Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2 | * linux/fs/ext4/inode.c |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 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 |
| 16 | * (sct@redhat.com), 1993, 1998 |
| 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 |
| 20 | * (jj@sunsite.ms.mff.cuni.cz) |
| 21 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 22 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 23 | */ |
| 24 | |
| 25 | #include <linux/module.h> |
| 26 | #include <linux/fs.h> |
| 27 | #include <linux/time.h> |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 28 | #include <linux/jbd2.h> |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 29 | #include <linux/highuid.h> |
| 30 | #include <linux/pagemap.h> |
| 31 | #include <linux/quotaops.h> |
| 32 | #include <linux/string.h> |
| 33 | #include <linux/buffer_head.h> |
| 34 | #include <linux/writeback.h> |
| 35 | #include <linux/mpage.h> |
| 36 | #include <linux/uio.h> |
| 37 | #include <linux/bio.h> |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 38 | #include "ext4_jbd2.h" |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 39 | #include "xattr.h" |
| 40 | #include "acl.h" |
| 41 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 42 | /* |
| 43 | * Test whether an inode is a fast symlink. |
| 44 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 45 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 46 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 47 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 48 | (inode->i_sb->s_blocksize >> 9) : 0; |
| 49 | |
| 50 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
| 51 | } |
| 52 | |
| 53 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 54 | * The ext4 forget function must perform a revoke if we are freeing data |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 55 | * which has been journaled. Metadata (eg. indirect blocks) must be |
| 56 | * revoked in all cases. |
| 57 | * |
| 58 | * "bh" may be NULL: a metadata block may have been freed from memory |
| 59 | * but there may still be a record of it in the journal, and that record |
| 60 | * still needs to be revoked. |
| 61 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 62 | int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, |
| 63 | struct buffer_head *bh, ext4_fsblk_t blocknr) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 64 | { |
| 65 | int err; |
| 66 | |
| 67 | might_sleep(); |
| 68 | |
| 69 | BUFFER_TRACE(bh, "enter"); |
| 70 | |
| 71 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " |
| 72 | "data mode %lx\n", |
| 73 | bh, is_metadata, inode->i_mode, |
| 74 | test_opt(inode->i_sb, DATA_FLAGS)); |
| 75 | |
| 76 | /* Never use the revoke function if we are doing full data |
| 77 | * journaling: there is no need to, and a V1 superblock won't |
| 78 | * support it. Otherwise, only skip the revoke on un-journaled |
| 79 | * data blocks. */ |
| 80 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 81 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
| 82 | (!is_metadata && !ext4_should_journal_data(inode))) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 83 | if (bh) { |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 84 | BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 85 | return ext4_journal_forget(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 86 | } |
| 87 | return 0; |
| 88 | } |
| 89 | |
| 90 | /* |
| 91 | * data!=journal && (is_metadata || should_journal_data(inode)) |
| 92 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 93 | BUFFER_TRACE(bh, "call ext4_journal_revoke"); |
| 94 | err = ext4_journal_revoke(handle, blocknr, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 95 | if (err) |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 96 | ext4_abort(inode->i_sb, __func__, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 97 | "error %d when attempting revoke", err); |
| 98 | BUFFER_TRACE(bh, "exit"); |
| 99 | return err; |
| 100 | } |
| 101 | |
| 102 | /* |
| 103 | * Work out how many blocks we need to proceed with the next chunk of a |
| 104 | * truncate transaction. |
| 105 | */ |
| 106 | static unsigned long blocks_for_truncate(struct inode *inode) |
| 107 | { |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 108 | ext4_lblk_t needed; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 109 | |
| 110 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); |
| 111 | |
| 112 | /* Give ourselves just enough room to cope with inodes in which |
| 113 | * i_blocks is corrupt: we've seen disk corruptions in the past |
| 114 | * which resulted in random data in an inode which looked enough |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 115 | * like a regular file for ext4 to try to delete it. Things |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 116 | * will go a bit crazy if that happens, but at least we should |
| 117 | * try not to panic the whole kernel. */ |
| 118 | if (needed < 2) |
| 119 | needed = 2; |
| 120 | |
| 121 | /* But we need to bound the transaction so we don't overflow the |
| 122 | * journal. */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 123 | if (needed > EXT4_MAX_TRANS_DATA) |
| 124 | needed = EXT4_MAX_TRANS_DATA; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 125 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 126 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 127 | } |
| 128 | |
| 129 | /* |
| 130 | * Truncate transactions can be complex and absolutely huge. So we need to |
| 131 | * be able to restart the transaction at a conventient checkpoint to make |
| 132 | * sure we don't overflow the journal. |
| 133 | * |
| 134 | * start_transaction gets us a new handle for a truncate transaction, |
| 135 | * and extend_transaction tries to extend the existing one a bit. If |
| 136 | * extend fails, we need to propagate the failure up and restart the |
| 137 | * transaction in the top-level truncate loop. --sct |
| 138 | */ |
| 139 | static handle_t *start_transaction(struct inode *inode) |
| 140 | { |
| 141 | handle_t *result; |
| 142 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 143 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 144 | if (!IS_ERR(result)) |
| 145 | return result; |
| 146 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 147 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 148 | return result; |
| 149 | } |
| 150 | |
| 151 | /* |
| 152 | * Try to extend this transaction for the purposes of truncation. |
| 153 | * |
| 154 | * Returns 0 if we managed to create more room. If we can't create more |
| 155 | * room, and the transaction must be restarted we return 1. |
| 156 | */ |
| 157 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) |
| 158 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 159 | if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 160 | return 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 161 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 162 | return 0; |
| 163 | return 1; |
| 164 | } |
| 165 | |
| 166 | /* |
| 167 | * Restart the transaction associated with *handle. This does a commit, |
| 168 | * so before we call here everything must be consistently dirtied against |
| 169 | * this transaction. |
| 170 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 171 | static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 172 | { |
| 173 | jbd_debug(2, "restarting handle %p\n", handle); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 174 | return ext4_journal_restart(handle, blocks_for_truncate(inode)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 175 | } |
| 176 | |
| 177 | /* |
| 178 | * Called at the last iput() if i_nlink is zero. |
| 179 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 180 | void ext4_delete_inode (struct inode * inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 181 | { |
| 182 | handle_t *handle; |
| 183 | |
| 184 | truncate_inode_pages(&inode->i_data, 0); |
| 185 | |
| 186 | if (is_bad_inode(inode)) |
| 187 | goto no_delete; |
| 188 | |
| 189 | handle = start_transaction(inode); |
| 190 | if (IS_ERR(handle)) { |
| 191 | /* |
| 192 | * If we're going to skip the normal cleanup, we still need to |
| 193 | * make sure that the in-core orphan linked list is properly |
| 194 | * cleaned up. |
| 195 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 196 | ext4_orphan_del(NULL, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 197 | goto no_delete; |
| 198 | } |
| 199 | |
| 200 | if (IS_SYNC(inode)) |
| 201 | handle->h_sync = 1; |
| 202 | inode->i_size = 0; |
| 203 | if (inode->i_blocks) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 204 | ext4_truncate(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 205 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 206 | * Kill off the orphan record which ext4_truncate created. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 207 | * AKPM: I think this can be inside the above `if'. |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 208 | * Note that ext4_orphan_del() has to be able to cope with the |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 209 | * deletion of a non-existent orphan - this is because we don't |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 210 | * know if ext4_truncate() actually created an orphan record. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 211 | * (Well, we could do this if we need to, but heck - it works) |
| 212 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 213 | ext4_orphan_del(handle, inode); |
| 214 | EXT4_I(inode)->i_dtime = get_seconds(); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 215 | |
| 216 | /* |
| 217 | * One subtle ordering requirement: if anything has gone wrong |
| 218 | * (transaction abort, IO errors, whatever), then we can still |
| 219 | * do these next steps (the fs will already have been marked as |
| 220 | * having errors), but we can't free the inode if the mark_dirty |
| 221 | * fails. |
| 222 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 223 | if (ext4_mark_inode_dirty(handle, inode)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 224 | /* If that failed, just do the required in-core inode clear. */ |
| 225 | clear_inode(inode); |
| 226 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 227 | ext4_free_inode(handle, inode); |
| 228 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 229 | return; |
| 230 | no_delete: |
| 231 | clear_inode(inode); /* We must guarantee clearing of inode... */ |
| 232 | } |
| 233 | |
| 234 | typedef struct { |
| 235 | __le32 *p; |
| 236 | __le32 key; |
| 237 | struct buffer_head *bh; |
| 238 | } Indirect; |
| 239 | |
| 240 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) |
| 241 | { |
| 242 | p->key = *(p->p = v); |
| 243 | p->bh = bh; |
| 244 | } |
| 245 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 246 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 247 | * ext4_block_to_path - parse the block number into array of offsets |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 248 | * @inode: inode in question (we are only interested in its superblock) |
| 249 | * @i_block: block number to be parsed |
| 250 | * @offsets: array to store the offsets in |
Dave Kleikamp | 8c55e20 | 2007-05-24 13:04:54 -0400 | [diff] [blame] | 251 | * @boundary: set this non-zero if the referred-to block is likely to be |
| 252 | * followed (on disk) by an indirect block. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 253 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 254 | * To store the locations of file's data ext4 uses a data structure common |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 255 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
| 256 | * data blocks at leaves and indirect blocks in intermediate nodes. |
| 257 | * This function translates the block number into path in that tree - |
| 258 | * return value is the path length and @offsets[n] is the offset of |
| 259 | * pointer to (n+1)th node in the nth one. If @block is out of range |
| 260 | * (negative or too large) warning is printed and zero returned. |
| 261 | * |
| 262 | * Note: function doesn't find node addresses, so no IO is needed. All |
| 263 | * we need to know is the capacity of indirect blocks (taken from the |
| 264 | * inode->i_sb). |
| 265 | */ |
| 266 | |
| 267 | /* |
| 268 | * Portability note: the last comparison (check that we fit into triple |
| 269 | * indirect block) is spelled differently, because otherwise on an |
| 270 | * architecture with 32-bit longs and 8Kb pages we might get into trouble |
| 271 | * if our filesystem had 8Kb blocks. We might use long long, but that would |
| 272 | * kill us on x86. Oh, well, at least the sign propagation does not matter - |
| 273 | * i_block would have to be negative in the very beginning, so we would not |
| 274 | * get there at all. |
| 275 | */ |
| 276 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 277 | static int ext4_block_to_path(struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 278 | ext4_lblk_t i_block, |
| 279 | ext4_lblk_t offsets[4], int *boundary) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 280 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 281 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
| 282 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); |
| 283 | const long direct_blocks = EXT4_NDIR_BLOCKS, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 284 | indirect_blocks = ptrs, |
| 285 | double_blocks = (1 << (ptrs_bits * 2)); |
| 286 | int n = 0; |
| 287 | int final = 0; |
| 288 | |
| 289 | if (i_block < 0) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 290 | ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0"); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 291 | } else if (i_block < direct_blocks) { |
| 292 | offsets[n++] = i_block; |
| 293 | final = direct_blocks; |
| 294 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 295 | offsets[n++] = EXT4_IND_BLOCK; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 296 | offsets[n++] = i_block; |
| 297 | final = ptrs; |
| 298 | } else if ((i_block -= indirect_blocks) < double_blocks) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 299 | offsets[n++] = EXT4_DIND_BLOCK; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 300 | offsets[n++] = i_block >> ptrs_bits; |
| 301 | offsets[n++] = i_block & (ptrs - 1); |
| 302 | final = ptrs; |
| 303 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 304 | offsets[n++] = EXT4_TIND_BLOCK; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 305 | offsets[n++] = i_block >> (ptrs_bits * 2); |
| 306 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); |
| 307 | offsets[n++] = i_block & (ptrs - 1); |
| 308 | final = ptrs; |
| 309 | } else { |
Eric Sandeen | e2b4657 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 310 | ext4_warning(inode->i_sb, "ext4_block_to_path", |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 311 | "block %lu > max", |
Eric Sandeen | e2b4657 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 312 | i_block + direct_blocks + |
| 313 | indirect_blocks + double_blocks); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 314 | } |
| 315 | if (boundary) |
| 316 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
| 317 | return n; |
| 318 | } |
| 319 | |
| 320 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 321 | * ext4_get_branch - read the chain of indirect blocks leading to data |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 322 | * @inode: inode in question |
| 323 | * @depth: depth of the chain (1 - direct pointer, etc.) |
| 324 | * @offsets: offsets of pointers in inode/indirect blocks |
| 325 | * @chain: place to store the result |
| 326 | * @err: here we store the error value |
| 327 | * |
| 328 | * Function fills the array of triples <key, p, bh> and returns %NULL |
| 329 | * if everything went OK or the pointer to the last filled triple |
| 330 | * (incomplete one) otherwise. Upon the return chain[i].key contains |
| 331 | * the number of (i+1)-th block in the chain (as it is stored in memory, |
| 332 | * i.e. little-endian 32-bit), chain[i].p contains the address of that |
| 333 | * number (it points into struct inode for i==0 and into the bh->b_data |
| 334 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect |
| 335 | * block for i>0 and NULL for i==0. In other words, it holds the block |
| 336 | * numbers of the chain, addresses they were taken from (and where we can |
| 337 | * verify that chain did not change) and buffer_heads hosting these |
| 338 | * numbers. |
| 339 | * |
| 340 | * Function stops when it stumbles upon zero pointer (absent block) |
| 341 | * (pointer to last triple returned, *@err == 0) |
| 342 | * or when it gets an IO error reading an indirect block |
| 343 | * (ditto, *@err == -EIO) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 344 | * or when it reads all @depth-1 indirect blocks successfully and finds |
| 345 | * the whole chain, all way to the data (returns %NULL, *err == 0). |
Aneesh Kumar K.V | c278bfe | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 346 | * |
| 347 | * Need to be called with |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 348 | * down_read(&EXT4_I(inode)->i_data_sem) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 349 | */ |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 350 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
| 351 | ext4_lblk_t *offsets, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 352 | Indirect chain[4], int *err) |
| 353 | { |
| 354 | struct super_block *sb = inode->i_sb; |
| 355 | Indirect *p = chain; |
| 356 | struct buffer_head *bh; |
| 357 | |
| 358 | *err = 0; |
| 359 | /* i_data is not going away, no lock needed */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 360 | add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 361 | if (!p->key) |
| 362 | goto no_block; |
| 363 | while (--depth) { |
| 364 | bh = sb_bread(sb, le32_to_cpu(p->key)); |
| 365 | if (!bh) |
| 366 | goto failure; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 367 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); |
| 368 | /* Reader: end */ |
| 369 | if (!p->key) |
| 370 | goto no_block; |
| 371 | } |
| 372 | return NULL; |
| 373 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 374 | failure: |
| 375 | *err = -EIO; |
| 376 | no_block: |
| 377 | return p; |
| 378 | } |
| 379 | |
| 380 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 381 | * ext4_find_near - find a place for allocation with sufficient locality |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 382 | * @inode: owner |
| 383 | * @ind: descriptor of indirect block. |
| 384 | * |
Benoit Boissinot | 1cc8dcf5 | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 385 | * This function returns the preferred place for block allocation. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 386 | * It is used when heuristic for sequential allocation fails. |
| 387 | * Rules are: |
| 388 | * + if there is a block to the left of our position - allocate near it. |
| 389 | * + if pointer will live in indirect block - allocate near that block. |
| 390 | * + if pointer will live in inode - allocate in the same |
| 391 | * cylinder group. |
| 392 | * |
| 393 | * In the latter case we colour the starting block by the callers PID to |
| 394 | * prevent it from clashing with concurrent allocations for a different inode |
| 395 | * in the same block group. The PID is used here so that functionally related |
| 396 | * files will be close-by on-disk. |
| 397 | * |
| 398 | * Caller must make sure that @ind is valid and will stay that way. |
| 399 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 400 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 401 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 402 | struct ext4_inode_info *ei = EXT4_I(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 403 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; |
| 404 | __le32 *p; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 405 | ext4_fsblk_t bg_start; |
Valerie Clement | 74d3487 | 2008-02-15 13:43:07 -0500 | [diff] [blame] | 406 | ext4_fsblk_t last_block; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 407 | ext4_grpblk_t colour; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 408 | |
| 409 | /* Try to find previous block */ |
| 410 | for (p = ind->p - 1; p >= start; p--) { |
| 411 | if (*p) |
| 412 | return le32_to_cpu(*p); |
| 413 | } |
| 414 | |
| 415 | /* No such thing, so let's try location of indirect block */ |
| 416 | if (ind->bh) |
| 417 | return ind->bh->b_blocknr; |
| 418 | |
| 419 | /* |
| 420 | * It is going to be referred to from the inode itself? OK, just put it |
| 421 | * into the same cylinder group then. |
| 422 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 423 | bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group); |
Valerie Clement | 74d3487 | 2008-02-15 13:43:07 -0500 | [diff] [blame] | 424 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
| 425 | |
| 426 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
| 427 | colour = (current->pid % 16) * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 428 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
Valerie Clement | 74d3487 | 2008-02-15 13:43:07 -0500 | [diff] [blame] | 429 | else |
| 430 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 431 | return bg_start + colour; |
| 432 | } |
| 433 | |
| 434 | /** |
Benoit Boissinot | 1cc8dcf5 | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 435 | * ext4_find_goal - find a preferred place for allocation. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 436 | * @inode: owner |
| 437 | * @block: block we want |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 438 | * @partial: pointer to the last triple within a chain |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 439 | * |
