blob: bd42ab29efec829a03d299e77606674990d5b613 [file] [log] [blame]
Alex Tomasa86c6182006-10-11 01:21:03 -07001/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
21 */
22
23/*
24 * Extents support for EXT4
25 *
26 * TODO:
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
30 */
31
32#include <linux/module.h>
33#include <linux/fs.h>
34#include <linux/time.h>
Mingming Caocd02ff02007-10-16 18:38:25 -040035#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070036#include <linux/highuid.h>
37#include <linux/pagemap.h>
38#include <linux/quotaops.h>
39#include <linux/string.h>
40#include <linux/slab.h>
Amit Aroraa2df2a62007-07-17 21:42:41 -040041#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070042#include <asm/uaccess.h>
Eric Sandeen6873fa02008-10-07 00:46:36 -040043#include <linux/fiemap.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040044#include "ext4_jbd2.h"
45#include "ext4_extents.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070046
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040047#include <trace/events/ext4.h>
48
Allison Hendersond583fb82011-05-25 07:41:43 -040049static int ext4_split_extent(handle_t *handle,
50 struct inode *inode,
51 struct ext4_ext_path *path,
52 struct ext4_map_blocks *map,
53 int split_flag,
54 int flags);
55
Jan Kara487caee2009-08-17 22:17:20 -040056static int ext4_ext_truncate_extend_restart(handle_t *handle,
57 struct inode *inode,
58 int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -070059{
60 int err;
61
Frank Mayhar03901312009-01-07 00:06:22 -050062 if (!ext4_handle_valid(handle))
63 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -070064 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -040065 return 0;
66 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -040067 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -040068 return err;
Jan Kara487caee2009-08-17 22:17:20 -040069 err = ext4_truncate_restart_trans(handle, inode, needed);
Dmitry Monakhov0617b832010-05-17 01:00:00 -040070 if (err == 0)
71 err = -EAGAIN;
Jan Kara487caee2009-08-17 22:17:20 -040072
73 return err;
Alex Tomasa86c6182006-10-11 01:21:03 -070074}
75
76/*
77 * could return:
78 * - EROFS
79 * - ENOMEM
80 */
81static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
82 struct ext4_ext_path *path)
83{
84 if (path->p_bh) {
85 /* path points to block */
86 return ext4_journal_get_write_access(handle, path->p_bh);
87 }
88 /* path points to leaf/index in inode body */
89 /* we use in-core data, no need to protect them */
90 return 0;
91}
92
93/*
94 * could return:
95 * - EROFS
96 * - ENOMEM
97 * - EIO
98 */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -040099#define ext4_ext_dirty(handle, inode, path) \
100 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
101static int __ext4_ext_dirty(const char *where, unsigned int line,
102 handle_t *handle, struct inode *inode,
103 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700104{
105 int err;
106 if (path->p_bh) {
107 /* path points to block */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400108 err = __ext4_handle_dirty_metadata(where, line, handle,
109 inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700110 } else {
111 /* path points to leaf/index in inode body */
112 err = ext4_mark_inode_dirty(handle, inode);
113 }
114 return err;
115}
116
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700117static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700118 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500119 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700120{
Alex Tomasa86c6182006-10-11 01:21:03 -0700121 int depth;
122
123 if (path) {
124 struct ext4_extent *ex;
125 depth = path->p_depth;
126
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500127 /*
128 * Try to predict block placement assuming that we are
129 * filling in a file which will eventually be
130 * non-sparse --- i.e., in the case of libbfd writing
131 * an ELF object sections out-of-order but in a way
132 * the eventually results in a contiguous object or
133 * executable file, or some database extending a table
134 * space file. However, this is actually somewhat
135 * non-ideal if we are writing a sparse file such as
136 * qemu or KVM writing a raw image file that is going
137 * to stay fairly sparse, since it will end up
138 * fragmenting the file system's free space. Maybe we
139 * should have some hueristics or some way to allow
140 * userspace to pass a hint to file system,
Tao Mab8d65682011-01-21 23:21:31 +0800141 * especially if the latter case turns out to be
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500142 * common.
143 */
Avantika Mathur7e028972006-12-06 20:41:33 -0800144 ex = path[depth].p_ext;
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500145 if (ex) {
146 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
147 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
148
149 if (block > ext_block)
150 return ext_pblk + (block - ext_block);
151 else
152 return ext_pblk - (ext_block - block);
153 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700154
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700155 /* it looks like index is empty;
156 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700157 if (path[depth].p_bh)
158 return path[depth].p_bh->b_blocknr;
159 }
160
161 /* OK. use inode's group */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400162 return ext4_inode_to_goal_block(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700163}
164
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400165/*
166 * Allocation for a meta data block
167 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700168static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400169ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700170 struct ext4_ext_path *path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400171 struct ext4_extent *ex, int *err, unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700172{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700173 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700174
175 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Allison Henderson55f020d2011-05-25 07:41:26 -0400176 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
177 NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700178 return newblock;
179}
180
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400181static inline int ext4_ext_space_block(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700182{
183 int size;
184
185 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
186 / sizeof(struct ext4_extent);
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400187 if (!check) {
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100188#ifdef AGGRESSIVE_TEST
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400189 if (size > 6)
190 size = 6;
Alex Tomasa86c6182006-10-11 01:21:03 -0700191#endif
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400192 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700193 return size;
194}
195
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400196static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700197{
198 int size;
199
200 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
201 / sizeof(struct ext4_extent_idx);
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400202 if (!check) {
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100203#ifdef AGGRESSIVE_TEST
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400204 if (size > 5)
205 size = 5;
Alex Tomasa86c6182006-10-11 01:21:03 -0700206#endif
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400207 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700208 return size;
209}
210
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400211static inline int ext4_ext_space_root(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700212{
213 int size;
214
215 size = sizeof(EXT4_I(inode)->i_data);
216 size -= sizeof(struct ext4_extent_header);
217 size /= sizeof(struct ext4_extent);
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400218 if (!check) {
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100219#ifdef AGGRESSIVE_TEST
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400220 if (size > 3)
221 size = 3;
Alex Tomasa86c6182006-10-11 01:21:03 -0700222#endif
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400223 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700224 return size;
225}
226
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400227static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700228{
229 int size;
230
231 size = sizeof(EXT4_I(inode)->i_data);
232 size -= sizeof(struct ext4_extent_header);
233 size /= sizeof(struct ext4_extent_idx);
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400234 if (!check) {
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100235#ifdef AGGRESSIVE_TEST
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400236 if (size > 4)
237 size = 4;
Alex Tomasa86c6182006-10-11 01:21:03 -0700238#endif
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400239 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700240 return size;
241}
242
Mingming Caod2a17632008-07-14 17:52:37 -0400243/*
244 * Calculate the number of metadata blocks needed
245 * to allocate @blocks
246 * Worse case is one block per extent
247 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500248int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400249{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500250 struct ext4_inode_info *ei = EXT4_I(inode);
251 int idxs, num = 0;
Mingming Caod2a17632008-07-14 17:52:37 -0400252
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500253 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
254 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400255
256 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500257 * If the new delayed allocation block is contiguous with the
258 * previous da block, it can share index blocks with the
259 * previous block, so we only need to allocate a new index
260 * block every idxs leaf blocks. At ldxs**2 blocks, we need
261 * an additional index block, and at ldxs**3 blocks, yet
262 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400263 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500264 if (ei->i_da_metadata_calc_len &&
265 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
266 if ((ei->i_da_metadata_calc_len % idxs) == 0)
267 num++;
268 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
269 num++;
270 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
271 num++;
272 ei->i_da_metadata_calc_len = 0;
273 } else
274 ei->i_da_metadata_calc_len++;
275 ei->i_da_metadata_calc_last_lblock++;
276 return num;
277 }
Mingming Caod2a17632008-07-14 17:52:37 -0400278
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500279 /*
280 * In the worst case we need a new set of index blocks at
281 * every level of the inode's extent tree.
