blob: bc15c2c17633079a54de855baf1272b0124f19eb [file] [log] [blame]
Amir Goldsteindae1e522011-06-27 19:40:50 -04001/*
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
Amir Goldsteindae1e522011-06-27 19:40:50 -040023#include "ext4_jbd2.h"
24#include "truncate.h"
Matthew Wilcoxc94c2ac2015-09-08 14:58:40 -070025#include <linux/dax.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080026#include <linux/uio.h>
Amir Goldsteindae1e522011-06-27 19:40:50 -040027
28#include <trace/events/ext4.h>
29
30typedef struct {
31 __le32 *p;
32 __le32 key;
33 struct buffer_head *bh;
34} Indirect;
35
36static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
37{
38 p->key = *(p->p = v);
39 p->bh = bh;
40}
41
42/**
43 * ext4_block_to_path - parse the block number into array of offsets
44 * @inode: inode in question (we are only interested in its superblock)
45 * @i_block: block number to be parsed
46 * @offsets: array to store the offsets in
47 * @boundary: set this non-zero if the referred-to block is likely to be
48 * followed (on disk) by an indirect block.
49 *
50 * To store the locations of file's data ext4 uses a data structure common
51 * for UNIX filesystems - tree of pointers anchored in the inode, with
52 * data blocks at leaves and indirect blocks in intermediate nodes.
53 * This function translates the block number into path in that tree -
54 * return value is the path length and @offsets[n] is the offset of
55 * pointer to (n+1)th node in the nth one. If @block is out of range
56 * (negative or too large) warning is printed and zero returned.
57 *
58 * Note: function doesn't find node addresses, so no IO is needed. All
59 * we need to know is the capacity of indirect blocks (taken from the
60 * inode->i_sb).
61 */
62
63/*
64 * Portability note: the last comparison (check that we fit into triple
65 * indirect block) is spelled differently, because otherwise on an
66 * architecture with 32-bit longs and 8Kb pages we might get into trouble
67 * if our filesystem had 8Kb blocks. We might use long long, but that would
68 * kill us on x86. Oh, well, at least the sign propagation does not matter -
69 * i_block would have to be negative in the very beginning, so we would not
70 * get there at all.
71 */
72
73static int ext4_block_to_path(struct inode *inode,
74 ext4_lblk_t i_block,
75 ext4_lblk_t offsets[4], int *boundary)
76{
77 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
78 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
79 const long direct_blocks = EXT4_NDIR_BLOCKS,
80 indirect_blocks = ptrs,
81 double_blocks = (1 << (ptrs_bits * 2));
82 int n = 0;
83 int final = 0;
84
85 if (i_block < direct_blocks) {
86 offsets[n++] = i_block;
87 final = direct_blocks;
88 } else if ((i_block -= direct_blocks) < indirect_blocks) {
89 offsets[n++] = EXT4_IND_BLOCK;
90 offsets[n++] = i_block;
91 final = ptrs;
92 } else if ((i_block -= indirect_blocks) < double_blocks) {
93 offsets[n++] = EXT4_DIND_BLOCK;
94 offsets[n++] = i_block >> ptrs_bits;
95 offsets[n++] = i_block & (ptrs - 1);
96 final = ptrs;
97 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
98 offsets[n++] = EXT4_TIND_BLOCK;
99 offsets[n++] = i_block >> (ptrs_bits * 2);
100 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
101 offsets[n++] = i_block & (ptrs - 1);
102 final = ptrs;
103 } else {
104 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
105 i_block + direct_blocks +
106 indirect_blocks + double_blocks, inode->i_ino);
107 }
108 if (boundary)
109 *boundary = final - 1 - (i_block & (ptrs - 1));
110 return n;
111}
112
113/**
114 * ext4_get_branch - read the chain of indirect blocks leading to data
115 * @inode: inode in question
116 * @depth: depth of the chain (1 - direct pointer, etc.)
117 * @offsets: offsets of pointers in inode/indirect blocks
118 * @chain: place to store the result
119 * @err: here we store the error value
120 *
121 * Function fills the array of triples <key, p, bh> and returns %NULL
122 * if everything went OK or the pointer to the last filled triple
123 * (incomplete one) otherwise. Upon the return chain[i].key contains
124 * the number of (i+1)-th block in the chain (as it is stored in memory,
125 * i.e. little-endian 32-bit), chain[i].p contains the address of that
126 * number (it points into struct inode for i==0 and into the bh->b_data
127 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
128 * block for i>0 and NULL for i==0. In other words, it holds the block
129 * numbers of the chain, addresses they were taken from (and where we can
130 * verify that chain did not change) and buffer_heads hosting these
131 * numbers.
132 *
133 * Function stops when it stumbles upon zero pointer (absent block)
134 * (pointer to last triple returned, *@err == 0)
135 * or when it gets an IO error reading an indirect block
136 * (ditto, *@err == -EIO)
137 * or when it reads all @depth-1 indirect blocks successfully and finds
138 * the whole chain, all way to the data (returns %NULL, *err == 0).
139 *
140 * Need to be called with
141 * down_read(&EXT4_I(inode)->i_data_sem)
142 */
143static Indirect *ext4_get_branch(struct inode *inode, int depth,
144 ext4_lblk_t *offsets,
145 Indirect chain[4], int *err)
146{
147 struct super_block *sb = inode->i_sb;
148 Indirect *p = chain;
149 struct buffer_head *bh;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500150 int ret = -EIO;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400151
152 *err = 0;
153 /* i_data is not going away, no lock needed */
154 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
155 if (!p->key)
156 goto no_block;
157 while (--depth) {
158 bh = sb_getblk(sb, le32_to_cpu(p->key));
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500159 if (unlikely(!bh)) {
160 ret = -ENOMEM;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400161 goto failure;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500162 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400163
164 if (!bh_uptodate_or_lock(bh)) {
165 if (bh_submit_read(bh) < 0) {
166 put_bh(bh);
167 goto failure;
168 }
169 /* validate block references */
170 if (ext4_check_indirect_blockref(inode, bh)) {
171 put_bh(bh);
172 goto failure;
173 }
174 }
175
176 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
177 /* Reader: end */
178 if (!p->key)
179 goto no_block;
180 }
181 return NULL;
182
183failure:
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500184 *err = ret;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400185no_block:
186 return p;
187}
188
189/**
190 * ext4_find_near - find a place for allocation with sufficient locality
191 * @inode: owner
192 * @ind: descriptor of indirect block.