Benoit Boissinot | 1cc8dcf5 | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 440 | * Normally this function find the preferred place for block allocation, |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 441 | * returns it. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 442 | */ |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 443 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 444 | Indirect *partial) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 445 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 446 | struct ext4_block_alloc_info *block_i; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 447 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 448 | block_i = EXT4_I(inode)->i_block_alloc_info; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 449 | |
| 450 | /* |
| 451 | * try the heuristic for sequential allocation, |
| 452 | * failing that at least try to get decent locality. |
| 453 | */ |
| 454 | if (block_i && (block == block_i->last_alloc_logical_block + 1) |
| 455 | && (block_i->last_alloc_physical_block != 0)) { |
| 456 | return block_i->last_alloc_physical_block + 1; |
| 457 | } |
| 458 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 459 | return ext4_find_near(inode, partial); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 460 | } |
| 461 | |
| 462 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 463 | * ext4_blks_to_allocate: Look up the block map and count the number |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 464 | * of direct blocks need to be allocated for the given branch. |
| 465 | * |
| 466 | * @branch: chain of indirect blocks |
| 467 | * @k: number of blocks need for indirect blocks |
| 468 | * @blks: number of data blocks to be mapped. |
| 469 | * @blocks_to_boundary: the offset in the indirect block |
| 470 | * |
| 471 | * return the total number of blocks to be allocate, including the |
| 472 | * direct and indirect blocks. |
| 473 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 474 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 475 | int blocks_to_boundary) |
| 476 | { |
| 477 | unsigned long count = 0; |
| 478 | |
| 479 | /* |
| 480 | * Simple case, [t,d]Indirect block(s) has not allocated yet |
| 481 | * then it's clear blocks on that path have not allocated |
| 482 | */ |
| 483 | if (k > 0) { |
| 484 | /* right now we don't handle cross boundary allocation */ |
| 485 | if (blks < blocks_to_boundary + 1) |
| 486 | count += blks; |
| 487 | else |
| 488 | count += blocks_to_boundary + 1; |
| 489 | return count; |
| 490 | } |
| 491 | |
| 492 | count++; |
| 493 | while (count < blks && count <= blocks_to_boundary && |
| 494 | le32_to_cpu(*(branch[0].p + count)) == 0) { |
| 495 | count++; |
| 496 | } |
| 497 | return count; |
| 498 | } |
| 499 | |
| 500 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 501 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 502 | * @indirect_blks: the number of blocks need to allocate for indirect |
| 503 | * blocks |
| 504 | * |
| 505 | * @new_blocks: on return it will store the new block numbers for |
| 506 | * the indirect blocks(if needed) and the first direct block, |
| 507 | * @blks: on return it will store the total number of allocated |
| 508 | * direct blocks |
| 509 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 510 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
| 511 | ext4_fsblk_t goal, int indirect_blks, int blks, |
| 512 | ext4_fsblk_t new_blocks[4], int *err) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 513 | { |
| 514 | int target, i; |
| 515 | unsigned long count = 0; |
| 516 | int index = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 517 | ext4_fsblk_t current_block = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 518 | int ret = 0; |
| 519 | |
| 520 | /* |
| 521 | * Here we try to allocate the requested multiple blocks at once, |
| 522 | * on a best-effort basis. |
| 523 | * To build a branch, we should allocate blocks for |
| 524 | * the indirect blocks(if not allocated yet), and at least |
| 525 | * the first direct block of this branch. That's the |
| 526 | * minimum number of blocks need to allocate(required) |
| 527 | */ |
| 528 | target = blks + indirect_blks; |
| 529 | |
| 530 | while (1) { |
| 531 | count = target; |
| 532 | /* allocating blocks for indirect blocks and direct blocks */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 533 | current_block = ext4_new_blocks(handle,inode,goal,&count,err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 534 | if (*err) |
| 535 | goto failed_out; |
| 536 | |
| 537 | target -= count; |
| 538 | /* allocate blocks for indirect blocks */ |
| 539 | while (index < indirect_blks && count) { |
| 540 | new_blocks[index++] = current_block++; |
| 541 | count--; |
| 542 | } |
| 543 | |
| 544 | if (count > 0) |
| 545 | break; |
| 546 | } |
| 547 | |
| 548 | /* save the new block number for the first direct block */ |
| 549 | new_blocks[index] = current_block; |
| 550 | |
| 551 | /* total number of blocks allocated for direct blocks */ |
| 552 | ret = count; |
| 553 | *err = 0; |
| 554 | return ret; |
| 555 | failed_out: |
| 556 | for (i = 0; i <index; i++) |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 557 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 558 | return ret; |
| 559 | } |
| 560 | |
| 561 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 562 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 563 | * @inode: owner |
| 564 | * @indirect_blks: number of allocated indirect blocks |
| 565 | * @blks: number of allocated direct blocks |
| 566 | * @offsets: offsets (in the blocks) to store the pointers to next. |
| 567 | * @branch: place to store the chain in. |
| 568 | * |
| 569 | * This function allocates blocks, zeroes out all but the last one, |
| 570 | * links them into chain and (if we are synchronous) writes them to disk. |
| 571 | * In other words, it prepares a branch that can be spliced onto the |
| 572 | * inode. It stores the information about that chain in the branch[], in |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 573 | * the same format as ext4_get_branch() would do. We are calling it after |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 574 | * we had read the existing part of chain and partial points to the last |
| 575 | * triple of that (one with zero ->key). Upon the exit we have the same |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 576 | * picture as after the successful ext4_get_block(), except that in one |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 577 | * place chain is disconnected - *branch->p is still zero (we did not |
| 578 | * set the last link), but branch->key contains the number that should |
| 579 | * be placed into *branch->p to fill that gap. |
| 580 | * |
| 581 | * If allocation fails we free all blocks we've allocated (and forget |
| 582 | * their buffer_heads) and return the error value the from failed |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 583 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 584 | * as described above and return 0. |
| 585 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 586 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
| 587 | int indirect_blks, int *blks, ext4_fsblk_t goal, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 588 | ext4_lblk_t *offsets, Indirect *branch) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 589 | { |
| 590 | int blocksize = inode->i_sb->s_blocksize; |
| 591 | int i, n = 0; |
| 592 | int err = 0; |
| 593 | struct buffer_head *bh; |
| 594 | int num; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 595 | ext4_fsblk_t new_blocks[4]; |
| 596 | ext4_fsblk_t current_block; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 597 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 598 | num = ext4_alloc_blocks(handle, inode, goal, indirect_blks, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 599 | *blks, new_blocks, &err); |
| 600 | if (err) |
| 601 | return err; |
| 602 | |
| 603 | branch[0].key = cpu_to_le32(new_blocks[0]); |
| 604 | /* |
| 605 | * metadata blocks and data blocks are allocated. |
| 606 | */ |
| 607 | for (n = 1; n <= indirect_blks; n++) { |
| 608 | /* |
| 609 | * Get buffer_head for parent block, zero it out |
| 610 | * and set the pointer to new one, then send |
| 611 | * parent to disk. |
| 612 | */ |
| 613 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); |
| 614 | branch[n].bh = bh; |
| 615 | lock_buffer(bh); |
| 616 | BUFFER_TRACE(bh, "call get_create_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 617 | err = ext4_journal_get_create_access(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 618 | if (err) { |
| 619 | unlock_buffer(bh); |
| 620 | brelse(bh); |
| 621 | goto failed; |
| 622 | } |
| 623 | |
| 624 | memset(bh->b_data, 0, blocksize); |
| 625 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; |
| 626 | branch[n].key = cpu_to_le32(new_blocks[n]); |
| 627 | *branch[n].p = branch[n].key; |
| 628 | if ( n == indirect_blks) { |
| 629 | current_block = new_blocks[n]; |
| 630 | /* |
| 631 | * End of chain, update the last new metablock of |
| 632 | * the chain to point to the new allocated |
| 633 | * data blocks numbers |
| 634 | */ |
| 635 | for (i=1; i < num; i++) |
| 636 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
| 637 | } |
| 638 | BUFFER_TRACE(bh, "marking uptodate"); |
| 639 | set_buffer_uptodate(bh); |
| 640 | unlock_buffer(bh); |
| 641 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 642 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
| 643 | err = ext4_journal_dirty_metadata(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 644 | if (err) |
| 645 | goto failed; |
| 646 | } |
| 647 | *blks = num; |
| 648 | return err; |
| 649 | failed: |
| 650 | /* Allocation failed, free what we already allocated */ |
| 651 | for (i = 1; i <= n ; i++) { |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 652 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 653 | ext4_journal_forget(handle, branch[i].bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 654 | } |
| 655 | for (i = 0; i <indirect_blks; i++) |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 656 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 657 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 658 | ext4_free_blocks(handle, inode, new_blocks[i], num, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 659 | |
| 660 | return err; |
| 661 | } |
| 662 | |
| 663 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 664 | * ext4_splice_branch - splice the allocated branch onto inode. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 665 | * @inode: owner |
| 666 | * @block: (logical) number of block we are adding |
| 667 | * @chain: chain of indirect blocks (with a missing link - see |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 668 | * ext4_alloc_branch) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 669 | * @where: location of missing link |
| 670 | * @num: number of indirect blocks we are adding |
| 671 | * @blks: number of direct blocks we are adding |
| 672 | * |
| 673 | * This function fills the missing link and does all housekeeping needed in |
| 674 | * inode (->i_blocks, etc.). In case of success we end up with the full |
| 675 | * chain to new block and return 0. |
| 676 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 677 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 678 | ext4_lblk_t block, Indirect *where, int num, int blks) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 679 | { |
| 680 | int i; |
| 681 | int err = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 682 | struct ext4_block_alloc_info *block_i; |
| 683 | ext4_fsblk_t current_block; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 684 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 685 | block_i = EXT4_I(inode)->i_block_alloc_info; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 686 | /* |
| 687 | * If we're splicing into a [td]indirect block (as opposed to the |
| 688 | * inode) then we need to get write access to the [td]indirect block |
| 689 | * before the splice. |
| 690 | */ |
| 691 | if (where->bh) { |
| 692 | BUFFER_TRACE(where->bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 693 | err = ext4_journal_get_write_access(handle, where->bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 694 | if (err) |
| 695 | goto err_out; |
| 696 | } |
| 697 | /* That's it */ |
| 698 | |
| 699 | *where->p = where->key; |
| 700 | |
| 701 | /* |
| 702 | * Update the host buffer_head or inode to point to more just allocated |
| 703 | * direct blocks blocks |
| 704 | */ |
| 705 | if (num == 0 && blks > 1) { |
| 706 | current_block = le32_to_cpu(where->key) + 1; |
| 707 | for (i = 1; i < blks; i++) |
| 708 | *(where->p + i ) = cpu_to_le32(current_block++); |
| 709 | } |
| 710 | |
| 711 | /* |
| 712 | * update the most recently allocated logical & physical block |
| 713 | * in i_block_alloc_info, to assist find the proper goal block for next |
| 714 | * allocation |
| 715 | */ |
| 716 | if (block_i) { |
| 717 | block_i->last_alloc_logical_block = block + blks - 1; |
| 718 | block_i->last_alloc_physical_block = |
| 719 | le32_to_cpu(where[num].key) + blks - 1; |
| 720 | } |
| 721 | |
| 722 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
| 723 | |
Kalpak Shah | ef7f383 | 2007-07-18 09:15:20 -0400 | [diff] [blame] | 724 | inode->i_ctime = ext4_current_time(inode); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 725 | ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 726 | |
| 727 | /* had we spliced it onto indirect block? */ |
| 728 | if (where->bh) { |
| 729 | /* |
| 730 | * If we spliced it onto an indirect block, we haven't |
| 731 | * altered the inode. Note however that if it is being spliced |
| 732 | * onto an indirect block at the very end of the file (the |
| 733 | * file is growing) then we *will* alter the inode to reflect |
| 734 | * the new i_size. But that is not done here - it is done in |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 735 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 736 | */ |
| 737 | jbd_debug(5, "splicing indirect only\n"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 738 | BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata"); |
| 739 | err = ext4_journal_dirty_metadata(handle, where->bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 740 | if (err) |
| 741 | goto err_out; |
| 742 | } else { |
| 743 | /* |
| 744 | * OK, we spliced it into the inode itself on a direct block. |
| 745 | * Inode was dirtied above. |
| 746 | */ |
| 747 | jbd_debug(5, "splicing direct\n"); |
| 748 | } |
| 749 | return err; |
| 750 | |
| 751 | err_out: |
| 752 | for (i = 1; i <= num; i++) { |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 753 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 754 | ext4_journal_forget(handle, where[i].bh); |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 755 | ext4_free_blocks(handle, inode, |
| 756 | le32_to_cpu(where[i-1].key), 1, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 757 | } |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 758 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 759 | |
| 760 | return err; |
| 761 | } |
| 762 | |
| 763 | /* |
| 764 | * Allocation strategy is simple: if we have to allocate something, we will |
| 765 | * have to go the whole way to leaf. So let's do it before attaching anything |
| 766 | * to tree, set linkage between the newborn blocks, write them if sync is |
| 767 | * required, recheck the path, free and repeat if check fails, otherwise |
| 768 | * set the last missing link (that will protect us from any truncate-generated |
| 769 | * removals - all blocks on the path are immune now) and possibly force the |
| 770 | * write on the parent block. |
| 771 | * That has a nice additional property: no special recovery from the failed |
| 772 | * allocations is needed - we simply release blocks and do not touch anything |
| 773 | * reachable from inode. |
| 774 | * |
| 775 | * `handle' can be NULL if create == 0. |
| 776 | * |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 777 | * return > 0, # of blocks mapped or allocated. |
| 778 | * return = 0, if plain lookup failed. |
| 779 | * return < 0, error case. |
Aneesh Kumar K.V | c278bfe | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 780 | * |
| 781 | * |
| 782 | * Need to be called with |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 783 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
| 784 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 785 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 786 | int ext4_get_blocks_handle(handle_t *handle, struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 787 | ext4_lblk_t iblock, unsigned long maxblocks, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 788 | struct buffer_head *bh_result, |
| 789 | int create, int extend_disksize) |
| 790 | { |
| 791 | int err = -EIO; |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 792 | ext4_lblk_t offsets[4]; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 793 | Indirect chain[4]; |
| 794 | Indirect *partial; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 795 | ext4_fsblk_t goal; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 796 | int indirect_blks; |
| 797 | int blocks_to_boundary = 0; |
| 798 | int depth; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 799 | struct ext4_inode_info *ei = EXT4_I(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 800 | int count = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 801 | ext4_fsblk_t first_block = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 802 | |
| 803 | |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 804 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 805 | J_ASSERT(handle != NULL || create == 0); |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 806 | depth = ext4_block_to_path(inode, iblock, offsets, |
| 807 | &blocks_to_boundary); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 808 | |
| 809 | if (depth == 0) |
| 810 | goto out; |
| 811 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 812 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 813 | |
| 814 | /* Simplest case - block found, no allocation needed */ |
| 815 | if (!partial) { |
| 816 | first_block = le32_to_cpu(chain[depth - 1].key); |
| 817 | clear_buffer_new(bh_result); |
| 818 | count++; |
| 819 | /*map more blocks*/ |
| 820 | while (count < maxblocks && count <= blocks_to_boundary) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 821 | ext4_fsblk_t blk; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 822 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 823 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
| 824 | |
| 825 | if (blk == first_block + count) |
| 826 | count++; |
| 827 | else |
| 828 | break; |
| 829 | } |
Aneesh Kumar K.V | c278bfe | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 830 | goto got_it; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 831 | } |
| 832 | |
| 833 | /* Next simple case - plain lookup or failed read of indirect block */ |
| 834 | if (!create || err == -EIO) |
| 835 | goto cleanup; |
| 836 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 837 | /* |
| 838 | * Okay, we need to do block allocation. Lazily initialize the block |
| 839 | * allocation info here if necessary |
| 840 | */ |
| 841 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 842 | ext4_init_block_alloc_info(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 843 | |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 844 | goal = ext4_find_goal(inode, iblock, partial); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 845 | |
| 846 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
| 847 | indirect_blks = (chain + depth) - partial - 1; |
| 848 | |
| 849 | /* |
| 850 | * Next look up the indirect map to count the totoal number of |
| 851 | * direct blocks to allocate for this branch. |
| 852 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 853 | count = ext4_blks_to_allocate(partial, indirect_blks, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 854 | maxblocks, blocks_to_boundary); |
| 855 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 856 | * Block out ext4_truncate while we alter the tree |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 857 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 858 | err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 859 | offsets + (partial - chain), partial); |
| 860 | |
| 861 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 862 | * The ext4_splice_branch call will free and forget any buffers |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 863 | * on the new chain if there is a failure, but that risks using |
| 864 | * up transaction credits, especially for bitmaps where the |
| 865 | * credits cannot be returned. Can we handle this somehow? We |
| 866 | * may need to return -EAGAIN upwards in the worst case. --sct |
| 867 | */ |
| 868 | if (!err) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 869 | err = ext4_splice_branch(handle, inode, iblock, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 870 | partial, indirect_blks, count); |
| 871 | /* |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 872 | * i_disksize growing is protected by i_data_sem. Don't forget to |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 873 | * protect it if you're about to implement concurrent |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 874 | * ext4_get_block() -bzzz |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 875 | */ |
| 876 | if (!err && extend_disksize && inode->i_size > ei->i_disksize) |
| 877 | ei->i_disksize = inode->i_size; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 878 | if (err) |