282 */
283 ei->i_da_metadata_calc_len = 1;
284 ei->i_da_metadata_calc_last_lblock = lblock;
285 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400286}
287
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400288static int
289ext4_ext_max_entries(struct inode *inode, int depth)
290{
291 int max;
292
293 if (depth == ext_depth(inode)) {
294 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400295 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400296 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400297 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400298 } else {
299 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400300 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400301 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400302 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400303 }
304
305 return max;
306}
307
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400308static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
309{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400310 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400311 int len = ext4_ext_get_actual_len(ext);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400312
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400313 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400314}
315
316static int ext4_valid_extent_idx(struct inode *inode,
317 struct ext4_extent_idx *ext_idx)
318{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400319 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400320
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400321 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400322}
323
324static int ext4_valid_extent_entries(struct inode *inode,
325 struct ext4_extent_header *eh,
326 int depth)
327{
328 struct ext4_extent *ext;
329 struct ext4_extent_idx *ext_idx;
330 unsigned short entries;
331 if (eh->eh_entries == 0)
332 return 1;
333
334 entries = le16_to_cpu(eh->eh_entries);
335
336 if (depth == 0) {
337 /* leaf entries */
338 ext = EXT_FIRST_EXTENT(eh);
339 while (entries) {
340 if (!ext4_valid_extent(inode, ext))
341 return 0;
342 ext++;
343 entries--;
344 }
345 } else {
346 ext_idx = EXT_FIRST_INDEX(eh);
347 while (entries) {
348 if (!ext4_valid_extent_idx(inode, ext_idx))
349 return 0;
350 ext_idx++;
351 entries--;
352 }
353 }
354 return 1;
355}
356
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400357static int __ext4_ext_check(const char *function, unsigned int line,
358 struct inode *inode, struct ext4_extent_header *eh,
359 int depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400360{
361 const char *error_msg;
362 int max = 0;
363
364 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
365 error_msg = "invalid magic";
366 goto corrupted;
367 }
368 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
369 error_msg = "unexpected eh_depth";
370 goto corrupted;
371 }
372 if (unlikely(eh->eh_max == 0)) {
373 error_msg = "invalid eh_max";
374 goto corrupted;
375 }
376 max = ext4_ext_max_entries(inode, depth);
377 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
378 error_msg = "too large eh_max";
379 goto corrupted;
380 }
381 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
382 error_msg = "invalid eh_entries";
383 goto corrupted;
384 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400385 if (!ext4_valid_extent_entries(inode, eh, depth)) {
386 error_msg = "invalid extent entries";
387 goto corrupted;
388 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400389 return 0;
390
391corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400392 ext4_error_inode(inode, function, line, 0,
Theodore Ts'o24676da2010-05-16 21:00:00 -0400393 "bad header/extent: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400394 "entries %u, max %u(%u), depth %u(%u)",
Theodore Ts'o24676da2010-05-16 21:00:00 -0400395 error_msg, le16_to_cpu(eh->eh_magic),
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400396 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
397 max, le16_to_cpu(eh->eh_depth), depth);
398
399 return -EIO;
400}
401
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400402#define ext4_ext_check(inode, eh, depth) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400403 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400404
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400405int ext4_ext_check_inode(struct inode *inode)
406{
407 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
408}
409
Alex Tomasa86c6182006-10-11 01:21:03 -0700410#ifdef EXT_DEBUG
411static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
412{
413 int k, l = path->p_depth;
414
415 ext_debug("path:");
416 for (k = 0; k <= l; k++, path++) {
417 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700418 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400419 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700420 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400421 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700422 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400423 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400424 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400425 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700426 } else
427 ext_debug(" []");
428 }
429 ext_debug("\n");
430}
431
432static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
433{
434 int depth = ext_depth(inode);
435 struct ext4_extent_header *eh;
436 struct ext4_extent *ex;
437 int i;
438
439 if (!path)
440 return;
441
442 eh = path[depth].p_hdr;
443 ex = EXT_FIRST_EXTENT(eh);
444
Mingming553f9002009-09-18 13:34:55 -0400445 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
446
Alex Tomasa86c6182006-10-11 01:21:03 -0700447 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400448 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
449 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400450 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700451 }
452 ext_debug("\n");
453}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400454
455static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
456 ext4_fsblk_t newblock, int level)
457{
458 int depth = ext_depth(inode);
459 struct ext4_extent *ex;
460
461 if (depth != level) {
462 struct ext4_extent_idx *idx;
463 idx = path[level].p_idx;
464 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
465 ext_debug("%d: move %d:%llu in new index %llu\n", level,
466 le32_to_cpu(idx->ei_block),
467 ext4_idx_pblock(idx),
468 newblock);
469 idx++;
470 }
471
472 return;
473 }
474
475 ex = path[depth].p_ext;
476 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
477 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
478 le32_to_cpu(ex->ee_block),
479 ext4_ext_pblock(ex),
480 ext4_ext_is_uninitialized(ex),
481 ext4_ext_get_actual_len(ex),
482 newblock);
483 ex++;
484 }
485}
486
Alex Tomasa86c6182006-10-11 01:21:03 -0700487#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400488#define ext4_ext_show_path(inode, path)
489#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400490#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700491#endif
492
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500493void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700494{
495 int depth = path->p_depth;
496 int i;
497
498 for (i = 0; i <= depth; i++, path++)
499 if (path->p_bh) {
500 brelse(path->p_bh);
501 path->p_bh = NULL;
502 }
503}
504
505/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700506 * ext4_ext_binsearch_idx:
507 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400508 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700509 */
510static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500511ext4_ext_binsearch_idx(struct inode *inode,
512 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700513{
514 struct ext4_extent_header *eh = path->p_hdr;
515 struct ext4_extent_idx *r, *l, *m;
516
Alex Tomasa86c6182006-10-11 01:21:03 -0700517
Eric Sandeenbba90742008-01-28 23:58:27 -0500518 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700519
520 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400521 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700522 while (l <= r) {
523 m = l + (r - l) / 2;
524 if (block < le32_to_cpu(m->ei_block))
525 r = m - 1;
526 else
527 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400528 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
529 m, le32_to_cpu(m->ei_block),
530 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700531 }
532
533 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700534 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400535 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700536
537#ifdef CHECK_BINSEARCH
538 {
539 struct ext4_extent_idx *chix, *ix;
540 int k;
541
542 chix = ix = EXT_FIRST_INDEX(eh);
543 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
544 if (k != 0 &&
545 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o47760042008-09-08 23:00:52 -0400546 printk(KERN_DEBUG "k=%d, ix=0x%p, "
547 "first=0x%p\n", k,
548 ix, EXT_FIRST_INDEX(eh));
549 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700550 le32_to_cpu(ix->ei_block),
551 le32_to_cpu(ix[-1].ei_block));
552 }
553 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400554 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700555 if (block < le32_to_cpu(ix->ei_block))
556 break;
557 chix = ix;
558 }
559 BUG_ON(chix != path->p_idx);
560 }
561#endif
562
563}
564
565/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700566 * ext4_ext_binsearch:
567 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400568 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700569 */
570static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500571ext4_ext_binsearch(struct inode *inode,
572 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700573{
574 struct ext4_extent_header *eh = path->p_hdr;
575 struct ext4_extent *r, *l, *m;
576
Alex Tomasa86c6182006-10-11 01:21:03 -0700577 if (eh->eh_entries == 0) {
578 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700579 * this leaf is empty:
580 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700581 */
582 return;
583 }
584
Eric Sandeenbba90742008-01-28 23:58:27 -0500585 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700586
587 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400588 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700589
590 while (l <= r) {
591 m = l + (r - l) / 2;
592 if (block < le32_to_cpu(m->ee_block))
593 r = m - 1;
594 else
595 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400596 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
597 m, le32_to_cpu(m->ee_block),
598 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700599 }
600
601 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400602 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400603 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400604 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400605 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400606 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700607
608#ifdef CHECK_BINSEARCH
609 {
610 struct ext4_extent *chex, *ex;
611 int k;
612
613 chex = ex = EXT_FIRST_EXTENT(eh);
614 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
615 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400616 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700617 if (block < le32_to_cpu(ex->ee_block))
618 break;
619 chex = ex;
620 }
621 BUG_ON(chex != path->p_ext);
622 }
623#endif
624
625}
626
627int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
628{
629 struct ext4_extent_header *eh;
630
631 eh = ext_inode_hdr(inode);
632 eh->eh_depth = 0;
633 eh->eh_entries = 0;
634 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400635 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700636 ext4_mark_inode_dirty(handle, inode);
637 ext4_ext_invalidate_cache(inode);
638 return 0;
639}
640
641struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500642ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
643 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700644{
645 struct ext4_extent_header *eh;
646 struct buffer_head *bh;
647 short int depth, i, ppos = 0, alloc = 0;
648
649 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400650 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700651
652 /* account possible depth increase */
653 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800654 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700655 GFP_NOFS);
656 if (!path)
657 return ERR_PTR(-ENOMEM);
658 alloc = 1;
659 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700660 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400661 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700662
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400663 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700664 /* walk through the tree */
665 while (i) {
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400666 int need_to_validate = 0;
667
Alex Tomasa86c6182006-10-11 01:21:03 -0700668 ext_debug("depth %d: num %d, max %d\n",
669 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400670
Alex Tomasa86c6182006-10-11 01:21:03 -0700671 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400672 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700673 path[ppos].p_depth = i;
674 path[ppos].p_ext = NULL;
675
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400676 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
677 if (unlikely(!bh))
Alex Tomasa86c6182006-10-11 01:21:03 -0700678 goto err;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400679 if (!bh_uptodate_or_lock(bh)) {
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400680 trace_ext4_ext_load_extent(inode, block,
681 path[ppos].p_block);
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400682 if (bh_submit_read(bh) < 0) {
683 put_bh(bh);
684 goto err;
685 }
686 /* validate the extent entries */
687 need_to_validate = 1;
688 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700689 eh = ext_block_hdr(bh);
690 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500691 if (unlikely(ppos > depth)) {
692 put_bh(bh);
693 EXT4_ERROR_INODE(inode,
694 "ppos %d > depth %d", ppos, depth);
695 goto err;
696 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700697 path[ppos].p_bh = bh;
698 path[ppos].p_hdr = eh;
699 i--;
700
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400701 if (need_to_validate && ext4_ext_check(inode, eh, i))
Alex Tomasa86c6182006-10-11 01:21:03 -0700702 goto err;
703 }
704
705 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700706 path[ppos].p_ext = NULL;
707 path[ppos].p_idx = NULL;
708
Alex Tomasa86c6182006-10-11 01:21:03 -0700709 /* find extent */
710 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400711 /* if not an empty leaf */
712 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400713 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700714
715 ext4_ext_show_path(inode, path);
716
717 return path;
718
719err:
720 ext4_ext_drop_refs(path);
721 if (alloc)
722 kfree(path);
723 return ERR_PTR(-EIO);
724}
725
726/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700727 * ext4_ext_insert_index:
728 * insert new index [@logical;@ptr] into the block at @curp;
729 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700730 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400731static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
732 struct ext4_ext_path *curp,
733 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700734{
735 struct ext4_extent_idx *ix;
736 int len, err;
737
Avantika Mathur7e028972006-12-06 20:41:33 -0800738 err = ext4_ext_get_access(handle, inode, curp);
739 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700740 return err;
741
Frank Mayhar273df552010-03-02 11:46:09 -0500742 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
743 EXT4_ERROR_INODE(inode,
744 "logical %d == ei_block %d!",
745 logical, le32_to_cpu(curp->p_idx->ei_block));
746 return -EIO;
747 }
Robin Dongd4620312011-07-17 23:43:42 -0400748
749 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
750 >= le16_to_cpu(curp->p_hdr->eh_max))) {
751 EXT4_ERROR_INODE(inode,
752 "eh_entries %d >= eh_max %d!",
753 le16_to_cpu(curp->p_hdr->eh_entries),
754 le16_to_cpu(curp->p_hdr->eh_max));
755 return -EIO;
756 }
757
Alex Tomasa86c6182006-10-11 01:21:03 -0700758 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
759 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
760 /* insert after */
761 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
762 len = (len - 1) * sizeof(struct ext4_extent_idx);
763 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400764 ext_debug("insert new index %d after: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700765 "move %d from 0x%p to 0x%p\n",
766 logical, ptr, len,
767 (curp->p_idx + 1), (curp->p_idx + 2));
768 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
769 }
770 ix = curp->p_idx + 1;
771 } else {
772 /* insert before */
773 len = len * sizeof(struct ext4_extent_idx);
774 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400775 ext_debug("insert new index %d before: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700776 "move %d from 0x%p to 0x%p\n",
777 logical, ptr, len,
778 curp->p_idx, (curp->p_idx + 1));
779 memmove(curp->p_idx + 1, curp->p_idx, len);
780 ix = curp->p_idx;
781 }
782
783 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700784 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400785 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700786
Frank Mayhar273df552010-03-02 11:46:09 -0500787 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
788 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
789 return -EIO;
790 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700791
792 err = ext4_ext_dirty(handle, inode, curp);
793 ext4_std_error(inode->i_sb, err);
794
795 return err;
796}
797
798/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700799 * ext4_ext_split:
800 * inserts new subtree into the path, using free index entry
801 * at depth @at:
802 * - allocates all needed blocks (new leaf and all intermediate index blocks)
803 * - makes decision where to split
804 * - moves remaining extents and index entries (right to the split point)
805 * into the newly allocated blocks
806 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700807 */
808static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400809 unsigned int flags,
810 struct ext4_ext_path *path,
811 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -0700812{
813 struct buffer_head *bh = NULL;
814 int depth = ext_depth(inode);
815 struct ext4_extent_header *neh;
816 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -0700817 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700818 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700819 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700820 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700821 int err = 0;
822
823 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700824 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700825
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700826 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700827 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -0500828 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
829 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
830 return -EIO;
831 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700832 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
833 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700834 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700835 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400836 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700837 } else {
838 border = newext->ee_block;
839 ext_debug("leaf will be added."
