blob: 29ed682061f6453341c48a0c0f5aab34ac42cd89 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext2/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@dcs.ed.ac.uk), 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 *
22 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
23 */
24
25#include <linux/smp_lock.h>
26#include <linux/time.h>
27#include <linux/highuid.h>
28#include <linux/pagemap.h>
29#include <linux/quotaops.h>
30#include <linux/module.h>
31#include <linux/writeback.h>
32#include <linux/buffer_head.h>
33#include <linux/mpage.h>
Josef Bacik68c9d702008-10-03 17:32:43 -040034#include <linux/fiemap.h>
Duane Griffin8d6d0c42008-12-19 20:47:13 +000035#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "ext2.h"
37#include "acl.h"
Carsten Otte6d791252005-06-23 22:05:26 -070038#include "xip.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40MODULE_AUTHOR("Remy Card and others");
41MODULE_DESCRIPTION("Second Extended Filesystem");
42MODULE_LICENSE("GPL");
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * Test whether an inode is a fast symlink.
46 */
47static inline int ext2_inode_is_fast_symlink(struct inode *inode)
48{
49 int ea_blocks = EXT2_I(inode)->i_file_acl ?
50 (inode->i_sb->s_blocksize >> 9) : 0;
51
52 return (S_ISLNK(inode->i_mode) &&
53 inode->i_blocks - ea_blocks == 0);
54}
55
56/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 * Called at the last iput() if i_nlink is zero.
58 */
59void ext2_delete_inode (struct inode * inode)
60{
Mark Fashehfef26652005-09-09 13:01:31 -070061 truncate_inode_pages(&inode->i_data, 0);
62
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 if (is_bad_inode(inode))
64 goto no_delete;
65 EXT2_I(inode)->i_dtime = get_seconds();
66 mark_inode_dirty(inode);
Al Viroe1740a42009-06-07 15:14:02 -040067 ext2_write_inode(inode, inode_needs_sync(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69 inode->i_size = 0;
70 if (inode->i_blocks)
71 ext2_truncate (inode);
72 ext2_free_inode (inode);
73
74 return;
75no_delete:
76 clear_inode(inode); /* We must guarantee clearing of inode... */
77}
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079typedef struct {
80 __le32 *p;
81 __le32 key;
82 struct buffer_head *bh;
83} Indirect;
84
85static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
86{
87 p->key = *(p->p = v);
88 p->bh = bh;
89}
90
91static inline int verify_chain(Indirect *from, Indirect *to)
92{
93 while (from <= to && from->key == *from->p)
94 from++;
95 return (from > to);
96}
97
98/**
99 * ext2_block_to_path - parse the block number into array of offsets
100 * @inode: inode in question (we are only interested in its superblock)
101 * @i_block: block number to be parsed
102 * @offsets: array to store the offsets in
103 * @boundary: set this non-zero if the referred-to block is likely to be
104 * followed (on disk) by an indirect block.
105 * To store the locations of file's data ext2 uses a data structure common
106 * for UNIX filesystems - tree of pointers anchored in the inode, with
107 * data blocks at leaves and indirect blocks in intermediate nodes.
108 * This function translates the block number into path in that tree -
109 * return value is the path length and @offsets[n] is the offset of
110 * pointer to (n+1)th node in the nth one. If @block is out of range
111 * (negative or too large) warning is printed and zero returned.
112 *
113 * Note: function doesn't find node addresses, so no IO is needed. All
114 * we need to know is the capacity of indirect blocks (taken from the
115 * inode->i_sb).
116 */
117
118/*
119 * Portability note: the last comparison (check that we fit into triple
120 * indirect block) is spelled differently, because otherwise on an
121 * architecture with 32-bit longs and 8Kb pages we might get into trouble
122 * if our filesystem had 8Kb blocks. We might use long long, but that would
123 * kill us on x86. Oh, well, at least the sign propagation does not matter -
124 * i_block would have to be negative in the very beginning, so we would not
125 * get there at all.
126 */
127
128static int ext2_block_to_path(struct inode *inode,
129 long i_block, int offsets[4], int *boundary)
130{
131 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
132 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
133 const long direct_blocks = EXT2_NDIR_BLOCKS,
134 indirect_blocks = ptrs,
135 double_blocks = (1 << (ptrs_bits * 2));
136 int n = 0;
137 int final = 0;
138
139 if (i_block < 0) {
140 ext2_warning (inode->i_sb, "ext2_block_to_path", "block < 0");
141 } else if (i_block < direct_blocks) {
142 offsets[n++] = i_block;
143 final = direct_blocks;
144 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
145 offsets[n++] = EXT2_IND_BLOCK;
146 offsets[n++] = i_block;
147 final = ptrs;
148 } else if ((i_block -= indirect_blocks) < double_blocks) {
149 offsets[n++] = EXT2_DIND_BLOCK;
150 offsets[n++] = i_block >> ptrs_bits;
151 offsets[n++] = i_block & (ptrs - 1);
152 final = ptrs;
153 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
154 offsets[n++] = EXT2_TIND_BLOCK;
155 offsets[n++] = i_block >> (ptrs_bits * 2);
156 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
157 offsets[n++] = i_block & (ptrs - 1);
158 final = ptrs;
159 } else {
160 ext2_warning (inode->i_sb, "ext2_block_to_path", "block > big");
161 }
162 if (boundary)
Martin J. Bligha686cd82007-10-16 23:30:46 -0700163 *boundary = final - 1 - (i_block & (ptrs - 1));
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 return n;
166}
167
168/**
169 * ext2_get_branch - read the chain of indirect blocks leading to data
170 * @inode: inode in question
171 * @depth: depth of the chain (1 - direct pointer, etc.)
172 * @offsets: offsets of pointers in inode/indirect blocks
173 * @chain: place to store the result
174 * @err: here we store the error value
175 *
176 * Function fills the array of triples <key, p, bh> and returns %NULL
177 * if everything went OK or the pointer to the last filled triple
178 * (incomplete one) otherwise. Upon the return chain[i].key contains
179 * the number of (i+1)-th block in the chain (as it is stored in memory,
180 * i.e. little-endian 32-bit), chain[i].p contains the address of that
181 * number (it points into struct inode for i==0 and into the bh->b_data
182 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
183 * block for i>0 and NULL for i==0. In other words, it holds the block
184 * numbers of the chain, addresses they were taken from (and where we can
185 * verify that chain did not change) and buffer_heads hosting these
186 * numbers.
