aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4
diff options
context:
space:
mode:
authorAmir Goldstein <amir73il@users.sf.net>2011-06-27 19:40:50 -0400
committerTheodore Ts'o <tytso@mit.edu>2011-06-27 19:40:50 -0400
commitdae1e52cb1267bf8f52e5e47a80fab566d7e8aa4 (patch)
tree2537e9f8f138e4935f88e605244174c9b2400bf4 /fs/ext4
parent9f125d641beb898f5bf2fe69583192c18043517a (diff)
ext4: move ext4_ind_* functions from inode.c to indirect.c
This patch moves functions from inode.c to indirect.c. The moved functions are ext4_ind_* functions and their helpers. Functions called from inode.c are declared extern. Signed-off-by: Amir Goldstein <amir73il@users.sf.net> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Diffstat (limited to 'fs/ext4')
-rw-r--r--fs/ext4/Makefile2
-rw-r--r--fs/ext4/block_validity.c1
-rw-r--r--fs/ext4/ext4.h9
-rw-r--r--fs/ext4/indirect.c1510
-rw-r--r--fs/ext4/inode.c1486
5 files changed, 1521 insertions, 1487 deletions
diff --git a/fs/ext4/Makefile b/fs/ext4/Makefile
index 04109460ba9e..56fd8f865930 100644
--- a/fs/ext4/Makefile
+++ b/fs/ext4/Makefile
@@ -7,7 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
7ext4-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \ 7ext4-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
8 ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \ 8 ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
9 ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \ 9 ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
10 mmp.o 10 mmp.o indirect.o
11 11
12ext4-$(CONFIG_EXT4_FS_XATTR) += xattr.o xattr_user.o xattr_trusted.o 12ext4-$(CONFIG_EXT4_FS_XATTR) += xattr.o xattr_user.o xattr_trusted.o
13ext4-$(CONFIG_EXT4_FS_POSIX_ACL) += acl.o 13ext4-$(CONFIG_EXT4_FS_POSIX_ACL) += acl.o
diff --git a/fs/ext4/block_validity.c b/fs/ext4/block_validity.c
index af103be491b0..8efb2f0a3447 100644
--- a/fs/ext4/block_validity.c
+++ b/fs/ext4/block_validity.c
@@ -266,3 +266,4 @@ int ext4_check_blockref(const char *function, unsigned int line,
266 } 266 }
267 return 0; 267 return 0;
268} 268}
269
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 82ba7eb7c4a5..ddaf5043fb38 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1834,6 +1834,15 @@ extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
1834extern qsize_t *ext4_get_reserved_space(struct inode *inode); 1834extern qsize_t *ext4_get_reserved_space(struct inode *inode);
1835extern void ext4_da_update_reserve_space(struct inode *inode, 1835extern void ext4_da_update_reserve_space(struct inode *inode,
1836 int used, int quota_claim); 1836 int used, int quota_claim);
1837
1838/* indirect.c */
1839extern int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
1840 struct ext4_map_blocks *map, int flags);
1841extern ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
1842 const struct iovec *iov, loff_t offset,
1843 unsigned long nr_segs);
1844extern int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock);
1845extern int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk);
1837extern void ext4_ind_truncate(struct inode *inode); 1846extern void ext4_ind_truncate(struct inode *inode);
1838 1847
1839/* ioctl.c */ 1848/* ioctl.c */
diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
new file mode 100644
index 000000000000..c3e85a86e821
--- /dev/null
+++ b/fs/ext4/indirect.c
@@ -0,0 +1,1510 @@
1/*
2 * linux/fs/ext4/indirect.c
3 *
4 * from
5 *
6 * linux/fs/ext4/inode.c
7 *
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 *
13 * from
14 *
15 * linux/fs/minix/inode.c
16 *
17 * Copyright (C) 1991, 1992 Linus Torvalds
18 *
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
21 */
22
23#include <linux/module.h>
24#include "ext4_jbd2.h"
25#include "truncate.h"
26
27#include <trace/events/ext4.h>
28
29typedef struct {
30 __le32 *p;
31 __le32 key;
32 struct buffer_head *bh;
33} Indirect;
34
35static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
36{
37 p->key = *(p->p = v);
38 p->bh = bh;
39}
40
41/**
42 * ext4_block_to_path - parse the block number into array of offsets
43 * @inode: inode in question (we are only interested in its superblock)
44 * @i_block: block number to be parsed
45 * @offsets: array to store the offsets in
46 * @boundary: set this non-zero if the referred-to block is likely to be
47 * followed (on disk) by an indirect block.
48 *
49 * To store the locations of file's data ext4 uses a data structure common
50 * for UNIX filesystems - tree of pointers anchored in the inode, with
51 * data blocks at leaves and indirect blocks in intermediate nodes.
52 * This function translates the block number into path in that tree -
53 * return value is the path length and @offsets[n] is the offset of
54 * pointer to (n+1)th node in the nth one. If @block is out of range
55 * (negative or too large) warning is printed and zero returned.
56 *
57 * Note: function doesn't find node addresses, so no IO is needed. All
58 * we need to know is the capacity of indirect blocks (taken from the
59 * inode->i_sb).
60 */
61
62/*
63 * Portability note: the last comparison (check that we fit into triple
64 * indirect block) is spelled differently, because otherwise on an
65 * architecture with 32-bit longs and 8Kb pages we might get into trouble
66 * if our filesystem had 8Kb blocks. We might use long long, but that would
67 * kill us on x86. Oh, well, at least the sign propagation does not matter -
68 * i_block would have to be negative in the very beginning, so we would not
69 * get there at all.
70 */
71
72static int ext4_block_to_path(struct inode *inode,
73 ext4_lblk_t i_block,
74 ext4_lblk_t offsets[4], int *boundary)
75{
76 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
77 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
78 const long direct_blocks = EXT4_NDIR_BLOCKS,
79 indirect_blocks = ptrs,
80 double_blocks = (1 << (ptrs_bits * 2));
81 int n = 0;
82 int final = 0;
83
84 if (i_block < direct_blocks) {
85 offsets[n++] = i_block;
86 final = direct_blocks;
87 } else if ((i_block -= direct_blocks) < indirect_blocks) {
88 offsets[n++] = EXT4_IND_BLOCK;
89 offsets[n++] = i_block;
90 final = ptrs;
91 } else if ((i_block -= indirect_blocks) < double_blocks) {
92 offsets[n++] = EXT4_DIND_BLOCK;
93 offsets[n++] = i_block >> ptrs_bits;
94 offsets[n++] = i_block & (ptrs - 1);
95 final = ptrs;
96 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
97 offsets[n++] = EXT4_TIND_BLOCK;
98 offsets[n++] = i_block >> (ptrs_bits * 2);
99 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
100 offsets[n++] = i_block & (ptrs - 1);
101 final = ptrs;
102 } else {
103 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
104 i_block + direct_blocks +
105 indirect_blocks + double_blocks, inode->i_ino);
106 }
107 if (boundary)
108 *boundary = final - 1 - (i_block & (ptrs - 1));
109 return n;
110}
111
112/**
113 * ext4_get_branch - read the chain of indirect blocks leading to data
114 * @inode: inode in question
115 * @depth: depth of the chain (1 - direct pointer, etc.)
116 * @offsets: offsets of pointers in inode/indirect blocks
117 * @chain: place to store the result
118 * @err: here we store the error value
119 *
120 * Function fills the array of triples <key, p, bh> and returns %NULL
121 * if everything went OK or the pointer to the last filled triple
122 * (incomplete one) otherwise. Upon the return chain[i].key contains
123 * the number of (i+1)-th block in the chain (as it is stored in memory,
124 * i.e. little-endian 32-bit), chain[i].p contains the address of that
125 * number (it points into struct inode for i==0 and into the bh->b_data
126 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
127 * block for i>0 and NULL for i==0. In other words, it holds the block
128 * numbers of the chain, addresses they were taken from (and where we can
129 * verify that chain did not change) and buffer_heads hosting these
130 * numbers.
131 *
132 * Function stops when it stumbles upon zero pointer (absent block)
133 * (pointer to last triple returned, *@err == 0)
134 * or when it gets an IO error reading an indirect block
135 * (ditto, *@err == -EIO)
136 * or when it reads all @depth-1 indirect blocks successfully and finds
137 * the whole chain, all way to the data (returns %NULL, *err == 0).
138 *
139 * Need to be called with
140 * down_read(&EXT4_I(inode)->i_data_sem)
141 */
142static Indirect *ext4_get_branch(struct inode *inode, int depth,
143 ext4_lblk_t *offsets,
144 Indirect chain[4], int *err)
145{
146 struct super_block *sb = inode->i_sb;
147 Indirect *p = chain;
148 struct buffer_head *bh;
149
150 *err = 0;
151 /* i_data is not going away, no lock needed */
152 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
153 if (!p->key)
154 goto no_block;
155 while (--depth) {
156 bh = sb_getblk(sb, le32_to_cpu(p->key));
157 if (unlikely(!bh))
158 goto failure;
159
160 if (!bh_uptodate_or_lock(bh)) {
161 if (bh_submit_read(bh) < 0) {
162 put_bh(bh);
163 goto failure;
164 }
165 /* validate block references */
166 if (ext4_check_indirect_blockref(inode, bh)) {
167 put_bh(bh);
168 goto failure;
169 }
170 }
171
172 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
173 /* Reader: end */
174 if (!p->key)
175 goto no_block;
176 }
177 return NULL;
178
179failure:
180 *err = -EIO;
181no_block:
182 return p;
183}
184
185/**
186 * ext4_find_near - find a place for allocation with sufficient locality
187 * @inode: owner
188 * @ind: descriptor of indirect block.
189 *
190 * This function returns the preferred place for block allocation.
191 * It is used when heuristic for sequential allocation fails.
192 * Rules are:
193 * + if there is a block to the left of our position - allocate near it.
194 * + if pointer will live in indirect block - allocate near that block.
195 * + if pointer will live in inode - allocate in the same
196 * cylinder group.
197 *
198 * In the latter case we colour the starting block by the callers PID to
199 * prevent it from clashing with concurrent allocations for a different inode
200 * in the same block group. The PID is used here so that functionally related
201 * files will be close-by on-disk.
202 *
203 * Caller must make sure that @ind is valid and will stay that way.
204 */
205static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
206{
207 struct ext4_inode_info *ei = EXT4_I(inode);
208 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
209 __le32 *p;
210 ext4_fsblk_t bg_start;
211 ext4_fsblk_t last_block;
212 ext4_grpblk_t colour;
213 ext4_group_t block_group;
214 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
215
216 /* Try to find previous block */
217 for (p = ind->p - 1; p >= start; p--) {
218 if (*p)
219 return le32_to_cpu(*p);
220 }
221
222 /* No such thing, so let's try location of indirect block */
223 if (ind->bh)
224 return ind->bh->b_blocknr;
225
226 /*
227 * It is going to be referred to from the inode itself? OK, just put it
228 * into the same cylinder group then.
229 */
230 block_group = ei->i_block_group;
231 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
232 block_group &= ~(flex_size-1);
233 if (S_ISREG(inode->i_mode))
234 block_group++;
235 }
236 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
237 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
238
239 /*
240 * If we are doing delayed allocation, we don't need take
241 * colour into account.
242 */
243 if (test_opt(inode->i_sb, DELALLOC))
244 return bg_start;
245
246 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
247 colour = (current->pid % 16) *
248 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
249 else
250 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
251 return bg_start + colour;
252}
253
254/**
255 * ext4_find_goal - find a preferred place for allocation.
256 * @inode: owner
257 * @block: block we want
258 * @partial: pointer to the last triple within a chain
259 *
260 * Normally this function find the preferred place for block allocation,
261 * returns it.
262 * Because this is only used for non-extent files, we limit the block nr
263 * to 32 bits.
264 */
265static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
266 Indirect *partial)
267{
268 ext4_fsblk_t goal;
269
270 /*
271 * XXX need to get goal block from mballoc's data structures
272 */
273
274 goal = ext4_find_near(inode, partial);
275 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
276 return goal;
277}
278
279/**
280 * ext4_blks_to_allocate - Look up the block map and count the number
281 * of direct blocks need to be allocated for the given branch.
282 *
283 * @branch: chain of indirect blocks
284 * @k: number of blocks need for indirect blocks
285 * @blks: number of data blocks to be mapped.
286 * @blocks_to_boundary: the offset in the indirect block
287 *
288 * return the total number of blocks to be allocate, including the
289 * direct and indirect blocks.