| 879 | goto cleanup; |
| 880 | |
| 881 | set_buffer_new(bh_result); |
| 882 | got_it: |
| 883 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); |
| 884 | if (count > blocks_to_boundary) |
| 885 | set_buffer_boundary(bh_result); |
| 886 | err = count; |
| 887 | /* Clean up and exit */ |
| 888 | partial = chain + depth - 1; /* the whole chain */ |
| 889 | cleanup: |
| 890 | while (partial > chain) { |
| 891 | BUFFER_TRACE(partial->bh, "call brelse"); |
| 892 | brelse(partial->bh); |
| 893 | partial--; |
| 894 | } |
| 895 | BUFFER_TRACE(bh_result, "returned"); |
| 896 | out: |
| 897 | return err; |
| 898 | } |
| 899 | |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 900 | /* Maximum number of blocks we map for direct IO at once. */ |
| 901 | #define DIO_MAX_BLOCKS 4096 |
| 902 | /* |
| 903 | * Number of credits we need for writing DIO_MAX_BLOCKS: |
| 904 | * We need sb + group descriptor + bitmap + inode -> 4 |
| 905 | * For B blocks with A block pointers per block we need: |
| 906 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). |
| 907 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. |
| 908 | */ |
| 909 | #define DIO_CREDITS 25 |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 910 | |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 911 | |
| 912 | /* |
| 913 | * |
| 914 | * |
| 915 | * ext4_ext4 get_block() wrapper function |
| 916 | * It will do a look up first, and returns if the blocks already mapped. |
| 917 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
| 918 | * and store the allocated blocks in the result buffer head and mark it |
| 919 | * mapped. |
| 920 | * |
| 921 | * If file type is extents based, it will call ext4_ext_get_blocks(), |
| 922 | * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping |
| 923 | * based files |
| 924 | * |
| 925 | * On success, it returns the number of blocks being mapped or allocate. |
| 926 | * if create==0 and the blocks are pre-allocated and uninitialized block, |
| 927 | * the result buffer head is unmapped. If the create ==1, it will make sure |
| 928 | * the buffer head is mapped. |
| 929 | * |
| 930 | * It returns 0 if plain look up failed (blocks have not been allocated), in |
| 931 | * that casem, buffer head is unmapped |
| 932 | * |
| 933 | * It returns the error in case of allocation failure. |
| 934 | */ |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 935 | int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block, |
| 936 | unsigned long max_blocks, struct buffer_head *bh, |
| 937 | int create, int extend_disksize) |
| 938 | { |
| 939 | int retval; |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 940 | |
| 941 | clear_buffer_mapped(bh); |
| 942 | |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 943 | /* |
| 944 | * Try to see if we can get the block without requesting |
| 945 | * for new file system block. |
| 946 | */ |
| 947 | down_read((&EXT4_I(inode)->i_data_sem)); |
| 948 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
| 949 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, |
| 950 | bh, 0, 0); |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 951 | } else { |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 952 | retval = ext4_get_blocks_handle(handle, |
| 953 | inode, block, max_blocks, bh, 0, 0); |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 954 | } |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 955 | up_read((&EXT4_I(inode)->i_data_sem)); |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 956 | |
| 957 | /* If it is only a block(s) look up */ |
| 958 | if (!create) |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 959 | return retval; |
| 960 | |
| 961 | /* |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 962 | * Returns if the blocks have already allocated |
| 963 | * |
| 964 | * Note that if blocks have been preallocated |
| 965 | * ext4_ext_get_block() returns th create = 0 |
| 966 | * with buffer head unmapped. |
| 967 | */ |
| 968 | if (retval > 0 && buffer_mapped(bh)) |
| 969 | return retval; |
| 970 | |
| 971 | /* |
| 972 | * New blocks allocate and/or writing to uninitialized extent |
| 973 | * will possibly result in updating i_data, so we take |
| 974 | * the write lock of i_data_sem, and call get_blocks() |
| 975 | * with create == 1 flag. |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 976 | */ |
| 977 | down_write((&EXT4_I(inode)->i_data_sem)); |
| 978 | /* |
| 979 | * We need to check for EXT4 here because migrate |
| 980 | * could have changed the inode type in between |
| 981 | */ |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 982 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
| 983 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, |
| 984 | bh, create, extend_disksize); |
| 985 | } else { |
| 986 | retval = ext4_get_blocks_handle(handle, inode, block, |
| 987 | max_blocks, bh, create, extend_disksize); |
Aneesh Kumar K.V | 267e4db | 2008-04-29 08:11:12 -0400 | [diff] [blame] | 988 | |
| 989 | if (retval > 0 && buffer_new(bh)) { |
| 990 | /* |
| 991 | * We allocated new blocks which will result in |
| 992 | * i_data's format changing. Force the migrate |
| 993 | * to fail by clearing migrate flags |
| 994 | */ |
| 995 | EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags & |
| 996 | ~EXT4_EXT_MIGRATE; |
| 997 | } |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 998 | } |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 999 | up_write((&EXT4_I(inode)->i_data_sem)); |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 1000 | return retval; |
| 1001 | } |
| 1002 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1003 | static int ext4_get_block(struct inode *inode, sector_t iblock, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1004 | struct buffer_head *bh_result, int create) |
| 1005 | { |
Dmitriy Monakhov | 3e4fdaf | 2007-02-10 01:46:35 -0800 | [diff] [blame] | 1006 | handle_t *handle = ext4_journal_current_handle(); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1007 | int ret = 0, started = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1008 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
| 1009 | |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1010 | if (create && !handle) { |
| 1011 | /* Direct IO write... */ |
| 1012 | if (max_blocks > DIO_MAX_BLOCKS) |
| 1013 | max_blocks = DIO_MAX_BLOCKS; |
| 1014 | handle = ext4_journal_start(inode, DIO_CREDITS + |
| 1015 | 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb)); |
| 1016 | if (IS_ERR(handle)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1017 | ret = PTR_ERR(handle); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1018 | goto out; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1019 | } |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1020 | started = 1; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1021 | } |
| 1022 | |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1023 | ret = ext4_get_blocks_wrap(handle, inode, iblock, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1024 | max_blocks, bh_result, create, 0); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1025 | if (ret > 0) { |
| 1026 | bh_result->b_size = (ret << inode->i_blkbits); |
| 1027 | ret = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1028 | } |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1029 | if (started) |
| 1030 | ext4_journal_stop(handle); |
| 1031 | out: |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1032 | return ret; |
| 1033 | } |
| 1034 | |
| 1035 | /* |
| 1036 | * `handle' can be NULL if create is zero |
| 1037 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1038 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1039 | ext4_lblk_t block, int create, int *errp) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1040 | { |
| 1041 | struct buffer_head dummy; |
| 1042 | int fatal = 0, err; |
| 1043 | |
| 1044 | J_ASSERT(handle != NULL || create == 0); |
| 1045 | |
| 1046 | dummy.b_state = 0; |
| 1047 | dummy.b_blocknr = -1000; |
| 1048 | buffer_trace_init(&dummy.b_history); |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 1049 | err = ext4_get_blocks_wrap(handle, inode, block, 1, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1050 | &dummy, create, 1); |
| 1051 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1052 | * ext4_get_blocks_handle() returns number of blocks |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1053 | * mapped. 0 in case of a HOLE. |
| 1054 | */ |
| 1055 | if (err > 0) { |
| 1056 | if (err > 1) |
| 1057 | WARN_ON(1); |
| 1058 | err = 0; |
| 1059 | } |
| 1060 | *errp = err; |
| 1061 | if (!err && buffer_mapped(&dummy)) { |
| 1062 | struct buffer_head *bh; |
| 1063 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); |
| 1064 | if (!bh) { |
| 1065 | *errp = -EIO; |
| 1066 | goto err; |
| 1067 | } |
| 1068 | if (buffer_new(&dummy)) { |
| 1069 | J_ASSERT(create != 0); |
Aneesh Kumar K.V | ac39849 | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 1070 | J_ASSERT(handle != NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1071 | |
| 1072 | /* |
| 1073 | * Now that we do not always journal data, we should |
| 1074 | * keep in mind whether this should always journal the |
| 1075 | * new buffer as metadata. For now, regular file |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1076 | * writes use ext4_get_block instead, so it's not a |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1077 | * problem. |
| 1078 | */ |
| 1079 | lock_buffer(bh); |
| 1080 | BUFFER_TRACE(bh, "call get_create_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1081 | fatal = ext4_journal_get_create_access(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1082 | if (!fatal && !buffer_uptodate(bh)) { |
| 1083 | memset(bh->b_data,0,inode->i_sb->s_blocksize); |
| 1084 | set_buffer_uptodate(bh); |
| 1085 | } |
| 1086 | unlock_buffer(bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1087 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
| 1088 | err = ext4_journal_dirty_metadata(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1089 | if (!fatal) |
| 1090 | fatal = err; |
| 1091 | } else { |
| 1092 | BUFFER_TRACE(bh, "not a new buffer"); |
| 1093 | } |
| 1094 | if (fatal) { |
| 1095 | *errp = fatal; |
| 1096 | brelse(bh); |
| 1097 | bh = NULL; |
| 1098 | } |
| 1099 | return bh; |
| 1100 | } |
| 1101 | err: |
| 1102 | return NULL; |
| 1103 | } |
| 1104 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1105 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1106 | ext4_lblk_t block, int create, int *err) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1107 | { |
| 1108 | struct buffer_head * bh; |
| 1109 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1110 | bh = ext4_getblk(handle, inode, block, create, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1111 | if (!bh) |
| 1112 | return bh; |
| 1113 | if (buffer_uptodate(bh)) |
| 1114 | return bh; |
| 1115 | ll_rw_block(READ_META, 1, &bh); |
| 1116 | wait_on_buffer(bh); |
| 1117 | if (buffer_uptodate(bh)) |
| 1118 | return bh; |
| 1119 | put_bh(bh); |
| 1120 | *err = -EIO; |
| 1121 | return NULL; |
| 1122 | } |
| 1123 | |
| 1124 | static int walk_page_buffers( handle_t *handle, |
| 1125 | struct buffer_head *head, |
| 1126 | unsigned from, |
| 1127 | unsigned to, |
| 1128 | int *partial, |
| 1129 | int (*fn)( handle_t *handle, |
| 1130 | struct buffer_head *bh)) |
| 1131 | { |
| 1132 | struct buffer_head *bh; |
| 1133 | unsigned block_start, block_end; |
| 1134 | unsigned blocksize = head->b_size; |
| 1135 | int err, ret = 0; |
| 1136 | struct buffer_head *next; |
| 1137 | |
| 1138 | for ( bh = head, block_start = 0; |
| 1139 | ret == 0 && (bh != head || !block_start); |
| 1140 | block_start = block_end, bh = next) |
| 1141 | { |
| 1142 | next = bh->b_this_page; |
| 1143 | block_end = block_start + blocksize; |
| 1144 | if (block_end <= from || block_start >= to) { |
| 1145 | if (partial && !buffer_uptodate(bh)) |
| 1146 | *partial = 1; |
| 1147 | continue; |
| 1148 | } |
| 1149 | err = (*fn)(handle, bh); |
| 1150 | if (!ret) |
| 1151 | ret = err; |
| 1152 | } |
| 1153 | return ret; |
| 1154 | } |
| 1155 | |
| 1156 | /* |
| 1157 | * To preserve ordering, it is essential that the hole instantiation and |
| 1158 | * the data write be encapsulated in a single transaction. We cannot |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1159 | * close off a transaction and start a new one between the ext4_get_block() |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1160 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1161 | * prepare_write() is the right place. |
| 1162 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1163 | * Also, this function can nest inside ext4_writepage() -> |
| 1164 | * block_write_full_page(). In that case, we *know* that ext4_writepage() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1165 | * has generated enough buffer credits to do the whole page. So we won't |
| 1166 | * block on the journal in that case, which is good, because the caller may |
| 1167 | * be PF_MEMALLOC. |
| 1168 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1169 | * By accident, ext4 can be reentered when a transaction is open via |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1170 | * quota file writes. If we were to commit the transaction while thus |
| 1171 | * reentered, there can be a deadlock - we would be holding a quota |
| 1172 | * lock, and the commit would never complete if another thread had a |
| 1173 | * transaction open and was blocking on the quota lock - a ranking |
| 1174 | * violation. |
| 1175 | * |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1176 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1177 | * will _not_ run commit under these circumstances because handle->h_ref |
| 1178 | * is elevated. We'll still have enough credits for the tiny quotafile |
| 1179 | * write. |
| 1180 | */ |
| 1181 | static int do_journal_get_write_access(handle_t *handle, |
| 1182 | struct buffer_head *bh) |
| 1183 | { |
| 1184 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1185 | return 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1186 | return ext4_journal_get_write_access(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1187 | } |
| 1188 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1189 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
| 1190 | loff_t pos, unsigned len, unsigned flags, |
| 1191 | struct page **pagep, void **fsdata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1192 | { |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1193 | struct inode *inode = mapping->host; |
Andrew Morton | 7479d2b | 2007-04-01 23:49:44 -0700 | [diff] [blame] | 1194 | int ret, needed_blocks = ext4_writepage_trans_blocks(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1195 | handle_t *handle; |
| 1196 | int retries = 0; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1197 | struct page *page; |
| 1198 | pgoff_t index; |
| 1199 | unsigned from, to; |
| 1200 | |
| 1201 | index = pos >> PAGE_CACHE_SHIFT; |
| 1202 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1203 | to = from + len; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1204 | |
| 1205 | retry: |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1206 | page = __grab_cache_page(mapping, index); |
| 1207 | if (!page) |
| 1208 | return -ENOMEM; |
| 1209 | *pagep = page; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1210 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1211 | handle = ext4_journal_start(inode, needed_blocks); |
| 1212 | if (IS_ERR(handle)) { |
| 1213 | unlock_page(page); |
| 1214 | page_cache_release(page); |
| 1215 | ret = PTR_ERR(handle); |
| 1216 | goto out; |
| 1217 | } |
| 1218 | |
| 1219 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
| 1220 | ext4_get_block); |
| 1221 | |
| 1222 | if (!ret && ext4_should_journal_data(inode)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1223 | ret = walk_page_buffers(handle, page_buffers(page), |
| 1224 | from, to, NULL, do_journal_get_write_access); |
| 1225 | } |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1226 | |
| 1227 | if (ret) { |
Andrew Morton | 7479d2b | 2007-04-01 23:49:44 -0700 | [diff] [blame] | 1228 | ext4_journal_stop(handle); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1229 | unlock_page(page); |
| 1230 | page_cache_release(page); |
| 1231 | } |
| 1232 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1233 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1234 | goto retry; |
Andrew Morton | 7479d2b | 2007-04-01 23:49:44 -0700 | [diff] [blame] | 1235 | out: |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1236 | return ret; |
| 1237 | } |
| 1238 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1239 | int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1240 | { |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1241 | int err = jbd2_journal_dirty_data(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1242 | if (err) |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 1243 | ext4_journal_abort_handle(__func__, __func__, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1244 | bh, handle, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1245 | return err; |
| 1246 | } |
| 1247 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1248 | /* For write_end() in data=journal mode */ |
| 1249 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1250 | { |
| 1251 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1252 | return 0; |
| 1253 | set_buffer_uptodate(bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1254 | return ext4_journal_dirty_metadata(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1255 | } |
| 1256 | |
| 1257 | /* |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1258 | * Generic write_end handler for ordered and writeback ext4 journal modes. |
| 1259 | * We can't use generic_write_end, because that unlocks the page and we need to |
| 1260 | * unlock the page after ext4_journal_stop, but ext4_journal_stop must run |
| 1261 | * after block_write_end. |
| 1262 | */ |
| 1263 | static int ext4_generic_write_end(struct file *file, |
| 1264 | struct address_space *mapping, |
| 1265 | loff_t pos, unsigned len, unsigned copied, |
| 1266 | struct page *page, void *fsdata) |
| 1267 | { |
| 1268 | struct inode *inode = file->f_mapping->host; |
| 1269 | |
| 1270 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 1271 | |
| 1272 | if (pos+copied > inode->i_size) { |
| 1273 | i_size_write(inode, pos+copied); |
| 1274 | mark_inode_dirty(inode); |
| 1275 | } |
| 1276 | |
| 1277 | return copied; |
| 1278 | } |
| 1279 | |
| 1280 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1281 | * We need to pick up the new inode size which generic_commit_write gave us |
| 1282 | * `file' can be NULL - eg, when called from page_symlink(). |
| 1283 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1284 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1285 | * buffers are managed internally. |
| 1286 | */ |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1287 | static int ext4_ordered_write_end(struct file *file, |
| 1288 | struct address_space *mapping, |
| 1289 | loff_t pos, unsigned len, unsigned copied, |
| 1290 | struct page *page, void *fsdata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1291 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1292 | handle_t *handle = ext4_journal_current_handle(); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1293 | struct inode *inode = file->f_mapping->host; |
| 1294 | unsigned from, to; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1295 | int ret = 0, ret2; |
| 1296 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1297 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1298 | to = from + len; |
| 1299 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1300 | ret = walk_page_buffers(handle, page_buffers(page), |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1301 | from, to, NULL, ext4_journal_dirty_data); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1302 | |
| 1303 | if (ret == 0) { |
| 1304 | /* |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1305 | * generic_write_end() will run mark_inode_dirty() if i_size |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1306 | * changes. So let's piggyback the i_disksize mark_inode_dirty |
| 1307 | * into that. |
| 1308 | */ |
| 1309 | loff_t new_i_size; |
| 1310 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1311 | new_i_size = pos + copied; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1312 | if (new_i_size > EXT4_I(inode)->i_disksize) |
| 1313 | EXT4_I(inode)->i_disksize = new_i_size; |
Roel Kluin | f8a87d8 | 2008-04-29 22:01:18 -0400 | [diff] [blame] | 1314 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1315 | page, fsdata); |
Roel Kluin | f8a87d8 | 2008-04-29 22:01:18 -0400 | [diff] [blame] | 1316 | copied = ret2; |
| 1317 | if (ret2 < 0) |
| 1318 | ret = ret2; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1319 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1320 | ret2 = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1321 | if (!ret) |
| 1322 | ret = ret2; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1323 | unlock_page(page); |
| 1324 | page_cache_release(page); |
| 1325 | |
| 1326 | return ret ? ret : copied; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1327 | } |
| 1328 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1329 | static int ext4_writeback_write_end(struct file *file, |
| 1330 | struct address_space *mapping, |
| 1331 | loff_t pos, unsigned len, unsigned copied, |
| 1332 | struct page *page, void *fsdata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1333 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1334 | handle_t *handle = ext4_journal_current_handle(); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1335 | struct inode *inode = file->f_mapping->host; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1336 | int ret = 0, ret2; |