187 *
188 * Function stops when it stumbles upon zero pointer (absent block)
189 * (pointer to last triple returned, *@err == 0)
190 * or when it gets an IO error reading an indirect block
191 * (ditto, *@err == -EIO)
192 * or when it notices that chain had been changed while it was reading
193 * (ditto, *@err == -EAGAIN)
194 * or when it reads all @depth-1 indirect blocks successfully and finds
195 * the whole chain, all way to the data (returns %NULL, *err == 0).
196 */
197static Indirect *ext2_get_branch(struct inode *inode,
198 int depth,
199 int *offsets,
200 Indirect chain[4],
201 int *err)
202{
203 struct super_block *sb = inode->i_sb;
204 Indirect *p = chain;
205 struct buffer_head *bh;
206
207 *err = 0;
208 /* i_data is not going away, no lock needed */
209 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
210 if (!p->key)
211 goto no_block;
212 while (--depth) {
213 bh = sb_bread(sb, le32_to_cpu(p->key));
214 if (!bh)
215 goto failure;
216 read_lock(&EXT2_I(inode)->i_meta_lock);
217 if (!verify_chain(chain, p))
218 goto changed;
219 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
220 read_unlock(&EXT2_I(inode)->i_meta_lock);
221 if (!p->key)
222 goto no_block;
223 }
224 return NULL;
225
226changed:
227 read_unlock(&EXT2_I(inode)->i_meta_lock);
228 brelse(bh);
229 *err = -EAGAIN;
230 goto no_block;
231failure:
232 *err = -EIO;
233no_block:
234 return p;
235}
236
237/**
238 * ext2_find_near - find a place for allocation with sufficient locality
239 * @inode: owner
240 * @ind: descriptor of indirect block.
241 *
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000242 * This function returns the preferred place for block allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 * It is used when heuristic for sequential allocation fails.
244 * Rules are:
245 * + if there is a block to the left of our position - allocate near it.
246 * + if pointer will live in indirect block - allocate near that block.
247 * + if pointer will live in inode - allocate in the same cylinder group.
248 *
249 * In the latter case we colour the starting block by the callers PID to
250 * prevent it from clashing with concurrent allocations for a different inode
251 * in the same block group. The PID is used here so that functionally related
252 * files will be close-by on-disk.
253 *
254 * Caller must make sure that @ind is valid and will stay that way.
255 */
256
Akinobu Mita4c8b3122008-04-28 02:16:02 -0700257static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
259 struct ext2_inode_info *ei = EXT2_I(inode);
260 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
261 __le32 *p;
Akinobu Mita4c8b3122008-04-28 02:16:02 -0700262 ext2_fsblk_t bg_start;
263 ext2_fsblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265 /* Try to find previous block */
266 for (p = ind->p - 1; p >= start; p--)
267 if (*p)
268 return le32_to_cpu(*p);
269
270 /* No such thing, so let's try location of indirect block */
271 if (ind->bh)
272 return ind->bh->b_blocknr;
273
274 /*
275 * It is going to be refered from inode itself? OK, just put it into
276 * the same cylinder group then.
277 */
Akinobu Mita24097d12008-04-28 02:16:01 -0700278 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 colour = (current->pid % 16) *
280 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
281 return bg_start + colour;
282}
283
284/**
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000285 * ext2_find_goal - find a preferred place for allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 * @inode: owner
287 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 *
Martin J. Bligha686cd82007-10-16 23:30:46 -0700290 * Returns preferred place for a block (the goal).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 */
292
Akinobu Mita4c8b3122008-04-28 02:16:02 -0700293static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
294 Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
Martin J. Bligha686cd82007-10-16 23:30:46 -0700296 struct ext2_block_alloc_info *block_i;
297
298 block_i = EXT2_I(inode)->i_block_alloc_info;
299
300 /*
301 * try the heuristic for sequential allocation,
302 * failing that at least try to get decent locality.
303 */
304 if (block_i && (block == block_i->last_alloc_logical_block + 1)
305 && (block_i->last_alloc_physical_block != 0)) {
306 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 }
Martin J. Bligha686cd82007-10-16 23:30:46 -0700308
309 return ext2_find_near(inode, partial);
310}
311
312/**
313 * ext2_blks_to_allocate: Look up the block map and count the number
314 * of direct blocks need to be allocated for the given branch.
315 *
316 * @branch: chain of indirect blocks
317 * @k: number of blocks need for indirect blocks
318 * @blks: number of data blocks to be mapped.
319 * @blocks_to_boundary: the offset in the indirect block
320 *
321 * return the total number of blocks to be allocate, including the
322 * direct and indirect blocks.
323 */
324static int
325ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
326 int blocks_to_boundary)
327{
328 unsigned long count = 0;
329
330 /*
331 * Simple case, [t,d]Indirect block(s) has not allocated yet
332 * then it's clear blocks on that path have not allocated
333 */
334 if (k > 0) {
335 /* right now don't hanel cross boundary allocation */
336 if (blks < blocks_to_boundary + 1)
337 count += blks;
338 else
339 count += blocks_to_boundary + 1;
340 return count;
341 }
342
343 count++;
344 while (count < blks && count <= blocks_to_boundary
345 && le32_to_cpu(*(branch[0].p + count)) == 0) {
346 count++;
347 }
348 return count;
349}
350
351/**
352 * ext2_alloc_blocks: multiple allocate blocks needed for a branch
353 * @indirect_blks: the number of blocks need to allocate for indirect
354 * blocks
355 *
356 * @new_blocks: on return it will store the new block numbers for
357 * the indirect blocks(if needed) and the first direct block,
358 * @blks: on return it will store the total number of allocated
359 * direct blocks
360 */
361static int ext2_alloc_blocks(struct inode *inode,
362 ext2_fsblk_t goal, int indirect_blks, int blks,
363 ext2_fsblk_t new_blocks[4], int *err)
364{
365 int target, i;
366 unsigned long count = 0;
367 int index = 0;
368 ext2_fsblk_t current_block = 0;
369 int ret = 0;
370
371 /*
372 * Here we try to allocate the requested multiple blocks at once,
373 * on a best-effort basis.