840 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400841 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700842 }
843
844 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700845 * If error occurs, then we break processing
846 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700847 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700848 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700849 */
850
851 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700852 * Get array to track all allocated blocks.
853 * We need this to handle errors and free blocks
854 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700855 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800856 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700857 if (!ablocks)
858 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700859
860 /* allocate all needed blocks */
861 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
862 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400863 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400864 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -0700865 if (newblock == 0)
866 goto cleanup;
867 ablocks[a] = newblock;
868 }
869
870 /* initialize new leaf */
871 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -0500872 if (unlikely(newblock == 0)) {
873 EXT4_ERROR_INODE(inode, "newblock == 0!");
874 err = -EIO;
875 goto cleanup;
876 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700877 bh = sb_getblk(inode->i_sb, newblock);
878 if (!bh) {
879 err = -EIO;
880 goto cleanup;
881 }
882 lock_buffer(bh);
883
Avantika Mathur7e028972006-12-06 20:41:33 -0800884 err = ext4_journal_get_create_access(handle, bh);
885 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700886 goto cleanup;
887
888 neh = ext_block_hdr(bh);
889 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400890 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700891 neh->eh_magic = EXT4_EXT_MAGIC;
892 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700893
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700894 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -0500895 if (unlikely(path[depth].p_hdr->eh_entries !=
896 path[depth].p_hdr->eh_max)) {
897 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
898 path[depth].p_hdr->eh_entries,
899 path[depth].p_hdr->eh_max);
900 err = -EIO;
901 goto cleanup;
902 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700903 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400904 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
905 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -0700906 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400907 struct ext4_extent *ex;
908 ex = EXT_FIRST_EXTENT(neh);
909 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400910 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700911 }
912
913 set_buffer_uptodate(bh);
914 unlock_buffer(bh);
915
Frank Mayhar03901312009-01-07 00:06:22 -0500916 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800917 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700918 goto cleanup;
919 brelse(bh);
920 bh = NULL;
921
922 /* correct old leaf */
923 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -0800924 err = ext4_ext_get_access(handle, inode, path + depth);
925 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700926 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400927 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -0800928 err = ext4_ext_dirty(handle, inode, path + depth);
929 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700930 goto cleanup;
931
932 }
933
934 /* create intermediate indexes */
935 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -0500936 if (unlikely(k < 0)) {
937 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
938 err = -EIO;
939 goto cleanup;
940 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700941 if (k)
942 ext_debug("create %d intermediate indices\n", k);
943 /* insert new index into current index block */
944 /* current depth stored in i var */
945 i = depth - 1;
946 while (k--) {
947 oldblock = newblock;
948 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -0500949 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700950 if (!bh) {
951 err = -EIO;
952 goto cleanup;
953 }
954 lock_buffer(bh);
955
Avantika Mathur7e028972006-12-06 20:41:33 -0800956 err = ext4_journal_get_create_access(handle, bh);
957 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700958 goto cleanup;
959
960 neh = ext_block_hdr(bh);
961 neh->eh_entries = cpu_to_le16(1);
962 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400963 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700964 neh->eh_depth = cpu_to_le16(depth - i);
965 fidx = EXT_FIRST_INDEX(neh);
966 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700967 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700968
Eric Sandeenbba90742008-01-28 23:58:27 -0500969 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
970 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700971
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400972 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -0500973 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
974 EXT_LAST_INDEX(path[i].p_hdr))) {
975 EXT4_ERROR_INODE(inode,
976 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
977 le32_to_cpu(path[i].p_ext->ee_block));
978 err = -EIO;
979 goto cleanup;
980 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400981 /* start copy indexes */
982 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
983 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
984 EXT_MAX_INDEX(path[i].p_hdr));
985 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -0700986 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400987 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -0700988 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400989 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700990 }
991 set_buffer_uptodate(bh);
992 unlock_buffer(bh);
993
Frank Mayhar03901312009-01-07 00:06:22 -0500994 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800995 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700996 goto cleanup;
997 brelse(bh);
998 bh = NULL;
999
1000 /* correct old index */
1001 if (m) {
1002 err = ext4_ext_get_access(handle, inode, path + i);
1003 if (err)
1004 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001005 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001006 err = ext4_ext_dirty(handle, inode, path + i);
1007 if (err)
1008 goto cleanup;
1009 }
1010
1011 i--;
1012 }
1013
1014 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001015 err = ext4_ext_insert_index(handle, inode, path + at,
1016 le32_to_cpu(border), newblock);
1017
1018cleanup:
1019 if (bh) {
1020 if (buffer_locked(bh))
1021 unlock_buffer(bh);
1022 brelse(bh);
1023 }
1024
1025 if (err) {
1026 /* free all allocated blocks in error case */
1027 for (i = 0; i < depth; i++) {
1028 if (!ablocks[i])
1029 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001030 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001031 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001032 }
1033 }
1034 kfree(ablocks);
1035
1036 return err;
1037}
1038
1039/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001040 * ext4_ext_grow_indepth:
1041 * implements tree growing procedure:
1042 * - allocates new block
1043 * - moves top-level data (index block or leaf) into the new block
1044 * - initializes new top-level, creating index that points to the
1045 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001046 */
1047static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001048 unsigned int flags,
1049 struct ext4_ext_path *path,
1050 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001051{
1052 struct ext4_ext_path *curp = path;
1053 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001054 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001055 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001056 int err = 0;
1057
Allison Henderson55f020d2011-05-25 07:41:26 -04001058 newblock = ext4_ext_new_meta_block(handle, inode, path,
1059 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001060 if (newblock == 0)
1061 return err;
1062
1063 bh = sb_getblk(inode->i_sb, newblock);
1064 if (!bh) {
1065 err = -EIO;
1066 ext4_std_error(inode->i_sb, err);
1067 return err;
1068 }
1069 lock_buffer(bh);
1070
Avantika Mathur7e028972006-12-06 20:41:33 -08001071 err = ext4_journal_get_create_access(handle, bh);
1072 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001073 unlock_buffer(bh);
1074 goto out;
1075 }
1076
1077 /* move top-level index/leaf into new block */
1078 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1079
1080 /* set size of new block */
1081 neh = ext_block_hdr(bh);
1082 /* old root could have indexes or leaves
1083 * so calculate e_max right way */
1084 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001085 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001086 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001087 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001088 neh->eh_magic = EXT4_EXT_MAGIC;
1089 set_buffer_uptodate(bh);
1090 unlock_buffer(bh);
1091
Frank Mayhar03901312009-01-07 00:06:22 -05001092 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001093 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001094 goto out;
1095
1096 /* create index in new top-level index: num,max,pointer */
Avantika Mathur7e028972006-12-06 20:41:33 -08001097 err = ext4_ext_get_access(handle, inode, curp);
1098 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001099 goto out;
1100
1101 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001102 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001103 curp->p_hdr->eh_entries = cpu_to_le16(1);
1104 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
Dmitry Monakhove9f410b2007-07-18 09:09:15 -04001105
1106 if (path[0].p_hdr->eh_depth)
1107 curp->p_idx->ei_block =
1108 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1109 else
1110 curp->p_idx->ei_block =
1111 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001112 ext4_idx_store_pblock(curp->p_idx, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001113
1114 neh = ext_inode_hdr(inode);
Mingming Cao2ae02102006-10-11 01:21:11 -07001115 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001116 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001117 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001118 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001119
1120 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1121 err = ext4_ext_dirty(handle, inode, curp);
1122out:
1123 brelse(bh);
1124
1125 return err;
1126}
1127
1128/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001129 * ext4_ext_create_new_leaf:
1130 * finds empty index and adds new leaf.
1131 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001132 */
1133static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001134 unsigned int flags,
1135 struct ext4_ext_path *path,
1136 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001137{
1138 struct ext4_ext_path *curp;
1139 int depth, i, err = 0;
1140
1141repeat:
1142 i = depth = ext_depth(inode);
1143
1144 /* walk up to the tree and look for free index entry */
1145 curp = path + depth;
1146 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1147 i--;
1148 curp--;
1149 }
1150
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001151 /* we use already allocated block for index block,
1152 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001153 if (EXT_HAS_FREE_INDEX(curp)) {
1154 /* if we found index with free entry, then use that
1155 * entry: create all needed subtree and add new leaf */
Allison Henderson55f020d2011-05-25 07:41:26 -04001156 err = ext4_ext_split(handle, inode, flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001157 if (err)
1158 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001159
1160 /* refill path */
1161 ext4_ext_drop_refs(path);
1162 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001163 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1164 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001165 if (IS_ERR(path))
1166 err = PTR_ERR(path);
1167 } else {
1168 /* tree is full, time to grow in depth */
Allison Henderson55f020d2011-05-25 07:41:26 -04001169 err = ext4_ext_grow_indepth(handle, inode, flags,
1170 path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001171 if (err)
1172 goto out;
1173
1174 /* refill path */
1175 ext4_ext_drop_refs(path);
1176 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001177 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1178 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001179 if (IS_ERR(path)) {
1180 err = PTR_ERR(path);
1181 goto out;
1182 }
1183
1184 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001185 * only first (depth 0 -> 1) produces free space;
1186 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001187 */
1188 depth = ext_depth(inode);
1189 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001190 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001191 goto repeat;
1192 }
1193 }
1194
1195out:
1196 return err;
1197}
1198
1199/*
Alex Tomas1988b512008-01-28 23:58:27 -05001200 * search the closest allocated block to the left for *logical
1201 * and returns it at @logical + it's physical address at @phys
1202 * if *logical is the smallest allocated block, the function
1203 * returns 0 at @phys
1204 * return value contains 0 (success) or error code
1205 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001206static int ext4_ext_search_left(struct inode *inode,
1207 struct ext4_ext_path *path,
1208 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001209{
1210 struct ext4_extent_idx *ix;
1211 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001212 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001213
Frank Mayhar273df552010-03-02 11:46:09 -05001214 if (unlikely(path == NULL)) {
1215 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1216 return -EIO;
1217 }
Alex Tomas1988b512008-01-28 23:58:27 -05001218 depth = path->p_depth;
1219 *phys = 0;
1220
1221 if (depth == 0 && path->p_ext == NULL)
1222 return 0;
1223
1224 /* usually extent in the path covers blocks smaller
1225 * then *logical, but it can be that extent is the
1226 * first one in the file */
1227
1228 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001229 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001230 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001231 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1232 EXT4_ERROR_INODE(inode,