290 */
291static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
292 int blocks_to_boundary)
293{
294 unsigned int count = 0;
295
296 /*
297 * Simple case, [t,d]Indirect block(s) has not allocated yet
298 * then it's clear blocks on that path have not allocated
299 */
300 if (k > 0) {
301 /* right now we don't handle cross boundary allocation */
302 if (blks < blocks_to_boundary + 1)
303 count += blks;
304 else
305 count += blocks_to_boundary + 1;
306 return count;
307 }
308
309 count++;
310 while (count < blks && count <= blocks_to_boundary &&
311 le32_to_cpu(*(branch[0].p + count)) == 0) {
312 count++;
313 }
314 return count;
315}
316
317/**
318 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
319 * @handle: handle for this transaction
320 * @inode: inode which needs allocated blocks
321 * @iblock: the logical block to start allocated at
322 * @goal: preferred physical block of allocation
323 * @indirect_blks: the number of blocks need to allocate for indirect
324 * blocks
325 * @blks: number of desired blocks
326 * @new_blocks: on return it will store the new block numbers for
327 * the indirect blocks(if needed) and the first direct block,
328 * @err: on return it will store the error code
329 *
330 * This function will return the number of blocks allocated as
331 * requested by the passed-in parameters.
332 */
333static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
334 ext4_lblk_t iblock, ext4_fsblk_t goal,
335 int indirect_blks, int blks,
336 ext4_fsblk_t new_blocks[4], int *err)
337{
338 struct ext4_allocation_request ar;
339 int target, i;
340 unsigned long count = 0, blk_allocated = 0;
341 int index = 0;
342 ext4_fsblk_t current_block = 0;
343 int ret = 0;
344
345 /*
346 * Here we try to allocate the requested multiple blocks at once,
347 * on a best-effort basis.
348 * To build a branch, we should allocate blocks for
349 * the indirect blocks(if not allocated yet), and at least
350 * the first direct block of this branch. That's the
351 * minimum number of blocks need to allocate(required)
352 */
353 /* first we try to allocate the indirect blocks */
354 target = indirect_blks;
355 while (target > 0) {
356 count = target;
357 /* allocating blocks for indirect blocks and direct blocks */
358 current_block = ext4_new_meta_blocks(handle, inode, goal,
359 0, &count, err);
360 if (*err)
361 goto failed_out;
362
363 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
364 EXT4_ERROR_INODE(inode,
365 "current_block %llu + count %lu > %d!",
366 current_block, count,
367 EXT4_MAX_BLOCK_FILE_PHYS);
368 *err = -EIO;
369 goto failed_out;
370 }
371
372 target -= count;
373 /* allocate blocks for indirect blocks */
374 while (index < indirect_blks && count) {
375 new_blocks[index++] = current_block++;
376 count--;
377 }
378 if (count > 0) {
379 /*
380 * save the new block number
381 * for the first direct block
382 */
383 new_blocks[index] = current_block;
384 printk(KERN_INFO "%s returned more blocks than "
385 "requested\n", __func__);
386 WARN_ON(1);
387 break;
388 }
389 }
390
391 target = blks - count ;
392 blk_allocated = count;
393 if (!target)
394 goto allocated;
395 /* Now allocate data blocks */
396 memset(&ar, 0, sizeof(ar));
397 ar.inode = inode;
398 ar.goal = goal;
399 ar.len = target;
400 ar.logical = iblock;
401 if (S_ISREG(inode->i_mode))
402 /* enable in-core preallocation only for regular files */
403 ar.flags = EXT4_MB_HINT_DATA;
404
405 current_block = ext4_mb_new_blocks(handle, &ar, err);
406 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
407 EXT4_ERROR_INODE(inode,
408 "current_block %llu + ar.len %d > %d!",
409 current_block, ar.len,
410 EXT4_MAX_BLOCK_FILE_PHYS);
411 *err = -EIO;
412 goto failed_out;
413 }
414
415 if (*err && (target == blks)) {
416 /*
417 * if the allocation failed and we didn't allocate
418 * any blocks before
419 */
420 goto failed_out;
421 }
422 if (!*err) {
423 if (target == blks) {
424 /*
425 * save the new block number
426 * for the first direct block
427 */
428 new_blocks[index] = current_block;
429 }
430 blk_allocated += ar.len;
431 }
432allocated:
433 /* total number of blocks allocated for direct blocks */
434 ret = blk_allocated;
435 *err = 0;
436 return ret;
437failed_out:
438 for (i = 0; i < index; i++)
439 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
440 return ret;
441}
442
443/**
444 * ext4_alloc_branch - allocate and set up a chain of blocks.
445 * @handle: handle for this transaction
446 * @inode: owner
447 * @indirect_blks: number of allocated indirect blocks
448 * @blks: number of allocated direct blocks
449 * @goal: preferred place for allocation
450 * @offsets: offsets (in the blocks) to store the pointers to next.
451 * @branch: place to store the chain in.
452 *
453 * This function allocates blocks, zeroes out all but the last one,
454 * links them into chain and (if we are synchronous) writes them to disk.
455 * In other words, it prepares a branch that can be spliced onto the
456 * inode. It stores the information about that chain in the branch[], in
457 * the same format as ext4_get_branch() would do. We are calling it after
458 * we had read the existing part of chain and partial points to the last
459 * triple of that (one with zero ->key). Upon the exit we have the same
460 * picture as after the successful ext4_get_block(), except that in one
461 * place chain is disconnected - *branch->p is still zero (we did not
462 * set the last link), but branch->key contains the number that should
463 * be placed into *branch->p to fill that gap.
464 *
465 * If allocation fails we free all blocks we've allocated (and forget
466 * their buffer_heads) and return the error value the from failed
467 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
468 * as described above and return 0.
469 */
470static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
471 ext4_lblk_t iblock, int indirect_blks,
472 int *blks, ext4_fsblk_t goal,
473 ext4_lblk_t *offsets, Indirect *branch)
474{
475 int blocksize = inode->i_sb->s_blocksize;
476 int i, n = 0;
477 int err = 0;
478 struct buffer_head *bh;
479 int num;
480 ext4_fsblk_t new_blocks[4];
481 ext4_fsblk_t current_block;
482
483 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
484 *blks, new_blocks, &err);
485 if (err)
486 return err;
487
488 branch[0].key = cpu_to_le32(new_blocks[0]);
489 /*
490 * metadata blocks and data blocks are allocated.
491 */
492 for (n = 1; n <= indirect_blks; n++) {
493 /*
494 * Get buffer_head for parent block, zero it out
495 * and set the pointer to new one, then send
496 * parent to disk.
497 */
498 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
499 if (unlikely(!bh)) {
500 err = -EIO;
501 goto failed;
502 }
503
504 branch[n].bh = bh;
505 lock_buffer(bh);
506 BUFFER_TRACE(bh, "call get_create_access");
507 err = ext4_journal_get_create_access(handle, bh);
508 if (err) {
509 /* Don't brelse(bh) here; it's done in
510 * ext4_journal_forget() below */
511 unlock_buffer(bh);
512 goto failed;
513 }
514
515 memset(bh->b_data, 0, blocksize);
516 branch[n].p = (__le32 *) bh->b_data + offsets[n];
517 branch[n].key = cpu_to_le32(new_blocks[n]);
518 *branch[n].p = branch[n].key;
519 if (n == indirect_blks) {
520 current_block = new_blocks[n];
521 /*
522 * End of chain, update the last new metablock of
523 * the chain to point to the new allocated
524 * data blocks numbers
525 */
526 for (i = 1; i < num; i++)
527 *(branch[n].p + i) = cpu_to_le32(++current_block);
528 }
529 BUFFER_TRACE(bh, "marking uptodate");
530 set_buffer_uptodate(bh);
531 unlock_buffer(bh);
532
533 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
534 err = ext4_handle_dirty_metadata(handle, inode, bh);
535 if (err)
536 goto failed;
537 }
538 *blks = num;
539 return err;
540failed:
541 /* Allocation failed, free what we already allocated */
542 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
543 for (i = 1; i <= n ; i++) {
544 /*
545 * branch[i].bh is newly allocated, so there is no
546 * need to revoke the block, which is why we don't
547 * need to set EXT4_FREE_BLOCKS_METADATA.
548 */
549 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
550 EXT4_FREE_BLOCKS_FORGET);
551 }
552 for (i = n+1; i < indirect_blks; i++)
553 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
554
555 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
556
557 return err;
558}
559
560/**
561 * ext4_splice_branch - splice the allocated branch onto inode.
562 * @handle: handle for this transaction
563 * @inode: owner
564 * @block: (logical) number of block we are adding
565 * @chain: chain of indirect blocks (with a missing link - see
566 * ext4_alloc_branch)
567 * @where: location of missing link
568 * @num: number of indirect blocks we are adding
569 * @blks: number of direct blocks we are adding
570 *
571 * This function fills the missing link and does all housekeeping needed in
572 * inode (->i_blocks, etc.). In case of success we end up with the full
573 * chain to new block and return 0.
574 */
575static int ext4_splice_branch(handle_t *handle, struct inode *inode,
576 ext4_lblk_t block, Indirect *where, int num,
577 int blks)
578{
579 int i;
580 int err = 0;
581 ext4_fsblk_t current_block;
582
583 /*
584 * If we're splicing into a [td]indirect block (as opposed to the
585 * inode) then we need to get write access to the [td]indirect block
586 * before the splice.
587 */
588 if (where->bh) {
589 BUFFER_TRACE(where->bh, "get_write_access");
590 err = ext4_journal_get_write_access(handle, where->bh);
591 if (err)
592 goto err_out;
593 }
594 /* That's it */
595
596 *where->p = where->key;
597
598 /*
599 * Update the host buffer_head or inode to point to more just allocated
600 * direct blocks blocks
601 */
602 if (num == 0 && blks > 1) {
603 current_block = le32_to_cpu(where->key) + 1;
604 for (i = 1; i < blks; i++)
605 *(where->p + i) = cpu_to_le32(current_block++);
606 }
607
608 /* We are done with atomic stuff, now do the rest of housekeeping */
609 /* had we spliced it onto indirect block? */
610 if (where->bh) {
611 /*
612 * If we spliced it onto an indirect block, we haven't
613 * altered the inode. Note however that if it is being spliced
614 * onto an indirect block at the very end of the file (the
615 * file is growing) then we *will* alter the inode to reflect
616 * the new i_size. But that is not done here - it is done in
617 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
618 */
619 jbd_debug(5, "splicing indirect only\n");
620 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
621 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
622 if (err)
623 goto err_out;
624 } else {
625 /*
626 * OK, we spliced it into the inode itself on a direct block.
627 */
628 ext4_mark_inode_dirty(handle, inode);
629 jbd_debug(5, "splicing direct\n");
630 }
631 return err;
632
633err_out:
634 for (i = 1; i <= num; i++) {
635 /*
636 * branch[i].bh is newly allocated, so there is no
637 * need to revoke the block, which is why we don't
638 * need to set EXT4_FREE_BLOCKS_METADATA.
639 */
640 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
641 EXT4_FREE_BLOCKS_FORGET);
642 }
643 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
644 blks, 0);
645
646 return err;
647}
648
649/*
650 * The ext4_ind_map_blocks() function handles non-extents inodes
651 * (i.e., using the traditional indirect/double-indirect i_blocks
652 * scheme) for ext4_map_blocks().
653 *
654 * Allocation strategy is simple: if we have to allocate something, we will
655 * have to go the whole way to leaf. So let's do it before attaching anything
656 * to tree, set linkage between the newborn blocks, write them if sync is
657 * required, recheck the path, free and repeat if check fails, otherwise
658 * set the last missing link (that will protect us from any truncate-generated
659 * removals - all blocks on the path are immune now) and possibly force the
660 * write on the parent block.
661 * That has a nice additional property: no special recovery from the failed
662 * allocations is needed - we simply release blocks and do not touch anything
663 * reachable from inode.
664 *
665 * `handle' can be NULL if create == 0.
666 *
667 * return > 0, # of blocks mapped or allocated.
668 * return = 0, if plain lookup failed.
669 * return < 0, error case.
670 *
671 * The ext4_ind_get_blocks() function should be called with
672 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
673 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
674 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
675 * blocks.
676 */
677int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
678 struct ext4_map_blocks *map,
679 int flags)
680{
681 int err = -EIO;
682 ext4_lblk_t offsets[4];
683 Indirect chain[4];
684 Indirect *partial;
685 ext4_fsblk_t goal;
686 int indirect_blks;
687 int blocks_to_boundary = 0;
688 int depth;
689 int count = 0;
690 ext4_fsblk_t first_block = 0;
691
692 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
693 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
694 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
695 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
696 &blocks_to_boundary);
697
698 if (depth == 0)
699 goto out;
700
701 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
702
703 /* Simplest case - block found, no allocation needed */
704 if (!partial) {
705 first_block = le32_to_cpu(chain[depth - 1].key);
706 count++;
707 /*map more blocks*/
708 while (count < map->m_len && count <= blocks_to_boundary) {
709 ext4_fsblk_t blk;
710
711 blk = le32_to_cpu(*(chain[depth-1].p + count));
712
713 if (blk == first_block + count)
714 count++;
715 else
716 break;
717 }
718 goto got_it;
719 }
720
721 /* Next simple case - plain lookup or failed read of indirect block */
722 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
723 goto cleanup;
724
725 /*
726 * Okay, we need to do block allocation.