374 * To build a branch, we should allocate blocks for
375 * the indirect blocks(if not allocated yet), and at least
376 * the first direct block of this branch. That's the
377 * minimum number of blocks need to allocate(required)
378 */
379 target = blks + indirect_blks;
380
381 while (1) {
382 count = target;
383 /* allocating blocks for indirect blocks and direct blocks */
384 current_block = ext2_new_blocks(inode,goal,&count,err);
385 if (*err)
386 goto failed_out;
387
388 target -= count;
389 /* allocate blocks for indirect blocks */
390 while (index < indirect_blks && count) {
391 new_blocks[index++] = current_block++;
392 count--;
393 }
394
395 if (count > 0)
396 break;
397 }
398
399 /* save the new block number for the first direct block */
400 new_blocks[index] = current_block;
401
402 /* total number of blocks allocated for direct blocks */
403 ret = count;
404 *err = 0;
405 return ret;
406failed_out:
407 for (i = 0; i <index; i++)
408 ext2_free_blocks(inode, new_blocks[i], 1);
409 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410}
411
412/**
413 * ext2_alloc_branch - allocate and set up a chain of blocks.
414 * @inode: owner
415 * @num: depth of the chain (number of blocks to allocate)
416 * @offsets: offsets (in the blocks) to store the pointers to next.
417 * @branch: place to store the chain in.
418 *
419 * This function allocates @num blocks, zeroes out all but the last one,
420 * links them into chain and (if we are synchronous) writes them to disk.
421 * In other words, it prepares a branch that can be spliced onto the
422 * inode. It stores the information about that chain in the branch[], in
423 * the same format as ext2_get_branch() would do. We are calling it after
424 * we had read the existing part of chain and partial points to the last
425 * triple of that (one with zero ->key). Upon the exit we have the same
426 * picture as after the successful ext2_get_block(), excpet that in one
427 * place chain is disconnected - *branch->p is still zero (we did not
428 * set the last link), but branch->key contains the number that should
429 * be placed into *branch->p to fill that gap.
430 *
431 * If allocation fails we free all blocks we've allocated (and forget
432 * their buffer_heads) and return the error value the from failed
433 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
434 * as described above and return 0.
435 */
436
437static int ext2_alloc_branch(struct inode *inode,
Martin J. Bligha686cd82007-10-16 23:30:46 -0700438 int indirect_blks, int *blks, ext2_fsblk_t goal,
439 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 int blocksize = inode->i_sb->s_blocksize;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700442 int i, n = 0;
443 int err = 0;
444 struct buffer_head *bh;
445 int num;
446 ext2_fsblk_t new_blocks[4];
447 ext2_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Martin J. Bligha686cd82007-10-16 23:30:46 -0700449 num = ext2_alloc_blocks(inode, goal, indirect_blks,
450 *blks, new_blocks, &err);
451 if (err)
452 return err;
453
454 branch[0].key = cpu_to_le32(new_blocks[0]);
455 /*
456 * metadata blocks and data blocks are allocated.
457 */
458 for (n = 1; n <= indirect_blks; n++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700460 * Get buffer_head for parent block, zero it out
461 * and set the pointer to new one, then send
462 * parent to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700464 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
465 branch[n].bh = bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 lock_buffer(bh);
467 memset(bh->b_data, 0, blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 branch[n].p = (__le32 *) bh->b_data + offsets[n];
Martin J. Bligha686cd82007-10-16 23:30:46 -0700469 branch[n].key = cpu_to_le32(new_blocks[n]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 *branch[n].p = branch[n].key;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700471 if ( n == indirect_blks) {
472 current_block = new_blocks[n];
473 /*
474 * End of chain, update the last new metablock of
475 * the chain to point to the new allocated
476 * data blocks numbers
477 */
478 for (i=1; i < num; i++)
479 *(branch[n].p + i) = cpu_to_le32(++current_block);
480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 set_buffer_uptodate(bh);
482 unlock_buffer(bh);
483 mark_buffer_dirty_inode(bh, inode);
484 /* We used to sync bh here if IS_SYNC(inode).
485 * But we now rely upon generic_osync_inode()
486 * and b_inode_buffers. But not for directories.
487 */
488 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
489 sync_dirty_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 }
Martin J. Bligha686cd82007-10-16 23:30:46 -0700491 *blks = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 return err;
493}
494
495/**
Martin J. Bligha686cd82007-10-16 23:30:46 -0700496 * ext2_splice_branch - splice the allocated branch onto inode.
497 * @inode: owner
498 * @block: (logical) number of block we are adding
Martin J. Bligha686cd82007-10-16 23:30:46 -0700499 * @where: location of missing link
500 * @num: number of indirect blocks we are adding
501 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 *
Martin J. Bligha686cd82007-10-16 23:30:46 -0700503 * This function fills the missing link and does all housekeeping needed in
504 * inode (->i_blocks, etc.). In case of success we end up with the full
505 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700507static void ext2_splice_branch(struct inode *inode,
508 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 int i;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700511 struct ext2_block_alloc_info *block_i;
512 ext2_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Martin J. Bligha686cd82007-10-16 23:30:46 -0700514 block_i = EXT2_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Martin J. Bligha686cd82007-10-16 23:30:46 -0700516 /* XXX LOCKING probably should have i_meta_lock ?*/
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 /* That's it */
518
519 *where->p = where->key;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Martin J. Bligha686cd82007-10-16 23:30:46 -0700521 /*
522 * Update the host buffer_head or inode to point to more just allocated
523 * direct blocks blocks
524 */
525 if (num == 0 && blks > 1) {
526 current_block = le32_to_cpu(where->key) + 1;
527 for (i = 1; i < blks; i++)
528 *(where->p + i ) = cpu_to_le32(current_block++);
529 }
530
531 /*
532 * update the most recently allocated logical & physical block
533 * in i_block_alloc_info, to assist find the proper goal block for next
534 * allocation
535 */
536 if (block_i) {
537 block_i->last_alloc_logical_block = block + blks - 1;
538 block_i->last_alloc_physical_block =
539 le32_to_cpu(where[num].key) + blks - 1;
540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 /* We are done with atomic stuff, now do the rest of housekeeping */
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 /* had we spliced it onto indirect block? */
545 if (where->bh)
546 mark_buffer_dirty_inode(where->bh, inode);
547
Martin J. Bligha686cd82007-10-16 23:30:46 -0700548 inode->i_ctime = CURRENT_TIME_SEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 mark_inode_dirty(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
552/*
553 * Allocation strategy is simple: if we have to allocate something, we will
554 * have to go the whole way to leaf. So let's do it before attaching anything
555 * to tree, set linkage between the newborn blocks, write them if sync is
556 * required, recheck the path, free and repeat if check fails, otherwise
557 * set the last missing link (that will protect us from any truncate-generated
558 * removals - all blocks on the path are immune now) and possibly force the
559 * write on the parent block.
