /* This task has a transaction open against a different fs */ |
| 3302 | printk(KERN_EMERG "%s: transactions do not match!\n", |
Harvey Harrison | e05b6b5 | 2008-04-28 02:16:15 -0700 | [diff] [blame] | 3303 | __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3304 | } else { |
| 3305 | jbd_debug(5, "marking dirty. outer handle=%p\n", |
| 3306 | current_handle); |
| 3307 | ext3_mark_inode_dirty(handle, inode); |
| 3308 | } |
| 3309 | ext3_journal_stop(handle); |
| 3310 | out: |
| 3311 | return; |
| 3312 | } |
| 3313 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3314 | #if 0 |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3315 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3316 | * Bind an inode's backing buffer_head into this transaction, to prevent |
| 3317 | * it from being flushed to disk early. Unlike |
| 3318 | * ext3_reserve_inode_write, this leaves behind no bh reference and |
| 3319 | * returns no iloc structure, so the caller needs to repeat the iloc |
| 3320 | * lookup to mark the inode dirty later. |
| 3321 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3322 | static int ext3_pin_inode(handle_t *handle, struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3323 | { |
| 3324 | struct ext3_iloc iloc; |
| 3325 | |
| 3326 | int err = 0; |
| 3327 | if (handle) { |
| 3328 | err = ext3_get_inode_loc(inode, &iloc); |
| 3329 | if (!err) { |
| 3330 | BUFFER_TRACE(iloc.bh, "get_write_access"); |
| 3331 | err = journal_get_write_access(handle, iloc.bh); |
| 3332 | if (!err) |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3333 | err = ext3_journal_dirty_metadata(handle, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3334 | iloc.bh); |
| 3335 | brelse(iloc.bh); |
| 3336 | } |
| 3337 | } |
| 3338 | ext3_std_error(inode->i_sb, err); |
| 3339 | return err; |
| 3340 | } |
| 3341 | #endif |
| 3342 | |
| 3343 | int ext3_change_inode_journal_flag(struct inode *inode, int val) |
| 3344 | { |
| 3345 | journal_t *journal; |
| 3346 | handle_t *handle; |
| 3347 | int err; |
| 3348 | |
| 3349 | /* |
| 3350 | * We have to be very careful here: changing a data block's |
| 3351 | * journaling status dynamically is dangerous. If we write a |
| 3352 | * data block to the journal, change the status and then delete |
| 3353 | * that block, we risk forgetting to revoke the old log record |
| 3354 | * from the journal and so a subsequent replay can corrupt data. |
| 3355 | * So, first we make sure that the journal is empty and that |
| 3356 | * nobody is changing anything. |
| 3357 | */ |
| 3358 | |
| 3359 | journal = EXT3_JOURNAL(inode); |
Dave Hansen | e3a68e3 | 2007-07-15 23:41:14 -0700 | [diff] [blame] | 3360 | if (is_journal_aborted(journal)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3361 | return -EROFS; |
| 3362 | |
| 3363 | journal_lock_updates(journal); |
| 3364 | journal_flush(journal); |
| 3365 | |
| 3366 | /* |
| 3367 | * OK, there are no updates running now, and all cached data is |
| 3368 | * synced to disk. We are now in a completely consistent state |
| 3369 | * which doesn't have anything in the journal, and we know that |
| 3370 | * no filesystem updates are running, so it is safe to modify |
| 3371 | * the inode's in-core data-journaling state flag now. |
| 3372 | */ |
| 3373 | |
| 3374 | if (val) |
| 3375 | EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL; |
| 3376 | else |
| 3377 | EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL; |
| 3378 | ext3_set_aops(inode); |
| 3379 | |
| 3380 | journal_unlock_updates(journal); |
| 3381 | |
| 3382 | /* Finally we can mark the inode as dirty. */ |
| 3383 | |
| 3384 | handle = ext3_journal_start(inode, 1); |
| 3385 | if (IS_ERR(handle)) |
| 3386 | return PTR_ERR(handle); |
| 3387 | |
| 3388 | err = ext3_mark_inode_dirty(handle, inode); |
| 3389 | handle->h_sync = 1; |
| 3390 | ext3_journal_stop(handle); |
| 3391 | ext3_std_error(inode->i_sb, err); |
| 3392 | |
| 3393 | return err; |
| 3394 | } |