727 */
728 goal = ext4_find_goal(inode, map->m_lblk, partial);
729
730 /* the number of blocks need to allocate for [d,t]indirect blocks */
731 indirect_blks = (chain + depth) - partial - 1;
732
733 /*
734 * Next look up the indirect map to count the totoal number of
735 * direct blocks to allocate for this branch.
736 */
737 count = ext4_blks_to_allocate(partial, indirect_blks,
738 map->m_len, blocks_to_boundary);
739 /*
740 * Block out ext4_truncate while we alter the tree
741 */
742 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
743 &count, goal,
744 offsets + (partial - chain), partial);
745
746 /*
747 * The ext4_splice_branch call will free and forget any buffers
748 * on the new chain if there is a failure, but that risks using
749 * up transaction credits, especially for bitmaps where the
750 * credits cannot be returned. Can we handle this somehow? We
751 * may need to return -EAGAIN upwards in the worst case. --sct
752 */
753 if (!err)
754 err = ext4_splice_branch(handle, inode, map->m_lblk,
755 partial, indirect_blks, count);
756 if (err)
757 goto cleanup;
758
759 map->m_flags |= EXT4_MAP_NEW;
760
761 ext4_update_inode_fsync_trans(handle, inode, 1);
762got_it:
763 map->m_flags |= EXT4_MAP_MAPPED;
764 map->m_pblk = le32_to_cpu(chain[depth-1].key);
765 map->m_len = count;
766 if (count > blocks_to_boundary)
767 map->m_flags |= EXT4_MAP_BOUNDARY;
768 err = count;
769 /* Clean up and exit */
770 partial = chain + depth - 1; /* the whole chain */
771cleanup:
772 while (partial > chain) {
773 BUFFER_TRACE(partial->bh, "call brelse");
774 brelse(partial->bh);
775 partial--;
776 }
777out:
778 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
779 map->m_pblk, map->m_len, err);
780 return err;
781}
782
783/*
784 * O_DIRECT for ext3 (or indirect map) based files
785 *
786 * If the O_DIRECT write will extend the file then add this inode to the
787 * orphan list. So recovery will truncate it back to the original size
788 * if the machine crashes during the write.
789 *
790 * If the O_DIRECT write is intantiating holes inside i_size and the machine
791 * crashes then stale disk data _may_ be exposed inside the file. But current
792 * VFS code falls back into buffered path in that case so we are safe.
793 */
794ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
795 const struct iovec *iov, loff_t offset,
796 unsigned long nr_segs)
797{
798 struct file *file = iocb->ki_filp;
799 struct inode *inode = file->f_mapping->host;
800 struct ext4_inode_info *ei = EXT4_I(inode);
801 handle_t *handle;
802 ssize_t ret;
803 int orphan = 0;
804 size_t count = iov_length(iov, nr_segs);
805 int retries = 0;
806
807 if (rw == WRITE) {
808 loff_t final_size = offset + count;
809
810 if (final_size > inode->i_size) {
811 /* Credits for sb + inode write */
812 handle = ext4_journal_start(inode, 2);
813 if (IS_ERR(handle)) {
814 ret = PTR_ERR(handle);
815 goto out;
816 }
817 ret = ext4_orphan_add(handle, inode);
818 if (ret) {
819 ext4_journal_stop(handle);
820 goto out;
821 }
822 orphan = 1;
823 ei->i_disksize = inode->i_size;
824 ext4_journal_stop(handle);
825 }
826 }
827
828retry:
829 if (rw == READ && ext4_should_dioread_nolock(inode))
830 ret = __blockdev_direct_IO(rw, iocb, inode,
831 inode->i_sb->s_bdev, iov,
832 offset, nr_segs,
833 ext4_get_block, NULL, NULL, 0);
834 else {
835 ret = blockdev_direct_IO(rw, iocb, inode,
836 inode->i_sb->s_bdev, iov,
837 offset, nr_segs,
838 ext4_get_block, NULL);
839
840 if (unlikely((rw & WRITE) && ret < 0)) {
841 loff_t isize = i_size_read(inode);
842 loff_t end = offset + iov_length(iov, nr_segs);
843
844 if (end > isize)
845 ext4_truncate_failed_write(inode);
846 }
847 }
848 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
849 goto retry;
850
851 if (orphan) {
852 int err;
853
854 /* Credits for sb + inode write */
855 handle = ext4_journal_start(inode, 2);
856 if (IS_ERR(handle)) {
857 /* This is really bad luck. We've written the data
858 * but cannot extend i_size. Bail out and pretend
859 * the write failed... */
860 ret = PTR_ERR(handle);
861 if (inode->i_nlink)
862 ext4_orphan_del(NULL, inode);
863
864 goto out;
865 }
866 if (inode->i_nlink)
867 ext4_orphan_del(handle, inode);
868 if (ret > 0) {
869 loff_t end = offset + ret;
870 if (end > inode->i_size) {
871 ei->i_disksize = end;
872 i_size_write(inode, end);
873 /*
874 * We're going to return a positive `ret'
875 * here due to non-zero-length I/O, so there's
876 * no way of reporting error returns from
877 * ext4_mark_inode_dirty() to userspace. So
878 * ignore it.
879 */
880 ext4_mark_inode_dirty(handle, inode);
881 }
882 }
883 err = ext4_journal_stop(handle);
884 if (ret == 0)
885 ret = err;
886 }
887out:
888 return ret;
889}
890
891/*
892 * Calculate the number of metadata blocks need to reserve
893 * to allocate a new block at @lblocks for non extent file based file
894 */
895int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
896{
897 struct ext4_inode_info *ei = EXT4_I(inode);
898 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
899 int blk_bits;
900
901 if (lblock < EXT4_NDIR_BLOCKS)
902 return 0;
903
904 lblock -= EXT4_NDIR_BLOCKS;
905
906 if (ei->i_da_metadata_calc_len &&
907 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
908 ei->i_da_metadata_calc_len++;
909 return 0;
910 }
911 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
912 ei->i_da_metadata_calc_len = 1;
913 blk_bits = order_base_2(lblock);
914 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
915}
916
917int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
918{
919 int indirects;
920
921 /* if nrblocks are contiguous */
922 if (chunk) {
923 /*
924 * With N contiguous data blocks, we need at most
925 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
926 * 2 dindirect blocks, and 1 tindirect block
927 */
928 return DIV_ROUND_UP(nrblocks,
929 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
930 }
931 /*
932 * if nrblocks are not contiguous, worse case, each block touch
933 * a indirect block, and each indirect block touch a double indirect
934 * block, plus a triple indirect block
935 */
936 indirects = nrblocks * 2 + 1;
937 return indirects;
938}
939
940/*
941 * Truncate transactions can be complex and absolutely huge. So we need to
942 * be able to restart the transaction at a conventient checkpoint to make
943 * sure we don't overflow the journal.
944 *
945 * start_transaction gets us a new handle for a truncate transaction,
946 * and extend_transaction tries to extend the existing one a bit. If
947 * extend fails, we need to propagate the failure up and restart the
948 * transaction in the top-level truncate loop. --sct
949 */
950static handle_t *start_transaction(struct inode *inode)
951{
952 handle_t *result;
953
954 result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode));
955 if (!IS_ERR(result))
956 return result;
957
958 ext4_std_error(inode->i_sb, PTR_ERR(result));
959 return result;
960}
961
962/*
963 * Try to extend this transaction for the purposes of truncation.
964 *
965 * Returns 0 if we managed to create more room. If we can't create more
966 * room, and the transaction must be restarted we return 1.
967 */
968static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
969{
970 if (!ext4_handle_valid(handle))
971 return 0;
972 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
973 return 0;
974 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
975 return 0;
976 return 1;
977}
978
979/*
980 * Probably it should be a library function... search for first non-zero word
981 * or memcmp with zero_page, whatever is better for particular architecture.
982 * Linus?
983 */
984static inline int all_zeroes(__le32 *p, __le32 *q)
985{
986 while (p < q)
987 if (*p++)
988 return 0;
989 return 1;
990}
991
992/**
993 * ext4_find_shared - find the indirect blocks for partial truncation.
994 * @inode: inode in question
995 * @depth: depth of the affected branch
996 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
997 * @chain: place to store the pointers to partial indirect blocks
998 * @top: place to the (detached) top of branch
999 *
1000 * This is a helper function used by ext4_truncate().
1001 *
1002 * When we do truncate() we may have to clean the ends of several
1003 * indirect blocks but leave the blocks themselves alive. Block is
1004 * partially truncated if some data below the new i_size is referred
1005 * from it (and it is on the path to the first completely truncated
1006 * data block, indeed). We have to free the top of that path along
1007 * with everything to the right of the path. Since no allocation
1008 * past the truncation point is possible until ext4_truncate()
1009 * finishes, we may safely do the latter, but top of branch may
1010 * require special attention - pageout below the truncation point
1011 * might try to populate it.
1012 *
1013 * We atomically detach the top of branch from the tree, store the
1014 * block number of its root in *@top, pointers to buffer_heads of
1015 * partially truncated blocks - in @chain[].bh and pointers to
1016 * their last elements that should not be removed - in
1017 * @chain[].p. Return value is the pointer to last filled element
1018 * of @chain.
1019 *
1020 * The work left to caller to do the actual freeing of subtrees:
1021 * a) free the subtree starting from *@top
1022 * b) free the subtrees whose roots are stored in
1023 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1024 * c) free the subtrees growing from the inode past the @chain[0].
1025 * (no partially truncated stuff there). */
1026
1027static Indirect *ext4_find_shared(struct inode *inode, int depth,
1028 ext4_lblk_t offsets[4], Indirect chain[4],
1029 __le32 *top)
1030{
1031 Indirect *partial, *p;
1032 int k, err;
1033
1034 *top = 0;
1035 /* Make k index the deepest non-null offset + 1 */
1036 for (k = depth; k > 1 && !offsets[k-1]; k--)
1037 ;
1038 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1039 /* Writer: pointers */
1040 if (!partial)
1041 partial = chain + k-1;
1042 /*
1043 * If the branch acquired continuation since we've looked at it -
1044 * fine, it should all survive and (new) top doesn't belong to us.
1045 */
1046 if (!partial->key && *partial->p)
1047 /* Writer: end */
1048 goto no_top;
1049 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
1050 ;
1051 /*
1052 * OK, we've found the last block that must survive. The rest of our
1053 * branch should be detached before unlocking. However, if that rest
1054 * of branch is all ours and does not grow immediately from the inode
1055 * it's easier to cheat and just decrement partial->p.
1056 */
1057 if (p == chain + k - 1 && p > chain) {
1058 p->p--;
1059 } else {
1060 *top = *p->p;
1061 /* Nope, don't do this in ext4. Must leave the tree intact */
1062#if 0
1063 *p->p = 0;
1064#endif
1065 }
1066 /* Writer: end */
1067
1068 while (partial > p) {
1069 brelse(partial->bh);
1070 partial--;
1071 }
1072no_top:
1073 return partial;
1074}
1075
1076/*
1077 * Zero a number of block pointers in either an inode or an indirect block.
1078 * If we restart the transaction we must again get write access to the
1079 * indirect block for further modification.
1080 *
1081 * We release `count' blocks on disk, but (last - first) may be greater
1082 * than `count' because there can be holes in there.
1083 *
1084 * Return 0 on success, 1 on invalid block range
1085 * and < 0 on fatal error.