560 * That has a nice additional property: no special recovery from the failed
561 * allocations is needed - we simply release blocks and do not touch anything
562 * reachable from inode.
Martin J. Bligha686cd82007-10-16 23:30:46 -0700563 *
564 * `handle' can be NULL if create == 0.
565 *
Martin J. Bligha686cd82007-10-16 23:30:46 -0700566 * return > 0, # of blocks mapped or allocated.
567 * return = 0, if plain lookup failed.
568 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700570static int ext2_get_blocks(struct inode *inode,
571 sector_t iblock, unsigned long maxblocks,
572 struct buffer_head *bh_result,
573 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
575 int err = -EIO;
576 int offsets[4];
577 Indirect chain[4];
578 Indirect *partial;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700579 ext2_fsblk_t goal;
580 int indirect_blks;
581 int blocks_to_boundary = 0;
582 int depth;
583 struct ext2_inode_info *ei = EXT2_I(inode);
584 int count = 0;
585 ext2_fsblk_t first_block = 0;
586
587 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 if (depth == 0)
Martin J. Bligha686cd82007-10-16 23:30:46 -0700590 return (err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Jan Kara316cb4ef32009-04-13 14:40:14 -0700592 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 /* Simplest case - block found, no allocation needed */
594 if (!partial) {
Martin J. Bligha686cd82007-10-16 23:30:46 -0700595 first_block = le32_to_cpu(chain[depth - 1].key);
596 clear_buffer_new(bh_result); /* What's this do? */
597 count++;
598 /*map more blocks*/
599 while (count < maxblocks && count <= blocks_to_boundary) {
600 ext2_fsblk_t blk;
601
Jan Kara316cb4ef32009-04-13 14:40:14 -0700602 if (!verify_chain(chain, chain + depth - 1)) {
Martin J. Bligha686cd82007-10-16 23:30:46 -0700603 /*
604 * Indirect block might be removed by
605 * truncate while we were reading it.
606 * Handling of that case: forget what we've
607 * got now, go to reread.
608 */
Jan Kara316cb4ef32009-04-13 14:40:14 -0700609 err = -EAGAIN;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700610 count = 0;
Jan Kara316cb4ef32009-04-13 14:40:14 -0700611 break;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700612 }
613 blk = le32_to_cpu(*(chain[depth-1].p + count));
614 if (blk == first_block + count)
615 count++;
616 else
617 break;
618 }
Jan Kara316cb4ef32009-04-13 14:40:14 -0700619 if (err != -EAGAIN)
620 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
622
623 /* Next simple case - plain lookup or failed read of indirect block */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700624 if (!create || err == -EIO)
625 goto cleanup;
626
627 mutex_lock(&ei->truncate_mutex);
Jan Kara316cb4ef32009-04-13 14:40:14 -0700628 /*
629 * If the indirect block is missing while we are reading
630 * the chain(ext3_get_branch() returns -EAGAIN err), or
631 * if the chain has been changed after we grab the semaphore,
632 * (either because another process truncated this branch, or
633 * another get_block allocated this branch) re-grab the chain to see if
634 * the request block has been allocated or not.
635 *
636 * Since we already block the truncate/other get_block
637 * at this point, we will have the current copy of the chain when we
638 * splice the branch into the tree.
639 */
640 if (err == -EAGAIN || !verify_chain(chain, partial)) {
641 while (partial > chain) {
642 brelse(partial->bh);
643 partial--;
644 }
645 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
646 if (!partial) {
647 count++;
648 mutex_unlock(&ei->truncate_mutex);
649 if (err)
650 goto cleanup;
651 clear_buffer_new(bh_result);
652 goto got_it;
653 }
654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
656 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700657 * Okay, we need to do block allocation. Lazily initialize the block
658 * allocation info here if necessary
659 */
660 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
661 ext2_init_block_alloc_info(inode);
662
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800663 goal = ext2_find_goal(inode, iblock, partial);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700664
665 /* the number of blocks need to allocate for [d,t]indirect blocks */
666 indirect_blks = (chain + depth) - partial - 1;
667 /*
668 * Next look up the indirect map to count the totoal number of
669 * direct blocks to allocate for this branch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700671 count = ext2_blks_to_allocate(partial, indirect_blks,
672 maxblocks, blocks_to_boundary);
673 /*
674 * XXX ???? Block out ext2_truncate while we alter the tree
675 */
676 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
677 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Martin J. Bligha686cd82007-10-16 23:30:46 -0700679 if (err) {
680 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 goto cleanup;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700682 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Carsten Otte6d791252005-06-23 22:05:26 -0700684 if (ext2_use_xip(inode->i_sb)) {
685 /*
686 * we need to clear the block
687 */
688 err = ext2_clear_xip_target (inode,
689 le32_to_cpu(chain[depth-1].key));
Martin J. Bligha686cd82007-10-16 23:30:46 -0700690 if (err) {
691 mutex_unlock(&ei->truncate_mutex);
Carsten Otte6d791252005-06-23 22:05:26 -0700692 goto cleanup;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700693 }
Carsten Otte6d791252005-06-23 22:05:26 -0700694 }
695
Martin J. Bligha686cd82007-10-16 23:30:46 -0700696 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
697 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 set_buffer_new(bh_result);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700699got_it:
700 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
701 if (count > blocks_to_boundary)
702 set_buffer_boundary(bh_result);
703 err = count;
704 /* Clean up and exit */
705 partial = chain + depth - 1; /* the whole chain */
706cleanup:
707 while (partial > chain) {
708 brelse(partial->bh);
709 partial--;
710 }
711 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712}
713
Martin J. Bligha686cd82007-10-16 23:30:46 -0700714int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
715{
716 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
717 int ret = ext2_get_blocks(inode, iblock, max_blocks,
718 bh_result, create);
719 if (ret > 0) {
720 bh_result->b_size = (ret << inode->i_blkbits);
721 ret = 0;
722 }
723 return ret;
724
725}
726
Josef Bacik68c9d702008-10-03 17:32:43 -0400727int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
728 u64 start, u64 len)
729{
730 return generic_block_fiemap(inode, fieinfo, start, len,
731 ext2_get_block);
732}
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734static int ext2_writepage(struct page *page, struct writeback_control *wbc)
735{
736 return block_write_full_page(page, ext2_get_block, wbc);
737}
738
739static int ext2_readpage(struct file *file, struct page *page)
740{
741 return mpage_readpage(page, ext2_get_block);
742}
743
744static int
745ext2_readpages(struct file *file, struct address_space *mapping,
746 struct list_head *pages, unsigned nr_pages)
747{
748 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
749}
750
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700751int __ext2_write_begin(struct file *file, struct address_space *mapping,
752 loff_t pos, unsigned len, unsigned flags,
753 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700755 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
756 ext2_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
759static int
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700760ext2_write_begin(struct file *file, struct address_space *mapping,
761 loff_t pos, unsigned len, unsigned flags,
762 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763{
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700764 *pagep = NULL;
765 return __ext2_write_begin(file, mapping, pos, len, flags, pagep,fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
Nick Piggin03158cd2007-10-16 01:25:25 -0700768static int
769ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
770 loff_t pos, unsigned len, unsigned flags,
771 struct page **pagep, void **fsdata)
772{
773 /*
774 * Dir-in-pagecache still uses ext2_write_begin. Would have to rework
775 * directory handling code to pass around offsets rather than struct
776 * pages in order to make this work easily.