| 1337 | loff_t new_i_size; |
| 1338 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1339 | new_i_size = pos + copied; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1340 | if (new_i_size > EXT4_I(inode)->i_disksize) |
| 1341 | EXT4_I(inode)->i_disksize = new_i_size; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1342 | |
Roel Kluin | f8a87d8 | 2008-04-29 22:01:18 -0400 | [diff] [blame] | 1343 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1344 | page, fsdata); |
Roel Kluin | f8a87d8 | 2008-04-29 22:01:18 -0400 | [diff] [blame] | 1345 | copied = ret2; |
| 1346 | if (ret2 < 0) |
| 1347 | ret = ret2; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1348 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1349 | ret2 = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1350 | if (!ret) |
| 1351 | ret = ret2; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1352 | unlock_page(page); |
| 1353 | page_cache_release(page); |
| 1354 | |
| 1355 | return ret ? ret : copied; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1356 | } |
| 1357 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1358 | static int ext4_journalled_write_end(struct file *file, |
| 1359 | struct address_space *mapping, |
| 1360 | loff_t pos, unsigned len, unsigned copied, |
| 1361 | struct page *page, void *fsdata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1362 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1363 | handle_t *handle = ext4_journal_current_handle(); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1364 | struct inode *inode = mapping->host; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1365 | int ret = 0, ret2; |
| 1366 | int partial = 0; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1367 | unsigned from, to; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1368 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1369 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1370 | to = from + len; |
| 1371 | |
| 1372 | if (copied < len) { |
| 1373 | if (!PageUptodate(page)) |
| 1374 | copied = 0; |
| 1375 | page_zero_new_buffers(page, from+copied, to); |
| 1376 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1377 | |
| 1378 | ret = walk_page_buffers(handle, page_buffers(page), from, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1379 | to, &partial, write_end_fn); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1380 | if (!partial) |
| 1381 | SetPageUptodate(page); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1382 | if (pos+copied > inode->i_size) |
| 1383 | i_size_write(inode, pos+copied); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1384 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
| 1385 | if (inode->i_size > EXT4_I(inode)->i_disksize) { |
| 1386 | EXT4_I(inode)->i_disksize = inode->i_size; |
| 1387 | ret2 = ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1388 | if (!ret) |
| 1389 | ret = ret2; |
| 1390 | } |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1391 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1392 | ret2 = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1393 | if (!ret) |
| 1394 | ret = ret2; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1395 | unlock_page(page); |
| 1396 | page_cache_release(page); |
| 1397 | |
| 1398 | return ret ? ret : copied; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1399 | } |
| 1400 | |
| 1401 | /* |
| 1402 | * bmap() is special. It gets used by applications such as lilo and by |
| 1403 | * the swapper to find the on-disk block of a specific piece of data. |
| 1404 | * |
| 1405 | * Naturally, this is dangerous if the block concerned is still in the |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1406 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1407 | * filesystem and enables swap, then they may get a nasty shock when the |
| 1408 | * data getting swapped to that swapfile suddenly gets overwritten by |
| 1409 | * the original zero's written out previously to the journal and |
| 1410 | * awaiting writeback in the kernel's buffer cache. |
| 1411 | * |
| 1412 | * So, if we see any bmap calls here on a modified, data-journaled file, |
| 1413 | * take extra steps to flush any blocks which might be in the cache. |
| 1414 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1415 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1416 | { |
| 1417 | struct inode *inode = mapping->host; |
| 1418 | journal_t *journal; |
| 1419 | int err; |
| 1420 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1421 | if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1422 | /* |
| 1423 | * This is a REALLY heavyweight approach, but the use of |
| 1424 | * bmap on dirty files is expected to be extremely rare: |
| 1425 | * only if we run lilo or swapon on a freshly made file |
| 1426 | * do we expect this to happen. |
| 1427 | * |
| 1428 | * (bmap requires CAP_SYS_RAWIO so this does not |
| 1429 | * represent an unprivileged user DOS attack --- we'd be |
| 1430 | * in trouble if mortal users could trigger this path at |
| 1431 | * will.) |
| 1432 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1433 | * NB. EXT4_STATE_JDATA is not set on files other than |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1434 | * regular files. If somebody wants to bmap a directory |
| 1435 | * or symlink and gets confused because the buffer |
| 1436 | * hasn't yet been flushed to disk, they deserve |
| 1437 | * everything they get. |
| 1438 | */ |
| 1439 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1440 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
| 1441 | journal = EXT4_JOURNAL(inode); |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1442 | jbd2_journal_lock_updates(journal); |
| 1443 | err = jbd2_journal_flush(journal); |
| 1444 | jbd2_journal_unlock_updates(journal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1445 | |
| 1446 | if (err) |
| 1447 | return 0; |
| 1448 | } |
| 1449 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1450 | return generic_block_bmap(mapping,block,ext4_get_block); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1451 | } |
| 1452 | |
| 1453 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
| 1454 | { |
| 1455 | get_bh(bh); |
| 1456 | return 0; |
| 1457 | } |
| 1458 | |
| 1459 | static int bput_one(handle_t *handle, struct buffer_head *bh) |
| 1460 | { |
| 1461 | put_bh(bh); |
| 1462 | return 0; |
| 1463 | } |
| 1464 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1465 | static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1466 | { |
| 1467 | if (buffer_mapped(bh)) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1468 | return ext4_journal_dirty_data(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1469 | return 0; |
| 1470 | } |
| 1471 | |
| 1472 | /* |
| 1473 | * Note that we always start a transaction even if we're not journalling |
| 1474 | * data. This is to preserve ordering: any hole instantiation within |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1475 | * __block_write_full_page -> ext4_get_block() should be journalled |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1476 | * along with the data so we don't crash and then get metadata which |
| 1477 | * refers to old data. |
| 1478 | * |
| 1479 | * In all journalling modes block_write_full_page() will start the I/O. |
| 1480 | * |
| 1481 | * Problem: |
| 1482 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1483 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
| 1484 | * ext4_writepage() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1485 | * |
| 1486 | * Similar for: |
| 1487 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1488 | * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1489 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1490 | * Same applies to ext4_get_block(). We will deadlock on various things like |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 1491 | * lock_journal and i_data_sem |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1492 | * |
| 1493 | * Setting PF_MEMALLOC here doesn't work - too many internal memory |
| 1494 | * allocations fail. |
| 1495 | * |
| 1496 | * 16May01: If we're reentered then journal_current_handle() will be |
| 1497 | * non-zero. We simply *return*. |
| 1498 | * |
| 1499 | * 1 July 2001: @@@ FIXME: |
| 1500 | * In journalled data mode, a data buffer may be metadata against the |
| 1501 | * current transaction. But the same file is part of a shared mapping |
| 1502 | * and someone does a writepage() on it. |
| 1503 | * |
| 1504 | * We will move the buffer onto the async_data list, but *after* it has |
| 1505 | * been dirtied. So there's a small window where we have dirty data on |
| 1506 | * BJ_Metadata. |
| 1507 | * |
| 1508 | * Note that this only applies to the last partial page in the file. The |
| 1509 | * bit which block_write_full_page() uses prepare/commit for. (That's |
| 1510 | * broken code anyway: it's wrong for msync()). |
| 1511 | * |
| 1512 | * It's a rare case: affects the final partial page, for journalled data |
| 1513 | * where the file is subject to bith write() and writepage() in the same |
| 1514 | * transction. To fix it we'll need a custom block_write_full_page(). |
| 1515 | * We'll probably need that anyway for journalling writepage() output. |
| 1516 | * |
| 1517 | * We don't honour synchronous mounts for writepage(). That would be |
| 1518 | * disastrous. Any write() or metadata operation will sync the fs for |
| 1519 | * us. |
| 1520 | * |
| 1521 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, |
| 1522 | * we don't need to open a transaction here. |
| 1523 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1524 | static int ext4_ordered_writepage(struct page *page, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1525 | struct writeback_control *wbc) |
| 1526 | { |
| 1527 | struct inode *inode = page->mapping->host; |
| 1528 | struct buffer_head *page_bufs; |
| 1529 | handle_t *handle = NULL; |
| 1530 | int ret = 0; |
| 1531 | int err; |
| 1532 | |
| 1533 | J_ASSERT(PageLocked(page)); |
| 1534 | |
| 1535 | /* |
| 1536 | * We give up here if we're reentered, because it might be for a |
| 1537 | * different filesystem. |
| 1538 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1539 | if (ext4_journal_current_handle()) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1540 | goto out_fail; |
| 1541 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1542 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1543 | |
| 1544 | if (IS_ERR(handle)) { |
| 1545 | ret = PTR_ERR(handle); |
| 1546 | goto out_fail; |
| 1547 | } |
| 1548 | |
| 1549 | if (!page_has_buffers(page)) { |
| 1550 | create_empty_buffers(page, inode->i_sb->s_blocksize, |
| 1551 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
| 1552 | } |
| 1553 | page_bufs = page_buffers(page); |
| 1554 | walk_page_buffers(handle, page_bufs, 0, |
| 1555 | PAGE_CACHE_SIZE, NULL, bget_one); |
| 1556 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1557 | ret = block_write_full_page(page, ext4_get_block, wbc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1558 | |
| 1559 | /* |
| 1560 | * The page can become unlocked at any point now, and |
| 1561 | * truncate can then come in and change things. So we |
| 1562 | * can't touch *page from now on. But *page_bufs is |
| 1563 | * safe due to elevated refcount. |
| 1564 | */ |
| 1565 | |
| 1566 | /* |
| 1567 | * And attach them to the current transaction. But only if |
| 1568 | * block_write_full_page() succeeded. Otherwise they are unmapped, |
| 1569 | * and generally junk. |
| 1570 | */ |
| 1571 | if (ret == 0) { |
| 1572 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1573 | NULL, jbd2_journal_dirty_data_fn); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1574 | if (!ret) |
| 1575 | ret = err; |
| 1576 | } |
| 1577 | walk_page_buffers(handle, page_bufs, 0, |
| 1578 | PAGE_CACHE_SIZE, NULL, bput_one); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1579 | err = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1580 | if (!ret) |
| 1581 | ret = err; |
| 1582 | return ret; |
| 1583 | |
| 1584 | out_fail: |
| 1585 | redirty_page_for_writepage(wbc, page); |
| 1586 | unlock_page(page); |
| 1587 | return ret; |
| 1588 | } |
| 1589 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1590 | static int ext4_writeback_writepage(struct page *page, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1591 | struct writeback_control *wbc) |
| 1592 | { |
| 1593 | struct inode *inode = page->mapping->host; |
| 1594 | handle_t *handle = NULL; |
| 1595 | int ret = 0; |
| 1596 | int err; |
| 1597 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1598 | if (ext4_journal_current_handle()) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1599 | goto out_fail; |
| 1600 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1601 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1602 | if (IS_ERR(handle)) { |
| 1603 | ret = PTR_ERR(handle); |
| 1604 | goto out_fail; |
| 1605 | } |
| 1606 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1607 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) |
| 1608 | ret = nobh_writepage(page, ext4_get_block, wbc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1609 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1610 | ret = block_write_full_page(page, ext4_get_block, wbc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1611 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1612 | err = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1613 | if (!ret) |
| 1614 | ret = err; |
| 1615 | return ret; |
| 1616 | |
| 1617 | out_fail: |
| 1618 | redirty_page_for_writepage(wbc, page); |
| 1619 | unlock_page(page); |
| 1620 | return ret; |
| 1621 | } |
| 1622 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1623 | static int ext4_journalled_writepage(struct page *page, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1624 | struct writeback_control *wbc) |
| 1625 | { |
| 1626 | struct inode *inode = page->mapping->host; |
| 1627 | handle_t *handle = NULL; |
| 1628 | int ret = 0; |
| 1629 | int err; |
| 1630 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1631 | if (ext4_journal_current_handle()) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1632 | goto no_write; |
| 1633 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1634 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1635 | if (IS_ERR(handle)) { |
| 1636 | ret = PTR_ERR(handle); |
| 1637 | goto no_write; |
| 1638 | } |
| 1639 | |
| 1640 | if (!page_has_buffers(page) || PageChecked(page)) { |
| 1641 | /* |
| 1642 | * It's mmapped pagecache. Add buffers and journal it. There |
| 1643 | * doesn't seem much point in redirtying the page here. |
| 1644 | */ |
| 1645 | ClearPageChecked(page); |
| 1646 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1647 | ext4_get_block); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1648 | if (ret != 0) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1649 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1650 | goto out_unlock; |
| 1651 | } |
| 1652 | ret = walk_page_buffers(handle, page_buffers(page), 0, |
| 1653 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); |
| 1654 | |
| 1655 | err = walk_page_buffers(handle, page_buffers(page), 0, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1656 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1657 | if (ret == 0) |
| 1658 | ret = err; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1659 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1660 | unlock_page(page); |
| 1661 | } else { |
| 1662 | /* |
| 1663 | * It may be a page full of checkpoint-mode buffers. We don't |
| 1664 | * really know unless we go poke around in the buffer_heads. |
| 1665 | * But block_write_full_page will do the right thing. |
| 1666 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1667 | ret = block_write_full_page(page, ext4_get_block, wbc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1668 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1669 | err = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1670 | if (!ret) |
| 1671 | ret = err; |
| 1672 | out: |
| 1673 | return ret; |
| 1674 | |
| 1675 | no_write: |
| 1676 | redirty_page_for_writepage(wbc, page); |
| 1677 | out_unlock: |
| 1678 | unlock_page(page); |
| 1679 | goto out; |
| 1680 | } |
| 1681 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1682 | static int ext4_readpage(struct file *file, struct page *page) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1683 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1684 | return mpage_readpage(page, ext4_get_block); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1685 | } |
| 1686 | |
| 1687 | static int |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1688 | ext4_readpages(struct file *file, struct address_space *mapping, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1689 | struct list_head *pages, unsigned nr_pages) |
| 1690 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1691 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1692 | } |
| 1693 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1694 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1695 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1696 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1697 | |
| 1698 | /* |
| 1699 | * If it's a full truncate we just forget about the pending dirtying |
| 1700 | */ |
| 1701 | if (offset == 0) |
| 1702 | ClearPageChecked(page); |
| 1703 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1704 | jbd2_journal_invalidatepage(journal, page, offset); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1705 | } |
| 1706 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1707 | static int ext4_releasepage(struct page *page, gfp_t wait) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1708 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1709 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1710 | |
| 1711 | WARN_ON(PageChecked(page)); |
| 1712 | if (!page_has_buffers(page)) |
| 1713 | return 0; |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 1714 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1715 | } |
| 1716 | |
| 1717 | /* |
| 1718 | * If the O_DIRECT write will extend the file then add this inode to the |
| 1719 | * orphan list. So recovery will truncate it back to the original size |
| 1720 | * if the machine crashes during the write. |
| 1721 | * |
| 1722 | * If the O_DIRECT write is intantiating holes inside i_size and the machine |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1723 | * crashes then stale disk data _may_ be exposed inside the file. But current |
| 1724 | * VFS code falls back into buffered path in that case so we are safe. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1725 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1726 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1727 | const struct iovec *iov, loff_t offset, |
| 1728 | unsigned long nr_segs) |
| 1729 | { |
| 1730 | struct file *file = iocb->ki_filp; |
| 1731 | struct inode *inode = file->f_mapping->host; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1732 | struct ext4_inode_info *ei = EXT4_I(inode); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1733 | handle_t *handle; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1734 | ssize_t ret; |
| 1735 | int orphan = 0; |
| 1736 | size_t count = iov_length(iov, nr_segs); |
| 1737 | |
| 1738 | if (rw == WRITE) { |
| 1739 | loff_t final_size = offset + count; |
| 1740 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1741 | if (final_size > inode->i_size) { |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1742 | /* Credits for sb + inode write */ |
| 1743 | handle = ext4_journal_start(inode, 2); |
| 1744 | if (IS_ERR(handle)) { |
| 1745 | ret = PTR_ERR(handle); |
| 1746 | goto out; |
| 1747 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1748 | ret = ext4_orphan_add(handle, inode); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1749 | if (ret) { |
| 1750 | ext4_journal_stop(handle); |
| 1751 | goto out; |
| 1752 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1753 | orphan = 1; |
| 1754 | ei->i_disksize = inode->i_size; |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1755 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1756 | } |
| 1757 | } |
| 1758 | |
| 1759 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, |
| 1760 | offset, nr_segs, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1761 | ext4_get_block, NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1762 | |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1763 | if (orphan) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1764 | int err; |
| 1765 | |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1766 | /* Credits for sb + inode write */ |
| 1767 | handle = ext4_journal_start(inode, 2); |
| 1768 | if (IS_ERR(handle)) { |
| 1769 | /* This is really bad luck. We've written the data |
| 1770 | * but cannot extend i_size. Bail out and pretend |
| 1771 | * the write failed... */ |
| 1772 | ret = PTR_ERR(handle); |
| 1773 | goto out; |
| 1774 | } |
| 1775 | if (inode->i_nlink) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1776 | ext4_orphan_del(handle, inode); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 1777 | if (ret > 0) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1778 | loff_t end = offset + ret; |
| 1779 | if (end > inode->i_size) { |
| 1780 | ei->i_disksize = end; |
| 1781 | i_size_write(inode, end); |
| 1782 | /* |
| 1783 | * We're going to return a positive `ret' |
| 1784 | * here due to non-zero-length I/O, so there's |
| 1785 | * no way of reporting error returns from |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1786 | * ext4_mark_inode_dirty() to userspace. So |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1787 | * ignore it. |