1086 */
1087static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
1088 struct buffer_head *bh,
1089 ext4_fsblk_t block_to_free,
1090 unsigned long count, __le32 *first,
1091 __le32 *last)
1092{
1093 __le32 *p;
1094 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
1095 int err;
1096
1097 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1098 flags |= EXT4_FREE_BLOCKS_METADATA;
1099
1100 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
1101 count)) {
1102 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
1103 "blocks %llu len %lu",
1104 (unsigned long long) block_to_free, count);
1105 return 1;
1106 }
1107
1108 if (try_to_extend_transaction(handle, inode)) {
1109 if (bh) {
1110 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1111 err = ext4_handle_dirty_metadata(handle, inode, bh);
1112 if (unlikely(err))
1113 goto out_err;
1114 }
1115 err = ext4_mark_inode_dirty(handle, inode);
1116 if (unlikely(err))
1117 goto out_err;
1118 err = ext4_truncate_restart_trans(handle, inode,
1119 ext4_blocks_for_truncate(inode));
1120 if (unlikely(err))
1121 goto out_err;
1122 if (bh) {
1123 BUFFER_TRACE(bh, "retaking write access");
1124 err = ext4_journal_get_write_access(handle, bh);
1125 if (unlikely(err))
1126 goto out_err;
1127 }
1128 }
1129
1130 for (p = first; p < last; p++)
1131 *p = 0;
1132
1133 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
1134 return 0;
1135out_err:
1136 ext4_std_error(inode->i_sb, err);
1137 return err;
1138}
1139
1140/**
1141 * ext4_free_data - free a list of data blocks
1142 * @handle: handle for this transaction
1143 * @inode: inode we are dealing with
1144 * @this_bh: indirect buffer_head which contains *@first and *@last
1145 * @first: array of block numbers
1146 * @last: points immediately past the end of array
1147 *
1148 * We are freeing all blocks referred from that array (numbers are stored as
1149 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1150 *
1151 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1152 * blocks are contiguous then releasing them at one time will only affect one
1153 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1154 * actually use a lot of journal space.
1155 *
1156 * @this_bh will be %NULL if @first and @last point into the inode's direct
1157 * block pointers.
1158 */
1159static void ext4_free_data(handle_t *handle, struct inode *inode,
1160 struct buffer_head *this_bh,
1161 __le32 *first, __le32 *last)
1162{
1163 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1164 unsigned long count = 0; /* Number of blocks in the run */
1165 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1166 corresponding to
1167 block_to_free */
1168 ext4_fsblk_t nr; /* Current block # */
1169 __le32 *p; /* Pointer into inode/ind
1170 for current block */
1171 int err = 0;
1172
1173 if (this_bh) { /* For indirect block */
1174 BUFFER_TRACE(this_bh, "get_write_access");
1175 err = ext4_journal_get_write_access(handle, this_bh);
1176 /* Important: if we can't update the indirect pointers
1177 * to the blocks, we can't free them. */
1178 if (err)
1179 return;
1180 }
1181
1182 for (p = first; p < last; p++) {
1183 nr = le32_to_cpu(*p);
1184 if (nr) {
1185 /* accumulate blocks to free if they're contiguous */
1186 if (count == 0) {
1187 block_to_free = nr;
1188 block_to_free_p = p;
1189 count = 1;
1190 } else if (nr == block_to_free + count) {
1191 count++;
1192 } else {
1193 err = ext4_clear_blocks(handle, inode, this_bh,
1194 block_to_free, count,
1195 block_to_free_p, p);
1196 if (err)
1197 break;
1198 block_to_free = nr;
1199 block_to_free_p = p;
1200 count = 1;
1201 }
1202 }
1203 }
1204
1205 if (!err && count > 0)
1206 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1207 count, block_to_free_p, p);
1208 if (err < 0)
1209 /* fatal error */
1210 return;
1211
1212 if (this_bh) {
1213 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1214
1215 /*
1216 * The buffer head should have an attached journal head at this
1217 * point. However, if the data is corrupted and an indirect
1218 * block pointed to itself, it would have been detached when
1219 * the block was cleared. Check for this instead of OOPSing.
1220 */
1221 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1222 ext4_handle_dirty_metadata(handle, inode, this_bh);
1223 else
1224 EXT4_ERROR_INODE(inode,
1225 "circular indirect block detected at "
1226 "block %llu",
1227 (unsigned long long) this_bh->b_blocknr);
1228 }
1229}
1230
1231/**
1232 * ext4_free_branches - free an array of branches
1233 * @handle: JBD handle for this transaction
1234 * @inode: inode we are dealing with
1235 * @parent_bh: the buffer_head which contains *@first and *@last
1236 * @first: array of block numbers
1237 * @last: pointer immediately past the end of array
1238 * @depth: depth of the branches to free
1239 *
1240 * We are freeing all blocks referred from these branches (numbers are
1241 * stored as little-endian 32-bit) and updating @inode->i_blocks
1242 * appropriately.
1243 */
1244static void ext4_free_branches(handle_t *handle, struct inode *inode,
1245 struct buffer_head *parent_bh,
1246 __le32 *first, __le32 *last, int depth)
1247{
1248 ext4_fsblk_t nr;
1249 __le32 *p;
1250
1251 if (ext4_handle_is_aborted(handle))
1252 return;
1253
1254 if (depth--) {
1255 struct buffer_head *bh;
1256 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1257 p = last;
1258 while (--p >= first) {
1259 nr = le32_to_cpu(*p);
1260 if (!nr)
1261 continue; /* A hole */
1262
1263 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1264 nr, 1)) {
1265 EXT4_ERROR_INODE(inode,
1266 "invalid indirect mapped "
1267 "block %lu (level %d)",
1268 (unsigned long) nr, depth);
1269 break;
1270 }
1271
1272 /* Go read the buffer for the next level down */
1273 bh = sb_bread(inode->i_sb, nr);
1274
1275 /*
1276 * A read failure? Report error and clear slot
1277 * (should be rare).
1278 */
1279 if (!bh) {
1280 EXT4_ERROR_INODE_BLOCK(inode, nr,
1281 "Read failure");
1282 continue;
1283 }
1284
1285 /* This zaps the entire block. Bottom up. */
1286 BUFFER_TRACE(bh, "free child branches");
1287 ext4_free_branches(handle, inode, bh,
1288 (__le32 *) bh->b_data,
1289 (__le32 *) bh->b_data + addr_per_block,
1290 depth);
1291 brelse(bh);
1292
1293 /*
1294 * Everything below this this pointer has been
1295 * released. Now let this top-of-subtree go.
1296 *
1297 * We want the freeing of this indirect block to be
1298 * atomic in the journal with the updating of the
1299 * bitmap block which owns it. So make some room in
1300 * the journal.
1301 *
1302 * We zero the parent pointer *after* freeing its
1303 * pointee in the bitmaps, so if extend_transaction()
1304 * for some reason fails to put the bitmap changes and
1305 * the release into the same transaction, recovery
1306 * will merely complain about releasing a free block,
1307 * rather than leaking blocks.
1308 */
1309 if (ext4_handle_is_aborted(handle))
1310 return;
1311 if (try_to_extend_transaction(handle, inode)) {
1312 ext4_mark_inode_dirty(handle, inode);
1313 ext4_truncate_restart_trans(handle, inode,
1314 ext4_blocks_for_truncate(inode));
1315 }
1316
1317 /*
1318 * The forget flag here is critical because if
1319 * we are journaling (and not doing data
1320 * journaling), we have to make sure a revoke
1321 * record is written to prevent the journal
1322 * replay from overwriting the (former)
1323 * indirect block if it gets reallocated as a
1324 * data block. This must happen in the same
1325 * transaction where the data blocks are
1326 * actually freed.
1327 */
1328 ext4_free_blocks(handle, inode, NULL, nr, 1,
1329 EXT4_FREE_BLOCKS_METADATA|
1330 EXT4_FREE_BLOCKS_FORGET);
1331
1332 if (parent_bh) {
1333 /*
1334 * The block which we have just freed is
1335 * pointed to by an indirect block: journal it
1336 */
1337 BUFFER_TRACE(parent_bh, "get_write_access");
1338 if (!ext4_journal_get_write_access(handle,
1339 parent_bh)){
1340 *p = 0;
1341 BUFFER_TRACE(parent_bh,
1342 "call ext4_handle_dirty_metadata");
1343 ext4_handle_dirty_metadata(handle,
1344 inode,
1345 parent_bh);
1346 }
1347 }
1348 }
1349 } else {
1350 /* We have reached the bottom of the tree. */
1351 BUFFER_TRACE(parent_bh, "free data blocks");
1352 ext4_free_data(handle, inode, parent_bh, first, last);
1353 }
1354}
1355
1356void ext4_ind_truncate(struct inode *inode)
1357{
1358 handle_t *handle;
1359 struct ext4_inode_info *ei = EXT4_I(inode);
1360 __le32 *i_data = ei->i_data;
1361 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1362 struct address_space *mapping = inode->i_mapping;
1363 ext4_lblk_t offsets[4];
1364 Indirect chain[4];
1365 Indirect *partial;
1366 __le32 nr = 0;
1367 int n = 0;
1368 ext4_lblk_t last_block, max_block;
1369 unsigned blocksize = inode->i_sb->s_blocksize;
1370
1371 handle = start_transaction(inode);
1372 if (IS_ERR(handle))
1373 return; /* AKPM: return what? */
1374
1375 last_block = (inode->i_size + blocksize-1)
1376 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1377 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1378 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1379
1380 if (inode->i_size & (blocksize - 1))
1381 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
1382 goto out_stop;
1383
1384 if (last_block != max_block) {
1385 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1386 if (n == 0)
1387 goto out_stop; /* error */
1388 }
1389
1390 /*
1391 * OK. This truncate is going to happen. We add the inode to the
1392 * orphan list, so that if this truncate spans multiple transactions,
1393 * and we crash, we will resume the truncate when the filesystem
1394 * recovers. It also marks the inode dirty, to catch the new size.
1395 *
1396 * Implication: the file must always be in a sane, consistent
1397 * truncatable state while each transaction commits.
1398 */
1399 if (ext4_orphan_add(handle, inode))
1400 goto out_stop;
1401
1402 /*
1403 * From here we block out all ext4_get_block() callers who want to
1404 * modify the block allocation tree.
1405 */
1406 down_write(&ei->i_data_sem);
1407
1408 ext4_discard_preallocations(inode);
1409
1410 /*
1411 * The orphan list entry will now protect us from any crash which
1412 * occurs before the truncate completes, so it is now safe to propagate
1413 * the new, shorter inode size (held for now in i_size) into the
1414 * on-disk inode. We do this via i_disksize, which is the value which
1415 * ext4 *really* writes onto the disk inode.
1416 */
1417 ei->i_disksize = inode->i_size;
1418
1419 if (last_block == max_block) {
1420 /*
1421 * It is unnecessary to free any data blocks if last_block is
1422 * equal to the indirect block limit.
1423 */
1424 goto out_unlock;
1425 } else if (n == 1) { /* direct blocks */
1426 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1427 i_data + EXT4_NDIR_BLOCKS);
1428 goto do_indirects;
1429 }
1430
1431 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1432 /* Kill the top of shared branch (not detached) */
1433 if (nr) {
1434 if (partial == chain) {
1435 /* Shared branch grows from the inode */
1436 ext4_free_branches(handle, inode, NULL,
1437 &nr, &nr+1, (chain+n-1) - partial);
1438 *partial->p = 0;
1439 /*
1440 * We mark the inode dirty prior to restart,
1441 * and prior to stop. No need for it here.
1442 */
1443 } else {
1444 /* Shared branch grows from an indirect block */
1445 BUFFER_TRACE(partial->bh, "get_write_access");
1446 ext4_free_branches(handle, inode, partial->bh,
1447 partial->p,
1448 partial->p+1, (chain+n-1) - partial);
1449 }
1450 }
1451 /* Clear the ends of indirect blocks on the shared branch */
1452 while (partial > chain) {
1453 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1454 (__le32*)partial->bh->b_data+addr_per_block,
1455 (chain+n-1) - partial);
1456 BUFFER_TRACE(partial->bh, "call brelse");
1457 brelse(partial->bh);
1458 partial--;
1459 }
1460do_indirects:
1461 /* Kill the remaining (whole) subtrees */
1462 switch (offsets[0]) {
1463 default:
1464 nr = i_data[EXT4_IND_BLOCK];
1465 if (nr) {
1466 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1467 i_data[EXT4_IND_BLOCK] = 0;
1468 }
1469 case EXT4_IND_BLOCK:
1470 nr = i_data[EXT4_DIND_BLOCK];
1471 if (nr) {
1472 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1473 i_data[EXT4_DIND_BLOCK] = 0;
1474 }
1475 case EXT4_DIND_BLOCK:
1476 nr = i_data[EXT4_TIND_BLOCK];
1477 if (nr) {
1478 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1479 i_data[EXT4_TIND_BLOCK] = 0;
1480 }
1481 case EXT4_TIND_BLOCK:
1482 ;
1483 }
1484
1485out_unlock:
1486 up_write(&ei->i_data_sem);
1487 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1488 ext4_mark_inode_dirty(handle, inode);
1489
1490 /*
1491 * In a multi-transaction truncate, we only make the final transaction
1492 * synchronous
1493 */
1494 if (IS_SYNC(inode))
1495 ext4_handle_sync(handle);
1496out_stop:
1497 /*
1498 * If this was a simple ftruncate(), and the file will remain alive
1499 * then we need to clear up the orphan record which we created above.
1500 * However, if this was a real unlink then we were called by
1501 * ext4_delete_inode(), and we allow that function to clean up the
1502 * orphan info for us.