193 *
194 * This function returns the preferred place for block allocation.
195 * It is used when heuristic for sequential allocation fails.
196 * Rules are:
197 * + if there is a block to the left of our position - allocate near it.
198 * + if pointer will live in indirect block - allocate near that block.
199 * + if pointer will live in inode - allocate in the same
200 * cylinder group.
201 *
202 * In the latter case we colour the starting block by the callers PID to
203 * prevent it from clashing with concurrent allocations for a different inode
204 * in the same block group. The PID is used here so that functionally related
205 * files will be close-by on-disk.
206 *
207 * Caller must make sure that @ind is valid and will stay that way.
208 */
209static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
210{
211 struct ext4_inode_info *ei = EXT4_I(inode);
212 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
213 __le32 *p;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400214
215 /* Try to find previous block */
216 for (p = ind->p - 1; p >= start; p--) {
217 if (*p)
218 return le32_to_cpu(*p);
219 }
220
221 /* No such thing, so let's try location of indirect block */
222 if (ind->bh)
223 return ind->bh->b_blocknr;
224
225 /*
226 * It is going to be referred to from the inode itself? OK, just put it
227 * into the same cylinder group then.
228 */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400229 return ext4_inode_to_goal_block(inode);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400230}
231
232/**
233 * ext4_find_goal - find a preferred place for allocation.
234 * @inode: owner
235 * @block: block we want
236 * @partial: pointer to the last triple within a chain
237 *
238 * Normally this function find the preferred place for block allocation,
239 * returns it.
240 * Because this is only used for non-extent files, we limit the block nr
241 * to 32 bits.
242 */
243static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
244 Indirect *partial)
245{
246 ext4_fsblk_t goal;
247
248 /*
249 * XXX need to get goal block from mballoc's data structures
250 */
251
252 goal = ext4_find_near(inode, partial);
253 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
254 return goal;
255}
256
257/**
258 * ext4_blks_to_allocate - Look up the block map and count the number
259 * of direct blocks need to be allocated for the given branch.
260 *
261 * @branch: chain of indirect blocks
262 * @k: number of blocks need for indirect blocks
263 * @blks: number of data blocks to be mapped.
264 * @blocks_to_boundary: the offset in the indirect block
265 *
266 * return the total number of blocks to be allocate, including the
267 * direct and indirect blocks.
268 */
269static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
270 int blocks_to_boundary)
271{
272 unsigned int count = 0;
273
274 /*
275 * Simple case, [t,d]Indirect block(s) has not allocated yet
276 * then it's clear blocks on that path have not allocated
277 */
278 if (k > 0) {
279 /* right now we don't handle cross boundary allocation */
280 if (blks < blocks_to_boundary + 1)
281 count += blks;
282 else
283 count += blocks_to_boundary + 1;
284 return count;
285 }
286
287 count++;
288 while (count < blks && count <= blocks_to_boundary &&
289 le32_to_cpu(*(branch[0].p + count)) == 0) {
290 count++;
291 }
292 return count;
293}
294
295/**
Amir Goldsteindae1e522011-06-27 19:40:50 -0400296 * ext4_alloc_branch - allocate and set up a chain of blocks.
297 * @handle: handle for this transaction
298 * @inode: owner
299 * @indirect_blks: number of allocated indirect blocks
300 * @blks: number of allocated direct blocks
301 * @goal: preferred place for allocation
302 * @offsets: offsets (in the blocks) to store the pointers to next.
303 * @branch: place to store the chain in.
304 *
305 * This function allocates blocks, zeroes out all but the last one,
306 * links them into chain and (if we are synchronous) writes them to disk.
307 * In other words, it prepares a branch that can be spliced onto the
308 * inode. It stores the information about that chain in the branch[], in
309 * the same format as ext4_get_branch() would do. We are calling it after
310 * we had read the existing part of chain and partial points to the last
311 * triple of that (one with zero ->key). Upon the exit we have the same
312 * picture as after the successful ext4_get_block(), except that in one
313 * place chain is disconnected - *branch->p is still zero (we did not
314 * set the last link), but branch->key contains the number that should
315 * be placed into *branch->p to fill that gap.
316 *
317 * If allocation fails we free all blocks we've allocated (and forget
318 * their buffer_heads) and return the error value the from failed
319 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
320 * as described above and return 0.
321 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400322static int ext4_alloc_branch(handle_t *handle,
323 struct ext4_allocation_request *ar,
324 int indirect_blks, ext4_lblk_t *offsets,
325 Indirect *branch)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400326{
Theodore Ts'o781f1432013-04-03 12:43:17 -0400327 struct buffer_head * bh;
328 ext4_fsblk_t b, new_blocks[4];
329 __le32 *p;
330 int i, j, err, len = 1;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400331
Theodore Ts'o781f1432013-04-03 12:43:17 -0400332 for (i = 0; i <= indirect_blks; i++) {
333 if (i == indirect_blks) {
Theodore Ts'oa5211002014-09-04 18:06:25 -0400334 new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
Theodore Ts'o781f1432013-04-03 12:43:17 -0400335 } else
Theodore Ts'oa5211002014-09-04 18:06:25 -0400336 ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -0400337 ar->inode, ar->goal,
338 ar->flags & EXT4_MB_DELALLOC_RESERVED,
339 NULL, &err);
Theodore Ts'o781f1432013-04-03 12:43:17 -0400340 if (err) {
341 i--;
342 goto failed;
343 }
344 branch[i].key = cpu_to_le32(new_blocks[i]);
345 if (i == 0)
346 continue;
347
Theodore Ts'oa5211002014-09-04 18:06:25 -0400348 bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400349 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500350 err = -ENOMEM;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400351 goto failed;
352 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400353 lock_buffer(bh);
354 BUFFER_TRACE(bh, "call get_create_access");
355 err = ext4_journal_get_create_access(handle, bh);
356 if (err) {
Amir Goldsteindae1e522011-06-27 19:40:50 -0400357 unlock_buffer(bh);
358 goto failed;
359 }
360
Theodore Ts'o781f1432013-04-03 12:43:17 -0400361 memset(bh->b_data, 0, bh->b_size);
362 p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
363 b = new_blocks[i];
364
365 if (i == indirect_blks)
Theodore Ts'oa5211002014-09-04 18:06:25 -0400366 len = ar->len;
Theodore Ts'o781f1432013-04-03 12:43:17 -0400367 for (j = 0; j < len; j++)
368 *p++ = cpu_to_le32(b++);
369
Amir Goldsteindae1e522011-06-27 19:40:50 -0400370 BUFFER_TRACE(bh, "marking uptodate");
371 set_buffer_uptodate(bh);
372 unlock_buffer(bh);
373
374 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Theodore Ts'oa5211002014-09-04 18:06:25 -0400375 err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400376 if (err)
377 goto failed;
378 }
Theodore Ts'o781f1432013-04-03 12:43:17 -0400379 return 0;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400380failed:
Theodore Ts'o781f1432013-04-03 12:43:17 -0400381 for (; i >= 0; i--) {
Jan Karac5c7b8d2014-06-15 23:46:28 -0400382 /*
383 * We want to ext4_forget() only freshly allocated indirect
384 * blocks. Buffer for new_blocks[i-1] is at branch[i].bh and
385 * buffer at branch[0].bh is indirect block / inode already
386 * existing before ext4_alloc_branch() was called.