| 1788 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1789 | ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1790 | } |
| 1791 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1792 | err = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1793 | if (ret == 0) |
| 1794 | ret = err; |
| 1795 | } |
| 1796 | out: |
| 1797 | return ret; |
| 1798 | } |
| 1799 | |
| 1800 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1801 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1802 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
| 1803 | * much here because ->set_page_dirty is called under VFS locks. The page is |
| 1804 | * not necessarily locked. |
| 1805 | * |
| 1806 | * We cannot just dirty the page and leave attached buffers clean, because the |
| 1807 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
| 1808 | * or jbddirty because all the journalling code will explode. |
| 1809 | * |
| 1810 | * So what we do is to mark the page "pending dirty" and next time writepage |
| 1811 | * is called, propagate that into the buffers appropriately. |
| 1812 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1813 | static int ext4_journalled_set_page_dirty(struct page *page) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1814 | { |
| 1815 | SetPageChecked(page); |
| 1816 | return __set_page_dirty_nobuffers(page); |
| 1817 | } |
| 1818 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1819 | static const struct address_space_operations ext4_ordered_aops = { |
| 1820 | .readpage = ext4_readpage, |
| 1821 | .readpages = ext4_readpages, |
| 1822 | .writepage = ext4_ordered_writepage, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1823 | .sync_page = block_sync_page, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1824 | .write_begin = ext4_write_begin, |
| 1825 | .write_end = ext4_ordered_write_end, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1826 | .bmap = ext4_bmap, |
| 1827 | .invalidatepage = ext4_invalidatepage, |
| 1828 | .releasepage = ext4_releasepage, |
| 1829 | .direct_IO = ext4_direct_IO, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1830 | .migratepage = buffer_migrate_page, |
| 1831 | }; |
| 1832 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1833 | static const struct address_space_operations ext4_writeback_aops = { |
| 1834 | .readpage = ext4_readpage, |
| 1835 | .readpages = ext4_readpages, |
| 1836 | .writepage = ext4_writeback_writepage, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1837 | .sync_page = block_sync_page, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1838 | .write_begin = ext4_write_begin, |
| 1839 | .write_end = ext4_writeback_write_end, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1840 | .bmap = ext4_bmap, |
| 1841 | .invalidatepage = ext4_invalidatepage, |
| 1842 | .releasepage = ext4_releasepage, |
| 1843 | .direct_IO = ext4_direct_IO, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1844 | .migratepage = buffer_migrate_page, |
| 1845 | }; |
| 1846 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1847 | static const struct address_space_operations ext4_journalled_aops = { |
| 1848 | .readpage = ext4_readpage, |
| 1849 | .readpages = ext4_readpages, |
| 1850 | .writepage = ext4_journalled_writepage, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1851 | .sync_page = block_sync_page, |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1852 | .write_begin = ext4_write_begin, |
| 1853 | .write_end = ext4_journalled_write_end, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1854 | .set_page_dirty = ext4_journalled_set_page_dirty, |
| 1855 | .bmap = ext4_bmap, |
| 1856 | .invalidatepage = ext4_invalidatepage, |
| 1857 | .releasepage = ext4_releasepage, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1858 | }; |
| 1859 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1860 | void ext4_set_aops(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1861 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1862 | if (ext4_should_order_data(inode)) |
| 1863 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
| 1864 | else if (ext4_should_writeback_data(inode)) |
| 1865 | inode->i_mapping->a_ops = &ext4_writeback_aops; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1866 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1867 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1868 | } |
| 1869 | |
| 1870 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1871 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1872 | * up to the end of the block which corresponds to `from'. |
| 1873 | * This required during truncate. We need to physically zero the tail end |
| 1874 | * of that block so it doesn't yield old data if the file is later grown. |
| 1875 | */ |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 1876 | int ext4_block_truncate_page(handle_t *handle, struct page *page, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1877 | struct address_space *mapping, loff_t from) |
| 1878 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1879 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1880 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 1881 | unsigned blocksize, length, pos; |
| 1882 | ext4_lblk_t iblock; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1883 | struct inode *inode = mapping->host; |
| 1884 | struct buffer_head *bh; |
| 1885 | int err = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1886 | |
| 1887 | blocksize = inode->i_sb->s_blocksize; |
| 1888 | length = blocksize - (offset & (blocksize - 1)); |
| 1889 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
| 1890 | |
| 1891 | /* |
| 1892 | * For "nobh" option, we can only work if we don't need to |
| 1893 | * read-in the page - otherwise we create buffers to do the IO. |
| 1894 | */ |
| 1895 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1896 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 1897 | zero_user(page, offset, length); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1898 | set_page_dirty(page); |
| 1899 | goto unlock; |
| 1900 | } |
| 1901 | |
| 1902 | if (!page_has_buffers(page)) |
| 1903 | create_empty_buffers(page, blocksize, 0); |
| 1904 | |
| 1905 | /* Find the buffer that contains "offset" */ |
| 1906 | bh = page_buffers(page); |
| 1907 | pos = blocksize; |
| 1908 | while (offset >= pos) { |
| 1909 | bh = bh->b_this_page; |
| 1910 | iblock++; |
| 1911 | pos += blocksize; |
| 1912 | } |
| 1913 | |
| 1914 | err = 0; |
| 1915 | if (buffer_freed(bh)) { |
| 1916 | BUFFER_TRACE(bh, "freed: skip"); |
| 1917 | goto unlock; |
| 1918 | } |
| 1919 | |
| 1920 | if (!buffer_mapped(bh)) { |
| 1921 | BUFFER_TRACE(bh, "unmapped"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1922 | ext4_get_block(inode, iblock, bh, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1923 | /* unmapped? It's a hole - nothing to do */ |
| 1924 | if (!buffer_mapped(bh)) { |
| 1925 | BUFFER_TRACE(bh, "still unmapped"); |
| 1926 | goto unlock; |
| 1927 | } |
| 1928 | } |
| 1929 | |
| 1930 | /* Ok, it's mapped. Make sure it's up-to-date */ |
| 1931 | if (PageUptodate(page)) |
| 1932 | set_buffer_uptodate(bh); |
| 1933 | |
| 1934 | if (!buffer_uptodate(bh)) { |
| 1935 | err = -EIO; |
| 1936 | ll_rw_block(READ, 1, &bh); |
| 1937 | wait_on_buffer(bh); |
| 1938 | /* Uhhuh. Read error. Complain and punt. */ |
| 1939 | if (!buffer_uptodate(bh)) |
| 1940 | goto unlock; |
| 1941 | } |
| 1942 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1943 | if (ext4_should_journal_data(inode)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1944 | BUFFER_TRACE(bh, "get write access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1945 | err = ext4_journal_get_write_access(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1946 | if (err) |
| 1947 | goto unlock; |
| 1948 | } |
| 1949 | |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 1950 | zero_user(page, offset, length); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1951 | |
| 1952 | BUFFER_TRACE(bh, "zeroed end of block"); |
| 1953 | |
| 1954 | err = 0; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1955 | if (ext4_should_journal_data(inode)) { |
| 1956 | err = ext4_journal_dirty_metadata(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1957 | } else { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1958 | if (ext4_should_order_data(inode)) |
| 1959 | err = ext4_journal_dirty_data(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1960 | mark_buffer_dirty(bh); |
| 1961 | } |
| 1962 | |
| 1963 | unlock: |
| 1964 | unlock_page(page); |
| 1965 | page_cache_release(page); |
| 1966 | return err; |
| 1967 | } |
| 1968 | |
| 1969 | /* |
| 1970 | * Probably it should be a library function... search for first non-zero word |
| 1971 | * or memcmp with zero_page, whatever is better for particular architecture. |
| 1972 | * Linus? |
| 1973 | */ |
| 1974 | static inline int all_zeroes(__le32 *p, __le32 *q) |
| 1975 | { |
| 1976 | while (p < q) |
| 1977 | if (*p++) |
| 1978 | return 0; |
| 1979 | return 1; |
| 1980 | } |
| 1981 | |
| 1982 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1983 | * ext4_find_shared - find the indirect blocks for partial truncation. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1984 | * @inode: inode in question |
| 1985 | * @depth: depth of the affected branch |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1986 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1987 | * @chain: place to store the pointers to partial indirect blocks |
| 1988 | * @top: place to the (detached) top of branch |
| 1989 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1990 | * This is a helper function used by ext4_truncate(). |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1991 | * |
| 1992 | * When we do truncate() we may have to clean the ends of several |
| 1993 | * indirect blocks but leave the blocks themselves alive. Block is |
| 1994 | * partially truncated if some data below the new i_size is refered |
| 1995 | * from it (and it is on the path to the first completely truncated |
| 1996 | * data block, indeed). We have to free the top of that path along |
| 1997 | * with everything to the right of the path. Since no allocation |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1998 | * past the truncation point is possible until ext4_truncate() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1999 | * finishes, we may safely do the latter, but top of branch may |
| 2000 | * require special attention - pageout below the truncation point |
| 2001 | * might try to populate it. |
| 2002 | * |
| 2003 | * We atomically detach the top of branch from the tree, store the |
| 2004 | * block number of its root in *@top, pointers to buffer_heads of |
| 2005 | * partially truncated blocks - in @chain[].bh and pointers to |
| 2006 | * their last elements that should not be removed - in |
| 2007 | * @chain[].p. Return value is the pointer to last filled element |
| 2008 | * of @chain. |
| 2009 | * |
| 2010 | * The work left to caller to do the actual freeing of subtrees: |
| 2011 | * a) free the subtree starting from *@top |
| 2012 | * b) free the subtrees whose roots are stored in |
| 2013 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) |
| 2014 | * c) free the subtrees growing from the inode past the @chain[0]. |
| 2015 | * (no partially truncated stuff there). */ |
| 2016 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2017 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2018 | ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2019 | { |
| 2020 | Indirect *partial, *p; |
| 2021 | int k, err; |
| 2022 | |
| 2023 | *top = 0; |
| 2024 | /* Make k index the deepest non-null offest + 1 */ |
| 2025 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
| 2026 | ; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2027 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2028 | /* Writer: pointers */ |
| 2029 | if (!partial) |
| 2030 | partial = chain + k-1; |
| 2031 | /* |
| 2032 | * If the branch acquired continuation since we've looked at it - |
| 2033 | * fine, it should all survive and (new) top doesn't belong to us. |
| 2034 | */ |
| 2035 | if (!partial->key && *partial->p) |
| 2036 | /* Writer: end */ |
| 2037 | goto no_top; |
| 2038 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) |
| 2039 | ; |
| 2040 | /* |
| 2041 | * OK, we've found the last block that must survive. The rest of our |
| 2042 | * branch should be detached before unlocking. However, if that rest |
| 2043 | * of branch is all ours and does not grow immediately from the inode |
| 2044 | * it's easier to cheat and just decrement partial->p. |
| 2045 | */ |
| 2046 | if (p == chain + k - 1 && p > chain) { |
| 2047 | p->p--; |
| 2048 | } else { |
| 2049 | *top = *p->p; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2050 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2051 | #if 0 |
| 2052 | *p->p = 0; |
| 2053 | #endif |
| 2054 | } |
| 2055 | /* Writer: end */ |
| 2056 | |
| 2057 | while(partial > p) { |
| 2058 | brelse(partial->bh); |
| 2059 | partial--; |
| 2060 | } |
| 2061 | no_top: |
| 2062 | return partial; |
| 2063 | } |
| 2064 | |
| 2065 | /* |
| 2066 | * Zero a number of block pointers in either an inode or an indirect block. |
| 2067 | * If we restart the transaction we must again get write access to the |
| 2068 | * indirect block for further modification. |
| 2069 | * |
| 2070 | * We release `count' blocks on disk, but (last - first) may be greater |
| 2071 | * than `count' because there can be holes in there. |
| 2072 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2073 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
| 2074 | struct buffer_head *bh, ext4_fsblk_t block_to_free, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2075 | unsigned long count, __le32 *first, __le32 *last) |
| 2076 | { |
| 2077 | __le32 *p; |
| 2078 | if (try_to_extend_transaction(handle, inode)) { |
| 2079 | if (bh) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2080 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
| 2081 | ext4_journal_dirty_metadata(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2082 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2083 | ext4_mark_inode_dirty(handle, inode); |
| 2084 | ext4_journal_test_restart(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2085 | if (bh) { |
| 2086 | BUFFER_TRACE(bh, "retaking write access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2087 | ext4_journal_get_write_access(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2088 | } |
| 2089 | } |
| 2090 | |
| 2091 | /* |
| 2092 | * Any buffers which are on the journal will be in memory. We find |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2093 | * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2094 | * on them. We've already detached each block from the file, so |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2095 | * bforget() in jbd2_journal_forget() should be safe. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2096 | * |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2097 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2098 | */ |
| 2099 | for (p = first; p < last; p++) { |
| 2100 | u32 nr = le32_to_cpu(*p); |
| 2101 | if (nr) { |
Aneesh Kumar K.V | 1d03ec9 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2102 | struct buffer_head *tbh; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2103 | |
| 2104 | *p = 0; |
Aneesh Kumar K.V | 1d03ec9 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2105 | tbh = sb_find_get_block(inode->i_sb, nr); |
| 2106 | ext4_forget(handle, 0, inode, tbh, nr); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2107 | } |
| 2108 | } |
| 2109 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 2110 | ext4_free_blocks(handle, inode, block_to_free, count, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2111 | } |
| 2112 | |
| 2113 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2114 | * ext4_free_data - free a list of data blocks |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2115 | * @handle: handle for this transaction |
| 2116 | * @inode: inode we are dealing with |
| 2117 | * @this_bh: indirect buffer_head which contains *@first and *@last |
| 2118 | * @first: array of block numbers |
| 2119 | * @last: points immediately past the end of array |
| 2120 | * |
| 2121 | * We are freeing all blocks refered from that array (numbers are stored as |
| 2122 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. |
| 2123 | * |
| 2124 | * We accumulate contiguous runs of blocks to free. Conveniently, if these |
| 2125 | * blocks are contiguous then releasing them at one time will only affect one |
| 2126 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't |
| 2127 | * actually use a lot of journal space. |
| 2128 | * |
| 2129 | * @this_bh will be %NULL if @first and @last point into the inode's direct |
| 2130 | * block pointers. |
| 2131 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2132 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2133 | struct buffer_head *this_bh, |
| 2134 | __le32 *first, __le32 *last) |
| 2135 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2136 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2137 | unsigned long count = 0; /* Number of blocks in the run */ |
| 2138 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
| 2139 | corresponding to |
| 2140 | block_to_free */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2141 | ext4_fsblk_t nr; /* Current block # */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2142 | __le32 *p; /* Pointer into inode/ind |
| 2143 | for current block */ |
| 2144 | int err; |
| 2145 | |
| 2146 | if (this_bh) { /* For indirect block */ |
| 2147 | BUFFER_TRACE(this_bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2148 | err = ext4_journal_get_write_access(handle, this_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2149 | /* Important: if we can't update the indirect pointers |
| 2150 | * to the blocks, we can't free them. */ |
| 2151 | if (err) |
| 2152 | return; |
| 2153 | } |
| 2154 | |
| 2155 | for (p = first; p < last; p++) { |
| 2156 | nr = le32_to_cpu(*p); |
| 2157 | if (nr) { |
| 2158 | /* accumulate blocks to free if they're contiguous */ |
| 2159 | if (count == 0) { |
| 2160 | block_to_free = nr; |
| 2161 | block_to_free_p = p; |
| 2162 | count = 1; |
| 2163 | } else if (nr == block_to_free + count) { |
| 2164 | count++; |
| 2165 | } else { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2166 | ext4_clear_blocks(handle, inode, this_bh, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2167 | block_to_free, |
| 2168 | count, block_to_free_p, p); |
| 2169 | block_to_free = nr; |
| 2170 | block_to_free_p = p; |
| 2171 | count = 1; |
| 2172 | } |
| 2173 | } |
| 2174 | } |
| 2175 | |
| 2176 | if (count > 0) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2177 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2178 | count, block_to_free_p, p); |
| 2179 | |
| 2180 | if (this_bh) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2181 | BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata"); |
| 2182 | ext4_journal_dirty_metadata(handle, this_bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2183 | } |
| 2184 | } |
| 2185 | |
| 2186 | /** |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2187 | * ext4_free_branches - free an array of branches |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2188 | * @handle: JBD handle for this transaction |
| 2189 | * @inode: inode we are dealing with |
| 2190 | * @parent_bh: the buffer_head which contains *@first and *@last |
| 2191 | * @first: array of block numbers |
| 2192 | * @last: pointer immediately past the end of array |
| 2193 | * @depth: depth of the branches to free |
| 2194 | * |
| 2195 | * We are freeing all blocks refered from these branches (numbers are |
| 2196 | * stored as little-endian 32-bit) and updating @inode->i_blocks |
| 2197 | * appropriately. |
| 2198 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2199 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2200 | struct buffer_head *parent_bh, |
| 2201 | __le32 *first, __le32 *last, int depth) |
| 2202 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2203 | ext4_fsblk_t nr; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2204 | __le32 *p; |
| 2205 | |
| 2206 | if (is_handle_aborted(handle)) |
| 2207 | return; |
| 2208 | |
| 2209 | if (depth--) { |
| 2210 | struct buffer_head *bh; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2211 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2212 | p = last; |
| 2213 | while (--p >= first) { |
| 2214 | nr = le32_to_cpu(*p); |
| 2215 | if (!nr) |
| 2216 | continue; /* A hole */ |
| 2217 | |
| 2218 | /* Go read the buffer for the next level down */ |
| 2219 | bh = sb_bread(inode->i_sb, nr); |
| 2220 | |
| 2221 | /* |
| 2222 | * A read failure? Report error and clear slot |
| 2223 | * (should be rare). |
| 2224 | */ |
| 2225 | if (!bh) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2226 | ext4_error(inode->i_sb, "ext4_free_branches", |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 2227 | "Read failure, inode=%lu, block=%llu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2228 | inode->i_ino, nr); |
| 2229 | continue; |
| 2230 | } |
| 2231 | |
| 2232 | /* This zaps the entire block. Bottom up. */ |
| 2233 | BUFFER_TRACE(bh, "free child branches"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2234 | ext4_free_branches(handle, inode, bh, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2235 | (__le32*)bh->b_data, |
| 2236 | (__le32*)bh->b_data + addr_per_block, |
| 2237 | depth); |
| 2238 | |
| 2239 | /* |
| 2240 | * We've probably journalled the indirect block several |
| 2241 | * times during the truncate. But it's no longer |