1503 */
1504 if (inode->i_nlink)
1505 ext4_orphan_del(handle, inode);
1506
1507 ext4_journal_stop(handle);
1508 trace_ext4_truncate_exit(inode);
1509}
1510
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 9b82ac7b0f55..de50b16a8f67 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -12,10 +12,6 @@
12 * 12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds 13 * Copyright (C) 1991, 1992 Linus Torvalds
14 * 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 15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz) 16 * (jj@sunsite.ms.mff.cuni.cz)
21 * 17 *
@@ -90,45 +86,6 @@ static int ext4_inode_is_fast_symlink(struct inode *inode)
90} 86}
91 87
92/* 88/*
93 * Truncate transactions can be complex and absolutely huge. So we need to
94 * be able to restart the transaction at a conventient checkpoint to make
95 * sure we don't overflow the journal.
96 *
97 * start_transaction gets us a new handle for a truncate transaction,
98 * and extend_transaction tries to extend the existing one a bit. If
99 * extend fails, we need to propagate the failure up and restart the
100 * transaction in the top-level truncate loop. --sct
101 */
102static handle_t *start_transaction(struct inode *inode)
103{
104 handle_t *result;
105
106 result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode));
107 if (!IS_ERR(result))
108 return result;
109
110 ext4_std_error(inode->i_sb, PTR_ERR(result));
111 return result;
112}
113
114/*
115 * Try to extend this transaction for the purposes of truncation.
116 *
117 * Returns 0 if we managed to create more room. If we can't create more
118 * room, and the transaction must be restarted we return 1.
119 */
120static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
121{
122 if (!ext4_handle_valid(handle))
123 return 0;
124 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
125 return 0;
126 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
127 return 0;
128 return 1;
129}
130
131/*
132 * Restart the transaction associated with *handle. This does a commit, 89 * Restart the transaction associated with *handle. This does a commit,
133 * so before we call here everything must be consistently dirtied against 90 * so before we call here everything must be consistently dirtied against
134 * this transaction. 91 * this transaction.
@@ -251,760 +208,6 @@ no_delete:
251 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ 208 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
252} 209}
253 210
254typedef struct {
255 __le32 *p;
256 __le32 key;
257 struct buffer_head *bh;
258} Indirect;
259
260static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
261{
262 p->key = *(p->p = v);
263 p->bh = bh;
264}
265
266/**
267 * ext4_block_to_path - parse the block number into array of offsets
268 * @inode: inode in question (we are only interested in its superblock)
269 * @i_block: block number to be parsed
270 * @offsets: array to store the offsets in
271 * @boundary: set this non-zero if the referred-to block is likely to be
272 * followed (on disk) by an indirect block.
273 *
274 * To store the locations of file's data ext4 uses a data structure common
275 * for UNIX filesystems - tree of pointers anchored in the inode, with
276 * data blocks at leaves and indirect blocks in intermediate nodes.
277 * This function translates the block number into path in that tree -
278 * return value is the path length and @offsets[n] is the offset of
279 * pointer to (n+1)th node in the nth one. If @block is out of range
280 * (negative or too large) warning is printed and zero returned.
281 *
282 * Note: function doesn't find node addresses, so no IO is needed. All
283 * we need to know is the capacity of indirect blocks (taken from the
284 * inode->i_sb).
285 */
286
287/*
288 * Portability note: the last comparison (check that we fit into triple
289 * indirect block) is spelled differently, because otherwise on an
290 * architecture with 32-bit longs and 8Kb pages we might get into trouble
291 * if our filesystem had 8Kb blocks. We might use long long, but that would
292 * kill us on x86. Oh, well, at least the sign propagation does not matter -
293 * i_block would have to be negative in the very beginning, so we would not
294 * get there at all.
295 */
296
297static int ext4_block_to_path(struct inode *inode,
298 ext4_lblk_t i_block,
299 ext4_lblk_t offsets[4], int *boundary)
300{
301 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
302 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
303 const long direct_blocks = EXT4_NDIR_BLOCKS,
304 indirect_blocks = ptrs,
305 double_blocks = (1 << (ptrs_bits * 2));
306 int n = 0;
307 int final = 0;
308
309 if (i_block < direct_blocks) {
310 offsets[n++] = i_block;
311 final = direct_blocks;
312 } else if ((i_block -= direct_blocks) < indirect_blocks) {
313 offsets[n++] = EXT4_IND_BLOCK;
314 offsets[n++] = i_block;
315 final = ptrs;
316 } else if ((i_block -= indirect_blocks) < double_blocks) {
317 offsets[n++] = EXT4_DIND_BLOCK;
318 offsets[n++] = i_block >> ptrs_bits;
319 offsets[n++] = i_block & (ptrs - 1);
320 final = ptrs;
321 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
322 offsets[n++] = EXT4_TIND_BLOCK;
323 offsets[n++] = i_block >> (ptrs_bits * 2);
324 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
325 offsets[n++] = i_block & (ptrs - 1);
326 final = ptrs;
327 } else {
328 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
329 i_block + direct_blocks +
330 indirect_blocks + double_blocks, inode->i_ino);
331 }
332 if (boundary)
333 *boundary = final - 1 - (i_block & (ptrs - 1));
334 return n;
335}
336
337/**
338 * ext4_get_branch - read the chain of indirect blocks leading to data
339 * @inode: inode in question
340 * @depth: depth of the chain (1 - direct pointer, etc.)
341 * @offsets: offsets of pointers in inode/indirect blocks
342 * @chain: place to store the result
343 * @err: here we store the error value
344 *
345 * Function fills the array of triples <key, p, bh> and returns %NULL
346 * if everything went OK or the pointer to the last filled triple
347 * (incomplete one) otherwise. Upon the return chain[i].key contains
348 * the number of (i+1)-th block in the chain (as it is stored in memory,
349 * i.e. little-endian 32-bit), chain[i].p contains the address of that
350 * number (it points into struct inode for i==0 and into the bh->b_data
351 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
352 * block for i>0 and NULL for i==0. In other words, it holds the block
353 * numbers of the chain, addresses they were taken from (and where we can
354 * verify that chain did not change) and buffer_heads hosting these
355 * numbers.
356 *
357 * Function stops when it stumbles upon zero pointer (absent block)
358 * (pointer to last triple returned, *@err == 0)
359 * or when it gets an IO error reading an indirect block
360 * (ditto, *@err == -EIO)
361 * or when it reads all @depth-1 indirect blocks successfully and finds
362 * the whole chain, all way to the data (returns %NULL, *err == 0).
363 *
364 * Need to be called with
365 * down_read(&EXT4_I(inode)->i_data_sem)
366 */
367static Indirect *ext4_get_branch(struct inode *inode, int depth,
368 ext4_lblk_t *offsets,
369 Indirect chain[4], int *err)
370{
371 struct super_block *sb = inode->i_sb;
372 Indirect *p = chain;
373 struct buffer_head *bh;
374
375 *err = 0;
376 /* i_data is not going away, no lock needed */
377 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
378 if (!p->key)
379 goto no_block;
380 while (--depth) {
381 bh = sb_getblk(sb, le32_to_cpu(p->key));
382 if (unlikely(!bh))
383 goto failure;
384
385 if (!bh_uptodate_or_lock(bh)) {
386 if (bh_submit_read(bh) < 0) {
387 put_bh(bh);
388 goto failure;
389 }
390 /* validate block references */
391 if (ext4_check_indirect_blockref(inode, bh)) {
392 put_bh(bh);
393 goto failure;
394 }
395 }
396
397 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
398 /* Reader: end */
399 if (!p->key)
400 goto no_block;
401 }
402 return NULL;
403
404failure:
405 *err = -EIO;
406no_block:
407 return p;
408}
409
410/**
411 * ext4_find_near - find a place for allocation with sufficient locality
412 * @inode: owner
413 * @ind: descriptor of indirect block.
414 *
415 * This function returns the preferred place for block allocation.
416 * It is used when heuristic for sequential allocation fails.
417 * Rules are:
418 * + if there is a block to the left of our position - allocate near it.
419 * + if pointer will live in indirect block - allocate near that block.
420 * + if pointer will live in inode - allocate in the same
421 * cylinder group.
422 *
423 * In the latter case we colour the starting block by the callers PID to
424 * prevent it from clashing with concurrent allocations for a different inode
425 * in the same block group. The PID is used here so that functionally related
426 * files will be close-by on-disk.
427 *
428 * Caller must make sure that @ind is valid and will stay that way.
429 */
430static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
431{
432 struct ext4_inode_info *ei = EXT4_I(inode);
433 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
434 __le32 *p;
435 ext4_fsblk_t bg_start;
436 ext4_fsblk_t last_block;
437 ext4_grpblk_t colour;
438 ext4_group_t block_group;
439 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
440
441 /* Try to find previous block */
442 for (p = ind->p - 1; p >= start; p--) {
443 if (*p)
444 return le32_to_cpu(*p);
445 }
446
447 /* No such thing, so let's try location of indirect block */
448 if (ind->bh)
449 return ind->bh->b_blocknr;
450
451 /*
452 * It is going to be referred to from the inode itself? OK, just put it
453 * into the same cylinder group then.
454 */
455 block_group = ei->i_block_group;
456 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
457 block_group &= ~(flex_size-1);
458 if (S_ISREG(inode->i_mode))
459 block_group++;
460 }
461 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
462 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
463
464 /*
465 * If we are doing delayed allocation, we don't need take
466 * colour into account.
467 */
468 if (test_opt(inode->i_sb, DELALLOC))
469 return bg_start;
470
471 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
472 colour = (current->pid % 16) *
473 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
474 else
475 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
476 return bg_start + colour;
477}
478
479/**
480 * ext4_find_goal - find a preferred place for allocation.
481 * @inode: owner
482 * @block: block we want
483 * @partial: pointer to the last triple within a chain
484 *
485 * Normally this function find the preferred place for block allocation,
486 * returns it.
487 * Because this is only used for non-extent files, we limit the block nr
488 * to 32 bits.
489 */
490static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
491 Indirect *partial)
492{
493 ext4_fsblk_t goal;
494
495 /*
496 * XXX need to get goal block from mballoc's data structures
497 */
498
499 goal = ext4_find_near(inode, partial);
500 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
501 return goal;
502}
503
504/**
505 * ext4_blks_to_allocate - Look up the block map and count the number
506 * of direct blocks need to be allocated for the given branch.
507 *
508 * @branch: chain of indirect blocks
509 * @k: number of blocks need for indirect blocks
510 * @blks: number of data blocks to be mapped.
511 * @blocks_to_boundary: the offset in the indirect block
512 *
513 * return the total number of blocks to be allocate, including the
514 * direct and indirect blocks.
515 */
516static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
517 int blocks_to_boundary)
518{
519 unsigned int count = 0;
520
521 /*
522 * Simple case, [t,d]Indirect block(s) has not allocated yet
523 * then it's clear blocks on that path have not allocated
524 */
525 if (k > 0) {
526 /* right now we don't handle cross boundary allocation */
527 if (blks < blocks_to_boundary + 1)
528 count += blks;
529 else
530 count += blocks_to_boundary + 1;
531 return count;
532 }
533
534 count++;
535 while (count < blks && count <= blocks_to_boundary &&
536 le32_to_cpu(*(branch[0].p + count)) == 0) {
537 count++;
538 }
539 return count;
540}
541
542/**
543 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
544 * @handle: handle for this transaction
545 * @inode: inode which needs allocated blocks
546 * @iblock: the logical block to start allocated at
547 * @goal: preferred physical block of allocation
548 * @indirect_blks: the number of blocks need to allocate for indirect
549 * blocks
550 * @blks: number of desired blocks
551 * @new_blocks: on return it will store the new block numbers for
552 * the indirect blocks(if needed) and the first direct block,
553 * @err: on return it will store the error code
554 *
555 * This function will return the number of blocks allocated as
556 * requested by the passed-in parameters.
557 */
558static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
559 ext4_lblk_t iblock, ext4_fsblk_t goal,
560 int indirect_blks, int blks,
561 ext4_fsblk_t new_blocks[4], int *err)
562{
563 struct ext4_allocation_request ar;
564 int target, i;
565 unsigned long count = 0, blk_allocated = 0;
566 int index = 0;
567 ext4_fsblk_t current_block = 0;
568 int ret = 0;
569
570 /*
571 * Here we try to allocate the requested multiple blocks at once,
572 * on a best-effort basis.