387 */
388 if (i > 0 && i != indirect_blks && branch[i].bh)
Theodore Ts'oa5211002014-09-04 18:06:25 -0400389 ext4_forget(handle, 1, ar->inode, branch[i].bh,
Theodore Ts'o781f1432013-04-03 12:43:17 -0400390 branch[i].bh->b_blocknr);
Theodore Ts'oa5211002014-09-04 18:06:25 -0400391 ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
392 (i == indirect_blks) ? ar->len : 1, 0);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400393 }
Amir Goldsteindae1e522011-06-27 19:40:50 -0400394 return err;
395}
396
397/**
398 * ext4_splice_branch - splice the allocated branch onto inode.
399 * @handle: handle for this transaction
400 * @inode: owner
401 * @block: (logical) number of block we are adding
402 * @chain: chain of indirect blocks (with a missing link - see
403 * ext4_alloc_branch)
404 * @where: location of missing link
405 * @num: number of indirect blocks we are adding
406 * @blks: number of direct blocks we are adding
407 *
408 * This function fills the missing link and does all housekeeping needed in
409 * inode (->i_blocks, etc.). In case of success we end up with the full
410 * chain to new block and return 0.
411 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400412static int ext4_splice_branch(handle_t *handle,
413 struct ext4_allocation_request *ar,
414 Indirect *where, int num)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400415{
416 int i;
417 int err = 0;
418 ext4_fsblk_t current_block;
419
420 /*
421 * If we're splicing into a [td]indirect block (as opposed to the
422 * inode) then we need to get write access to the [td]indirect block
423 * before the splice.
424 */
425 if (where->bh) {
426 BUFFER_TRACE(where->bh, "get_write_access");
427 err = ext4_journal_get_write_access(handle, where->bh);
428 if (err)
429 goto err_out;
430 }
431 /* That's it */
432
433 *where->p = where->key;
434
435 /*
436 * Update the host buffer_head or inode to point to more just allocated
437 * direct blocks blocks
438 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400439 if (num == 0 && ar->len > 1) {
Amir Goldsteindae1e522011-06-27 19:40:50 -0400440 current_block = le32_to_cpu(where->key) + 1;
Theodore Ts'oa5211002014-09-04 18:06:25 -0400441 for (i = 1; i < ar->len; i++)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400442 *(where->p + i) = cpu_to_le32(current_block++);
443 }
444
445 /* We are done with atomic stuff, now do the rest of housekeeping */
446 /* had we spliced it onto indirect block? */
447 if (where->bh) {
448 /*
449 * If we spliced it onto an indirect block, we haven't
450 * altered the inode. Note however that if it is being spliced
451 * onto an indirect block at the very end of the file (the
452 * file is growing) then we *will* alter the inode to reflect
453 * the new i_size. But that is not done here - it is done in
454 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
455 */
456 jbd_debug(5, "splicing indirect only\n");
457 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
Theodore Ts'oa5211002014-09-04 18:06:25 -0400458 err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400459 if (err)
460 goto err_out;
461 } else {
462 /*
463 * OK, we spliced it into the inode itself on a direct block.
464 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400465 ext4_mark_inode_dirty(handle, ar->inode);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400466 jbd_debug(5, "splicing direct\n");
467 }
468 return err;
469
470err_out:
471 for (i = 1; i <= num; i++) {
472 /*
473 * branch[i].bh is newly allocated, so there is no
474 * need to revoke the block, which is why we don't
475 * need to set EXT4_FREE_BLOCKS_METADATA.
476 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400477 ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
Amir Goldsteindae1e522011-06-27 19:40:50 -0400478 EXT4_FREE_BLOCKS_FORGET);
479 }
Theodore Ts'oa5211002014-09-04 18:06:25 -0400480 ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
481 ar->len, 0);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400482
483 return err;
484}
485
486/*
487 * The ext4_ind_map_blocks() function handles non-extents inodes
488 * (i.e., using the traditional indirect/double-indirect i_blocks
489 * scheme) for ext4_map_blocks().
490 *
491 * Allocation strategy is simple: if we have to allocate something, we will
492 * have to go the whole way to leaf. So let's do it before attaching anything
493 * to tree, set linkage between the newborn blocks, write them if sync is
494 * required, recheck the path, free and repeat if check fails, otherwise
495 * set the last missing link (that will protect us from any truncate-generated
496 * removals - all blocks on the path are immune now) and possibly force the
497 * write on the parent block.
498 * That has a nice additional property: no special recovery from the failed
499 * allocations is needed - we simply release blocks and do not touch anything
500 * reachable from inode.
501 *
502 * `handle' can be NULL if create == 0.
503 *
504 * return > 0, # of blocks mapped or allocated.
505 * return = 0, if plain lookup failed.
506 * return < 0, error case.
507 *
508 * The ext4_ind_get_blocks() function should be called with
509 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
510 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
511 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
512 * blocks.