| 2242 | * needed and we now drop it from the transaction via |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2243 | * jbd2_journal_revoke(). |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2244 | * |
| 2245 | * That's easy if it's exclusively part of this |
| 2246 | * transaction. But if it's part of the committing |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2247 | * transaction then jbd2_journal_forget() will simply |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2248 | * brelse() it. That means that if the underlying |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2249 | * block is reallocated in ext4_get_block(), |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2250 | * unmap_underlying_metadata() will find this block |
| 2251 | * and will try to get rid of it. damn, damn. |
| 2252 | * |
| 2253 | * If this block has already been committed to the |
| 2254 | * journal, a revoke record will be written. And |
| 2255 | * revoke records must be emitted *before* clearing |
| 2256 | * this block's bit in the bitmaps. |
| 2257 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2258 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2259 | |
| 2260 | /* |
| 2261 | * Everything below this this pointer has been |
| 2262 | * released. Now let this top-of-subtree go. |
| 2263 | * |
| 2264 | * We want the freeing of this indirect block to be |
| 2265 | * atomic in the journal with the updating of the |
| 2266 | * bitmap block which owns it. So make some room in |
| 2267 | * the journal. |
| 2268 | * |
| 2269 | * We zero the parent pointer *after* freeing its |
| 2270 | * pointee in the bitmaps, so if extend_transaction() |
| 2271 | * for some reason fails to put the bitmap changes and |
| 2272 | * the release into the same transaction, recovery |
| 2273 | * will merely complain about releasing a free block, |
| 2274 | * rather than leaking blocks. |
| 2275 | */ |
| 2276 | if (is_handle_aborted(handle)) |
| 2277 | return; |
| 2278 | if (try_to_extend_transaction(handle, inode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2279 | ext4_mark_inode_dirty(handle, inode); |
| 2280 | ext4_journal_test_restart(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2281 | } |
| 2282 | |
Alex Tomas | c9de560 | 2008-01-29 00:19:52 -0500 | [diff] [blame] | 2283 | ext4_free_blocks(handle, inode, nr, 1, 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2284 | |
| 2285 | if (parent_bh) { |
| 2286 | /* |
| 2287 | * The block which we have just freed is |
| 2288 | * pointed to by an indirect block: journal it |
| 2289 | */ |
| 2290 | BUFFER_TRACE(parent_bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2291 | if (!ext4_journal_get_write_access(handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2292 | parent_bh)){ |
| 2293 | *p = 0; |
| 2294 | BUFFER_TRACE(parent_bh, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2295 | "call ext4_journal_dirty_metadata"); |
| 2296 | ext4_journal_dirty_metadata(handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2297 | parent_bh); |
| 2298 | } |
| 2299 | } |
| 2300 | } |
| 2301 | } else { |
| 2302 | /* We have reached the bottom of the tree. */ |
| 2303 | BUFFER_TRACE(parent_bh, "free data blocks"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2304 | ext4_free_data(handle, inode, parent_bh, first, last); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2305 | } |
| 2306 | } |
| 2307 | |
| 2308 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2309 | * ext4_truncate() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2310 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2311 | * We block out ext4_get_block() block instantiations across the entire |
| 2312 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2313 | * simultaneously on behalf of the same inode. |
| 2314 | * |
| 2315 | * As we work through the truncate and commmit bits of it to the journal there |
| 2316 | * is one core, guiding principle: the file's tree must always be consistent on |
| 2317 | * disk. We must be able to restart the truncate after a crash. |
| 2318 | * |
| 2319 | * The file's tree may be transiently inconsistent in memory (although it |
| 2320 | * probably isn't), but whenever we close off and commit a journal transaction, |
| 2321 | * the contents of (the filesystem + the journal) must be consistent and |
| 2322 | * restartable. It's pretty simple, really: bottom up, right to left (although |
| 2323 | * left-to-right works OK too). |
| 2324 | * |
| 2325 | * Note that at recovery time, journal replay occurs *before* the restart of |
| 2326 | * truncate against the orphan inode list. |
| 2327 | * |
| 2328 | * The committed inode has the new, desired i_size (which is the same as |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2329 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2330 | * that this inode's truncate did not complete and it will again call |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2331 | * ext4_truncate() to have another go. So there will be instantiated blocks |
| 2332 | * to the right of the truncation point in a crashed ext4 filesystem. But |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2333 | * that's fine - as long as they are linked from the inode, the post-crash |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2334 | * ext4_truncate() run will find them and release them. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2335 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2336 | void ext4_truncate(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2337 | { |
| 2338 | handle_t *handle; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2339 | struct ext4_inode_info *ei = EXT4_I(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2340 | __le32 *i_data = ei->i_data; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2341 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2342 | struct address_space *mapping = inode->i_mapping; |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2343 | ext4_lblk_t offsets[4]; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2344 | Indirect chain[4]; |
| 2345 | Indirect *partial; |
| 2346 | __le32 nr = 0; |
| 2347 | int n; |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2348 | ext4_lblk_t last_block; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2349 | unsigned blocksize = inode->i_sb->s_blocksize; |
| 2350 | struct page *page; |
| 2351 | |
| 2352 | if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 2353 | S_ISLNK(inode->i_mode))) |
| 2354 | return; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2355 | if (ext4_inode_is_fast_symlink(inode)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2356 | return; |
| 2357 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) |
| 2358 | return; |
| 2359 | |
| 2360 | /* |
| 2361 | * We have to lock the EOF page here, because lock_page() nests |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2362 | * outside jbd2_journal_start(). |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2363 | */ |
| 2364 | if ((inode->i_size & (blocksize - 1)) == 0) { |
| 2365 | /* Block boundary? Nothing to do */ |
| 2366 | page = NULL; |
| 2367 | } else { |
| 2368 | page = grab_cache_page(mapping, |
| 2369 | inode->i_size >> PAGE_CACHE_SHIFT); |
| 2370 | if (!page) |
| 2371 | return; |
| 2372 | } |
| 2373 | |
Aneesh Kumar K.V | 1d03ec9 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2374 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
| 2375 | ext4_ext_truncate(inode, page); |
| 2376 | return; |
| 2377 | } |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 2378 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2379 | handle = start_transaction(inode); |
| 2380 | if (IS_ERR(handle)) { |
| 2381 | if (page) { |
| 2382 | clear_highpage(page); |
| 2383 | flush_dcache_page(page); |
| 2384 | unlock_page(page); |
| 2385 | page_cache_release(page); |
| 2386 | } |
| 2387 | return; /* AKPM: return what? */ |
| 2388 | } |
| 2389 | |
| 2390 | last_block = (inode->i_size + blocksize-1) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2391 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2392 | |
| 2393 | if (page) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2394 | ext4_block_truncate_page(handle, page, mapping, inode->i_size); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2395 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2396 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2397 | if (n == 0) |
| 2398 | goto out_stop; /* error */ |
| 2399 | |
| 2400 | /* |
| 2401 | * OK. This truncate is going to happen. We add the inode to the |
| 2402 | * orphan list, so that if this truncate spans multiple transactions, |
| 2403 | * and we crash, we will resume the truncate when the filesystem |
| 2404 | * recovers. It also marks the inode dirty, to catch the new size. |
| 2405 | * |
| 2406 | * Implication: the file must always be in a sane, consistent |
| 2407 | * truncatable state while each transaction commits. |
| 2408 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2409 | if (ext4_orphan_add(handle, inode)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2410 | goto out_stop; |
| 2411 | |
| 2412 | /* |
| 2413 | * The orphan list entry will now protect us from any crash which |
| 2414 | * occurs before the truncate completes, so it is now safe to propagate |
| 2415 | * the new, shorter inode size (held for now in i_size) into the |
| 2416 | * on-disk inode. We do this via i_disksize, which is the value which |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2417 | * ext4 *really* writes onto the disk inode. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2418 | */ |
| 2419 | ei->i_disksize = inode->i_size; |
| 2420 | |
| 2421 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2422 | * From here we block out all ext4_get_block() callers who want to |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2423 | * modify the block allocation tree. |
| 2424 | */ |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2425 | down_write(&ei->i_data_sem); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2426 | |
| 2427 | if (n == 1) { /* direct blocks */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2428 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
| 2429 | i_data + EXT4_NDIR_BLOCKS); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2430 | goto do_indirects; |
| 2431 | } |
| 2432 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2433 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2434 | /* Kill the top of shared branch (not detached) */ |
| 2435 | if (nr) { |
| 2436 | if (partial == chain) { |
| 2437 | /* Shared branch grows from the inode */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2438 | ext4_free_branches(handle, inode, NULL, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2439 | &nr, &nr+1, (chain+n-1) - partial); |
| 2440 | *partial->p = 0; |
| 2441 | /* |
| 2442 | * We mark the inode dirty prior to restart, |
| 2443 | * and prior to stop. No need for it here. |
| 2444 | */ |
| 2445 | } else { |
| 2446 | /* Shared branch grows from an indirect block */ |
| 2447 | BUFFER_TRACE(partial->bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2448 | ext4_free_branches(handle, inode, partial->bh, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2449 | partial->p, |
| 2450 | partial->p+1, (chain+n-1) - partial); |
| 2451 | } |
| 2452 | } |
| 2453 | /* Clear the ends of indirect blocks on the shared branch */ |
| 2454 | while (partial > chain) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2455 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2456 | (__le32*)partial->bh->b_data+addr_per_block, |
| 2457 | (chain+n-1) - partial); |
| 2458 | BUFFER_TRACE(partial->bh, "call brelse"); |
| 2459 | brelse (partial->bh); |
| 2460 | partial--; |
| 2461 | } |
| 2462 | do_indirects: |
| 2463 | /* Kill the remaining (whole) subtrees */ |
| 2464 | switch (offsets[0]) { |
| 2465 | default: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2466 | nr = i_data[EXT4_IND_BLOCK]; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2467 | if (nr) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2468 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
| 2469 | i_data[EXT4_IND_BLOCK] = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2470 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2471 | case EXT4_IND_BLOCK: |
| 2472 | nr = i_data[EXT4_DIND_BLOCK]; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2473 | if (nr) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2474 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
| 2475 | i_data[EXT4_DIND_BLOCK] = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2476 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2477 | case EXT4_DIND_BLOCK: |
| 2478 | nr = i_data[EXT4_TIND_BLOCK]; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2479 | if (nr) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2480 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
| 2481 | i_data[EXT4_TIND_BLOCK] = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2482 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2483 | case EXT4_TIND_BLOCK: |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2484 | ; |
| 2485 | } |
| 2486 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2487 | ext4_discard_reservation(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2488 | |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2489 | up_write(&ei->i_data_sem); |
Kalpak Shah | ef7f383 | 2007-07-18 09:15:20 -0400 | [diff] [blame] | 2490 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2491 | ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2492 | |
| 2493 | /* |
| 2494 | * In a multi-transaction truncate, we only make the final transaction |
| 2495 | * synchronous |
| 2496 | */ |
| 2497 | if (IS_SYNC(inode)) |
| 2498 | handle->h_sync = 1; |
| 2499 | out_stop: |
| 2500 | /* |
| 2501 | * If this was a simple ftruncate(), and the file will remain alive |
| 2502 | * then we need to clear up the orphan record which we created above. |
| 2503 | * However, if this was a real unlink then we were called by |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2504 | * ext4_delete_inode(), and we allow that function to clean up the |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2505 | * orphan info for us. |
| 2506 | */ |
| 2507 | if (inode->i_nlink) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2508 | ext4_orphan_del(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2509 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2510 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2511 | } |
| 2512 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2513 | static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb, |
| 2514 | unsigned long ino, struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2515 | { |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2516 | ext4_group_t block_group; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2517 | unsigned long offset; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2518 | ext4_fsblk_t block; |
Akinobu Mita | c0a4ef3 | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 2519 | struct ext4_group_desc *gdp; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2520 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2521 | if (!ext4_valid_inum(sb, ino)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2522 | /* |
| 2523 | * This error is already checked for in namei.c unless we are |
| 2524 | * looking at an NFS filehandle, in which case no error |
| 2525 | * report is needed |
| 2526 | */ |
| 2527 | return 0; |
| 2528 | } |
| 2529 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2530 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
Akinobu Mita | c0a4ef3 | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 2531 | gdp = ext4_get_group_desc(sb, block_group, NULL); |
| 2532 | if (!gdp) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2533 | return 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2534 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2535 | /* |
| 2536 | * Figure out the offset within the block group inode table |
| 2537 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2538 | offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) * |
| 2539 | EXT4_INODE_SIZE(sb); |
Alexandre Ratchov | 8fadc14 | 2006-10-11 01:21:15 -0700 | [diff] [blame] | 2540 | block = ext4_inode_table(sb, gdp) + |
| 2541 | (offset >> EXT4_BLOCK_SIZE_BITS(sb)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2542 | |
| 2543 | iloc->block_group = block_group; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2544 | iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2545 | return block; |
| 2546 | } |
| 2547 | |
| 2548 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2549 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2550 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
| 2551 | * data in memory that is needed to recreate the on-disk version of this |
| 2552 | * inode. |
| 2553 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2554 | static int __ext4_get_inode_loc(struct inode *inode, |
| 2555 | struct ext4_iloc *iloc, int in_mem) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2556 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2557 | ext4_fsblk_t block; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2558 | struct buffer_head *bh; |
| 2559 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2560 | block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2561 | if (!block) |
| 2562 | return -EIO; |
| 2563 | |
| 2564 | bh = sb_getblk(inode->i_sb, block); |
| 2565 | if (!bh) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2566 | ext4_error (inode->i_sb, "ext4_get_inode_loc", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2567 | "unable to read inode block - " |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 2568 | "inode=%lu, block=%llu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2569 | inode->i_ino, block); |
| 2570 | return -EIO; |
| 2571 | } |
| 2572 | if (!buffer_uptodate(bh)) { |
| 2573 | lock_buffer(bh); |
| 2574 | if (buffer_uptodate(bh)) { |
| 2575 | /* someone brought it uptodate while we waited */ |
| 2576 | unlock_buffer(bh); |
| 2577 | goto has_buffer; |
| 2578 | } |
| 2579 | |
| 2580 | /* |
| 2581 | * If we have all information of the inode in memory and this |
| 2582 | * is the only valid inode in the block, we need not read the |
| 2583 | * block. |
| 2584 | */ |
| 2585 | if (in_mem) { |
| 2586 | struct buffer_head *bitmap_bh; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2587 | struct ext4_group_desc *desc; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2588 | int inodes_per_buffer; |
| 2589 | int inode_offset, i; |
Avantika Mathur | fd2d429 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2590 | ext4_group_t block_group; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2591 | int start; |
| 2592 | |
| 2593 | block_group = (inode->i_ino - 1) / |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2594 | EXT4_INODES_PER_GROUP(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2595 | inodes_per_buffer = bh->b_size / |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2596 | EXT4_INODE_SIZE(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2597 | inode_offset = ((inode->i_ino - 1) % |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2598 | EXT4_INODES_PER_GROUP(inode->i_sb)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2599 | start = inode_offset & ~(inodes_per_buffer - 1); |
| 2600 | |
| 2601 | /* Is the inode bitmap in cache? */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2602 | desc = ext4_get_group_desc(inode->i_sb, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2603 | block_group, NULL); |
| 2604 | if (!desc) |
| 2605 | goto make_io; |
| 2606 | |
| 2607 | bitmap_bh = sb_getblk(inode->i_sb, |
Alexandre Ratchov | 8fadc14 | 2006-10-11 01:21:15 -0700 | [diff] [blame] | 2608 | ext4_inode_bitmap(inode->i_sb, desc)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2609 | if (!bitmap_bh) |
| 2610 | goto make_io; |
| 2611 | |
| 2612 | /* |
| 2613 | * If the inode bitmap isn't in cache then the |
| 2614 | * optimisation may end up performing two reads instead |
| 2615 | * of one, so skip it. |
| 2616 | */ |
| 2617 | if (!buffer_uptodate(bitmap_bh)) { |
| 2618 | brelse(bitmap_bh); |
| 2619 | goto make_io; |
| 2620 | } |
| 2621 | for (i = start; i < start + inodes_per_buffer; i++) { |
| 2622 | if (i == inode_offset) |
| 2623 | continue; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2624 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2625 | break; |
| 2626 | } |
| 2627 | brelse(bitmap_bh); |
| 2628 | if (i == start + inodes_per_buffer) { |
| 2629 | /* all other inodes are free, so skip I/O */ |
| 2630 | memset(bh->b_data, 0, bh->b_size); |
| 2631 | set_buffer_uptodate(bh); |
| 2632 | unlock_buffer(bh); |
| 2633 | goto has_buffer; |
| 2634 | } |
| 2635 | } |
| 2636 | |
| 2637 | make_io: |
| 2638 | /* |
| 2639 | * There are other valid inodes in the buffer, this inode |
| 2640 | * has in-inode xattrs, or we don't have this inode in memory. |
| 2641 | * Read the block from disk. |
| 2642 | */ |
| 2643 | get_bh(bh); |
| 2644 | bh->b_end_io = end_buffer_read_sync; |
| 2645 | submit_bh(READ_META, bh); |
| 2646 | wait_on_buffer(bh); |
| 2647 | if (!buffer_uptodate(bh)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2648 | ext4_error(inode->i_sb, "ext4_get_inode_loc", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2649 | "unable to read inode block - " |
Mingming Cao | 2ae0210 | 2006-10-11 01:21:11 -0700 | [diff] [blame] | 2650 | "inode=%lu, block=%llu", |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2651 | inode->i_ino, block); |
| 2652 | brelse(bh); |
| 2653 | return -EIO; |
| 2654 | } |
| 2655 | } |
| 2656 | has_buffer: |