1233 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1234 *logical, le32_to_cpu(ex->ee_block));
1235 return -EIO;
1236 }
Alex Tomas1988b512008-01-28 23:58:27 -05001237 while (--depth >= 0) {
1238 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001239 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1240 EXT4_ERROR_INODE(inode,
1241 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1242 ix != NULL ? ix->ei_block : 0,
1243 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1244 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1245 depth);
1246 return -EIO;
1247 }
Alex Tomas1988b512008-01-28 23:58:27 -05001248 }
1249 return 0;
1250 }
1251
Frank Mayhar273df552010-03-02 11:46:09 -05001252 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1253 EXT4_ERROR_INODE(inode,
1254 "logical %d < ee_block %d + ee_len %d!",
1255 *logical, le32_to_cpu(ex->ee_block), ee_len);
1256 return -EIO;
1257 }
Alex Tomas1988b512008-01-28 23:58:27 -05001258
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001259 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001260 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001261 return 0;
1262}
1263
1264/*
1265 * search the closest allocated block to the right for *logical
1266 * and returns it at @logical + it's physical address at @phys
1267 * if *logical is the smallest allocated block, the function
1268 * returns 0 at @phys
1269 * return value contains 0 (success) or error code
1270 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001271static int ext4_ext_search_right(struct inode *inode,
1272 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001273 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1274 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001275{
1276 struct buffer_head *bh = NULL;
1277 struct ext4_extent_header *eh;
1278 struct ext4_extent_idx *ix;
1279 struct ext4_extent *ex;
1280 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001281 int depth; /* Note, NOT eh_depth; depth from top of tree */
1282 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001283
Frank Mayhar273df552010-03-02 11:46:09 -05001284 if (unlikely(path == NULL)) {
1285 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1286 return -EIO;
1287 }
Alex Tomas1988b512008-01-28 23:58:27 -05001288 depth = path->p_depth;
1289 *phys = 0;
1290
1291 if (depth == 0 && path->p_ext == NULL)
1292 return 0;
1293
1294 /* usually extent in the path covers blocks smaller
1295 * then *logical, but it can be that extent is the
1296 * first one in the file */
1297
1298 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001299 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001300 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001301 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1302 EXT4_ERROR_INODE(inode,
1303 "first_extent(path[%d].p_hdr) != ex",
1304 depth);
1305 return -EIO;
1306 }
Alex Tomas1988b512008-01-28 23:58:27 -05001307 while (--depth >= 0) {
1308 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001309 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1310 EXT4_ERROR_INODE(inode,
1311 "ix != EXT_FIRST_INDEX *logical %d!",
1312 *logical);
1313 return -EIO;
1314 }
Alex Tomas1988b512008-01-28 23:58:27 -05001315 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001316 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001317 }
1318
Frank Mayhar273df552010-03-02 11:46:09 -05001319 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1320 EXT4_ERROR_INODE(inode,
1321 "logical %d < ee_block %d + ee_len %d!",
1322 *logical, le32_to_cpu(ex->ee_block), ee_len);
1323 return -EIO;
1324 }
Alex Tomas1988b512008-01-28 23:58:27 -05001325
1326 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1327 /* next allocated block in this leaf */
1328 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001329 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001330 }
1331
1332 /* go up and search for index to the right */
1333 while (--depth >= 0) {
1334 ix = path[depth].p_idx;
1335 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001336 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001337 }
1338
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001339 /* we've gone up to the root and found no index to the right */
1340 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001341
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001342got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001343 /* we've found index to the right, let's
1344 * follow it and find the closest allocated
1345 * block to the right */
1346 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001347 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001348 while (++depth < path->p_depth) {
1349 bh = sb_bread(inode->i_sb, block);
1350 if (bh == NULL)
1351 return -EIO;
1352 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001353 /* subtract from p_depth to get proper eh_depth */
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04001354 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001355 put_bh(bh);
1356 return -EIO;
1357 }
1358 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001359 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001360 put_bh(bh);
1361 }
1362
1363 bh = sb_bread(inode->i_sb, block);
1364 if (bh == NULL)
1365 return -EIO;
1366 eh = ext_block_hdr(bh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04001367 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001368 put_bh(bh);
1369 return -EIO;
1370 }
1371 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001372found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001373 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001374 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001375 *ret_ex = ex;
1376 if (bh)
1377 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001378 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001379}
1380
1381/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001382 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001383 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001384 * NOTE: it considers block number from index entry as
1385 * allocated block. Thus, index entries have to be consistent
1386 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001387 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001388static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001389ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1390{
1391 int depth;
1392
1393 BUG_ON(path == NULL);
1394 depth = path->p_depth;
1395
1396 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001397 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001398
1399 while (depth >= 0) {
1400 if (depth == path->p_depth) {
1401 /* leaf */
1402 if (path[depth].p_ext !=
1403 EXT_LAST_EXTENT(path[depth].p_hdr))
1404 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1405 } else {
1406 /* index */
1407 if (path[depth].p_idx !=
1408 EXT_LAST_INDEX(path[depth].p_hdr))
1409 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1410 }
1411 depth--;
1412 }
1413
Lukas Czernerf17722f2011-06-06 00:05:17 -04001414 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001415}
1416
1417/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001418 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001419 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001420 */
Robin Dong57187892011-07-23 21:49:07 -04001421static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001422{
1423 int depth;
1424
1425 BUG_ON(path == NULL);
1426 depth = path->p_depth;
1427
1428 /* zero-tree has no leaf blocks at all */
1429 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001430 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001431
1432 /* go to index block */
1433 depth--;
1434
1435 while (depth >= 0) {
1436 if (path[depth].p_idx !=
1437 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001438 return (ext4_lblk_t)
1439 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001440 depth--;
1441 }
1442
Lukas Czernerf17722f2011-06-06 00:05:17 -04001443 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001444}
1445
1446/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001447 * ext4_ext_correct_indexes:
1448 * if leaf gets modified and modified extent is first in the leaf,
1449 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001450 * TODO: do we need to correct tree in all cases?
1451 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001452static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001453 struct ext4_ext_path *path)
1454{
1455 struct ext4_extent_header *eh;
1456 int depth = ext_depth(inode);
1457 struct ext4_extent *ex;
1458 __le32 border;
1459 int k, err = 0;
1460
1461 eh = path[depth].p_hdr;
1462 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001463
1464 if (unlikely(ex == NULL || eh == NULL)) {
1465 EXT4_ERROR_INODE(inode,
1466 "ex %p == NULL or eh %p == NULL", ex, eh);
1467 return -EIO;
1468 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001469
1470 if (depth == 0) {
1471 /* there is no tree at all */
1472 return 0;
1473 }
1474
1475 if (ex != EXT_FIRST_EXTENT(eh)) {
1476 /* we correct tree if first leaf got modified only */
1477 return 0;
1478 }
1479
1480 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001481 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001482 */
1483 k = depth - 1;
1484 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001485 err = ext4_ext_get_access(handle, inode, path + k);
1486 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001487 return err;
1488 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001489 err = ext4_ext_dirty(handle, inode, path + k);
1490 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001491 return err;
1492
1493 while (k--) {
1494 /* change all left-side indexes */
1495 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1496 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001497 err = ext4_ext_get_access(handle, inode, path + k);
1498 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001499 break;
1500 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001501 err = ext4_ext_dirty(handle, inode, path + k);
1502 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001503 break;
1504 }
1505
1506 return err;
1507}
1508
Akira Fujita748de672009-06-17 19:24:03 -04001509int
Alex Tomasa86c6182006-10-11 01:21:03 -07001510ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1511 struct ext4_extent *ex2)
1512{
Amit Arora749269f2007-07-18 09:02:56 -04001513 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001514
1515 /*
1516 * Make sure that either both extents are uninitialized, or
1517 * both are _not_.
1518 */
1519 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1520 return 0;
1521
Amit Arora749269f2007-07-18 09:02:56 -04001522 if (ext4_ext_is_uninitialized(ex1))
1523 max_len = EXT_UNINIT_MAX_LEN;
1524 else
1525 max_len = EXT_INIT_MAX_LEN;
1526
Amit Aroraa2df2a62007-07-17 21:42:41 -04001527 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1528 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1529
1530 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001531 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001532 return 0;
1533
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001534 /*
1535 * To allow future support for preallocated extents to be added
1536 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001537 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001538 */
Amit Arora749269f2007-07-18 09:02:56 -04001539 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001540 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001541#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001542 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001543 return 0;
1544#endif
1545
Theodore Ts'obf89d162010-10-27 21:30:14 -04001546 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001547 return 1;
1548 return 0;
1549}
1550
1551/*
Amit Arora56055d32007-07-17 21:42:38 -04001552 * This function tries to merge the "ex" extent to the next extent in the tree.
1553 * It always tries to merge towards right. If you want to merge towards
1554 * left, pass "ex - 1" as argument instead of "ex".
1555 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1556 * 1 if they got merged.
1557 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001558static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001559 struct ext4_ext_path *path,
1560 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001561{
1562 struct ext4_extent_header *eh;
1563 unsigned int depth, len;
1564 int merge_done = 0;
1565 int uninitialized = 0;
1566
1567 depth = ext_depth(inode);
1568 BUG_ON(path[depth].p_hdr == NULL);
1569 eh = path[depth].p_hdr;
1570
1571 while (ex < EXT_LAST_EXTENT(eh)) {
1572 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1573 break;
1574 /* merge with next extent! */
1575 if (ext4_ext_is_uninitialized(ex))
1576 uninitialized = 1;
1577 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1578 + ext4_ext_get_actual_len(ex + 1));
1579 if (uninitialized)
1580 ext4_ext_mark_uninitialized(ex);
1581
1582 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1583 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1584 * sizeof(struct ext4_extent);
1585 memmove(ex + 1, ex + 2, len);
1586 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001587 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001588 merge_done = 1;
1589 WARN_ON(eh->eh_entries == 0);
1590 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001591 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001592 }
1593
1594 return merge_done;
1595}
1596
1597/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001598 * This function tries to merge the @ex extent to neighbours in the tree.
1599 * return 1 if merge left else 0.
1600 */
1601static int ext4_ext_try_to_merge(struct inode *inode,
1602 struct ext4_ext_path *path,
1603 struct ext4_extent *ex) {
1604 struct ext4_extent_header *eh;
1605 unsigned int depth;
1606 int merge_done = 0;
1607 int ret = 0;
1608
1609 depth = ext_depth(inode);
1610 BUG_ON(path[depth].p_hdr == NULL);
1611 eh = path[depth].p_hdr;
1612
1613 if (ex > EXT_FIRST_EXTENT(eh))
1614 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1615
1616 if (!merge_done)
1617 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1618
1619 return ret;
1620}
1621
1622/*
Amit Arora25d14f92007-05-24 13:04:13 -04001623 * check if a portion of the "newext" extent overlaps with an
1624 * existing extent.
1625 *
1626 * If there is an overlap discovered, it updates the length of the newext
1627 * such that there will be no overlap, and then returns 1.
1628 * If there is no overlap found, it returns 0.