573 * To build a branch, we should allocate blocks for
574 * the indirect blocks(if not allocated yet), and at least
575 * the first direct block of this branch. That's the
576 * minimum number of blocks need to allocate(required)
577 */
578 /* first we try to allocate the indirect blocks */
579 target = indirect_blks;
580 while (target > 0) {
581 count = target;
582 /* allocating blocks for indirect blocks and direct blocks */
583 current_block = ext4_new_meta_blocks(handle, inode, goal,
584 0, &count, err);
585 if (*err)
586 goto failed_out;
587
588 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
589 EXT4_ERROR_INODE(inode,
590 "current_block %llu + count %lu > %d!",
591 current_block, count,
592 EXT4_MAX_BLOCK_FILE_PHYS);
593 *err = -EIO;
594 goto failed_out;
595 }
596
597 target -= count;
598 /* allocate blocks for indirect blocks */
599 while (index < indirect_blks && count) {
600 new_blocks[index++] = current_block++;
601 count--;
602 }
603 if (count > 0) {
604 /*
605 * save the new block number
606 * for the first direct block
607 */
608 new_blocks[index] = current_block;
609 printk(KERN_INFO "%s returned more blocks than "
610 "requested\n", __func__);
611 WARN_ON(1);
612 break;
613 }
614 }
615
616 target = blks - count ;
617 blk_allocated = count;
618 if (!target)
619 goto allocated;
620 /* Now allocate data blocks */
621 memset(&ar, 0, sizeof(ar));
622 ar.inode = inode;
623 ar.goal = goal;
624 ar.len = target;
625 ar.logical = iblock;
626 if (S_ISREG(inode->i_mode))
627 /* enable in-core preallocation only for regular files */
628 ar.flags = EXT4_MB_HINT_DATA;
629
630 current_block = ext4_mb_new_blocks(handle, &ar, err);
631 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
632 EXT4_ERROR_INODE(inode,
633 "current_block %llu + ar.len %d > %d!",
634 current_block, ar.len,
635 EXT4_MAX_BLOCK_FILE_PHYS);
636 *err = -EIO;
637 goto failed_out;
638 }
639
640 if (*err && (target == blks)) {
641 /*
642 * if the allocation failed and we didn't allocate
643 * any blocks before
644 */
645 goto failed_out;
646 }
647 if (!*err) {
648 if (target == blks) {
649 /*
650 * save the new block number
651 * for the first direct block
652 */
653 new_blocks[index] = current_block;
654 }
655 blk_allocated += ar.len;
656 }
657allocated:
658 /* total number of blocks allocated for direct blocks */
659 ret = blk_allocated;
660 *err = 0;
661 return ret;
662failed_out:
663 for (i = 0; i < index; i++)
664 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
665 return ret;
666}
667
668/**
669 * ext4_alloc_branch - allocate and set up a chain of blocks.
670 * @handle: handle for this transaction
671 * @inode: owner
672 * @indirect_blks: number of allocated indirect blocks
673 * @blks: number of allocated direct blocks
674 * @goal: preferred place for allocation
675 * @offsets: offsets (in the blocks) to store the pointers to next.
676 * @branch: place to store the chain in.
677 *
678 * This function allocates blocks, zeroes out all but the last one,
679 * links them into chain and (if we are synchronous) writes them to disk.
680 * In other words, it prepares a branch that can be spliced onto the
681 * inode. It stores the information about that chain in the branch[], in
682 * the same format as ext4_get_branch() would do. We are calling it after
683 * we had read the existing part of chain and partial points to the last
684 * triple of that (one with zero ->key). Upon the exit we have the same
685 * picture as after the successful ext4_get_block(), except that in one
686 * place chain is disconnected - *branch->p is still zero (we did not
687 * set the last link), but branch->key contains the number that should
688 * be placed into *branch->p to fill that gap.
689 *
690 * If allocation fails we free all blocks we've allocated (and forget
691 * their buffer_heads) and return the error value the from failed
692 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
693 * as described above and return 0.
694 */
695static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
696 ext4_lblk_t iblock, int indirect_blks,
697 int *blks, ext4_fsblk_t goal,
698 ext4_lblk_t *offsets, Indirect *branch)
699{
700 int blocksize = inode->i_sb->s_blocksize;
701 int i, n = 0;
702 int err = 0;
703 struct buffer_head *bh;
704 int num;
705 ext4_fsblk_t new_blocks[4];
706 ext4_fsblk_t current_block;
707
708 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
709 *blks, new_blocks, &err);
710 if (err)
711 return err;
712
713 branch[0].key = cpu_to_le32(new_blocks[0]);
714 /*
715 * metadata blocks and data blocks are allocated.
716 */
717 for (n = 1; n <= indirect_blks; n++) {
718 /*
719 * Get buffer_head for parent block, zero it out
720 * and set the pointer to new one, then send
721 * parent to disk.
722 */
723 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
724 if (unlikely(!bh)) {
725 err = -EIO;
726 goto failed;
727 }
728
729 branch[n].bh = bh;
730 lock_buffer(bh);
731 BUFFER_TRACE(bh, "call get_create_access");
732 err = ext4_journal_get_create_access(handle, bh);
733 if (err) {
734 /* Don't brelse(bh) here; it's done in
735 * ext4_journal_forget() below */
736 unlock_buffer(bh);
737 goto failed;
738 }
739
740 memset(bh->b_data, 0, blocksize);
741 branch[n].p = (__le32 *) bh->b_data + offsets[n];
742 branch[n].key = cpu_to_le32(new_blocks[n]);
743 *branch[n].p = branch[n].key;
744 if (n == indirect_blks) {
745 current_block = new_blocks[n];
746 /*
747 * End of chain, update the last new metablock of
748 * the chain to point to the new allocated
749 * data blocks numbers
750 */
751 for (i = 1; i < num; i++)
752 *(branch[n].p + i) = cpu_to_le32(++current_block);
753 }
754 BUFFER_TRACE(bh, "marking uptodate");
755 set_buffer_uptodate(bh);
756 unlock_buffer(bh);
757
758 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
759 err = ext4_handle_dirty_metadata(handle, inode, bh);
760 if (err)
761 goto failed;
762 }
763 *blks = num;
764 return err;
765failed:
766 /* Allocation failed, free what we already allocated */
767 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
768 for (i = 1; i <= n ; i++) {
769 /*
770 * branch[i].bh is newly allocated, so there is no
771 * need to revoke the block, which is why we don't
772 * need to set EXT4_FREE_BLOCKS_METADATA.
773 */
774 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
775 EXT4_FREE_BLOCKS_FORGET);
776 }
777 for (i = n+1; i < indirect_blks; i++)
778 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
779
780 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
781
782 return err;
783}
784
785/**
786 * ext4_splice_branch - splice the allocated branch onto inode.
787 * @handle: handle for this transaction
788 * @inode: owner
789 * @block: (logical) number of block we are adding
790 * @chain: chain of indirect blocks (with a missing link - see
791 * ext4_alloc_branch)
792 * @where: location of missing link
793 * @num: number of indirect blocks we are adding
794 * @blks: number of direct blocks we are adding
795 *
796 * This function fills the missing link and does all housekeeping needed in
797 * inode (->i_blocks, etc.). In case of success we end up with the full
798 * chain to new block and return 0.
799 */
800static int ext4_splice_branch(handle_t *handle, struct inode *inode,
801 ext4_lblk_t block, Indirect *where, int num,
802 int blks)
803{
804 int i;
805 int err = 0;
806 ext4_fsblk_t current_block;
807
808 /*
809 * If we're splicing into a [td]indirect block (as opposed to the
810 * inode) then we need to get write access to the [td]indirect block
811 * before the splice.
812 */
813 if (where->bh) {
814 BUFFER_TRACE(where->bh, "get_write_access");
815 err = ext4_journal_get_write_access(handle, where->bh);
816 if (err)
817 goto err_out;
818 }
819 /* That's it */
820
821 *where->p = where->key;
822
823 /*
824 * Update the host buffer_head or inode to point to more just allocated
825 * direct blocks blocks
826 */
827 if (num == 0 && blks > 1) {
828 current_block = le32_to_cpu(where->key) + 1;
829 for (i = 1; i < blks; i++)
830 *(where->p + i) = cpu_to_le32(current_block++);
831 }
832
833 /* We are done with atomic stuff, now do the rest of housekeeping */
834 /* had we spliced it onto indirect block? */
835 if (where->bh) {
836 /*
837 * If we spliced it onto an indirect block, we haven't
838 * altered the inode. Note however that if it is being spliced
839 * onto an indirect block at the very end of the file (the
840 * file is growing) then we *will* alter the inode to reflect
841 * the new i_size. But that is not done here - it is done in
842 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
843 */
844 jbd_debug(5, "splicing indirect only\n");
845 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
846 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
847 if (err)
848 goto err_out;
849 } else {
850 /*
851 * OK, we spliced it into the inode itself on a direct block.
852 */
853 ext4_mark_inode_dirty(handle, inode);
854 jbd_debug(5, "splicing direct\n");
855 }
856 return err;
857
858err_out:
859 for (i = 1; i <= num; i++) {
860 /*
861 * branch[i].bh is newly allocated, so there is no
862 * need to revoke the block, which is why we don't
863 * need to set EXT4_FREE_BLOCKS_METADATA.
864 */
865 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
866 EXT4_FREE_BLOCKS_FORGET);
867 }
868 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
869 blks, 0);
870
871 return err;
872}
873
874/*
875 * The ext4_ind_map_blocks() function handles non-extents inodes
876 * (i.e., using the traditional indirect/double-indirect i_blocks
877 * scheme) for ext4_map_blocks().
878 *
879 * Allocation strategy is simple: if we have to allocate something, we will
880 * have to go the whole way to leaf. So let's do it before attaching anything
881 * to tree, set linkage between the newborn blocks, write them if sync is
882 * required, recheck the path, free and repeat if check fails, otherwise
883 * set the last missing link (that will protect us from any truncate-generated
884 * removals - all blocks on the path are immune now) and possibly force the
885 * write on the parent block.
886 * That has a nice additional property: no special recovery from the failed
887 * allocations is needed - we simply release blocks and do not touch anything
888 * reachable from inode.
889 *
890 * `handle' can be NULL if create == 0.
891 *
892 * return > 0, # of blocks mapped or allocated.
893 * return = 0, if plain lookup failed.
894 * return < 0, error case.
895 *
896 * The ext4_ind_get_blocks() function should be called with
897 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
898 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
899 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
900 * blocks.
901 */
902static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
903 struct ext4_map_blocks *map,
904 int flags)
905{
906 int err = -EIO;
907 ext4_lblk_t offsets[4];
908 Indirect chain[4];
909 Indirect *partial;
910 ext4_fsblk_t goal;
911 int indirect_blks;
912 int blocks_to_boundary = 0;
913 int depth;
914 int count = 0;
915 ext4_fsblk_t first_block = 0;
916
917 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
918 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
919 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
920 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
921 &blocks_to_boundary);
922
923 if (depth == 0)
924 goto out;
925
926 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
927
928 /* Simplest case - block found, no allocation needed */
929 if (!partial) {
930 first_block = le32_to_cpu(chain[depth - 1].key);
931 count++;
932 /*map more blocks*/
933 while (count < map->m_len && count <= blocks_to_boundary) {
934 ext4_fsblk_t blk;
935
936 blk = le32_to_cpu(*(chain[depth-1].p + count));
937
938 if (blk == first_block + count)
939 count++;
940 else
941 break;
942 }
943 goto got_it;
944 }
945
946 /* Next simple case - plain lookup or failed read of indirect block */
947 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
948 goto cleanup;
949
950 /*
951 * Okay, we need to do block allocation.
952 */
953 goal = ext4_find_goal(inode, map->m_lblk, partial);
954
955 /* the number of blocks need to allocate for [d,t]indirect blocks */
956 indirect_blks = (chain + depth) - partial - 1;
957
958 /*
959 * Next look up the indirect map to count the totoal number of
960 * direct blocks to allocate for this branch.