| 2657 | iloc->bh = bh; |
| 2658 | return 0; |
| 2659 | } |
| 2660 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2661 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2662 | { |
| 2663 | /* We have all inode data except xattrs in memory here. */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2664 | return __ext4_get_inode_loc(inode, iloc, |
| 2665 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2666 | } |
| 2667 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2668 | void ext4_set_inode_flags(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2669 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2670 | unsigned int flags = EXT4_I(inode)->i_flags; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2671 | |
| 2672 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2673 | if (flags & EXT4_SYNC_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2674 | inode->i_flags |= S_SYNC; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2675 | if (flags & EXT4_APPEND_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2676 | inode->i_flags |= S_APPEND; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2677 | if (flags & EXT4_IMMUTABLE_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2678 | inode->i_flags |= S_IMMUTABLE; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2679 | if (flags & EXT4_NOATIME_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2680 | inode->i_flags |= S_NOATIME; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2681 | if (flags & EXT4_DIRSYNC_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2682 | inode->i_flags |= S_DIRSYNC; |
| 2683 | } |
| 2684 | |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 2685 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
| 2686 | void ext4_get_inode_flags(struct ext4_inode_info *ei) |
| 2687 | { |
| 2688 | unsigned int flags = ei->vfs_inode.i_flags; |
| 2689 | |
| 2690 | ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL| |
| 2691 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL); |
| 2692 | if (flags & S_SYNC) |
| 2693 | ei->i_flags |= EXT4_SYNC_FL; |
| 2694 | if (flags & S_APPEND) |
| 2695 | ei->i_flags |= EXT4_APPEND_FL; |
| 2696 | if (flags & S_IMMUTABLE) |
| 2697 | ei->i_flags |= EXT4_IMMUTABLE_FL; |
| 2698 | if (flags & S_NOATIME) |
| 2699 | ei->i_flags |= EXT4_NOATIME_FL; |
| 2700 | if (flags & S_DIRSYNC) |
| 2701 | ei->i_flags |= EXT4_DIRSYNC_FL; |
| 2702 | } |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2703 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
| 2704 | struct ext4_inode_info *ei) |
| 2705 | { |
| 2706 | blkcnt_t i_blocks ; |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2707 | struct inode *inode = &(ei->vfs_inode); |
| 2708 | struct super_block *sb = inode->i_sb; |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2709 | |
| 2710 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 2711 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { |
| 2712 | /* we are using combined 48 bit field */ |
| 2713 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | |
| 2714 | le32_to_cpu(raw_inode->i_blocks_lo); |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2715 | if (ei->i_flags & EXT4_HUGE_FILE_FL) { |
| 2716 | /* i_blocks represent file system block size */ |
| 2717 | return i_blocks << (inode->i_blkbits - 9); |
| 2718 | } else { |
| 2719 | return i_blocks; |
| 2720 | } |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2721 | } else { |
| 2722 | return le32_to_cpu(raw_inode->i_blocks_lo); |
| 2723 | } |
| 2724 | } |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 2725 | |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2726 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2727 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2728 | struct ext4_iloc iloc; |
| 2729 | struct ext4_inode *raw_inode; |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2730 | struct ext4_inode_info *ei; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2731 | struct buffer_head *bh; |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2732 | struct inode *inode; |
| 2733 | long ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2734 | int block; |
| 2735 | |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2736 | inode = iget_locked(sb, ino); |
| 2737 | if (!inode) |
| 2738 | return ERR_PTR(-ENOMEM); |
| 2739 | if (!(inode->i_state & I_NEW)) |
| 2740 | return inode; |
| 2741 | |
| 2742 | ei = EXT4_I(inode); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2743 | #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL |
| 2744 | ei->i_acl = EXT4_ACL_NOT_CACHED; |
| 2745 | ei->i_default_acl = EXT4_ACL_NOT_CACHED; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2746 | #endif |
| 2747 | ei->i_block_alloc_info = NULL; |
| 2748 | |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2749 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
| 2750 | if (ret < 0) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2751 | goto bad_inode; |
| 2752 | bh = iloc.bh; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2753 | raw_inode = ext4_raw_inode(&iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2754 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
| 2755 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 2756 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
| 2757 | if(!(test_opt (inode->i_sb, NO_UID32))) { |
| 2758 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 2759 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
| 2760 | } |
| 2761 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2762 | |
| 2763 | ei->i_state = 0; |
| 2764 | ei->i_dir_start_lookup = 0; |
| 2765 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 2766 | /* We now have enough fields to check if the inode was active or not. |
| 2767 | * This is needed because nfsd might try to access dead inodes |
| 2768 | * the test is that same one that e2fsck uses |
| 2769 | * NeilBrown 1999oct15 |
| 2770 | */ |
| 2771 | if (inode->i_nlink == 0) { |
| 2772 | if (inode->i_mode == 0 || |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2773 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2774 | /* this inode is deleted */ |
| 2775 | brelse (bh); |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2776 | ret = -ESTALE; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2777 | goto bad_inode; |
| 2778 | } |
| 2779 | /* The only unlinked inodes we let through here have |
| 2780 | * valid i_mode and are being read by the orphan |
| 2781 | * recovery code: that's fine, we're about to complete |
| 2782 | * the process of deleting those. */ |
| 2783 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2784 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2785 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
Aneesh Kumar K.V | 7973c0c | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2786 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
Mingming Cao | 9b8f1f0 | 2006-10-11 01:21:13 -0700 | [diff] [blame] | 2787 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
Aneesh Kumar K.V | a48380f | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2788 | cpu_to_le32(EXT4_OS_HURD)) { |
Badari Pulavarty | a1ddeb7 | 2006-10-11 01:21:09 -0700 | [diff] [blame] | 2789 | ei->i_file_acl |= |
| 2790 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2791 | } |
Aneesh Kumar K.V | a48380f | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2792 | inode->i_size = ext4_isize(raw_inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2793 | ei->i_disksize = inode->i_size; |
| 2794 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 2795 | ei->i_block_group = iloc.block_group; |
| 2796 | /* |
| 2797 | * NOTE! The in-memory inode i_data array is in little-endian order |
| 2798 | * even on big-endian machines: we do NOT byteswap the block numbers! |
| 2799 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2800 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2801 | ei->i_data[block] = raw_inode->i_block[block]; |
| 2802 | INIT_LIST_HEAD(&ei->i_orphan); |
| 2803 | |
Eric Sandeen | 0040d98 | 2008-02-05 22:36:43 -0500 | [diff] [blame] | 2804 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2805 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2806 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
Kirill Korotaev | e5d2861 | 2007-06-23 17:16:51 -0700 | [diff] [blame] | 2807 | EXT4_INODE_SIZE(inode->i_sb)) { |
| 2808 | brelse (bh); |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2809 | ret = -EIO; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2810 | goto bad_inode; |
Kirill Korotaev | e5d2861 | 2007-06-23 17:16:51 -0700 | [diff] [blame] | 2811 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2812 | if (ei->i_extra_isize == 0) { |
| 2813 | /* The extra space is currently unused. Use it. */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2814 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
| 2815 | EXT4_GOOD_OLD_INODE_SIZE; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2816 | } else { |
| 2817 | __le32 *magic = (void *)raw_inode + |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2818 | EXT4_GOOD_OLD_INODE_SIZE + |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2819 | ei->i_extra_isize; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2820 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
| 2821 | ei->i_state |= EXT4_STATE_XATTR; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2822 | } |
| 2823 | } else |
| 2824 | ei->i_extra_isize = 0; |
| 2825 | |
Kalpak Shah | ef7f383 | 2007-07-18 09:15:20 -0400 | [diff] [blame] | 2826 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
| 2827 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); |
| 2828 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); |
| 2829 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); |
| 2830 | |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2831 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
| 2832 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 2833 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 2834 | inode->i_version |= |
| 2835 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
| 2836 | } |
| 2837 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2838 | if (S_ISREG(inode->i_mode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2839 | inode->i_op = &ext4_file_inode_operations; |
| 2840 | inode->i_fop = &ext4_file_operations; |
| 2841 | ext4_set_aops(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2842 | } else if (S_ISDIR(inode->i_mode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2843 | inode->i_op = &ext4_dir_inode_operations; |
| 2844 | inode->i_fop = &ext4_dir_operations; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2845 | } else if (S_ISLNK(inode->i_mode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2846 | if (ext4_inode_is_fast_symlink(inode)) |
| 2847 | inode->i_op = &ext4_fast_symlink_inode_operations; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2848 | else { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2849 | inode->i_op = &ext4_symlink_inode_operations; |
| 2850 | ext4_set_aops(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2851 | } |
| 2852 | } else { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2853 | inode->i_op = &ext4_special_inode_operations; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2854 | if (raw_inode->i_block[0]) |
| 2855 | init_special_inode(inode, inode->i_mode, |
| 2856 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
| 2857 | else |
| 2858 | init_special_inode(inode, inode->i_mode, |
| 2859 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
| 2860 | } |
| 2861 | brelse (iloc.bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2862 | ext4_set_inode_flags(inode); |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2863 | unlock_new_inode(inode); |
| 2864 | return inode; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2865 | |
| 2866 | bad_inode: |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 2867 | iget_failed(inode); |
| 2868 | return ERR_PTR(ret); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2869 | } |
| 2870 | |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2871 | static int ext4_inode_blocks_set(handle_t *handle, |
| 2872 | struct ext4_inode *raw_inode, |
| 2873 | struct ext4_inode_info *ei) |
| 2874 | { |
| 2875 | struct inode *inode = &(ei->vfs_inode); |
| 2876 | u64 i_blocks = inode->i_blocks; |
| 2877 | struct super_block *sb = inode->i_sb; |
| 2878 | int err = 0; |
| 2879 | |
| 2880 | if (i_blocks <= ~0U) { |
| 2881 | /* |
| 2882 | * i_blocks can be represnted in a 32 bit variable |
| 2883 | * as multiple of 512 bytes |
| 2884 | */ |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2885 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2886 | raw_inode->i_blocks_high = 0; |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2887 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2888 | } else if (i_blocks <= 0xffffffffffffULL) { |
| 2889 | /* |
| 2890 | * i_blocks can be represented in a 48 bit variable |
| 2891 | * as multiple of 512 bytes |
| 2892 | */ |
| 2893 | err = ext4_update_rocompat_feature(handle, sb, |
| 2894 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE); |
| 2895 | if (err) |
| 2896 | goto err_out; |
| 2897 | /* i_block is stored in the split 48 bit fields */ |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2898 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2899 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2900 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2901 | } else { |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2902 | /* |
| 2903 | * i_blocks should be represented in a 48 bit variable |
| 2904 | * as multiple of file system block size |
| 2905 | */ |
| 2906 | err = ext4_update_rocompat_feature(handle, sb, |
| 2907 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE); |
| 2908 | if (err) |
| 2909 | goto err_out; |
| 2910 | ei->i_flags |= EXT4_HUGE_FILE_FL; |
| 2911 | /* i_block is stored in file system block size */ |
| 2912 | i_blocks = i_blocks >> (inode->i_blkbits - 9); |
| 2913 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 2914 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2915 | } |
| 2916 | err_out: |
| 2917 | return err; |
| 2918 | } |
| 2919 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2920 | /* |
| 2921 | * Post the struct inode info into an on-disk inode location in the |
| 2922 | * buffer-cache. This gobbles the caller's reference to the |
| 2923 | * buffer_head in the inode location struct. |
| 2924 | * |
| 2925 | * The caller must have write access to iloc->bh. |
| 2926 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2927 | static int ext4_do_update_inode(handle_t *handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2928 | struct inode *inode, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2929 | struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2930 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2931 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
| 2932 | struct ext4_inode_info *ei = EXT4_I(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2933 | struct buffer_head *bh = iloc->bh; |
| 2934 | int err = 0, rc, block; |
| 2935 | |
| 2936 | /* For fields not not tracking in the in-memory inode, |
| 2937 | * initialise them to zero for new inodes. */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2938 | if (ei->i_state & EXT4_STATE_NEW) |
| 2939 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2940 | |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 2941 | ext4_get_inode_flags(ei); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2942 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
| 2943 | if(!(test_opt(inode->i_sb, NO_UID32))) { |
| 2944 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
| 2945 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); |
| 2946 | /* |
| 2947 | * Fix up interoperability with old kernels. Otherwise, old inodes get |
| 2948 | * re-used with the upper 16 bits of the uid/gid intact |
| 2949 | */ |
| 2950 | if(!ei->i_dtime) { |
| 2951 | raw_inode->i_uid_high = |
| 2952 | cpu_to_le16(high_16_bits(inode->i_uid)); |
| 2953 | raw_inode->i_gid_high = |
| 2954 | cpu_to_le16(high_16_bits(inode->i_gid)); |
| 2955 | } else { |
| 2956 | raw_inode->i_uid_high = 0; |
| 2957 | raw_inode->i_gid_high = 0; |
| 2958 | } |
| 2959 | } else { |
| 2960 | raw_inode->i_uid_low = |
| 2961 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); |
| 2962 | raw_inode->i_gid_low = |
| 2963 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); |
| 2964 | raw_inode->i_uid_high = 0; |
| 2965 | raw_inode->i_gid_high = 0; |
| 2966 | } |
| 2967 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
Kalpak Shah | ef7f383 | 2007-07-18 09:15:20 -0400 | [diff] [blame] | 2968 | |
| 2969 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
| 2970 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); |
| 2971 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); |
| 2972 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); |
| 2973 | |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 2974 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
| 2975 | goto out_brelse; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2976 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
Aneesh Kumar K.V | 267e4db | 2008-04-29 08:11:12 -0400 | [diff] [blame] | 2977 | /* clear the migrate flag in the raw_inode */ |
| 2978 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE); |
Mingming Cao | 9b8f1f0 | 2006-10-11 01:21:13 -0700 | [diff] [blame] | 2979 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 2980 | cpu_to_le32(EXT4_OS_HURD)) |
Badari Pulavarty | a1ddeb7 | 2006-10-11 01:21:09 -0700 | [diff] [blame] | 2981 | raw_inode->i_file_acl_high = |
| 2982 | cpu_to_le16(ei->i_file_acl >> 32); |
Aneesh Kumar K.V | 7973c0c | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2983 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
Aneesh Kumar K.V | a48380f | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 2984 | ext4_isize_set(raw_inode, ei->i_disksize); |
| 2985 | if (ei->i_disksize > 0x7fffffffULL) { |
| 2986 | struct super_block *sb = inode->i_sb; |
| 2987 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 2988 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || |
| 2989 | EXT4_SB(sb)->s_es->s_rev_level == |
| 2990 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { |
| 2991 | /* If this is the first large file |
| 2992 | * created, add a flag to the superblock. |
| 2993 | */ |
| 2994 | err = ext4_journal_get_write_access(handle, |
| 2995 | EXT4_SB(sb)->s_sbh); |
| 2996 | if (err) |
| 2997 | goto out_brelse; |
| 2998 | ext4_update_dynamic_rev(sb); |
| 2999 | EXT4_SET_RO_COMPAT_FEATURE(sb, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3000 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
Aneesh Kumar K.V | a48380f | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 3001 | sb->s_dirt = 1; |
| 3002 | handle->h_sync = 1; |
| 3003 | err = ext4_journal_dirty_metadata(handle, |
| 3004 | EXT4_SB(sb)->s_sbh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3005 | } |
| 3006 | } |
| 3007 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 3008 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 3009 | if (old_valid_dev(inode->i_rdev)) { |
| 3010 | raw_inode->i_block[0] = |
| 3011 | cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 3012 | raw_inode->i_block[1] = 0; |
| 3013 | } else { |
| 3014 | raw_inode->i_block[0] = 0; |
| 3015 | raw_inode->i_block[1] = |
| 3016 | cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 3017 | raw_inode->i_block[2] = 0; |
| 3018 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3019 | } else for (block = 0; block < EXT4_N_BLOCKS; block++) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3020 | raw_inode->i_block[block] = ei->i_data[block]; |
| 3021 | |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 3022 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
| 3023 | if (ei->i_extra_isize) { |
| 3024 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 3025 | raw_inode->i_version_hi = |
| 3026 | cpu_to_le32(inode->i_version >> 32); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3027 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 3028 | } |
| 3029 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3030 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3031 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
| 3032 | rc = ext4_journal_dirty_metadata(handle, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3033 | if (!err) |
| 3034 | err = rc; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3035 | ei->i_state &= ~EXT4_STATE_NEW; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3036 | |
| 3037 | out_brelse: |
| 3038 | brelse (bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3039 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3040 | return err; |
| 3041 | } |
| 3042 | |
| 3043 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3044 | * ext4_write_inode() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3045 | * |
| 3046 | * We are called from a few places: |
| 3047 | * |
| 3048 | * - Within generic_file_write() for O_SYNC files. |
| 3049 | * Here, there will be no transaction running. We wait for any running |
| 3050 | * trasnaction to commit. |
| 3051 | * |
| 3052 | * - Within sys_sync(), kupdate and such. |
| 3053 | * We wait on commit, if tol to. |
| 3054 | * |
| 3055 | * - Within prune_icache() (PF_MEMALLOC == true) |
| 3056 | * Here we simply return. We can't afford to block kswapd on the |
| 3057 | * journal commit. |
| 3058 | * |
| 3059 | * In all cases it is actually safe for us to return without doing anything, |
| 3060 | * because the inode has been copied into a raw inode buffer in |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3061 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3062 | * knfsd. |