1629 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001630static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1631 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001632 struct ext4_extent *newext,
1633 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001634{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001635 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001636 unsigned int depth, len1;
1637 unsigned int ret = 0;
1638
1639 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001640 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001641 depth = ext_depth(inode);
1642 if (!path[depth].p_ext)
1643 goto out;
1644 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001645 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001646
1647 /*
1648 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001649 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001650 */
1651 if (b2 < b1) {
1652 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001653 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001654 goto out;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001655 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001656 }
1657
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001658 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001659 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001660 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001661 newext->ee_len = cpu_to_le16(len1);
1662 ret = 1;
1663 }
1664
1665 /* check for overlap */
1666 if (b1 + len1 > b2) {
1667 newext->ee_len = cpu_to_le16(b2 - b1);
1668 ret = 1;
1669 }
1670out:
1671 return ret;
1672}
1673
1674/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001675 * ext4_ext_insert_extent:
1676 * tries to merge requsted extent into the existing extent or
1677 * inserts requested extent as new one into the tree,
1678 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001679 */
1680int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1681 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04001682 struct ext4_extent *newext, int flag)
Alex Tomasa86c6182006-10-11 01:21:03 -07001683{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001684 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001685 struct ext4_extent *ex, *fex;
1686 struct ext4_extent *nearex; /* nearest extent */
1687 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001688 int depth, len, err;
1689 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001690 unsigned uninitialized = 0;
Allison Henderson55f020d2011-05-25 07:41:26 -04001691 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001692
Frank Mayhar273df552010-03-02 11:46:09 -05001693 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1694 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1695 return -EIO;
1696 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001697 depth = ext_depth(inode);
1698 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001699 if (unlikely(path[depth].p_hdr == NULL)) {
1700 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1701 return -EIO;
1702 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001703
1704 /* try to insert block into found extent and return */
Jiaying Zhang744692d2010-03-04 16:14:02 -05001705 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
Mingming Cao00314622009-09-28 15:49:08 -04001706 && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming553f9002009-09-18 13:34:55 -04001707 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04001708 ext4_ext_is_uninitialized(newext),
1709 ext4_ext_get_actual_len(newext),
1710 le32_to_cpu(ex->ee_block),
1711 ext4_ext_is_uninitialized(ex),
1712 ext4_ext_get_actual_len(ex),
1713 ext4_ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001714 err = ext4_ext_get_access(handle, inode, path + depth);
1715 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001716 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001717
1718 /*
1719 * ext4_can_extents_be_merged should have checked that either
1720 * both extents are uninitialized, or both aren't. Thus we
1721 * need to check only one of them here.
1722 */
1723 if (ext4_ext_is_uninitialized(ex))
1724 uninitialized = 1;
1725 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1726 + ext4_ext_get_actual_len(newext));
1727 if (uninitialized)
1728 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001729 eh = path[depth].p_hdr;
1730 nearex = ex;
1731 goto merge;
1732 }
1733
Alex Tomasa86c6182006-10-11 01:21:03 -07001734 depth = ext_depth(inode);
1735 eh = path[depth].p_hdr;
1736 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1737 goto has_space;
1738
1739 /* probably next leaf has space for us? */
1740 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001741 next = EXT_MAX_BLOCKS;
1742 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001743 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001744 if (next != EXT_MAX_BLOCKS) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001745 ext_debug("next leaf block - %d\n", next);
1746 BUG_ON(npath != NULL);
1747 npath = ext4_ext_find_extent(inode, next, NULL);
1748 if (IS_ERR(npath))
1749 return PTR_ERR(npath);
1750 BUG_ON(npath->p_depth != path->p_depth);
1751 eh = npath[depth].p_hdr;
1752 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001753 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001754 le16_to_cpu(eh->eh_entries));
1755 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04001756 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07001757 }
1758 ext_debug("next leaf has no free space(%d,%d)\n",
1759 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1760 }
1761
1762 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001763 * There is no free space in the found leaf.
1764 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001765 */
Allison Henderson55f020d2011-05-25 07:41:26 -04001766 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1767 flags = EXT4_MB_USE_ROOT_BLOCKS;
1768 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001769 if (err)
1770 goto cleanup;
1771 depth = ext_depth(inode);
1772 eh = path[depth].p_hdr;
1773
1774has_space:
1775 nearex = path[depth].p_ext;
1776
Avantika Mathur7e028972006-12-06 20:41:33 -08001777 err = ext4_ext_get_access(handle, inode, path + depth);
1778 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001779 goto cleanup;
1780
1781 if (!nearex) {
1782 /* there is no extent in this leaf, create first one */
Mingming553f9002009-09-18 13:34:55 -04001783 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001784 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001785 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001786 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001787 ext4_ext_get_actual_len(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001788 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1789 } else if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001790 > le32_to_cpu(nearex->ee_block)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001791/* BUG_ON(newext->ee_block == nearex->ee_block); */
1792 if (nearex != EXT_LAST_EXTENT(eh)) {
1793 len = EXT_MAX_EXTENT(eh) - nearex;
1794 len = (len - 1) * sizeof(struct ext4_extent);
1795 len = len < 0 ? 0 : len;
Mingming553f9002009-09-18 13:34:55 -04001796 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001797 "move %d from 0x%p to 0x%p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001798 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001799 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001800 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001801 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001802 nearex, len, nearex + 1, nearex + 2);
1803 memmove(nearex + 2, nearex + 1, len);
1804 }
1805 path[depth].p_ext = nearex + 1;
1806 } else {
1807 BUG_ON(newext->ee_block == nearex->ee_block);
1808 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1809 len = len < 0 ? 0 : len;
Mingming553f9002009-09-18 13:34:55 -04001810 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001811 "move %d from 0x%p to 0x%p\n",
1812 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001813 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001814 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001815 ext4_ext_get_actual_len(newext),
Robin Dong07379642011-07-23 21:51:07 -04001816 nearex, len, nearex, nearex + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001817 memmove(nearex + 1, nearex, len);
1818 path[depth].p_ext = nearex;
1819 }
1820
Marcin Slusarze8546d02008-04-17 10:38:59 -04001821 le16_add_cpu(&eh->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001822 nearex = path[depth].p_ext;
1823 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001824 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001825 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001826
1827merge:
1828 /* try to merge extents to the right */
Jiaying Zhang744692d2010-03-04 16:14:02 -05001829 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
Mingming Cao00314622009-09-28 15:49:08 -04001830 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001831
1832 /* try to merge extents to the left */
1833
1834 /* time to correct all indexes above */
1835 err = ext4_ext_correct_indexes(handle, inode, path);
1836 if (err)
1837 goto cleanup;
1838
1839 err = ext4_ext_dirty(handle, inode, path + depth);
1840
1841cleanup:
1842 if (npath) {
1843 ext4_ext_drop_refs(npath);
1844 kfree(npath);
1845 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001846 ext4_ext_invalidate_cache(inode);
1847 return err;
1848}
1849
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001850static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1851 ext4_lblk_t num, ext_prepare_callback func,
1852 void *cbdata)
Eric Sandeen6873fa02008-10-07 00:46:36 -04001853{
1854 struct ext4_ext_path *path = NULL;
1855 struct ext4_ext_cache cbex;
1856 struct ext4_extent *ex;
1857 ext4_lblk_t next, start = 0, end = 0;
1858 ext4_lblk_t last = block + num;
1859 int depth, exists, err = 0;
1860
1861 BUG_ON(func == NULL);
1862 BUG_ON(inode == NULL);
1863
Lukas Czernerf17722f2011-06-06 00:05:17 -04001864 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04001865 num = last - block;
1866 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05001867 down_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001868 path = ext4_ext_find_extent(inode, block, path);
Theodore Ts'ofab3a542009-12-09 21:30:02 -05001869 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001870 if (IS_ERR(path)) {
1871 err = PTR_ERR(path);
1872 path = NULL;
1873 break;
1874 }
1875
1876 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05001877 if (unlikely(path[depth].p_hdr == NULL)) {
1878 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1879 err = -EIO;
1880 break;
1881 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04001882 ex = path[depth].p_ext;
1883 next = ext4_ext_next_allocated_block(path);
1884
1885 exists = 0;
1886 if (!ex) {
1887 /* there is no extent yet, so try to allocate
1888 * all requested space */
1889 start = block;
1890 end = block + num;
1891 } else if (le32_to_cpu(ex->ee_block) > block) {
1892 /* need to allocate space before found extent */
1893 start = block;
1894 end = le32_to_cpu(ex->ee_block);
1895 if (block + num < end)
1896 end = block + num;
1897 } else if (block >= le32_to_cpu(ex->ee_block)
1898 + ext4_ext_get_actual_len(ex)) {
1899 /* need to allocate space after found extent */
1900 start = block;
1901 end = block + num;
1902 if (end >= next)
1903 end = next;
1904 } else if (block >= le32_to_cpu(ex->ee_block)) {
1905 /*
1906 * some part of requested space is covered
1907 * by found extent
1908 */
1909 start = block;
1910 end = le32_to_cpu(ex->ee_block)
1911 + ext4_ext_get_actual_len(ex);
1912 if (block + num < end)
1913 end = block + num;
1914 exists = 1;
1915 } else {
1916 BUG();
1917 }
1918 BUG_ON(end <= start);
1919
1920 if (!exists) {
1921 cbex.ec_block = start;
1922 cbex.ec_len = end - start;
1923 cbex.ec_start = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04001924 } else {
1925 cbex.ec_block = le32_to_cpu(ex->ee_block);
1926 cbex.ec_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001927 cbex.ec_start = ext4_ext_pblock(ex);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001928 }
1929
Frank Mayhar273df552010-03-02 11:46:09 -05001930 if (unlikely(cbex.ec_len == 0)) {
1931 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1932 err = -EIO;
1933 break;
1934 }
Lukas Czernerc03f8aa2011-06-06 00:06:52 -04001935 err = func(inode, next, &cbex, ex, cbdata);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001936 ext4_ext_drop_refs(path);
1937
1938 if (err < 0)
1939 break;
1940
1941 if (err == EXT_REPEAT)
1942 continue;
1943 else if (err == EXT_BREAK) {
1944 err = 0;
1945 break;
1946 }
1947
1948 if (ext_depth(inode) != depth) {
1949 /* depth was changed. we have to realloc path */
1950 kfree(path);
1951 path = NULL;
1952 }
1953
1954 block = cbex.ec_block + cbex.ec_len;
1955 }
1956
1957 if (path) {
1958 ext4_ext_drop_refs(path);
1959 kfree(path);
1960 }
1961
1962 return err;
1963}
1964
Avantika Mathur09b88252006-12-06 20:41:36 -08001965static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001966ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05001967 __u32 len, ext4_fsblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001968{
1969 struct ext4_ext_cache *cex;
1970 BUG_ON(len == 0);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001971 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001972 cex = &EXT4_I(inode)->i_cached_extent;
Alex Tomasa86c6182006-10-11 01:21:03 -07001973 cex->ec_block = block;
1974 cex->ec_len = len;
1975 cex->ec_start = start;
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001976 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001977}
1978
1979/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001980 * ext4_ext_put_gap_in_cache:
1981 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001982 * and cache this gap
1983 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001984static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001985ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001986 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001987{
1988 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001989 unsigned long len;
1990 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001991 struct ext4_extent *ex;
1992
1993 ex = path[depth].p_ext;
1994 if (ex == NULL) {
1995 /* there is no extent yet, so gap is [0;-] */
1996 lblock = 0;
Lukas Czernerf17722f2011-06-06 00:05:17 -04001997 len = EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001998 ext_debug("cache gap(whole file):");
1999 } else if (block < le32_to_cpu(ex->ee_block)) {
2000 lblock = block;
2001 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002002 ext_debug("cache gap(before): %u [%u:%u]",
2003 block,
2004 le32_to_cpu(ex->ee_block),
2005 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002006 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002007 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002008 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002009 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002010 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002011
2012 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002013 ext_debug("cache gap(after): [%u:%u] %u",
2014 le32_to_cpu(ex->ee_block),
2015 ext4_ext_get_actual_len(ex),
2016 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002017 BUG_ON(next == lblock);
2018 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002019 } else {
2020 lblock = len = 0;
2021 BUG();
2022 }
2023
Eric Sandeenbba90742008-01-28 23:58:27 -05002024 ext_debug(" -> %u:%lu\n", lblock, len);
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002025 ext4_ext_put_in_cache(inode, lblock, len, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002026}
2027
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002028/*
Robin Dongb7ca1e82011-07-23 21:53:25 -04002029 * ext4_ext_check_cache()
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002030 * Checks to see if the given block is in the cache.