513 */
514int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
515 struct ext4_map_blocks *map,
516 int flags)
517{
Theodore Ts'oa5211002014-09-04 18:06:25 -0400518 struct ext4_allocation_request ar;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400519 int err = -EIO;
520 ext4_lblk_t offsets[4];
521 Indirect chain[4];
522 Indirect *partial;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400523 int indirect_blks;
524 int blocks_to_boundary = 0;
525 int depth;
526 int count = 0;
527 ext4_fsblk_t first_block = 0;
528
529 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
530 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
531 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
532 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
533 &blocks_to_boundary);
534
535 if (depth == 0)
536 goto out;
537
538 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
539
540 /* Simplest case - block found, no allocation needed */
541 if (!partial) {
542 first_block = le32_to_cpu(chain[depth - 1].key);
543 count++;
544 /*map more blocks*/
545 while (count < map->m_len && count <= blocks_to_boundary) {
546 ext4_fsblk_t blk;
547
548 blk = le32_to_cpu(*(chain[depth-1].p + count));
549
550 if (blk == first_block + count)
551 count++;
552 else
553 break;
554 }
555 goto got_it;
556 }
557
Jan Karafacab4d2016-03-09 22:54:00 -0500558 /* Next simple case - plain lookup failed */
559 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
560 unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
561 int i;
562
563 /* Count number blocks in a subtree under 'partial' */
564 count = 1;
565 for (i = 0; partial + i != chain + depth - 1; i++)
566 count *= epb;
567 /* Fill in size of a hole we found */
568 map->m_pblk = 0;
569 map->m_len = min_t(unsigned int, map->m_len, count);
570 goto cleanup;
571 }
572
573 /* Failed read of indirect block */
574 if (err == -EIO)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400575 goto cleanup;
576
577 /*
578 * Okay, we need to do block allocation.
579 */
Darrick J. Wonge2b911c2015-10-17 16:18:43 -0400580 if (ext4_has_feature_bigalloc(inode->i_sb)) {
Theodore Ts'obab08ab2011-09-09 18:36:51 -0400581 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
582 "non-extent mapped inodes with bigalloc");
Darrick J. Wong6a797d22015-10-17 16:16:04 -0400583 return -EFSCORRUPTED;
Theodore Ts'obab08ab2011-09-09 18:36:51 -0400584 }
585
Theodore Ts'oa5211002014-09-04 18:06:25 -0400586 /* Set up for the direct block allocation */
587 memset(&ar, 0, sizeof(ar));
588 ar.inode = inode;
589 ar.logical = map->m_lblk;
590 if (S_ISREG(inode->i_mode))
591 ar.flags = EXT4_MB_HINT_DATA;
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -0400592 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
593 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
Theodore Ts'oc5e298a2015-06-21 01:25:29 -0400594 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
595 ar.flags |= EXT4_MB_USE_RESERVED;
Theodore Ts'oa5211002014-09-04 18:06:25 -0400596
597 ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400598
599 /* the number of blocks need to allocate for [d,t]indirect blocks */
600 indirect_blks = (chain + depth) - partial - 1;
601
602 /*
603 * Next look up the indirect map to count the totoal number of
604 * direct blocks to allocate for this branch.
605 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400606 ar.len = ext4_blks_to_allocate(partial, indirect_blks,
607 map->m_len, blocks_to_boundary);
608
Amir Goldsteindae1e522011-06-27 19:40:50 -0400609 /*
610 * Block out ext4_truncate while we alter the tree
611 */
Theodore Ts'oa5211002014-09-04 18:06:25 -0400612 err = ext4_alloc_branch(handle, &ar, indirect_blks,
Amir Goldsteindae1e522011-06-27 19:40:50 -0400613 offsets + (partial - chain), partial);
614
615 /*
616 * The ext4_splice_branch call will free and forget any buffers
617 * on the new chain if there is a failure, but that risks using
618 * up transaction credits, especially for bitmaps where the
619 * credits cannot be returned. Can we handle this somehow? We
620 * may need to return -EAGAIN upwards in the worst case. --sct
621 */
622 if (!err)
Theodore Ts'oa5211002014-09-04 18:06:25 -0400623 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400624 if (err)
625 goto cleanup;
626
627 map->m_flags |= EXT4_MAP_NEW;
628
629 ext4_update_inode_fsync_trans(handle, inode, 1);
Theodore Ts'oa5211002014-09-04 18:06:25 -0400630 count = ar.len;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400631got_it:
632 map->m_flags |= EXT4_MAP_MAPPED;
633 map->m_pblk = le32_to_cpu(chain[depth-1].key);
634 map->m_len = count;
635 if (count > blocks_to_boundary)
636 map->m_flags |= EXT4_MAP_BOUNDARY;
637 err = count;
638 /* Clean up and exit */
639 partial = chain + depth - 1; /* the whole chain */
640cleanup:
641 while (partial > chain) {
642 BUFFER_TRACE(partial->bh, "call brelse");
643 brelse(partial->bh);
644 partial--;
645 }
646out:
Theodore Ts'o21ddd562013-07-01 08:12:40 -0400647 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
Amir Goldsteindae1e522011-06-27 19:40:50 -0400648 return err;
649}
650
651/*
Amir Goldsteindae1e522011-06-27 19:40:50 -0400652 * Calculate the number of metadata blocks need to reserve
653 * to allocate a new block at @lblocks for non extent file based file
654 */
655int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
656{
657 struct ext4_inode_info *ei = EXT4_I(inode);
658 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
659 int blk_bits;
660
661 if (lblock < EXT4_NDIR_BLOCKS)
662 return 0;
663
664 lblock -= EXT4_NDIR_BLOCKS;
665
666 if (ei->i_da_metadata_calc_len &&
667 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
668 ei->i_da_metadata_calc_len++;
669 return 0;
670 }
671 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
672 ei->i_da_metadata_calc_len = 1;
673 blk_bits = order_base_2(lblock);
674 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
675}
676
Jan Karafa55a0e2013-06-04 12:56:55 -0400677/*
678 * Calculate number of indirect blocks touched by mapping @nrblocks logically
679 * contiguous blocks
680 */
681int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
Amir Goldsteindae1e522011-06-27 19:40:50 -0400682{
Amir Goldsteindae1e522011-06-27 19:40:50 -0400683 /*
Jan Karafa55a0e2013-06-04 12:56:55 -0400684 * With N contiguous data blocks, we need at most
685 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
686 * 2 dindirect blocks, and 1 tindirect block
Amir Goldsteindae1e522011-06-27 19:40:50 -0400687 */
Jan Karafa55a0e2013-06-04 12:56:55 -0400688 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400689}
690
691/*
692 * Truncate transactions can be complex and absolutely huge. So we need to
693 * be able to restart the transaction at a conventient checkpoint to make
694 * sure we don't overflow the journal.
695 *
Theodore Ts'o819c4922013-04-03 12:47:17 -0400696 * Try to extend this transaction for the purposes of truncation. If
Amir Goldsteindae1e522011-06-27 19:40:50 -0400697 * extend fails, we need to propagate the failure up and restart the
698 * transaction in the top-level truncate loop. --sct
Amir Goldsteindae1e522011-06-27 19:40:50 -0400699 *
700 * Returns 0 if we managed to create more room. If we can't create more
701 * room, and the transaction must be restarted we return 1.