| 3063 | * |
| 3064 | * Note that we are absolutely dependent upon all inode dirtiers doing the |
| 3065 | * right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 3066 | * which we are interested. |
| 3067 | * |
| 3068 | * It would be a bug for them to not do this. The code: |
| 3069 | * |
| 3070 | * mark_inode_dirty(inode) |
| 3071 | * stuff(); |
| 3072 | * inode->i_size = expr; |
| 3073 | * |
| 3074 | * is in error because a kswapd-driven write_inode() could occur while |
| 3075 | * `stuff()' is running, and the new i_size will be lost. Plus the inode |
| 3076 | * will no longer be on the superblock's dirty inode list. |
| 3077 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3078 | int ext4_write_inode(struct inode *inode, int wait) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3079 | { |
| 3080 | if (current->flags & PF_MEMALLOC) |
| 3081 | return 0; |
| 3082 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3083 | if (ext4_journal_current_handle()) { |
Mingming Cao | b38bd33 | 2007-07-19 01:48:35 -0700 | [diff] [blame] | 3084 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3085 | dump_stack(); |
| 3086 | return -EIO; |
| 3087 | } |
| 3088 | |
| 3089 | if (!wait) |
| 3090 | return 0; |
| 3091 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3092 | return ext4_force_commit(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3093 | } |
| 3094 | |
| 3095 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3096 | * ext4_setattr() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3097 | * |
| 3098 | * Called from notify_change. |
| 3099 | * |
| 3100 | * We want to trap VFS attempts to truncate the file as soon as |
| 3101 | * possible. In particular, we want to make sure that when the VFS |
| 3102 | * shrinks i_size, we put the inode on the orphan list and modify |
| 3103 | * i_disksize immediately, so that during the subsequent flushing of |
| 3104 | * dirty pages and freeing of disk blocks, we can guarantee that any |
| 3105 | * commit will leave the blocks being flushed in an unused state on |
| 3106 | * disk. (On recovery, the inode will get truncated and the blocks will |
| 3107 | * be freed, so we have a strong guarantee that no future commit will |
| 3108 | * leave these blocks visible to the user.) |
| 3109 | * |
| 3110 | * Called with inode->sem down. |
| 3111 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3112 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3113 | { |
| 3114 | struct inode *inode = dentry->d_inode; |
| 3115 | int error, rc = 0; |
| 3116 | const unsigned int ia_valid = attr->ia_valid; |
| 3117 | |
| 3118 | error = inode_change_ok(inode, attr); |
| 3119 | if (error) |
| 3120 | return error; |
| 3121 | |
| 3122 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
| 3123 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { |
| 3124 | handle_t *handle; |
| 3125 | |
| 3126 | /* (user+group)*(old+new) structure, inode write (sb, |
| 3127 | * inode block, ? - but truncate inode update has it) */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3128 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
| 3129 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3130 | if (IS_ERR(handle)) { |
| 3131 | error = PTR_ERR(handle); |
| 3132 | goto err_out; |
| 3133 | } |
| 3134 | error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0; |
| 3135 | if (error) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3136 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3137 | return error; |
| 3138 | } |
| 3139 | /* Update corresponding info in inode so that everything is in |
| 3140 | * one transaction */ |
| 3141 | if (attr->ia_valid & ATTR_UID) |
| 3142 | inode->i_uid = attr->ia_uid; |
| 3143 | if (attr->ia_valid & ATTR_GID) |
| 3144 | inode->i_gid = attr->ia_gid; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3145 | error = ext4_mark_inode_dirty(handle, inode); |
| 3146 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3147 | } |
| 3148 | |
Eric Sandeen | e2b4657 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 3149 | if (attr->ia_valid & ATTR_SIZE) { |
| 3150 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { |
| 3151 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 3152 | |
| 3153 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
| 3154 | error = -EFBIG; |
| 3155 | goto err_out; |
| 3156 | } |
| 3157 | } |
| 3158 | } |
| 3159 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3160 | if (S_ISREG(inode->i_mode) && |
| 3161 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { |
| 3162 | handle_t *handle; |
| 3163 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3164 | handle = ext4_journal_start(inode, 3); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3165 | if (IS_ERR(handle)) { |
| 3166 | error = PTR_ERR(handle); |
| 3167 | goto err_out; |
| 3168 | } |
| 3169 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3170 | error = ext4_orphan_add(handle, inode); |
| 3171 | EXT4_I(inode)->i_disksize = attr->ia_size; |
| 3172 | rc = ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3173 | if (!error) |
| 3174 | error = rc; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3175 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3176 | } |
| 3177 | |
| 3178 | rc = inode_setattr(inode, attr); |
| 3179 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3180 | /* If inode_setattr's call to ext4_truncate failed to get a |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3181 | * transaction handle at all, we need to clean up the in-core |
| 3182 | * orphan list manually. */ |
| 3183 | if (inode->i_nlink) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3184 | ext4_orphan_del(NULL, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3185 | |
| 3186 | if (!rc && (ia_valid & ATTR_MODE)) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3187 | rc = ext4_acl_chmod(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3188 | |
| 3189 | err_out: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3190 | ext4_std_error(inode->i_sb, error); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3191 | if (!error) |
| 3192 | error = rc; |
| 3193 | return error; |
| 3194 | } |
| 3195 | |
| 3196 | |
| 3197 | /* |
| 3198 | * How many blocks doth make a writepage()? |
| 3199 | * |
| 3200 | * With N blocks per page, it may be: |
| 3201 | * N data blocks |
| 3202 | * 2 indirect block |
| 3203 | * 2 dindirect |
| 3204 | * 1 tindirect |
| 3205 | * N+5 bitmap blocks (from the above) |
| 3206 | * N+5 group descriptor summary blocks |
| 3207 | * 1 inode block |
| 3208 | * 1 superblock. |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3209 | * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3210 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3211 | * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3212 | * |
| 3213 | * With ordered or writeback data it's the same, less the N data blocks. |
| 3214 | * |
| 3215 | * If the inode's direct blocks can hold an integral number of pages then a |
| 3216 | * page cannot straddle two indirect blocks, and we can only touch one indirect |
| 3217 | * and dindirect block, and the "5" above becomes "3". |
| 3218 | * |
| 3219 | * This still overestimates under most circumstances. If we were to pass the |
| 3220 | * start and end offsets in here as well we could do block_to_path() on each |
| 3221 | * block and work out the exact number of indirects which are touched. Pah. |
| 3222 | */ |
| 3223 | |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 3224 | int ext4_writepage_trans_blocks(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3225 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3226 | int bpp = ext4_journal_blocks_per_page(inode); |
| 3227 | int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3228 | int ret; |
| 3229 | |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 3230 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
| 3231 | return ext4_ext_writepage_trans_blocks(inode, bpp); |
| 3232 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3233 | if (ext4_should_journal_data(inode)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3234 | ret = 3 * (bpp + indirects) + 2; |
| 3235 | else |
| 3236 | ret = 2 * (bpp + indirects) + 2; |
| 3237 | |
| 3238 | #ifdef CONFIG_QUOTA |
| 3239 | /* We know that structure was already allocated during DQUOT_INIT so |
| 3240 | * we will be updating only the data blocks + inodes */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3241 | ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3242 | #endif |
| 3243 | |
| 3244 | return ret; |
| 3245 | } |
| 3246 | |
| 3247 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3248 | * The caller must have previously called ext4_reserve_inode_write(). |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3249 | * Give this, we know that the caller already has write access to iloc->bh. |
| 3250 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3251 | int ext4_mark_iloc_dirty(handle_t *handle, |
| 3252 | struct inode *inode, struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3253 | { |
| 3254 | int err = 0; |
| 3255 | |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 3256 | if (test_opt(inode->i_sb, I_VERSION)) |
| 3257 | inode_inc_iversion(inode); |
| 3258 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3259 | /* the do_update_inode consumes one bh->b_count */ |
| 3260 | get_bh(iloc->bh); |
| 3261 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 3262 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3263 | err = ext4_do_update_inode(handle, inode, iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3264 | put_bh(iloc->bh); |
| 3265 | return err; |
| 3266 | } |
| 3267 | |
| 3268 | /* |
| 3269 | * On success, We end up with an outstanding reference count against |
| 3270 | * iloc->bh. This _must_ be cleaned up later. |
| 3271 | */ |
| 3272 | |
| 3273 | int |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3274 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
| 3275 | struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3276 | { |
| 3277 | int err = 0; |
| 3278 | if (handle) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3279 | err = ext4_get_inode_loc(inode, iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3280 | if (!err) { |
| 3281 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3282 | err = ext4_journal_get_write_access(handle, iloc->bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3283 | if (err) { |
| 3284 | brelse(iloc->bh); |
| 3285 | iloc->bh = NULL; |
| 3286 | } |
| 3287 | } |
| 3288 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3289 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3290 | return err; |
| 3291 | } |
| 3292 | |
| 3293 | /* |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 3294 | * Expand an inode by new_extra_isize bytes. |
| 3295 | * Returns 0 on success or negative error number on failure. |
| 3296 | */ |
Aneesh Kumar K.V | 1d03ec9 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 3297 | static int ext4_expand_extra_isize(struct inode *inode, |
| 3298 | unsigned int new_extra_isize, |
| 3299 | struct ext4_iloc iloc, |
| 3300 | handle_t *handle) |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 3301 | { |
| 3302 | struct ext4_inode *raw_inode; |
| 3303 | struct ext4_xattr_ibody_header *header; |
| 3304 | struct ext4_xattr_entry *entry; |
| 3305 | |
| 3306 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) |
| 3307 | return 0; |
| 3308 | |
| 3309 | raw_inode = ext4_raw_inode(&iloc); |
| 3310 | |
| 3311 | header = IHDR(inode, raw_inode); |
| 3312 | entry = IFIRST(header); |
| 3313 | |
| 3314 | /* No extended attributes present */ |
| 3315 | if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) || |
| 3316 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 3317 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
| 3318 | new_extra_isize); |
| 3319 | EXT4_I(inode)->i_extra_isize = new_extra_isize; |
| 3320 | return 0; |
| 3321 | } |
| 3322 | |
| 3323 | /* try to expand with EAs present */ |
| 3324 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, |
| 3325 | raw_inode, handle); |
| 3326 | } |
| 3327 | |
| 3328 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3329 | * What we do here is to mark the in-core inode as clean with respect to inode |
| 3330 | * dirtiness (it may still be data-dirty). |
| 3331 | * This means that the in-core inode may be reaped by prune_icache |
| 3332 | * without having to perform any I/O. This is a very good thing, |
| 3333 | * because *any* task may call prune_icache - even ones which |
| 3334 | * have a transaction open against a different journal. |
| 3335 | * |
| 3336 | * Is this cheating? Not really. Sure, we haven't written the |
| 3337 | * inode out, but prune_icache isn't a user-visible syncing function. |
| 3338 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 3339 | * we start and wait on commits. |
| 3340 | * |
| 3341 | * Is this efficient/effective? Well, we're being nice to the system |
| 3342 | * by cleaning up our inodes proactively so they can be reaped |
| 3343 | * without I/O. But we are potentially leaving up to five seconds' |
| 3344 | * worth of inodes floating about which prune_icache wants us to |
| 3345 | * write out. One way to fix that would be to get prune_icache() |
| 3346 | * to do a write_super() to free up some memory. It has the desired |
| 3347 | * effect. |
| 3348 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3349 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3350 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3351 | struct ext4_iloc iloc; |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 3352 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 3353 | static unsigned int mnt_count; |
| 3354 | int err, ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3355 | |
| 3356 | might_sleep(); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3357 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 3358 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
| 3359 | !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { |
| 3360 | /* |
| 3361 | * We need extra buffer credits since we may write into EA block |
| 3362 | * with this same handle. If journal_extend fails, then it will |
| 3363 | * only result in a minor loss of functionality for that inode. |
| 3364 | * If this is felt to be critical, then e2fsck should be run to |
| 3365 | * force a large enough s_min_extra_isize. |
| 3366 | */ |
| 3367 | if ((jbd2_journal_extend(handle, |
| 3368 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { |
| 3369 | ret = ext4_expand_extra_isize(inode, |
| 3370 | sbi->s_want_extra_isize, |
| 3371 | iloc, handle); |
| 3372 | if (ret) { |
| 3373 | EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND; |
Aneesh Kumar K.V | c1bddad | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 3374 | if (mnt_count != |
| 3375 | le16_to_cpu(sbi->s_es->s_mnt_count)) { |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 3376 | ext4_warning(inode->i_sb, __func__, |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 3377 | "Unable to expand inode %lu. Delete" |
| 3378 | " some EAs or run e2fsck.", |
| 3379 | inode->i_ino); |
Aneesh Kumar K.V | c1bddad | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 3380 | mnt_count = |
| 3381 | le16_to_cpu(sbi->s_es->s_mnt_count); |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 3382 | } |
| 3383 | } |
| 3384 | } |
| 3385 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3386 | if (!err) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3387 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3388 | return err; |
| 3389 | } |
| 3390 | |
| 3391 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3392 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3393 | * |
| 3394 | * We're really interested in the case where a file is being extended. |
| 3395 | * i_size has been changed by generic_commit_write() and we thus need |
| 3396 | * to include the updated inode in the current transaction. |
| 3397 | * |
| 3398 | * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks |
| 3399 | * are allocated to the file. |
| 3400 | * |
| 3401 | * If the inode is marked synchronous, we don't honour that here - doing |
| 3402 | * so would cause a commit on atime updates, which we don't bother doing. |
| 3403 | * We handle synchronous inodes at the highest possible level. |
| 3404 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3405 | void ext4_dirty_inode(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3406 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3407 | handle_t *current_handle = ext4_journal_current_handle(); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3408 | handle_t *handle; |
| 3409 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3410 | handle = ext4_journal_start(inode, 2); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3411 | if (IS_ERR(handle)) |
| 3412 | goto out; |
| 3413 | if (current_handle && |
| 3414 | current_handle->h_transaction != handle->h_transaction) { |
| 3415 | /* This task has a transaction open against a different fs */ |
| 3416 | printk(KERN_EMERG "%s: transactions do not match!\n", |
Harvey Harrison | 46e665e | 2008-04-17 10:38:59 -0400 | [diff] [blame] | 3417 | __func__); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3418 | } else { |
| 3419 | jbd_debug(5, "marking dirty. outer handle=%p\n", |
| 3420 | current_handle); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3421 | ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3422 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3423 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3424 | out: |
| 3425 | return; |
| 3426 | } |
| 3427 | |
| 3428 | #if 0 |
| 3429 | /* |
| 3430 | * Bind an inode's backing buffer_head into this transaction, to prevent |
| 3431 | * it from being flushed to disk early. Unlike |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3432 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3433 | * returns no iloc structure, so the caller needs to repeat the iloc |
| 3434 | * lookup to mark the inode dirty later. |
| 3435 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3436 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3437 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3438 | struct ext4_iloc iloc; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3439 | |
| 3440 | int err = 0; |
| 3441 | if (handle) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3442 | err = ext4_get_inode_loc(inode, &iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3443 | if (!err) { |
| 3444 | BUFFER_TRACE(iloc.bh, "get_write_access"); |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 3445 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3446 | if (!err) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3447 | err = ext4_journal_dirty_metadata(handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3448 | iloc.bh); |
| 3449 | brelse(iloc.bh); |
| 3450 | } |
| 3451 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3452 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3453 | return err; |
| 3454 | } |
| 3455 | #endif |
| 3456 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3457 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3458 | { |
| 3459 | journal_t *journal; |
| 3460 | handle_t *handle; |
| 3461 | int err; |
| 3462 | |
| 3463 | /* |
| 3464 | * We have to be very careful here: changing a data block's |
| 3465 | * journaling status dynamically is dangerous. If we write a |
| 3466 | * data block to the journal, change the status and then delete |
| 3467 | * that block, we risk forgetting to revoke the old log record |
| 3468 | * from the journal and so a subsequent replay can corrupt data. |
| 3469 | * So, first we make sure that the journal is empty and that |
| 3470 | * nobody is changing anything. |
| 3471 | */ |
| 3472 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3473 | journal = EXT4_JOURNAL(inode); |
Dave Hansen | d699594 | 2007-07-18 08:33:51 -0400 | [diff] [blame] | 3474 | if (is_journal_aborted(journal)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3475 | return -EROFS; |
| 3476 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 3477 | jbd2_journal_lock_updates(journal); |
| 3478 | jbd2_journal_flush(journal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3479 | |
| 3480 | /* |
| 3481 | * OK, there are no updates running now, and all cached data is |
| 3482 | * synced to disk. We are now in a completely consistent state |
| 3483 | * which doesn't have anything in the journal, and we know that |
| 3484 | * no filesystem updates are running, so it is safe to modify |
| 3485 | * the inode's in-core data-journaling state flag now. |
| 3486 | */ |
| 3487 | |
| 3488 | if (val) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3489 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3490 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3491 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
| 3492 | ext4_set_aops(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3493 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 3494 | jbd2_journal_unlock_updates(journal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3495 | |
| 3496 | /* Finally we can mark the inode as dirty. */ |
| 3497 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3498 | handle = ext4_journal_start(inode, 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3499 | if (IS_ERR(handle)) |
| 3500 | return PTR_ERR(handle); |
| 3501 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3502 | err = ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3503 | handle->h_sync = 1; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3504 | ext4_journal_stop(handle); |
| 3505 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3506 | |
| 3507 | return err; |
| 3508 | } |