2031 * If it is, the cached extent is stored in the given
2032 * cache extent pointer. If the cached extent is a hole,
2033 * this routine should be used instead of
2034 * ext4_ext_in_cache if the calling function needs to
2035 * know the size of the hole.
2036 *
2037 * @inode: The files inode
2038 * @block: The block to look for in the cache
2039 * @ex: Pointer where the cached extent will be stored
2040 * if it contains block
2041 *
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002042 * Return 0 if cache is invalid; 1 if the cache is valid
2043 */
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002044static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2045 struct ext4_ext_cache *ex){
Alex Tomasa86c6182006-10-11 01:21:03 -07002046 struct ext4_ext_cache *cex;
Vivek Haldar77f41352011-05-22 21:24:16 -04002047 struct ext4_sb_info *sbi;
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002048 int ret = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002049
Theodore Ts'o60e66792010-05-17 07:00:00 -04002050 /*
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002051 * We borrow i_block_reservation_lock to protect i_cached_extent
2052 */
2053 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07002054 cex = &EXT4_I(inode)->i_cached_extent;
Vivek Haldar77f41352011-05-22 21:24:16 -04002055 sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002056
2057 /* has cache valid data? */
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002058 if (cex->ec_len == 0)
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002059 goto errout;
Alex Tomasa86c6182006-10-11 01:21:03 -07002060
Akinobu Mita731eb1a2010-03-03 23:55:01 -05002061 if (in_range(block, cex->ec_block, cex->ec_len)) {
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002062 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
Eric Sandeenbba90742008-01-28 23:58:27 -05002063 ext_debug("%u cached by %u:%u:%llu\n",
2064 block,
2065 cex->ec_block, cex->ec_len, cex->ec_start);
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002066 ret = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07002067 }
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002068errout:
Vivek Haldar77f41352011-05-22 21:24:16 -04002069 if (!ret)
2070 sbi->extent_cache_misses++;
2071 else
2072 sbi->extent_cache_hits++;
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002073 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2074 return ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002075}
2076
2077/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002078 * ext4_ext_in_cache()
2079 * Checks to see if the given block is in the cache.
2080 * If it is, the cached extent is stored in the given
2081 * extent pointer.
2082 *
2083 * @inode: The files inode
2084 * @block: The block to look for in the cache
2085 * @ex: Pointer where the cached extent will be stored
2086 * if it contains block
2087 *
2088 * Return 0 if cache is invalid; 1 if the cache is valid
2089 */
2090static int
2091ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2092 struct ext4_extent *ex)
2093{
2094 struct ext4_ext_cache cex;
2095 int ret = 0;
2096
2097 if (ext4_ext_check_cache(inode, block, &cex)) {
2098 ex->ee_block = cpu_to_le32(cex.ec_block);
2099 ext4_ext_store_pblock(ex, cex.ec_start);
2100 ex->ee_len = cpu_to_le16(cex.ec_len);
2101 ret = 1;
2102 }
2103
2104 return ret;
2105}
2106
2107
2108/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002109 * ext4_ext_rm_idx:
2110 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002111 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002112static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07002113 struct ext4_ext_path *path)
2114{
Alex Tomasa86c6182006-10-11 01:21:03 -07002115 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002116 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002117
2118 /* free index block */
2119 path--;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002120 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002121 if (unlikely(path->p_hdr->eh_entries == 0)) {
2122 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2123 return -EIO;
2124 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002125 err = ext4_ext_get_access(handle, inode, path);
2126 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002127 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002128
2129 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2130 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2131 len *= sizeof(struct ext4_extent_idx);
2132 memmove(path->p_idx, path->p_idx + 1, len);
2133 }
2134
Marcin Slusarze8546d02008-04-17 10:38:59 -04002135 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002136 err = ext4_ext_dirty(handle, inode, path);
2137 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002138 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002139 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Peter Huewe7dc57612011-02-21 21:01:42 -05002140 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002141 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Alex Tomasa86c6182006-10-11 01:21:03 -07002142 return err;
2143}
2144
2145/*
Mingming Caoee12b632008-08-19 22:16:05 -04002146 * ext4_ext_calc_credits_for_single_extent:
2147 * This routine returns max. credits that needed to insert an extent
2148 * to the extent tree.
2149 * When pass the actual path, the caller should calculate credits
2150 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002151 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002152int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002153 struct ext4_ext_path *path)
2154{
Alex Tomasa86c6182006-10-11 01:21:03 -07002155 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002156 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002157 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002158
Alex Tomasa86c6182006-10-11 01:21:03 -07002159 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002160 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002161 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2162
2163 /*
2164 * There are some space in the leaf tree, no
2165 * need to account for leaf block credit
2166 *
2167 * bitmaps and block group descriptor blocks
2168 * and other metadat blocks still need to be
2169 * accounted.
2170 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002171 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002172 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002173 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002174 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002175 }
2176
Mingming Cao525f4ed2008-08-19 22:15:58 -04002177 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002178}
Alex Tomasa86c6182006-10-11 01:21:03 -07002179
Mingming Caoee12b632008-08-19 22:16:05 -04002180/*
2181 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2182 *
2183 * if nrblocks are fit in a single extent (chunk flag is 1), then
2184 * in the worse case, each tree level index/leaf need to be changed
2185 * if the tree split due to insert a new extent, then the old tree
2186 * index/leaf need to be updated too
2187 *
2188 * If the nrblocks are discontiguous, they could cause
2189 * the whole tree split more than once, but this is really rare.
2190 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002191int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04002192{
2193 int index;
2194 int depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002195
Mingming Caoee12b632008-08-19 22:16:05 -04002196 if (chunk)
2197 index = depth * 2;
2198 else
2199 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002200
Mingming Caoee12b632008-08-19 22:16:05 -04002201 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002202}
2203
2204static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2205 struct ext4_extent *ex,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002206 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002207{
Amit Aroraa2df2a62007-07-17 21:42:41 -04002208 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'oe6362602009-11-23 07:17:05 -05002209 int flags = EXT4_FREE_BLOCKS_FORGET;
Alex Tomasa86c6182006-10-11 01:21:03 -07002210
Alex Tomasc9de5602008-01-29 00:19:52 -05002211 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Theodore Ts'oe6362602009-11-23 07:17:05 -05002212 flags |= EXT4_FREE_BLOCKS_METADATA;
Alex Tomasa86c6182006-10-11 01:21:03 -07002213#ifdef EXTENTS_STATS
2214 {
2215 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002216 spin_lock(&sbi->s_ext_stats_lock);
2217 sbi->s_ext_blocks += ee_len;
2218 sbi->s_ext_extents++;
2219 if (ee_len < sbi->s_ext_min)
2220 sbi->s_ext_min = ee_len;
2221 if (ee_len > sbi->s_ext_max)
2222 sbi->s_ext_max = ee_len;
2223 if (ext_depth(inode) > sbi->s_depth_max)
2224 sbi->s_depth_max = ext_depth(inode);
2225 spin_unlock(&sbi->s_ext_stats_lock);
2226 }
2227#endif
2228 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002229 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002230 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002231 ext4_lblk_t num;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002232 ext4_fsblk_t start;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002233
Amit Aroraa2df2a62007-07-17 21:42:41 -04002234 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002235 start = ext4_ext_pblock(ex) + ee_len - num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002236 ext_debug("free last %u blocks starting %llu\n", num, start);
Peter Huewe7dc57612011-02-21 21:01:42 -05002237 ext4_free_blocks(handle, inode, NULL, start, num, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07002238 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002239 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002240 /* head removal */
2241 ext4_lblk_t num;
2242 ext4_fsblk_t start;
2243
2244 num = to - from;
2245 start = ext4_ext_pblock(ex);
2246
2247 ext_debug("free first %u blocks starting %llu\n", num, start);
2248 ext4_free_blocks(handle, inode, 0, start, num, flags);
2249
Alex Tomasa86c6182006-10-11 01:21:03 -07002250 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002251 printk(KERN_INFO "strange request: removal(2) "
2252 "%u-%u from %u:%u\n",
2253 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002254 }
2255 return 0;
2256}
2257
Allison Hendersond583fb82011-05-25 07:41:43 -04002258
2259/*
2260 * ext4_ext_rm_leaf() Removes the extents associated with the
2261 * blocks appearing between "start" and "end", and splits the extents
2262 * if "start" and "end" appear in the same extent
2263 *
2264 * @handle: The journal handle
2265 * @inode: The files inode
2266 * @path: The path to the leaf
2267 * @start: The first block to remove
2268 * @end: The last block to remove
2269 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002270static int
2271ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Allison Hendersond583fb82011-05-25 07:41:43 -04002272 struct ext4_ext_path *path, ext4_lblk_t start,
2273 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002274{
2275 int err = 0, correct_index = 0;
2276 int depth = ext_depth(inode), credits;
2277 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002278 ext4_lblk_t a, b, block;
2279 unsigned num;
2280 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002281 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002282 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002283 struct ext4_extent *ex;
Allison Hendersond583fb82011-05-25 07:41:43 -04002284 struct ext4_map_blocks map;
Alex Tomasa86c6182006-10-11 01:21:03 -07002285
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002286 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002287 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002288 if (!path[depth].p_hdr)
2289 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2290 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002291 if (unlikely(path[depth].p_hdr == NULL)) {
2292 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2293 return -EIO;
2294 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002295 /* find where to start removing */
2296 ex = EXT_LAST_EXTENT(eh);
2297
2298 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002299 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002300
2301 while (ex >= EXT_FIRST_EXTENT(eh) &&
2302 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002303
2304 if (ext4_ext_is_uninitialized(ex))
2305 uninitialized = 1;
2306 else
2307 uninitialized = 0;
2308
Mingming553f9002009-09-18 13:34:55 -04002309 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2310 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002311 path[depth].p_ext = ex;
2312
2313 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002314 b = ex_ee_block+ex_ee_len - 1 < end ?