702 */
703static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
704{
705 if (!ext4_handle_valid(handle))
706 return 0;
707 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
708 return 0;
709 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
710 return 0;
711 return 1;
712}
713
714/*
715 * Probably it should be a library function... search for first non-zero word
716 * or memcmp with zero_page, whatever is better for particular architecture.
717 * Linus?
718 */
719static inline int all_zeroes(__le32 *p, __le32 *q)
720{
721 while (p < q)
722 if (*p++)
723 return 0;
724 return 1;
725}
726
727/**
728 * ext4_find_shared - find the indirect blocks for partial truncation.
729 * @inode: inode in question
730 * @depth: depth of the affected branch
731 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
732 * @chain: place to store the pointers to partial indirect blocks
733 * @top: place to the (detached) top of branch
734 *
735 * This is a helper function used by ext4_truncate().
736 *
737 * When we do truncate() we may have to clean the ends of several
738 * indirect blocks but leave the blocks themselves alive. Block is
739 * partially truncated if some data below the new i_size is referred
740 * from it (and it is on the path to the first completely truncated
741 * data block, indeed). We have to free the top of that path along
742 * with everything to the right of the path. Since no allocation
743 * past the truncation point is possible until ext4_truncate()
744 * finishes, we may safely do the latter, but top of branch may
745 * require special attention - pageout below the truncation point
746 * might try to populate it.
747 *
748 * We atomically detach the top of branch from the tree, store the
749 * block number of its root in *@top, pointers to buffer_heads of
750 * partially truncated blocks - in @chain[].bh and pointers to
751 * their last elements that should not be removed - in
752 * @chain[].p. Return value is the pointer to last filled element
753 * of @chain.
754 *
755 * The work left to caller to do the actual freeing of subtrees:
756 * a) free the subtree starting from *@top
757 * b) free the subtrees whose roots are stored in
758 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
759 * c) free the subtrees growing from the inode past the @chain[0].
760 * (no partially truncated stuff there). */
761
762static Indirect *ext4_find_shared(struct inode *inode, int depth,
763 ext4_lblk_t offsets[4], Indirect chain[4],
764 __le32 *top)
765{
766 Indirect *partial, *p;
767 int k, err;
768
769 *top = 0;
770 /* Make k index the deepest non-null offset + 1 */
771 for (k = depth; k > 1 && !offsets[k-1]; k--)
772 ;
773 partial = ext4_get_branch(inode, k, offsets, chain, &err);
774 /* Writer: pointers */
775 if (!partial)
776 partial = chain + k-1;
777 /*
778 * If the branch acquired continuation since we've looked at it -
779 * fine, it should all survive and (new) top doesn't belong to us.
780 */
781 if (!partial->key && *partial->p)
782 /* Writer: end */
783 goto no_top;
784 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
785 ;
786 /*
787 * OK, we've found the last block that must survive. The rest of our
788 * branch should be detached before unlocking. However, if that rest
789 * of branch is all ours and does not grow immediately from the inode
790 * it's easier to cheat and just decrement partial->p.
791 */
792 if (p == chain + k - 1 && p > chain) {
793 p->p--;
794 } else {
795 *top = *p->p;
796 /* Nope, don't do this in ext4. Must leave the tree intact */
797#if 0
798 *p->p = 0;
799#endif
800 }
801 /* Writer: end */
802
803 while (partial > p) {
804 brelse(partial->bh);
805 partial--;
806 }
807no_top:
808 return partial;
809}
810
811/*
812 * Zero a number of block pointers in either an inode or an indirect block.
813 * If we restart the transaction we must again get write access to the
814 * indirect block for further modification.
815 *
816 * We release `count' blocks on disk, but (last - first) may be greater
817 * than `count' because there can be holes in there.
818 *
819 * Return 0 on success, 1 on invalid block range
820 * and < 0 on fatal error.
821 */
822static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
823 struct buffer_head *bh,
824 ext4_fsblk_t block_to_free,
825 unsigned long count, __le32 *first,
826 __le32 *last)
827{
828 __le32 *p;
Theodore Ts'o981250c2013-06-12 11:48:29 -0400829 int flags = EXT4_FREE_BLOCKS_VALIDATED;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400830 int err;
831
832 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Theodore Ts'o981250c2013-06-12 11:48:29 -0400833 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
834 else if (ext4_should_journal_data(inode))
835 flags |= EXT4_FREE_BLOCKS_FORGET;
Amir Goldsteindae1e522011-06-27 19:40:50 -0400836
837 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
838 count)) {
839 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
840 "blocks %llu len %lu",
841 (unsigned long long) block_to_free, count);
842 return 1;
843 }
844
845 if (try_to_extend_transaction(handle, inode)) {
846 if (bh) {
847 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
848 err = ext4_handle_dirty_metadata(handle, inode, bh);
849 if (unlikely(err))
850 goto out_err;
851 }
852 err = ext4_mark_inode_dirty(handle, inode);
853 if (unlikely(err))
854 goto out_err;
855 err = ext4_truncate_restart_trans(handle, inode,
856 ext4_blocks_for_truncate(inode));
857 if (unlikely(err))
858 goto out_err;
859 if (bh) {
860 BUFFER_TRACE(bh, "retaking write access");
861 err = ext4_journal_get_write_access(handle, bh);
862 if (unlikely(err))
863 goto out_err;
864 }
865 }
866
867 for (p = first; p < last; p++)
868 *p = 0;
869
870 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
871 return 0;
872out_err:
873 ext4_std_error(inode->i_sb, err);
874 return err;
875}
876
877/**
878 * ext4_free_data - free a list of data blocks
879 * @handle: handle for this transaction
880 * @inode: inode we are dealing with
881 * @this_bh: indirect buffer_head which contains *@first and *@last
882 * @first: array of block numbers
883 * @last: points immediately past the end of array
884 *
885 * We are freeing all blocks referred from that array (numbers are stored as
886 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
887 *
888 * We accumulate contiguous runs of blocks to free. Conveniently, if these
889 * blocks are contiguous then releasing them at one time will only affect one
890 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
891 * actually use a lot of journal space.
892 *
893 * @this_bh will be %NULL if @first and @last point into the inode's direct
894 * block pointers.