777 */
778 return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
779 ext2_get_block);
780}
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782static int ext2_nobh_writepage(struct page *page,
783 struct writeback_control *wbc)
784{
785 return nobh_writepage(page, ext2_get_block, wbc);
786}
787
788static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
789{
790 return generic_block_bmap(mapping,block,ext2_get_block);
791}
792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793static ssize_t
794ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
795 loff_t offset, unsigned long nr_segs)
796{
797 struct file *file = iocb->ki_filp;
798 struct inode *inode = file->f_mapping->host;
799
800 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800801 offset, nr_segs, ext2_get_block, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
803
804static int
805ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
806{
807 return mpage_writepages(mapping, wbc, ext2_get_block);
808}
809
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700810const struct address_space_operations ext2_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 .readpage = ext2_readpage,
812 .readpages = ext2_readpages,
813 .writepage = ext2_writepage,
814 .sync_page = block_sync_page,
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700815 .write_begin = ext2_write_begin,
816 .write_end = generic_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 .bmap = ext2_bmap,
818 .direct_IO = ext2_direct_IO,
819 .writepages = ext2_writepages,
Christoph Lametere965f962006-02-01 03:05:41 -0800820 .migratepage = buffer_migrate_page,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -0700821 .is_partially_uptodate = block_is_partially_uptodate,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822};
823
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700824const struct address_space_operations ext2_aops_xip = {
Carsten Otte6d791252005-06-23 22:05:26 -0700825 .bmap = ext2_bmap,
Nick Piggin70688e42008-04-28 02:13:02 -0700826 .get_xip_mem = ext2_get_xip_mem,
Carsten Otte6d791252005-06-23 22:05:26 -0700827};
828
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700829const struct address_space_operations ext2_nobh_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 .readpage = ext2_readpage,
831 .readpages = ext2_readpages,
832 .writepage = ext2_nobh_writepage,
833 .sync_page = block_sync_page,
Nick Piggin03158cd2007-10-16 01:25:25 -0700834 .write_begin = ext2_nobh_write_begin,
835 .write_end = nobh_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 .bmap = ext2_bmap,
837 .direct_IO = ext2_direct_IO,
838 .writepages = ext2_writepages,
Christoph Lametere965f962006-02-01 03:05:41 -0800839 .migratepage = buffer_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840};
841
842/*
843 * Probably it should be a library function... search for first non-zero word
844 * or memcmp with zero_page, whatever is better for particular architecture.
845 * Linus?
846 */
847static inline int all_zeroes(__le32 *p, __le32 *q)
848{
849 while (p < q)
850 if (*p++)
851 return 0;
852 return 1;
853}
854
855/**
856 * ext2_find_shared - find the indirect blocks for partial truncation.
857 * @inode: inode in question
858 * @depth: depth of the affected branch
859 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
860 * @chain: place to store the pointers to partial indirect blocks
861 * @top: place to the (detached) top of branch
862 *
863 * This is a helper function used by ext2_truncate().
864 *
865 * When we do truncate() we may have to clean the ends of several indirect
866 * blocks but leave the blocks themselves alive. Block is partially
867 * truncated if some data below the new i_size is refered from it (and
868 * it is on the path to the first completely truncated data block, indeed).
869 * We have to free the top of that path along with everything to the right
870 * of the path. Since no allocation past the truncation point is possible
871 * until ext2_truncate() finishes, we may safely do the latter, but top
872 * of branch may require special attention - pageout below the truncation
873 * point might try to populate it.
874 *
875 * We atomically detach the top of branch from the tree, store the block
876 * number of its root in *@top, pointers to buffer_heads of partially
877 * truncated blocks - in @chain[].bh and pointers to their last elements
878 * that should not be removed - in @chain[].p. Return value is the pointer
879 * to last filled element of @chain.
880 *
881 * The work left to caller to do the actual freeing of subtrees:
882 * a) free the subtree starting from *@top
883 * b) free the subtrees whose roots are stored in
884 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
885 * c) free the subtrees growing from the inode past the @chain[0].p
886 * (no partially truncated stuff there).
887 */
888
889static Indirect *ext2_find_shared(struct inode *inode,
890 int depth,
891 int offsets[4],
892 Indirect chain[4],
893 __le32 *top)
894{
895 Indirect *partial, *p;
896 int k, err;
897
898 *top = 0;
899 for (k = depth; k > 1 && !offsets[k-1]; k--)
900 ;
901 partial = ext2_get_branch(inode, k, offsets, chain, &err);
902 if (!partial)
903 partial = chain + k-1;
904 /*
905 * If the branch acquired continuation since we've looked at it -
906 * fine, it should all survive and (new) top doesn't belong to us.
907 */
908 write_lock(&EXT2_I(inode)->i_meta_lock);
909 if (!partial->key && *partial->p) {
910 write_unlock(&EXT2_I(inode)->i_meta_lock);
911 goto no_top;
912 }
913 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
914 ;
915 /*
916 * OK, we've found the last block that must survive. The rest of our
917 * branch should be detached before unlocking. However, if that rest
918 * of branch is all ours and does not grow immediately from the inode
919 * it's easier to cheat and just decrement partial->p.