2315 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002316
2317 ext_debug(" border %u:%u\n", a, b);
2318
Allison Hendersond583fb82011-05-25 07:41:43 -04002319 /* If this extent is beyond the end of the hole, skip it */
2320 if (end <= ex_ee_block) {
2321 ex--;
2322 ex_ee_block = le32_to_cpu(ex->ee_block);
2323 ex_ee_len = ext4_ext_get_actual_len(ex);
2324 continue;
2325 } else if (a != ex_ee_block &&
2326 b != ex_ee_block + ex_ee_len - 1) {
2327 /*
2328 * If this is a truncate, then this condition should
2329 * never happen because at least one of the end points
2330 * needs to be on the edge of the extent.
2331 */
Lukas Czernerf17722f2011-06-06 00:05:17 -04002332 if (end == EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002333 ext_debug(" bad truncate %u:%u\n",
2334 start, end);
2335 block = 0;
2336 num = 0;
2337 err = -EIO;
2338 goto out;
2339 }
2340 /*
2341 * else this is a hole punch, so the extent needs to
2342 * be split since neither edge of the hole is on the
2343 * extent edge
2344 */
2345 else{
2346 map.m_pblk = ext4_ext_pblock(ex);
2347 map.m_lblk = ex_ee_block;
2348 map.m_len = b - ex_ee_block;
2349
2350 err = ext4_split_extent(handle,
2351 inode, path, &map, 0,
2352 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2353 EXT4_GET_BLOCKS_PRE_IO);
2354
2355 if (err < 0)
2356 goto out;
2357
2358 ex_ee_len = ext4_ext_get_actual_len(ex);
2359
2360 b = ex_ee_block+ex_ee_len - 1 < end ?
2361 ex_ee_block+ex_ee_len - 1 : end;
2362
2363 /* Then remove tail of this extent */
2364 block = ex_ee_block;
2365 num = a - block;
2366 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002367 } else if (a != ex_ee_block) {
2368 /* remove tail of the extent */
2369 block = ex_ee_block;
2370 num = a - block;
2371 } else if (b != ex_ee_block + ex_ee_len - 1) {
2372 /* remove head of the extent */
Allison Hendersond583fb82011-05-25 07:41:43 -04002373 block = b;
2374 num = ex_ee_block + ex_ee_len - b;
2375
2376 /*
2377 * If this is a truncate, this condition
2378 * should never happen
2379 */
Lukas Czernerf17722f2011-06-06 00:05:17 -04002380 if (end == EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002381 ext_debug(" bad truncate %u:%u\n",
2382 start, end);
2383 err = -EIO;
2384 goto out;
2385 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002386 } else {
2387 /* remove whole extent: excellent! */
2388 block = ex_ee_block;
2389 num = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002390 if (a != ex_ee_block) {
2391 ext_debug(" bad truncate %u:%u\n",
2392 start, end);
2393 err = -EIO;
2394 goto out;
2395 }
2396
2397 if (b != ex_ee_block + ex_ee_len - 1) {
2398 ext_debug(" bad truncate %u:%u\n",
2399 start, end);
2400 err = -EIO;
2401 goto out;
2402 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002403 }
2404
Theodore Ts'o34071da2008-08-01 21:59:19 -04002405 /*
2406 * 3 for leaf, sb, and inode plus 2 (bmap and group
2407 * descriptor) for each block group; assume two block
2408 * groups plus ex_ee_len/blocks_per_block_group for
2409 * the worst case
2410 */
2411 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002412 if (ex == EXT_FIRST_EXTENT(eh)) {
2413 correct_index = 1;
2414 credits += (ext_depth(inode)) + 1;
2415 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002416 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002417
Jan Kara487caee2009-08-17 22:17:20 -04002418 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002419 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002420 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002421
2422 err = ext4_ext_get_access(handle, inode, path + depth);
2423 if (err)
2424 goto out;
2425
2426 err = ext4_remove_blocks(handle, inode, ex, a, b);
2427 if (err)
2428 goto out;
2429
2430 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002431 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002432 ext4_ext_store_pblock(ex, 0);
Allison Hendersond583fb82011-05-25 07:41:43 -04002433 } else if (block != ex_ee_block) {
2434 /*
2435 * If this was a head removal, then we need to update
2436 * the physical block since it is now at a different
2437 * location
2438 */
2439 ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
Alex Tomasa86c6182006-10-11 01:21:03 -07002440 }
2441
2442 ex->ee_block = cpu_to_le32(block);
2443 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002444 /*
2445 * Do not mark uninitialized if all the blocks in the
2446 * extent have been removed.
2447 */
2448 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002449 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002450
2451 err = ext4_ext_dirty(handle, inode, path + depth);
2452 if (err)
2453 goto out;
2454
Allison Hendersond583fb82011-05-25 07:41:43 -04002455 /*
2456 * If the extent was completely released,
2457 * we need to remove it from the leaf
2458 */
2459 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002460 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002461 /*
2462 * For hole punching, we need to scoot all the
2463 * extents up when an extent is removed so that
2464 * we dont have blank extents in the middle
2465 */
2466 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2467 sizeof(struct ext4_extent));
2468
2469 /* Now get rid of the one at the end */
2470 memset(EXT_LAST_EXTENT(eh), 0,
2471 sizeof(struct ext4_extent));
2472 }
2473 le16_add_cpu(&eh->eh_entries, -1);
2474 }
2475
Mingming Cao2ae02102006-10-11 01:21:11 -07002476 ext_debug("new extent: %u:%u:%llu\n", block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002477 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002478 ex--;
2479 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002480 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002481 }
2482
2483 if (correct_index && eh->eh_entries)
2484 err = ext4_ext_correct_indexes(handle, inode, path);
2485
2486 /* if this leaf is free, then we should
2487 * remove it from index block above */
2488 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2489 err = ext4_ext_rm_idx(handle, inode, path + depth);
2490
2491out:
2492 return err;
2493}
2494
2495/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002496 * ext4_ext_more_to_rm:
2497 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002498 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002499static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002500ext4_ext_more_to_rm(struct ext4_ext_path *path)
2501{
2502 BUG_ON(path->p_idx == NULL);
2503
2504 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2505 return 0;
2506
2507 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002508 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002509 * so we have to consider current index for truncation
2510 */
2511 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2512 return 0;
2513 return 1;
2514}
2515
Allison Hendersonc6a03712011-07-17 23:21:03 -04002516static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07002517{
2518 struct super_block *sb = inode->i_sb;
2519 int depth = ext_depth(inode);
2520 struct ext4_ext_path *path;
2521 handle_t *handle;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002522 int i, err;
Alex Tomasa86c6182006-10-11 01:21:03 -07002523
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002524 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002525
2526 /* probably first extent we're gonna free will be last in block */
2527 handle = ext4_journal_start(inode, depth + 1);
2528 if (IS_ERR(handle))
2529 return PTR_ERR(handle);
2530
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002531again:
Alex Tomasa86c6182006-10-11 01:21:03 -07002532 ext4_ext_invalidate_cache(inode);
2533
2534 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002535 * We start scanning from right side, freeing all the blocks
2536 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002537 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002538 depth = ext_depth(inode);
Josef Bacik216553c2008-04-29 22:02:02 -04002539 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07002540 if (path == NULL) {
2541 ext4_journal_stop(handle);
2542 return -ENOMEM;
2543 }
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002544 path[0].p_depth = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -07002545 path[0].p_hdr = ext_inode_hdr(inode);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04002546 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002547 err = -EIO;
2548 goto out;
2549 }
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002550 i = err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002551
2552 while (i >= 0 && err == 0) {
2553 if (i == depth) {
2554 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002555 err = ext4_ext_rm_leaf(handle, inode, path,
Allison Hendersonc6a03712011-07-17 23:21:03 -04002556 start, EXT_MAX_BLOCKS - 1);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002557 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002558 brelse(path[i].p_bh);
2559 path[i].p_bh = NULL;
2560 i--;
2561 continue;
2562 }
2563
2564 /* this is index block */
2565 if (!path[i].p_hdr) {
2566 ext_debug("initialize header\n");
2567 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002568 }
2569
Alex Tomasa86c6182006-10-11 01:21:03 -07002570 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002571 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002572 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2573 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2574 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2575 path[i].p_hdr,
2576 le16_to_cpu(path[i].p_hdr->eh_entries));
2577 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002578 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002579 path[i].p_idx--;
2580 }
2581
2582 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2583 i, EXT_FIRST_INDEX(path[i].p_hdr),
2584 path[i].p_idx);
2585 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002586 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002587 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002588 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002589 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002590 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'obf89d162010-10-27 21:30:14 -04002591 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002592 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002593 /* should we reset i_size? */
2594 err = -EIO;
2595 break;
2596 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002597 if (WARN_ON(i + 1 > depth)) {
2598 err = -EIO;
2599 break;
2600 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04002601 if (ext4_ext_check(inode, ext_block_hdr(bh),
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002602 depth - i - 1)) {
2603 err = -EIO;
2604 break;
2605 }
2606 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002607
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002608 /* save actual number of indexes since this
2609 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002610 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2611 i++;
2612 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002613 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002614 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002615 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002616 * handle must be already prepared by the
2617 * truncatei_leaf() */
2618 err = ext4_ext_rm_idx(handle, inode, path + i);
2619 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002620 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002621 brelse(path[i].p_bh);
2622 path[i].p_bh = NULL;
2623 i--;
2624 ext_debug("return to level %d\n", i);
2625 }
2626 }
2627
2628 /* TODO: flexible tree reduction should be here */
2629 if (path->p_hdr->eh_entries == 0) {
2630 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002631 * truncate to zero freed all the tree,
2632 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002633 */
2634 err = ext4_ext_get_access(handle, inode, path);
2635 if (err == 0) {
2636 ext_inode_hdr(inode)->eh_depth = 0;
2637 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002638 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002639 err = ext4_ext_dirty(handle, inode, path);
2640 }
2641 }
2642out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002643 ext4_ext_drop_refs(path);
2644 kfree(path);
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002645 if (err == -EAGAIN)
2646 goto again;
Alex Tomasa86c6182006-10-11 01:21:03 -07002647 ext4_journal_stop(handle);
2648
2649 return err;
2650}
2651
2652/*
2653 * called at mount time
2654 */
2655void ext4_ext_init(struct super_block *sb)
2656{
2657 /*
2658 * possible initialization would be here
2659 */
2660
Theodore Ts'o83982b62009-01-06 14:53:16 -05002661 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002662#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o47760042008-09-08 23:00:52 -04002663 printk(KERN_INFO "EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002664#ifdef AGGRESSIVE_TEST
2665 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07002666#endif
2667#ifdef CHECK_BINSEARCH
2668 printk(", check binsearch");
2669#endif
2670#ifdef EXTENTS_STATS
2671 printk(", stats");
2672#endif
2673 printk("\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04002674#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07002675#ifdef EXTENTS_STATS
2676 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2677 EXT4_SB(sb)->s_ext_min = 1 << 30;
2678 EXT4_SB(sb)->s_ext_max = 0;
2679#endif
2680 }
2681}
2682
2683/*
2684 * called at umount time
2685 */
2686void ext4_ext_release(struct super_block *sb)
2687{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002688 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002689 return;
2690
2691#ifdef EXTENTS_STATS
2692 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2693 struct ext4_sb_info *sbi = EXT4_SB(sb);
2694 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2695 sbi->s_ext_blocks, sbi->s_ext_extents,
2696 sbi->s_ext_blocks / sbi->s_ext_extents);
2697 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2698 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2699 }
2700#endif
2701}
2702
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002703/* FIXME!! we need to try to merge to left or right after zero-out */
2704static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2705{
Lukas Czerner24075182010-10-27 21:30:06 -04002706 ext4_fsblk_t ee_pblock;
2707 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04002708 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002709
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002710 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04002711 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002712
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002713 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04002714 if (ret > 0)
2715 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002716
Lukas Czerner24075182010-10-27 21:30:06 -04002717 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002718}
2719
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002720/*
2721 * used by extent splitting.