895 */
896static void ext4_free_data(handle_t *handle, struct inode *inode,
897 struct buffer_head *this_bh,
898 __le32 *first, __le32 *last)
899{
900 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
901 unsigned long count = 0; /* Number of blocks in the run */
902 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
903 corresponding to
904 block_to_free */
905 ext4_fsblk_t nr; /* Current block # */
906 __le32 *p; /* Pointer into inode/ind
907 for current block */
908 int err = 0;
909
910 if (this_bh) { /* For indirect block */
911 BUFFER_TRACE(this_bh, "get_write_access");
912 err = ext4_journal_get_write_access(handle, this_bh);
913 /* Important: if we can't update the indirect pointers
914 * to the blocks, we can't free them. */
915 if (err)
916 return;
917 }
918
919 for (p = first; p < last; p++) {
920 nr = le32_to_cpu(*p);
921 if (nr) {
922 /* accumulate blocks to free if they're contiguous */
923 if (count == 0) {
924 block_to_free = nr;
925 block_to_free_p = p;
926 count = 1;
927 } else if (nr == block_to_free + count) {
928 count++;
929 } else {
930 err = ext4_clear_blocks(handle, inode, this_bh,
931 block_to_free, count,
932 block_to_free_p, p);
933 if (err)
934 break;
935 block_to_free = nr;
936 block_to_free_p = p;
937 count = 1;
938 }
939 }
940 }
941
942 if (!err && count > 0)
943 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
944 count, block_to_free_p, p);
945 if (err < 0)
946 /* fatal error */
947 return;
948
949 if (this_bh) {
950 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
951
952 /*
953 * The buffer head should have an attached journal head at this
954 * point. However, if the data is corrupted and an indirect
955 * block pointed to itself, it would have been detached when
956 * the block was cleared. Check for this instead of OOPSing.
957 */
958 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
959 ext4_handle_dirty_metadata(handle, inode, this_bh);
960 else
961 EXT4_ERROR_INODE(inode,
962 "circular indirect block detected at "
963 "block %llu",
964 (unsigned long long) this_bh->b_blocknr);
965 }
966}
967
968/**
969 * ext4_free_branches - free an array of branches
970 * @handle: JBD handle for this transaction
971 * @inode: inode we are dealing with
972 * @parent_bh: the buffer_head which contains *@first and *@last
973 * @first: array of block numbers
974 * @last: pointer immediately past the end of array
975 * @depth: depth of the branches to free
976 *
977 * We are freeing all blocks referred from these branches (numbers are
978 * stored as little-endian 32-bit) and updating @inode->i_blocks
979 * appropriately.
980 */
981static void ext4_free_branches(handle_t *handle, struct inode *inode,
982 struct buffer_head *parent_bh,
983 __le32 *first, __le32 *last, int depth)
984{
985 ext4_fsblk_t nr;
986 __le32 *p;
987
988 if (ext4_handle_is_aborted(handle))
989 return;
990
991 if (depth--) {
992 struct buffer_head *bh;
993 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
994 p = last;
995 while (--p >= first) {
996 nr = le32_to_cpu(*p);
997 if (!nr)
998 continue; /* A hole */
999
1000 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1001 nr, 1)) {
1002 EXT4_ERROR_INODE(inode,
1003 "invalid indirect mapped "
1004 "block %lu (level %d)",
1005 (unsigned long) nr, depth);
1006 break;
1007 }
1008
1009 /* Go read the buffer for the next level down */
1010 bh = sb_bread(inode->i_sb, nr);
1011
1012 /*
1013 * A read failure? Report error and clear slot
1014 * (should be rare).
1015 */
1016 if (!bh) {
1017 EXT4_ERROR_INODE_BLOCK(inode, nr,
1018 "Read failure");
1019 continue;
1020 }
1021
1022 /* This zaps the entire block. Bottom up. */
1023 BUFFER_TRACE(bh, "free child branches");
1024 ext4_free_branches(handle, inode, bh,
1025 (__le32 *) bh->b_data,
1026 (__le32 *) bh->b_data + addr_per_block,
1027 depth);
1028 brelse(bh);
1029
1030 /*
1031 * Everything below this this pointer has been
1032 * released. Now let this top-of-subtree go.
1033 *
1034 * We want the freeing of this indirect block to be
1035 * atomic in the journal with the updating of the
1036 * bitmap block which owns it. So make some room in
1037 * the journal.
1038 *
1039 * We zero the parent pointer *after* freeing its
1040 * pointee in the bitmaps, so if extend_transaction()
1041 * for some reason fails to put the bitmap changes and
1042 * the release into the same transaction, recovery
1043 * will merely complain about releasing a free block,
1044 * rather than leaking blocks.
1045 */
1046 if (ext4_handle_is_aborted(handle))
1047 return;
1048 if (try_to_extend_transaction(handle, inode)) {
1049 ext4_mark_inode_dirty(handle, inode);
1050 ext4_truncate_restart_trans(handle, inode,
1051 ext4_blocks_for_truncate(inode));
1052 }
1053
1054 /*
1055 * The forget flag here is critical because if
1056 * we are journaling (and not doing data
1057 * journaling), we have to make sure a revoke
1058 * record is written to prevent the journal
1059 * replay from overwriting the (former)
1060 * indirect block if it gets reallocated as a
1061 * data block. This must happen in the same
1062 * transaction where the data blocks are
1063 * actually freed.