920 */
921 if (p == chain + k - 1 && p > chain) {
922 p->p--;
923 } else {
924 *top = *p->p;
925 *p->p = 0;
926 }
927 write_unlock(&EXT2_I(inode)->i_meta_lock);
928
929 while(partial > p)
930 {
931 brelse(partial->bh);
932 partial--;
933 }
934no_top:
935 return partial;
936}
937
938/**
939 * ext2_free_data - free a list of data blocks
940 * @inode: inode we are dealing with
941 * @p: array of block numbers
942 * @q: points immediately past the end of array
943 *
944 * We are freeing all blocks refered from that array (numbers are
945 * stored as little-endian 32-bit) and updating @inode->i_blocks
946 * appropriately.
947 */
948static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
949{
950 unsigned long block_to_free = 0, count = 0;
951 unsigned long nr;
952
953 for ( ; p < q ; p++) {
954 nr = le32_to_cpu(*p);
955 if (nr) {
956 *p = 0;
957 /* accumulate blocks to free if they're contiguous */
958 if (count == 0)
959 goto free_this;
960 else if (block_to_free == nr - count)
961 count++;
962 else {
963 mark_inode_dirty(inode);
964 ext2_free_blocks (inode, block_to_free, count);
965 free_this:
966 block_to_free = nr;
967 count = 1;
968 }
969 }
970 }
971 if (count > 0) {
972 mark_inode_dirty(inode);
973 ext2_free_blocks (inode, block_to_free, count);
974 }
975}
976
977/**
978 * ext2_free_branches - free an array of branches
979 * @inode: inode we are dealing with
980 * @p: array of block numbers
981 * @q: pointer immediately past the end of array
982 * @depth: depth of the branches to free
983 *
984 * We are freeing all blocks refered from these branches (numbers are
985 * stored as little-endian 32-bit) and updating @inode->i_blocks
986 * appropriately.
987 */
988static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
989{
990 struct buffer_head * bh;
991 unsigned long nr;
992
993 if (depth--) {
994 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
995 for ( ; p < q ; p++) {
996 nr = le32_to_cpu(*p);
997 if (!nr)
998 continue;
999 *p = 0;
1000 bh = sb_bread(inode->i_sb, nr);
1001 /*
1002 * A read failure? Report error and clear slot
1003 * (should be rare).
1004 */
1005 if (!bh) {
1006 ext2_error(inode->i_sb, "ext2_free_branches",
1007 "Read failure, inode=%ld, block=%ld",
1008 inode->i_ino, nr);
1009 continue;
1010 }
1011 ext2_free_branches(inode,
1012 (__le32*)bh->b_data,
1013 (__le32*)bh->b_data + addr_per_block,
1014 depth);
1015 bforget(bh);
1016 ext2_free_blocks(inode, nr, 1);
1017 mark_inode_dirty(inode);
1018 }
1019 } else
1020 ext2_free_data(inode, p, q);
1021}
1022
Martin J. Bligha686cd82007-10-16 23:30:46 -07001023void ext2_truncate(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024{
1025 __le32 *i_data = EXT2_I(inode)->i_data;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001026 struct ext2_inode_info *ei = EXT2_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1028 int offsets[4];
1029 Indirect chain[4];
1030 Indirect *partial;
1031 __le32 nr = 0;
1032 int n;
1033 long iblock;
1034 unsigned blocksize;
1035
1036 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1037 S_ISLNK(inode->i_mode)))
1038 return;
1039 if (ext2_inode_is_fast_symlink(inode))
1040 return;
1041 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1042 return;
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 blocksize = inode->i_sb->s_blocksize;
1045 iblock = (inode->i_size + blocksize-1)
1046 >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1047
Carsten Otte6d791252005-06-23 22:05:26 -07001048 if (mapping_is_xip(inode->i_mapping))
1049 xip_truncate_page(inode->i_mapping, inode->i_size);
1050 else if (test_opt(inode->i_sb, NOBH))
Nick Piggin03158cd2007-10-16 01:25:25 -07001051 nobh_truncate_page(inode->i_mapping,
1052 inode->i_size, ext2_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 else
1054 block_truncate_page(inode->i_mapping,
1055 inode->i_size, ext2_get_block);
1056
1057 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1058 if (n == 0)
1059 return;
1060
Martin J. Bligha686cd82007-10-16 23:30:46 -07001061 /*
1062 * From here we block out all ext2_get_block() callers who want to
1063 * modify the block allocation tree.