961 */
962 count = ext4_blks_to_allocate(partial, indirect_blks,
963 map->m_len, blocks_to_boundary);
964 /*
965 * Block out ext4_truncate while we alter the tree
966 */
967 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
968 &count, goal,
969 offsets + (partial - chain), partial);
970
971 /*
972 * The ext4_splice_branch call will free and forget any buffers
973 * on the new chain if there is a failure, but that risks using
974 * up transaction credits, especially for bitmaps where the
975 * credits cannot be returned. Can we handle this somehow? We
976 * may need to return -EAGAIN upwards in the worst case. --sct
977 */
978 if (!err)
979 err = ext4_splice_branch(handle, inode, map->m_lblk,
980 partial, indirect_blks, count);
981 if (err)
982 goto cleanup;
983
984 map->m_flags |= EXT4_MAP_NEW;
985
986 ext4_update_inode_fsync_trans(handle, inode, 1);
987got_it:
988 map->m_flags |= EXT4_MAP_MAPPED;
989 map->m_pblk = le32_to_cpu(chain[depth-1].key);
990 map->m_len = count;
991 if (count > blocks_to_boundary)
992 map->m_flags |= EXT4_MAP_BOUNDARY;
993 err = count;
994 /* Clean up and exit */
995 partial = chain + depth - 1; /* the whole chain */
996cleanup:
997 while (partial > chain) {
998 BUFFER_TRACE(partial->bh, "call brelse");
999 brelse(partial->bh);
1000 partial--;
1001 }
1002out:
1003 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
1004 map->m_pblk, map->m_len, err);
1005 return err;
1006}
1007
1008#ifdef CONFIG_QUOTA 211#ifdef CONFIG_QUOTA
1009qsize_t *ext4_get_reserved_space(struct inode *inode) 212qsize_t *ext4_get_reserved_space(struct inode *inode)
1010{ 213{
@@ -1014,32 +217,6 @@ qsize_t *ext4_get_reserved_space(struct inode *inode)
1014 217
1015/* 218/*
1016 * Calculate the number of metadata blocks need to reserve 219 * Calculate the number of metadata blocks need to reserve
1017 * to allocate a new block at @lblocks for non extent file based file
1018 */
1019static int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
1020{
1021 struct ext4_inode_info *ei = EXT4_I(inode);
1022 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1023 int blk_bits;
1024
1025 if (lblock < EXT4_NDIR_BLOCKS)
1026 return 0;
1027
1028 lblock -= EXT4_NDIR_BLOCKS;
1029
1030 if (ei->i_da_metadata_calc_len &&
1031 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1032 ei->i_da_metadata_calc_len++;
1033 return 0;
1034 }
1035 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1036 ei->i_da_metadata_calc_len = 1;
1037 blk_bits = order_base_2(lblock);
1038 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1039}
1040
1041/*
1042 * Calculate the number of metadata blocks need to reserve
1043 * to allocate a block located at @lblock 220 * to allocate a block located at @lblock
1044 */ 221 */
1045static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) 222static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
@@ -3380,114 +2557,6 @@ static int ext4_releasepage(struct page *page, gfp_t wait)
3380} 2557}
3381 2558
3382/* 2559/*
3383 * O_DIRECT for ext3 (or indirect map) based files
3384 *
3385 * If the O_DIRECT write will extend the file then add this inode to the
3386 * orphan list. So recovery will truncate it back to the original size
3387 * if the machine crashes during the write.
3388 *
3389 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3390 * crashes then stale disk data _may_ be exposed inside the file. But current
3391 * VFS code falls back into buffered path in that case so we are safe.
3392 */
3393static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3394 const struct iovec *iov, loff_t offset,
3395 unsigned long nr_segs)
3396{
3397 struct file *file = iocb->ki_filp;
3398 struct inode *inode = file->f_mapping->host;
3399 struct ext4_inode_info *ei = EXT4_I(inode);
3400 handle_t *handle;
3401 ssize_t ret;
3402 int orphan = 0;
3403 size_t count = iov_length(iov, nr_segs);
3404 int retries = 0;
3405
3406 if (rw == WRITE) {
3407 loff_t final_size = offset + count;
3408
3409 if (final_size > inode->i_size) {
3410 /* Credits for sb + inode write */
3411 handle = ext4_journal_start(inode, 2);
3412 if (IS_ERR(handle)) {
3413 ret = PTR_ERR(handle);
3414 goto out;
3415 }
3416 ret = ext4_orphan_add(handle, inode);
3417 if (ret) {
3418 ext4_journal_stop(handle);
3419 goto out;
3420 }
3421 orphan = 1;
3422 ei->i_disksize = inode->i_size;
3423 ext4_journal_stop(handle);
3424 }
3425 }
3426
3427retry:
3428 if (rw == READ && ext4_should_dioread_nolock(inode))
3429 ret = __blockdev_direct_IO(rw, iocb, inode,
3430 inode->i_sb->s_bdev, iov,
3431 offset, nr_segs,
3432 ext4_get_block, NULL, NULL, 0);
3433 else {
3434 ret = blockdev_direct_IO(rw, iocb, inode,
3435 inode->i_sb->s_bdev, iov,
3436 offset, nr_segs,
3437 ext4_get_block, NULL);
3438
3439 if (unlikely((rw & WRITE) && ret < 0)) {
3440 loff_t isize = i_size_read(inode);
3441 loff_t end = offset + iov_length(iov, nr_segs);
3442
3443 if (end > isize)
3444 ext4_truncate_failed_write(inode);
3445 }
3446 }
3447 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3448 goto retry;
3449
3450 if (orphan) {
3451 int err;
3452
3453 /* Credits for sb + inode write */
3454 handle = ext4_journal_start(inode, 2);
3455 if (IS_ERR(handle)) {
3456 /* This is really bad luck. We've written the data
3457 * but cannot extend i_size. Bail out and pretend
3458 * the write failed... */
3459 ret = PTR_ERR(handle);
3460 if (inode->i_nlink)
3461 ext4_orphan_del(NULL, inode);
3462
3463 goto out;
3464 }
3465 if (inode->i_nlink)
3466 ext4_orphan_del(handle, inode);
3467 if (ret > 0) {
3468 loff_t end = offset + ret;
3469 if (end > inode->i_size) {
3470 ei->i_disksize = end;
3471 i_size_write(inode, end);
3472 /*
3473 * We're going to return a positive `ret'
3474 * here due to non-zero-length I/O, so there's
3475 * no way of reporting error returns from
3476 * ext4_mark_inode_dirty() to userspace. So
3477 * ignore it.
3478 */
3479 ext4_mark_inode_dirty(handle, inode);
3480 }
3481 }
3482 err = ext4_journal_stop(handle);
3483 if (ret == 0)
3484 ret = err;
3485 }
3486out:
3487 return ret;
3488}
3489
3490/*
3491 * ext4_get_block used when preparing for a DIO write or buffer write. 2560 * ext4_get_block used when preparing for a DIO write or buffer write.
3492 * We allocate an uinitialized extent if blocks haven't been allocated. 2561 * We allocate an uinitialized extent if blocks haven't been allocated.
3493 * The extent will be converted to initialized after the IO is complete. 2562 * The extent will be converted to initialized after the IO is complete.
@@ -3958,383 +3027,6 @@ unlock:
3958 return err; 3027 return err;
3959} 3028}
3960 3029
3961/*
3962 * Probably it should be a library function... search for first non-zero word
3963 * or memcmp with zero_page, whatever is better for particular architecture.
3964 * Linus?
3965 */
3966static inline int all_zeroes(__le32 *p, __le32 *q)
3967{
3968 while (p < q)
3969 if (*p++)
3970 return 0;
3971 return 1;
3972}
3973
3974/**
3975 * ext4_find_shared - find the indirect blocks for partial truncation.
3976 * @inode: inode in question
3977 * @depth: depth of the affected branch
3978 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3979 * @chain: place to store the pointers to partial indirect blocks
3980 * @top: place to the (detached) top of branch
3981 *
3982 * This is a helper function used by ext4_truncate().
3983 *
3984 * When we do truncate() we may have to clean the ends of several
3985 * indirect blocks but leave the blocks themselves alive. Block is
3986 * partially truncated if some data below the new i_size is referred
3987 * from it (and it is on the path to the first completely truncated
3988 * data block, indeed). We have to free the top of that path along
3989 * with everything to the right of the path. Since no allocation
3990 * past the truncation point is possible until ext4_truncate()
3991 * finishes, we may safely do the latter, but top of branch may
3992 * require special attention - pageout below the truncation point
3993 * might try to populate it.
3994 *
3995 * We atomically detach the top of branch from the tree, store the
3996 * block number of its root in *@top, pointers to buffer_heads of
3997 * partially truncated blocks - in @chain[].bh and pointers to
3998 * their last elements that should not be removed - in
3999 * @chain[].p. Return value is the pointer to last filled element
4000 * of @chain.
4001 *
4002 * The work left to caller to do the actual freeing of subtrees:
4003 * a) free the subtree starting from *@top
4004 * b) free the subtrees whose roots are stored in
4005 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4006 * c) free the subtrees growing from the inode past the @chain[0].
4007 * (no partially truncated stuff there). */
4008
4009static Indirect *ext4_find_shared(struct inode *inode, int depth,
4010 ext4_lblk_t offsets[4], Indirect chain[4],
4011 __le32 *top)
4012{
4013 Indirect *partial, *p;
4014 int k, err;
4015
4016 *top = 0;
4017 /* Make k index the deepest non-null offset + 1 */
4018 for (k = depth; k > 1 && !offsets[k-1]; k--)
4019 ;
4020 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4021 /* Writer: pointers */
4022 if (!partial)
4023 partial = chain + k-1;
4024 /*
4025 * If the branch acquired continuation since we've looked at it -
4026 * fine, it should all survive and (new) top doesn't belong to us.
4027 */
4028 if (!partial->key && *partial->p)
4029 /* Writer: end */
4030 goto no_top;
4031 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4032 ;
4033 /*
4034 * OK, we've found the last block that must survive. The rest of our
4035 * branch should be detached before unlocking. However, if that rest
4036 * of branch is all ours and does not grow immediately from the inode
4037 * it's easier to cheat and just decrement partial->p.
4038 */
4039 if (p == chain + k - 1 && p > chain) {
4040 p->p--;
4041 } else {
4042 *top = *p->p;
4043 /* Nope, don't do this in ext4. Must leave the tree intact */
4044#if 0
4045 *p->p = 0;
4046#endif
4047 }
4048 /* Writer: end */
4049
4050 while (partial > p) {
4051 brelse(partial->bh);
4052 partial--;
4053 }
4054no_top:
4055 return partial;
4056}
4057
4058/*
4059 * Zero a number of block pointers in either an inode or an indirect block.
4060 * If we restart the transaction we must again get write access to the
4061 * indirect block for further modification.
4062 *
4063 * We release `count' blocks on disk, but (last - first) may be greater
4064 * than `count' because there can be holes in there.
4065 *
4066 * Return 0 on success, 1 on invalid block range
4067 * and < 0 on fatal error.
4068 */
4069static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4070 struct buffer_head *bh,
4071 ext4_fsblk_t block_to_free,
4072 unsigned long count, __le32 *first,
4073 __le32 *last)
4074{
4075 __le32 *p;
4076 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4077 int err;
4078
4079 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4080 flags |= EXT4_FREE_BLOCKS_METADATA;
4081
4082 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4083 count)) {
4084 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4085 "blocks %llu len %lu",
4086 (unsigned long long) block_to_free, count);
4087 return 1;
4088 }
4089
4090 if (try_to_extend_transaction(handle, inode)) {
4091 if (bh) {
4092 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4093 err = ext4_handle_dirty_metadata(handle, inode, bh);
4094 if (unlikely(err))
4095 goto out_err;
4096 }
4097 err = ext4_mark_inode_dirty(handle, inode);
4098 if (unlikely(err))
4099 goto out_err;
4100 err = ext4_truncate_restart_trans(handle, inode,
4101 ext4_blocks_for_truncate(inode));
4102 if (unlikely(err))
4103 goto out_err;
4104 if (bh) {
4105 BUFFER_TRACE(bh, "retaking write access");
4106 err = ext4_journal_get_write_access(handle, bh);
4107 if (unlikely(err))
4108 goto out_err;
4109 }
4110 }
4111
4112 for (p = first; p < last; p++)
4113 *p = 0;
4114
4115 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
4116 return 0;
4117out_err:
4118 ext4_std_error(inode->i_sb, err);
4119 return err;
4120}
4121
4122/**
4123 * ext4_free_data - free a list of data blocks
4124 * @handle: handle for this transaction
4125 * @inode: inode we are dealing with
4126 * @this_bh: indirect buffer_head which contains *@first and *@last
4127 * @first: array of block numbers
4128 * @last: points immediately past the end of array
4129 *
4130 * We are freeing all blocks referred from that array (numbers are stored as
4131 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4132 *
4133 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4134 * blocks are contiguous then releasing them at one time will only affect one
4135 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4136 * actually use a lot of journal space.
4137 *
4138 * @this_bh will be %NULL if @first and @last point into the inode's direct
4139 * block pointers.