1064 */
1065 ext4_free_blocks(handle, inode, NULL, nr, 1,
1066 EXT4_FREE_BLOCKS_METADATA|
1067 EXT4_FREE_BLOCKS_FORGET);
1068
1069 if (parent_bh) {
1070 /*
1071 * The block which we have just freed is
1072 * pointed to by an indirect block: journal it
1073 */
1074 BUFFER_TRACE(parent_bh, "get_write_access");
1075 if (!ext4_journal_get_write_access(handle,
1076 parent_bh)){
1077 *p = 0;
1078 BUFFER_TRACE(parent_bh,
1079 "call ext4_handle_dirty_metadata");
1080 ext4_handle_dirty_metadata(handle,
1081 inode,
1082 parent_bh);
1083 }
1084 }
1085 }
1086 } else {
1087 /* We have reached the bottom of the tree. */
1088 BUFFER_TRACE(parent_bh, "free data blocks");
1089 ext4_free_data(handle, inode, parent_bh, first, last);
1090 }
1091}
1092
Theodore Ts'o819c4922013-04-03 12:47:17 -04001093void ext4_ind_truncate(handle_t *handle, struct inode *inode)
Amir Goldsteindae1e522011-06-27 19:40:50 -04001094{
Amir Goldsteindae1e522011-06-27 19:40:50 -04001095 struct ext4_inode_info *ei = EXT4_I(inode);
1096 __le32 *i_data = ei->i_data;
1097 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Amir Goldsteindae1e522011-06-27 19:40:50 -04001098 ext4_lblk_t offsets[4];
1099 Indirect chain[4];
1100 Indirect *partial;
1101 __le32 nr = 0;
1102 int n = 0;
1103 ext4_lblk_t last_block, max_block;
1104 unsigned blocksize = inode->i_sb->s_blocksize;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001105
1106 last_block = (inode->i_size + blocksize-1)
1107 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1108 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1109 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1110
Amir Goldsteindae1e522011-06-27 19:40:50 -04001111 if (last_block != max_block) {
1112 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1113 if (n == 0)
Theodore Ts'o819c4922013-04-03 12:47:17 -04001114 return;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001115 }
1116
Zheng Liu51865fd2012-11-08 21:57:32 -05001117 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
Amir Goldsteindae1e522011-06-27 19:40:50 -04001118
1119 /*
1120 * The orphan list entry will now protect us from any crash which
1121 * occurs before the truncate completes, so it is now safe to propagate
1122 * the new, shorter inode size (held for now in i_size) into the
1123 * on-disk inode. We do this via i_disksize, which is the value which
1124 * ext4 *really* writes onto the disk inode.
1125 */
1126 ei->i_disksize = inode->i_size;
1127
1128 if (last_block == max_block) {
1129 /*
1130 * It is unnecessary to free any data blocks if last_block is
1131 * equal to the indirect block limit.
1132 */
Theodore Ts'o819c4922013-04-03 12:47:17 -04001133 return;
Amir Goldsteindae1e522011-06-27 19:40:50 -04001134 } else if (n == 1) { /* direct blocks */
1135 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1136 i_data + EXT4_NDIR_BLOCKS);
1137 goto do_indirects;
1138 }
1139
1140 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1141 /* Kill the top of shared branch (not detached) */
1142 if (nr) {
1143 if (partial == chain) {
1144 /* Shared branch grows from the inode */
1145 ext4_free_branches(handle, inode, NULL,
1146 &nr, &nr+1, (chain+n-1) - partial);
1147 *partial->p = 0;
1148 /*
1149 * We mark the inode dirty prior to restart,
1150 * and prior to stop. No need for it here.
1151 */
1152 } else {
1153 /* Shared branch grows from an indirect block */
1154 BUFFER_TRACE(partial->bh, "get_write_access");
1155 ext4_free_branches(handle, inode, partial->bh,
1156 partial->p,
1157 partial->p+1, (chain+n-1) - partial);
1158 }
1159 }
1160 /* Clear the ends of indirect blocks on the shared branch */
1161 while (partial > chain) {
1162 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1163 (__le32*)partial->bh->b_data+addr_per_block,
1164 (chain+n-1) - partial);
1165 BUFFER_TRACE(partial->bh, "call brelse");
1166 brelse(partial->bh);
1167 partial--;
1168 }
1169do_indirects:
1170 /* Kill the remaining (whole) subtrees */
1171 switch (offsets[0]) {
1172 default:
1173 nr = i_data[EXT4_IND_BLOCK];
1174 if (nr) {
1175 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1176 i_data[EXT4_IND_BLOCK] = 0;
1177 }
1178 case EXT4_IND_BLOCK:
1179 nr = i_data[EXT4_DIND_BLOCK];
1180 if (nr) {
1181 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1182 i_data[EXT4_DIND_BLOCK] = 0;
1183 }
1184 case EXT4_DIND_BLOCK:
1185 nr = i_data[EXT4_TIND_BLOCK];
1186 if (nr) {
1187 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1188 i_data[EXT4_TIND_BLOCK] = 0;
1189 }
1190 case EXT4_TIND_BLOCK:
1191 ;
1192 }
Amir Goldsteindae1e522011-06-27 19:40:50 -04001193}
1194
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001195/**
1196 * ext4_ind_remove_space - remove space from the range
1197 * @handle: JBD handle for this transaction
1198 * @inode: inode we are dealing with
1199 * @start: First block to remove
1200 * @end: One block after the last block to remove (exclusive)
1201 *
1202 * Free the blocks in the defined range (end is exclusive endpoint of
1203 * range). This is used by ext4_punch_hole().
1204 */
1205int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1206 ext4_lblk_t start, ext4_lblk_t end)
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001207{
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001208 struct ext4_inode_info *ei = EXT4_I(inode);
1209 __le32 *i_data = ei->i_data;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001210 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001211 ext4_lblk_t offsets[4], offsets2[4];
1212 Indirect chain[4], chain2[4];
1213 Indirect *partial, *partial2;
1214 ext4_lblk_t max_block;
1215 __le32 nr = 0, nr2 = 0;
1216 int n = 0, n2 = 0;
1217 unsigned blocksize = inode->i_sb->s_blocksize;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001218
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001219 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1220 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1221 if (end >= max_block)
1222 end = max_block;
1223 if ((start >= end) || (start > max_block))
1224 return 0;
Jan Karaa93cd4c2014-06-26 12:30:54 -04001225
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001226 n = ext4_block_to_path(inode, start, offsets, NULL);
1227 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1228
1229 BUG_ON(n > n2);
1230
1231 if ((n == 1) && (n == n2)) {
1232 /* We're punching only within direct block range */
1233 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1234 i_data + offsets2[0]);
1235 return 0;
1236 } else if (n2 > n) {
1237 /*
1238 * Start and end are on a different levels so we're going to
1239 * free partial block at start, and partial block at end of
1240 * the range. If there are some levels in between then
1241 * do_indirects label will take care of that.
1242 */
1243
1244 if (n == 1) {
1245 /*
1246 * Start is at the direct block level, free
1247 * everything to the end of the level.