1064 */
1065 mutex_lock(&ei->truncate_mutex);
1066
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 if (n == 1) {
1068 ext2_free_data(inode, i_data+offsets[0],
1069 i_data + EXT2_NDIR_BLOCKS);
1070 goto do_indirects;
1071 }
1072
1073 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1074 /* Kill the top of shared branch (already detached) */
1075 if (nr) {
1076 if (partial == chain)
1077 mark_inode_dirty(inode);
1078 else
1079 mark_buffer_dirty_inode(partial->bh, inode);
1080 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1081 }
1082 /* Clear the ends of indirect blocks on the shared branch */
1083 while (partial > chain) {
1084 ext2_free_branches(inode,
1085 partial->p + 1,
1086 (__le32*)partial->bh->b_data+addr_per_block,
1087 (chain+n-1) - partial);
1088 mark_buffer_dirty_inode(partial->bh, inode);
1089 brelse (partial->bh);
1090 partial--;
1091 }
1092do_indirects:
1093 /* Kill the remaining (whole) subtrees */
1094 switch (offsets[0]) {
1095 default:
1096 nr = i_data[EXT2_IND_BLOCK];
1097 if (nr) {
1098 i_data[EXT2_IND_BLOCK] = 0;
1099 mark_inode_dirty(inode);
1100 ext2_free_branches(inode, &nr, &nr+1, 1);
1101 }
1102 case EXT2_IND_BLOCK:
1103 nr = i_data[EXT2_DIND_BLOCK];
1104 if (nr) {
1105 i_data[EXT2_DIND_BLOCK] = 0;
1106 mark_inode_dirty(inode);
1107 ext2_free_branches(inode, &nr, &nr+1, 2);
1108 }
1109 case EXT2_DIND_BLOCK:
1110 nr = i_data[EXT2_TIND_BLOCK];
1111 if (nr) {
1112 i_data[EXT2_TIND_BLOCK] = 0;
1113 mark_inode_dirty(inode);
1114 ext2_free_branches(inode, &nr, &nr+1, 3);
1115 }
1116 case EXT2_TIND_BLOCK:
1117 ;
1118 }
Martin J. Bligha686cd82007-10-16 23:30:46 -07001119
1120 ext2_discard_reservation(inode);
1121
1122 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1124 if (inode_needs_sync(inode)) {
1125 sync_mapping_buffers(inode->i_mapping);
1126 ext2_sync_inode (inode);
1127 } else {
1128 mark_inode_dirty(inode);
1129 }
1130}
1131
1132static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1133 struct buffer_head **p)
1134{
1135 struct buffer_head * bh;
1136 unsigned long block_group;
1137 unsigned long block;
1138 unsigned long offset;
1139 struct ext2_group_desc * gdp;
1140
1141 *p = NULL;
1142 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1143 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1144 goto Einval;
1145
1146 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
Eric Sandeenef2fb672007-10-16 23:26:30 -07001147 gdp = ext2_get_group_desc(sb, block_group, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 if (!gdp)
1149 goto Egdp;
1150 /*
1151 * Figure out the offset within the block group inode table
1152 */
1153 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1154 block = le32_to_cpu(gdp->bg_inode_table) +
1155 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1156 if (!(bh = sb_bread(sb, block)))
1157 goto Eio;
1158
1159 *p = bh;
1160 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1161 return (struct ext2_inode *) (bh->b_data + offset);
1162
1163Einval:
1164 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1165 (unsigned long) ino);
1166 return ERR_PTR(-EINVAL);
1167Eio:
1168 ext2_error(sb, "ext2_get_inode",
1169 "unable to read inode block - inode=%lu, block=%lu",
1170 (unsigned long) ino, block);
1171Egdp:
1172 return ERR_PTR(-EIO);
1173}
1174
1175void ext2_set_inode_flags(struct inode *inode)
1176{
1177 unsigned int flags = EXT2_I(inode)->i_flags;
1178
1179 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1180 if (flags & EXT2_SYNC_FL)
1181 inode->i_flags |= S_SYNC;
1182 if (flags & EXT2_APPEND_FL)
1183 inode->i_flags |= S_APPEND;
1184 if (flags & EXT2_IMMUTABLE_FL)
1185 inode->i_flags |= S_IMMUTABLE;
1186 if (flags & EXT2_NOATIME_FL)
1187 inode->i_flags |= S_NOATIME;
1188 if (flags & EXT2_DIRSYNC_FL)
1189 inode->i_flags |= S_DIRSYNC;
1190}
1191
Jan Kara4f99ed62007-05-08 00:31:04 -07001192/* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1193void ext2_get_inode_flags(struct ext2_inode_info *ei)
1194{
1195 unsigned int flags = ei->vfs_inode.i_flags;
1196
1197 ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1198 EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1199 if (flags & S_SYNC)
1200 ei->i_flags |= EXT2_SYNC_FL;
1201 if (flags & S_APPEND)
1202 ei->i_flags |= EXT2_APPEND_FL;
1203 if (flags & S_IMMUTABLE)
1204 ei->i_flags |= EXT2_IMMUTABLE_FL;
1205 if (flags & S_NOATIME)
1206 ei->i_flags |= EXT2_NOATIME_FL;
1207 if (flags & S_DIRSYNC)
1208 ei->i_flags |= EXT2_DIRSYNC_FL;
1209}
1210
David Howells52fcf702008-02-07 00:15:35 -08001211struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212{
David Howells52fcf702008-02-07 00:15:35 -08001213 struct ext2_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 struct buffer_head * bh;
David Howells52fcf702008-02-07 00:15:35 -08001215 struct ext2_inode *raw_inode;
1216 struct inode *inode;
1217 long ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 int n;
1219
David Howells52fcf702008-02-07 00:15:35 -08001220 inode = iget_locked(sb, ino);
1221 if (!inode)
1222 return ERR_PTR(-ENOMEM);
1223 if (!(inode->i_state & I_NEW))
1224 return inode;
1225
1226 ei = EXT2_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227#ifdef CONFIG_EXT2_FS_POSIX_ACL
1228 ei->i_acl = EXT2_ACL_NOT_CACHED;
1229 ei->i_default_acl = EXT2_ACL_NOT_CACHED;
1230#endif
Martin J. Bligha686cd82007-10-16 23:30:46 -07001231 ei->i_block_alloc_info = NULL;
1232
David Howells52fcf702008-02-07 00:15:35 -08001233 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1234 if (IS_ERR(raw_inode)) {
1235 ret = PTR_ERR(raw_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 goto bad_inode;
David Howells52fcf702008-02-07 00:15:35 -08001237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
1239 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1240 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1241 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1242 if (!(test_opt (inode->i_sb, NO_UID32))) {
1243 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1244 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1245 }
1246 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1247 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07001248 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1249 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1250 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1252 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1253 /* We now have enough fields to check if the inode was active or not.
1254 * This is needed because nfsd might try to access dead inodes
1255 * the test is that same one that e2fsck uses
1256 * NeilBrown 1999oct15
1257 */
1258 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1259 /* this inode is deleted */
1260 brelse (bh);
David Howells52fcf702008-02-07 00:15:35 -08001261 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 goto bad_inode;
1263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1265 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1266 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1267 ei->i_frag_no = raw_inode->i_frag;
1268 ei->i_frag_size = raw_inode->i_fsize;
1269 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1270 ei->i_dir_acl = 0;
1271 if (S_ISREG(inode->i_mode))
1272 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1273 else
1274 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1275 ei->i_dtime = 0;
1276 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1277 ei->i_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1279 ei->i_dir_start_lookup = 0;
1280
1281 /*
1282 * NOTE! The in-memory inode i_data array is in little-endian order
1283 * even on big-endian machines: we do NOT byteswap the block numbers!