4140 */
4141static void ext4_free_data(handle_t *handle, struct inode *inode,
4142 struct buffer_head *this_bh,
4143 __le32 *first, __le32 *last)
4144{
4145 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4146 unsigned long count = 0; /* Number of blocks in the run */
4147 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4148 corresponding to
4149 block_to_free */
4150 ext4_fsblk_t nr; /* Current block # */
4151 __le32 *p; /* Pointer into inode/ind
4152 for current block */
4153 int err = 0;
4154
4155 if (this_bh) { /* For indirect block */
4156 BUFFER_TRACE(this_bh, "get_write_access");
4157 err = ext4_journal_get_write_access(handle, this_bh);
4158 /* Important: if we can't update the indirect pointers
4159 * to the blocks, we can't free them. */
4160 if (err)
4161 return;
4162 }
4163
4164 for (p = first; p < last; p++) {
4165 nr = le32_to_cpu(*p);
4166 if (nr) {
4167 /* accumulate blocks to free if they're contiguous */
4168 if (count == 0) {
4169 block_to_free = nr;
4170 block_to_free_p = p;
4171 count = 1;
4172 } else if (nr == block_to_free + count) {
4173 count++;
4174 } else {
4175 err = ext4_clear_blocks(handle, inode, this_bh,
4176 block_to_free, count,
4177 block_to_free_p, p);
4178 if (err)
4179 break;
4180 block_to_free = nr;
4181 block_to_free_p = p;
4182 count = 1;
4183 }
4184 }
4185 }
4186
4187 if (!err && count > 0)
4188 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4189 count, block_to_free_p, p);
4190 if (err < 0)
4191 /* fatal error */
4192 return;
4193
4194 if (this_bh) {
4195 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4196
4197 /*
4198 * The buffer head should have an attached journal head at this
4199 * point. However, if the data is corrupted and an indirect
4200 * block pointed to itself, it would have been detached when
4201 * the block was cleared. Check for this instead of OOPSing.
4202 */
4203 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4204 ext4_handle_dirty_metadata(handle, inode, this_bh);
4205 else
4206 EXT4_ERROR_INODE(inode,
4207 "circular indirect block detected at "
4208 "block %llu",
4209 (unsigned long long) this_bh->b_blocknr);
4210 }
4211}
4212
4213/**
4214 * ext4_free_branches - free an array of branches
4215 * @handle: JBD handle for this transaction
4216 * @inode: inode we are dealing with
4217 * @parent_bh: the buffer_head which contains *@first and *@last
4218 * @first: array of block numbers
4219 * @last: pointer immediately past the end of array
4220 * @depth: depth of the branches to free
4221 *
4222 * We are freeing all blocks referred from these branches (numbers are
4223 * stored as little-endian 32-bit) and updating @inode->i_blocks
4224 * appropriately.
4225 */
4226static void ext4_free_branches(handle_t *handle, struct inode *inode,
4227 struct buffer_head *parent_bh,
4228 __le32 *first, __le32 *last, int depth)
4229{
4230 ext4_fsblk_t nr;
4231 __le32 *p;
4232
4233 if (ext4_handle_is_aborted(handle))
4234 return;
4235
4236 if (depth--) {
4237 struct buffer_head *bh;
4238 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4239 p = last;
4240 while (--p >= first) {
4241 nr = le32_to_cpu(*p);
4242 if (!nr)
4243 continue; /* A hole */
4244
4245 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4246 nr, 1)) {
4247 EXT4_ERROR_INODE(inode,
4248 "invalid indirect mapped "
4249 "block %lu (level %d)",
4250 (unsigned long) nr, depth);
4251 break;
4252 }
4253
4254 /* Go read the buffer for the next level down */
4255 bh = sb_bread(inode->i_sb, nr);
4256
4257 /*
4258 * A read failure? Report error and clear slot
4259 * (should be rare).
4260 */
4261 if (!bh) {
4262 EXT4_ERROR_INODE_BLOCK(inode, nr,
4263 "Read failure");
4264 continue;
4265 }
4266
4267 /* This zaps the entire block. Bottom up. */
4268 BUFFER_TRACE(bh, "free child branches");
4269 ext4_free_branches(handle, inode, bh,
4270 (__le32 *) bh->b_data,
4271 (__le32 *) bh->b_data + addr_per_block,
4272 depth);
4273 brelse(bh);
4274
4275 /*
4276 * Everything below this this pointer has been
4277 * released. Now let this top-of-subtree go.
4278 *
4279 * We want the freeing of this indirect block to be
4280 * atomic in the journal with the updating of the
4281 * bitmap block which owns it. So make some room in
4282 * the journal.
4283 *
4284 * We zero the parent pointer *after* freeing its
4285 * pointee in the bitmaps, so if extend_transaction()
4286 * for some reason fails to put the bitmap changes and
4287 * the release into the same transaction, recovery
4288 * will merely complain about releasing a free block,
4289 * rather than leaking blocks.
4290 */
4291 if (ext4_handle_is_aborted(handle))
4292 return;
4293 if (try_to_extend_transaction(handle, inode)) {
4294 ext4_mark_inode_dirty(handle, inode);
4295 ext4_truncate_restart_trans(handle, inode,
4296 ext4_blocks_for_truncate(inode));
4297 }
4298
4299 /*
4300 * The forget flag here is critical because if
4301 * we are journaling (and not doing data
4302 * journaling), we have to make sure a revoke
4303 * record is written to prevent the journal
4304 * replay from overwriting the (former)
4305 * indirect block if it gets reallocated as a
4306 * data block. This must happen in the same
4307 * transaction where the data blocks are
4308 * actually freed.
4309 */
4310 ext4_free_blocks(handle, inode, NULL, nr, 1,
4311 EXT4_FREE_BLOCKS_METADATA|
4312 EXT4_FREE_BLOCKS_FORGET);
4313
4314 if (parent_bh) {
4315 /*
4316 * The block which we have just freed is
4317 * pointed to by an indirect block: journal it
4318 */
4319 BUFFER_TRACE(parent_bh, "get_write_access");
4320 if (!ext4_journal_get_write_access(handle,
4321 parent_bh)){
4322 *p = 0;
4323 BUFFER_TRACE(parent_bh,
4324 "call ext4_handle_dirty_metadata");
4325 ext4_handle_dirty_metadata(handle,
4326 inode,
4327 parent_bh);
4328 }
4329 }
4330 }
4331 } else {
4332 /* We have reached the bottom of the tree. */
4333 BUFFER_TRACE(parent_bh, "free data blocks");
4334 ext4_free_data(handle, inode, parent_bh, first, last);
4335 }
4336}
4337
4338int ext4_can_truncate(struct inode *inode) 3030int ext4_can_truncate(struct inode *inode)
4339{ 3031{
4340 if (S_ISREG(inode->i_mode)) 3032 if (S_ISREG(inode->i_mode))
@@ -4419,161 +3111,6 @@ void ext4_truncate(struct inode *inode)
4419 trace_ext4_truncate_exit(inode); 3111 trace_ext4_truncate_exit(inode);
4420} 3112}
4421 3113
4422void ext4_ind_truncate(struct inode *inode)
4423{
4424 handle_t *handle;
4425 struct ext4_inode_info *ei = EXT4_I(inode);
4426 __le32 *i_data = ei->i_data;
4427 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4428 struct address_space *mapping = inode->i_mapping;
4429 ext4_lblk_t offsets[4];
4430 Indirect chain[4];
4431 Indirect *partial;
4432 __le32 nr = 0;
4433 int n = 0;
4434 ext4_lblk_t last_block, max_block;
4435 unsigned blocksize = inode->i_sb->s_blocksize;
4436
4437 handle = start_transaction(inode);
4438 if (IS_ERR(handle))
4439 return; /* AKPM: return what? */
4440
4441 last_block = (inode->i_size + blocksize-1)
4442 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4443 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
4444 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4445
4446 if (inode->i_size & (blocksize - 1))
4447 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4448 goto out_stop;
4449
4450 if (last_block != max_block) {
4451 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4452 if (n == 0)
4453 goto out_stop; /* error */
4454 }
4455
4456 /*
4457 * OK. This truncate is going to happen. We add the inode to the
4458 * orphan list, so that if this truncate spans multiple transactions,
4459 * and we crash, we will resume the truncate when the filesystem
4460 * recovers. It also marks the inode dirty, to catch the new size.
4461 *
4462 * Implication: the file must always be in a sane, consistent
4463 * truncatable state while each transaction commits.
4464 */
4465 if (ext4_orphan_add(handle, inode))
4466 goto out_stop;
4467
4468 /*
4469 * From here we block out all ext4_get_block() callers who want to
4470 * modify the block allocation tree.
4471 */
4472 down_write(&ei->i_data_sem);
4473
4474 ext4_discard_preallocations(inode);
4475
4476 /*
4477 * The orphan list entry will now protect us from any crash which
4478 * occurs before the truncate completes, so it is now safe to propagate
4479 * the new, shorter inode size (held for now in i_size) into the
4480 * on-disk inode. We do this via i_disksize, which is the value which
4481 * ext4 *really* writes onto the disk inode.
4482 */
4483 ei->i_disksize = inode->i_size;
4484
4485 if (last_block == max_block) {
4486 /*
4487 * It is unnecessary to free any data blocks if last_block is
4488 * equal to the indirect block limit.
4489 */
4490 goto out_unlock;
4491 } else if (n == 1) { /* direct blocks */
4492 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4493 i_data + EXT4_NDIR_BLOCKS);
4494 goto do_indirects;
4495 }
4496
4497 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4498 /* Kill the top of shared branch (not detached) */
4499 if (nr) {
4500 if (partial == chain) {
4501 /* Shared branch grows from the inode */
4502 ext4_free_branches(handle, inode, NULL,
4503 &nr, &nr+1, (chain+n-1) - partial);
4504 *partial->p = 0;
4505 /*
4506 * We mark the inode dirty prior to restart,
4507 * and prior to stop. No need for it here.
4508 */
4509 } else {
4510 /* Shared branch grows from an indirect block */
4511 BUFFER_TRACE(partial->bh, "get_write_access");
4512 ext4_free_branches(handle, inode, partial->bh,
4513 partial->p,
4514 partial->p+1, (chain+n-1) - partial);
4515 }
4516 }
4517 /* Clear the ends of indirect blocks on the shared branch */
4518 while (partial > chain) {
4519 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4520 (__le32*)partial->bh->b_data+addr_per_block,
4521 (chain+n-1) - partial);
4522 BUFFER_TRACE(partial->bh, "call brelse");
4523 brelse(partial->bh);
4524 partial--;
4525 }
4526do_indirects:
4527 /* Kill the remaining (whole) subtrees */
4528 switch (offsets[0]) {
4529 default:
4530 nr = i_data[EXT4_IND_BLOCK];
4531 if (nr) {
4532 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4533 i_data[EXT4_IND_BLOCK] = 0;
4534 }
4535 case EXT4_IND_BLOCK:
4536 nr = i_data[EXT4_DIND_BLOCK];
4537 if (nr) {
4538 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4539 i_data[EXT4_DIND_BLOCK] = 0;
4540 }
4541 case EXT4_DIND_BLOCK:
4542 nr = i_data[EXT4_TIND_BLOCK];
4543 if (nr) {
4544 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4545 i_data[EXT4_TIND_BLOCK] = 0;
4546 }
4547 case EXT4_TIND_BLOCK:
4548 ;
4549 }
4550
4551out_unlock:
4552 up_write(&ei->i_data_sem);
4553 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4554 ext4_mark_inode_dirty(handle, inode);
4555
4556 /*
4557 * In a multi-transaction truncate, we only make the final transaction
4558 * synchronous
4559 */
4560 if (IS_SYNC(inode))
4561 ext4_handle_sync(handle);
4562out_stop:
4563 /*
4564 * If this was a simple ftruncate(), and the file will remain alive
4565 * then we need to clear up the orphan record which we created above.
4566 * However, if this was a real unlink then we were called by
4567 * ext4_delete_inode(), and we allow that function to clean up the
4568 * orphan info for us.
4569 */
4570 if (inode->i_nlink)
4571 ext4_orphan_del(handle, inode);
4572
4573 ext4_journal_stop(handle);
4574 trace_ext4_truncate_exit(inode);
4575}
4576
4577/* 3114/*
4578 * ext4_get_inode_loc returns with an extra refcount against the inode's 3115 * ext4_get_inode_loc returns with an extra refcount against the inode's
4579 * underlying buffer_head on success. If 'in_mem' is true, we have all 3116 * underlying buffer_head on success. If 'in_mem' is true, we have all
@@ -5386,29 +3923,6 @@ int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5386 return 0; 3923 return 0;
5387} 3924}
5388 3925
5389static int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5390{
5391 int indirects;
5392
5393 /* if nrblocks are contiguous */
5394 if (chunk) {
5395 /*
5396 * With N contiguous data blocks, we need at most
5397 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
5398 * 2 dindirect blocks, and 1 tindirect block
5399 */
5400 return DIV_ROUND_UP(nrblocks,
5401 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
5402 }
5403 /*
5404 * if nrblocks are not contiguous, worse case, each block touch
5405 * a indirect block, and each indirect block touch a double indirect
5406 * block, plus a triple indirect block
5407 */
5408 indirects = nrblocks * 2 + 1;
5409 return indirects;
5410}
5411
5412static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) 3926static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5413{ 3927{
5414 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 3928 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-21 01:24:04 -0500