1248 */
1249 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1250 i_data + EXT4_NDIR_BLOCKS);
1251 goto end_range;
1252 }
1253
1254
1255 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1256 if (nr) {
1257 if (partial == chain) {
1258 /* Shared branch grows from the inode */
1259 ext4_free_branches(handle, inode, NULL,
1260 &nr, &nr+1, (chain+n-1) - partial);
1261 *partial->p = 0;
Jan Karaa93cd4c2014-06-26 12:30:54 -04001262 } else {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001263 /* Shared branch grows from an indirect block */
1264 BUFFER_TRACE(partial->bh, "get_write_access");
1265 ext4_free_branches(handle, inode, partial->bh,
1266 partial->p,
1267 partial->p+1, (chain+n-1) - partial);
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001268 }
1269 }
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001270
1271 /*
1272 * Clear the ends of indirect blocks on the shared branch
1273 * at the start of the range
1274 */
1275 while (partial > chain) {
1276 ext4_free_branches(handle, inode, partial->bh,
1277 partial->p + 1,
1278 (__le32 *)partial->bh->b_data+addr_per_block,
1279 (chain+n-1) - partial);
1280 BUFFER_TRACE(partial->bh, "call brelse");
1281 brelse(partial->bh);
1282 partial--;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001283 }
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001284
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001285end_range:
1286 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1287 if (nr2) {
1288 if (partial2 == chain2) {
1289 /*
1290 * Remember, end is exclusive so here we're at
1291 * the start of the next level we're not going
1292 * to free. Everything was covered by the start
1293 * of the range.
1294 */
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001295 goto do_indirects;
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001296 }
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001297 } else {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001298 /*
1299 * ext4_find_shared returns Indirect structure which
1300 * points to the last element which should not be
1301 * removed by truncate. But this is end of the range
1302 * in punch_hole so we need to point to the next element
1303 */
1304 partial2->p++;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001305 }
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001306
1307 /*
1308 * Clear the ends of indirect blocks on the shared branch
1309 * at the end of the range
1310 */
1311 while (partial2 > chain2) {
1312 ext4_free_branches(handle, inode, partial2->bh,
1313 (__le32 *)partial2->bh->b_data,
1314 partial2->p,
1315 (chain2+n2-1) - partial2);
1316 BUFFER_TRACE(partial2->bh, "call brelse");
1317 brelse(partial2->bh);
1318 partial2--;
1319 }
1320 goto do_indirects;
1321 }
1322
1323 /* Punch happened within the same level (n == n2) */
1324 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1325 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001326
1327 /* Free top, but only if partial2 isn't its subtree. */
1328 if (nr) {
1329 int level = min(partial - chain, partial2 - chain2);
1330 int i;
1331 int subtree = 1;
1332
1333 for (i = 0; i <= level; i++) {
1334 if (offsets[i] != offsets2[i]) {
1335 subtree = 0;
1336 break;
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001337 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001338 }
1339
1340 if (!subtree) {
1341 if (partial == chain) {
1342 /* Shared branch grows from the inode */
1343 ext4_free_branches(handle, inode, NULL,
1344 &nr, &nr+1,
1345 (chain+n-1) - partial);
1346 *partial->p = 0;
1347 } else {
1348 /* Shared branch grows from an indirect block */
1349 BUFFER_TRACE(partial->bh, "get_write_access");
1350 ext4_free_branches(handle, inode, partial->bh,
1351 partial->p,
1352 partial->p+1,
1353 (chain+n-1) - partial);
1354 }
1355 }
1356 }
1357
1358 if (!nr2) {
1359 /*
1360 * ext4_find_shared returns Indirect structure which
1361 * points to the last element which should not be
1362 * removed by truncate. But this is end of the range
1363 * in punch_hole so we need to point to the next element
1364 */
1365 partial2->p++;
1366 }
1367
1368 while (partial > chain || partial2 > chain2) {
1369 int depth = (chain+n-1) - partial;
1370 int depth2 = (chain2+n2-1) - partial2;
1371
1372 if (partial > chain && partial2 > chain2 &&
1373 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1374 /*
1375 * We've converged on the same block. Clear the range,
1376 * then we're done.
1377 */
1378 ext4_free_branches(handle, inode, partial->bh,
1379 partial->p + 1,
1380 partial2->p,
1381 (chain+n-1) - partial);
1382 BUFFER_TRACE(partial->bh, "call brelse");
1383 brelse(partial->bh);
1384 BUFFER_TRACE(partial2->bh, "call brelse");
1385 brelse(partial2->bh);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001386 return 0;
1387 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001388
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001389 /*
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001390 * The start and end partial branches may not be at the same
1391 * level even though the punch happened within one level. So, we
1392 * give them a chance to arrive at the same level, then walk
1393 * them in step with each other until we converge on the same
1394 * block.
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001395 */
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001396 if (partial > chain && depth <= depth2) {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001397 ext4_free_branches(handle, inode, partial->bh,
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001398 partial->p + 1,
1399 (__le32 *)partial->bh->b_data+addr_per_block,
1400 (chain+n-1) - partial);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001401 BUFFER_TRACE(partial->bh, "call brelse");
1402 brelse(partial->bh);
1403 partial--;
1404 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001405 if (partial2 > chain2 && depth2 <= depth) {
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001406 ext4_free_branches(handle, inode, partial2->bh,
1407 (__le32 *)partial2->bh->b_data,
1408 partial2->p,
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001409 (chain2+n2-1) - partial2);
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001410 BUFFER_TRACE(partial2->bh, "call brelse");
1411 brelse(partial2->bh);
1412 partial2--;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001413 }
1414 }
Omar Sandoval6f30b7e2015-02-14 20:08:51 -05001415 return 0;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001416
Lukas Czerner4f579ae2014-07-15 06:03:38 -04001417do_indirects:
1418 /* Kill the remaining (whole) subtrees */
1419 switch (offsets[0]) {
1420 default:
1421 if (++n >= n2)
1422 return 0;
1423 nr = i_data[EXT4_IND_BLOCK];
1424 if (nr) {
1425 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1426 i_data[EXT4_IND_BLOCK] = 0;
1427 }
1428 case EXT4_IND_BLOCK:
1429 if (++n >= n2)
1430 return 0;
1431 nr = i_data[EXT4_DIND_BLOCK];
1432 if (nr) {
1433 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1434 i_data[EXT4_DIND_BLOCK] = 0;
1435 }
1436 case EXT4_DIND_BLOCK:
1437 if (++n >= n2)
1438 return 0;
1439 nr = i_data[EXT4_TIND_BLOCK];
1440 if (nr) {
1441 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1442 i_data[EXT4_TIND_BLOCK] = 0;
1443 }
1444 case EXT4_TIND_BLOCK:
1445 ;
1446 }
1447 return 0;
Zheng Liu8bad6fc2013-01-28 09:21:37 -05001448}