1284 */
1285 for (n = 0; n < EXT2_N_BLOCKS; n++)
1286 ei->i_data[n] = raw_inode->i_block[n];
1287
1288 if (S_ISREG(inode->i_mode)) {
1289 inode->i_op = &ext2_file_inode_operations;
Carsten Otte6d791252005-06-23 22:05:26 -07001290 if (ext2_use_xip(inode->i_sb)) {
1291 inode->i_mapping->a_ops = &ext2_aops_xip;
1292 inode->i_fop = &ext2_xip_file_operations;
1293 } else if (test_opt(inode->i_sb, NOBH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 inode->i_mapping->a_ops = &ext2_nobh_aops;
Carsten Otte6d791252005-06-23 22:05:26 -07001295 inode->i_fop = &ext2_file_operations;
1296 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 inode->i_mapping->a_ops = &ext2_aops;
Carsten Otte6d791252005-06-23 22:05:26 -07001298 inode->i_fop = &ext2_file_operations;
1299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 } else if (S_ISDIR(inode->i_mode)) {
1301 inode->i_op = &ext2_dir_inode_operations;
1302 inode->i_fop = &ext2_dir_operations;
1303 if (test_opt(inode->i_sb, NOBH))
1304 inode->i_mapping->a_ops = &ext2_nobh_aops;
1305 else
1306 inode->i_mapping->a_ops = &ext2_aops;
1307 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffin8d6d0c42008-12-19 20:47:13 +00001308 if (ext2_inode_is_fast_symlink(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 inode->i_op = &ext2_fast_symlink_inode_operations;
Duane Griffin8d6d0c42008-12-19 20:47:13 +00001310 nd_terminate_link(ei->i_data, inode->i_size,
1311 sizeof(ei->i_data) - 1);
1312 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 inode->i_op = &ext2_symlink_inode_operations;
1314 if (test_opt(inode->i_sb, NOBH))
1315 inode->i_mapping->a_ops = &ext2_nobh_aops;
1316 else
1317 inode->i_mapping->a_ops = &ext2_aops;
1318 }
1319 } else {
1320 inode->i_op = &ext2_special_inode_operations;
1321 if (raw_inode->i_block[0])
1322 init_special_inode(inode, inode->i_mode,
1323 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1324 else
1325 init_special_inode(inode, inode->i_mode,
1326 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1327 }
1328 brelse (bh);
1329 ext2_set_inode_flags(inode);
David Howells52fcf702008-02-07 00:15:35 -08001330 unlock_new_inode(inode);
1331 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333bad_inode:
David Howells52fcf702008-02-07 00:15:35 -08001334 iget_failed(inode);
1335 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337
Al Viroe1740a42009-06-07 15:14:02 -04001338int ext2_write_inode(struct inode *inode, int do_sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339{
1340 struct ext2_inode_info *ei = EXT2_I(inode);
1341 struct super_block *sb = inode->i_sb;
1342 ino_t ino = inode->i_ino;
1343 uid_t uid = inode->i_uid;
1344 gid_t gid = inode->i_gid;
1345 struct buffer_head * bh;
1346 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1347 int n;
1348 int err = 0;
1349
1350 if (IS_ERR(raw_inode))
1351 return -EIO;
1352
1353 /* For fields not not tracking in the in-memory inode,
1354 * initialise them to zero for new inodes. */
1355 if (ei->i_state & EXT2_STATE_NEW)
1356 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1357
Jan Kara4f99ed62007-05-08 00:31:04 -07001358 ext2_get_inode_flags(ei);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1360 if (!(test_opt(sb, NO_UID32))) {
1361 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1362 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1363/*
1364 * Fix up interoperability with old kernels. Otherwise, old inodes get
1365 * re-used with the upper 16 bits of the uid/gid intact
1366 */
1367 if (!ei->i_dtime) {
1368 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1369 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1370 } else {
1371 raw_inode->i_uid_high = 0;
1372 raw_inode->i_gid_high = 0;
1373 }
1374 } else {
1375 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1376 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1377 raw_inode->i_uid_high = 0;
1378 raw_inode->i_gid_high = 0;
1379 }
1380 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1381 raw_inode->i_size = cpu_to_le32(inode->i_size);
1382 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1383 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1384 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1385
1386 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1387 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1388 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1389 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1390 raw_inode->i_frag = ei->i_frag_no;
1391 raw_inode->i_fsize = ei->i_frag_size;
1392 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1393 if (!S_ISREG(inode->i_mode))
1394 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1395 else {
1396 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1397 if (inode->i_size > 0x7fffffffULL) {
1398 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1399 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1400 EXT2_SB(sb)->s_es->s_rev_level ==
1401 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1402 /* If this is the first large file
1403 * created, add a flag to the superblock.
1404 */
1405 lock_kernel();
1406 ext2_update_dynamic_rev(sb);
1407 EXT2_SET_RO_COMPAT_FEATURE(sb,
1408 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1409 unlock_kernel();
1410 ext2_write_super(sb);
1411 }
1412 }
1413 }
1414
1415 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1416 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1417 if (old_valid_dev(inode->i_rdev)) {
1418 raw_inode->i_block[0] =
1419 cpu_to_le32(old_encode_dev(inode->i_rdev));
1420 raw_inode->i_block[1] = 0;
1421 } else {
1422 raw_inode->i_block[0] = 0;
1423 raw_inode->i_block[1] =
1424 cpu_to_le32(new_encode_dev(inode->i_rdev));
1425 raw_inode->i_block[2] = 0;
1426 }
1427 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1428 raw_inode->i_block[n] = ei->i_data[n];
1429 mark_buffer_dirty(bh);
1430 if (do_sync) {
1431 sync_dirty_buffer(bh);
1432 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1433 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1434 sb->s_id, (unsigned long) ino);
1435 err = -EIO;
1436 }
1437 }
1438 ei->i_state &= ~EXT2_STATE_NEW;
1439 brelse (bh);
1440 return err;
1441}
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443int ext2_sync_inode(struct inode *inode)
1444{
1445 struct writeback_control wbc = {
1446 .sync_mode = WB_SYNC_ALL,
1447 .nr_to_write = 0, /* sys_fsync did this */
1448 };
1449 return sync_inode(inode, &wbc);
1450}
1451
1452int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1453{
1454 struct inode *inode = dentry->d_inode;
1455 int error;
1456
1457 error = inode_change_ok(inode, iattr);
1458 if (error)
1459 return error;
1460 if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1461 (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
Jan Kara6f90bee2009-01-26 16:52:06 +01001462 error = vfs_dq_transfer(inode, iattr) ? -EDQUOT : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 if (error)
1464 return error;
1465 }
1466 error = inode_setattr(inode, iattr);
1467 if (!error && (iattr->ia_valid & ATTR_MODE))
1468 error = ext2_acl_chmod(inode);
1469 return error;
1470}