2722 */
2723#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2724 due to ENOSPC */
2725#define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2726#define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2727
2728/*
2729 * ext4_split_extent_at() splits an extent at given block.
2730 *
2731 * @handle: the journal handle
2732 * @inode: the file inode
2733 * @path: the path to the extent
2734 * @split: the logical block where the extent is splitted.
2735 * @split_flags: indicates if the extent could be zeroout if split fails, and
2736 * the states(init or uninit) of new extents.
2737 * @flags: flags used to insert new extent to extent tree.
2738 *
2739 *
2740 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2741 * of which are deterimined by split_flag.
2742 *
2743 * There are two cases:
2744 * a> the extent are splitted into two extent.
2745 * b> split is not needed, and just mark the extent.
2746 *
2747 * return 0 on success.
2748 */
2749static int ext4_split_extent_at(handle_t *handle,
2750 struct inode *inode,
2751 struct ext4_ext_path *path,
2752 ext4_lblk_t split,
2753 int split_flag,
2754 int flags)
2755{
2756 ext4_fsblk_t newblock;
2757 ext4_lblk_t ee_block;
2758 struct ext4_extent *ex, newex, orig_ex;
2759 struct ext4_extent *ex2 = NULL;
2760 unsigned int ee_len, depth;
2761 int err = 0;
2762
2763 ext_debug("ext4_split_extents_at: inode %lu, logical"
2764 "block %llu\n", inode->i_ino, (unsigned long long)split);
2765
2766 ext4_ext_show_leaf(inode, path);
2767
2768 depth = ext_depth(inode);
2769 ex = path[depth].p_ext;
2770 ee_block = le32_to_cpu(ex->ee_block);
2771 ee_len = ext4_ext_get_actual_len(ex);
2772 newblock = split - ee_block + ext4_ext_pblock(ex);
2773
2774 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2775
2776 err = ext4_ext_get_access(handle, inode, path + depth);
2777 if (err)
2778 goto out;
2779
2780 if (split == ee_block) {
2781 /*
2782 * case b: block @split is the block that the extent begins with
2783 * then we just change the state of the extent, and splitting
2784 * is not needed.
2785 */
2786 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2787 ext4_ext_mark_uninitialized(ex);
2788 else
2789 ext4_ext_mark_initialized(ex);
2790
2791 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2792 ext4_ext_try_to_merge(inode, path, ex);
2793
2794 err = ext4_ext_dirty(handle, inode, path + depth);
2795 goto out;
2796 }
2797
2798 /* case a */
2799 memcpy(&orig_ex, ex, sizeof(orig_ex));
2800 ex->ee_len = cpu_to_le16(split - ee_block);
2801 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2802 ext4_ext_mark_uninitialized(ex);
2803
2804 /*
2805 * path may lead to new leaf, not to original leaf any more
2806 * after ext4_ext_insert_extent() returns,
2807 */
2808 err = ext4_ext_dirty(handle, inode, path + depth);
2809 if (err)
2810 goto fix_extent_len;
2811
2812 ex2 = &newex;
2813 ex2->ee_block = cpu_to_le32(split);
2814 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2815 ext4_ext_store_pblock(ex2, newblock);
2816 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2817 ext4_ext_mark_uninitialized(ex2);
2818
2819 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2820 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2821 err = ext4_ext_zeroout(inode, &orig_ex);
2822 if (err)
2823 goto fix_extent_len;
2824 /* update the extent length and mark as initialized */
2825 ex->ee_len = cpu_to_le32(ee_len);
2826 ext4_ext_try_to_merge(inode, path, ex);
2827 err = ext4_ext_dirty(handle, inode, path + depth);
2828 goto out;
2829 } else if (err)
2830 goto fix_extent_len;
2831
2832out:
2833 ext4_ext_show_leaf(inode, path);
2834 return err;
2835
2836fix_extent_len:
2837 ex->ee_len = orig_ex.ee_len;
2838 ext4_ext_dirty(handle, inode, path + depth);
2839 return err;
2840}
2841
2842/*
2843 * ext4_split_extents() splits an extent and mark extent which is covered
2844 * by @map as split_flags indicates
2845 *
2846 * It may result in splitting the extent into multiple extents (upto three)
2847 * There are three possibilities:
2848 * a> There is no split required
2849 * b> Splits in two extents: Split is happening at either end of the extent
2850 * c> Splits in three extents: Somone is splitting in middle of the extent
2851 *
2852 */
2853static int ext4_split_extent(handle_t *handle,
2854 struct inode *inode,
2855 struct ext4_ext_path *path,
2856 struct ext4_map_blocks *map,
2857 int split_flag,
2858 int flags)
2859{
2860 ext4_lblk_t ee_block;
2861 struct ext4_extent *ex;
2862 unsigned int ee_len, depth;
2863 int err = 0;
2864 int uninitialized;
2865 int split_flag1, flags1;
2866
2867 depth = ext_depth(inode);
2868 ex = path[depth].p_ext;
2869 ee_block = le32_to_cpu(ex->ee_block);
2870 ee_len = ext4_ext_get_actual_len(ex);
2871 uninitialized = ext4_ext_is_uninitialized(ex);
2872
2873 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2874 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2875 EXT4_EXT_MAY_ZEROOUT : 0;
2876 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2877 if (uninitialized)
2878 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2879 EXT4_EXT_MARK_UNINIT2;
2880 err = ext4_split_extent_at(handle, inode, path,
2881 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04002882 if (err)
2883 goto out;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002884 }
2885
2886 ext4_ext_drop_refs(path);
2887 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2888 if (IS_ERR(path))
2889 return PTR_ERR(path);
2890
2891 if (map->m_lblk >= ee_block) {
2892 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2893 EXT4_EXT_MAY_ZEROOUT : 0;
2894 if (uninitialized)
2895 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2896 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2897 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2898 err = ext4_split_extent_at(handle, inode, path,
2899 map->m_lblk, split_flag1, flags);
2900 if (err)
2901 goto out;
2902 }
2903
2904 ext4_ext_show_leaf(inode, path);
2905out:
2906 return err ? err : map->m_len;
2907}
2908
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002909#define EXT4_EXT_ZERO_LEN 7
Amit Arora56055d32007-07-17 21:42:38 -04002910/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002911 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04002912 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002913 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04002914 * uninitialized).
2915 * There are three possibilities:
2916 * a> There is no split required: Entire extent should be initialized
2917 * b> Splits in two extents: Write is happening at either end of the extent
2918 * c> Splits in three extents: Somone is writing in middle of the extent
2919 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002920static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002921 struct inode *inode,
2922 struct ext4_map_blocks *map,
2923 struct ext4_ext_path *path)
Amit Arora56055d32007-07-17 21:42:38 -04002924{
Yongqiang Yang667eff32011-05-03 12:25:07 -04002925 struct ext4_map_blocks split_map;
2926 struct ext4_extent zero_ex;
2927 struct ext4_extent *ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002928 ext4_lblk_t ee_block, eof_block;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002929 unsigned int allocated, ee_len, depth;
Amit Arora56055d32007-07-17 21:42:38 -04002930 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04002931 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002932
2933 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2934 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002935 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002936
2937 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2938 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002939 if (eof_block < map->m_lblk + map->m_len)
2940 eof_block = map->m_lblk + map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04002941
2942 depth = ext_depth(inode);
Amit Arora56055d32007-07-17 21:42:38 -04002943 ex = path[depth].p_ext;
2944 ee_block = le32_to_cpu(ex->ee_block);
2945 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002946 allocated = ee_len - (map->m_lblk - ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002947
Yongqiang Yang667eff32011-05-03 12:25:07 -04002948 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002949 /*
2950 * It is safe to convert extent to initialized via explicit
2951 * zeroout only if extent is fully insde i_size or new_size.
2952 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04002953 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002954
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002955 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
Yongqiang Yang667eff32011-05-03 12:25:07 -04002956 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
2957 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2958 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002959 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002960 goto out;
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002961
2962 err = ext4_ext_get_access(handle, inode, path + depth);
2963 if (err)
2964 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04002965 ext4_ext_mark_initialized(ex);
2966 ext4_ext_try_to_merge(inode, path, ex);
2967 err = ext4_ext_dirty(handle, inode, path + depth);
2968 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04002969 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04002970
Amit Arora56055d32007-07-17 21:42:38 -04002971 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04002972 * four cases:
2973 * 1. split the extent into three extents.
2974 * 2. split the extent into two extents, zeroout the first half.
2975 * 3. split the extent into two extents, zeroout the second half.
2976 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04002977 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04002978 split_map.m_lblk = map->m_lblk;
2979 split_map.m_len = map->m_len;
2980
2981 if (allocated > map->m_len) {
2982 if (allocated <= EXT4_EXT_ZERO_LEN &&
2983 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2984 /* case 3 */
2985 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04002986 cpu_to_le32(map->m_lblk);
2987 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04002988 ext4_ext_store_pblock(&zero_ex,
2989 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
2990 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04002991 if (err)
2992 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04002993 split_map.m_lblk = map->m_lblk;
2994 split_map.m_len = allocated;
2995 } else if ((map->m_lblk - ee_block + map->m_len <
2996 EXT4_EXT_ZERO_LEN) &&
2997 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2998 /* case 2 */
2999 if (map->m_lblk != ee_block) {
3000 zero_ex.ee_block = ex->ee_block;
3001 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3002 ee_block);
3003 ext4_ext_store_pblock(&zero_ex,
3004 ext4_ext_pblock(ex));
3005 err = ext4_ext_zeroout(inode, &zero_ex);
3006 if (err)
3007 goto out;
3008 }
3009
Yongqiang Yang667eff32011-05-03 12:25:07 -04003010 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003011 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3012 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003013 }
3014 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003015
3016 allocated = ext4_split_extent(handle, inode, path,
3017 &split_map, split_flag, 0);
3018 if (allocated < 0)
3019 err = allocated;
3020
Amit Arora56055d32007-07-17 21:42:38 -04003021out:
3022 return err ? err : allocated;
3023}
3024
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003025/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003026 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003027 * ext4_get_blocks_dio_write() when DIO to write
3028 * to an uninitialized extent.
3029 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003030 * Writing to an uninitialized extent may result in splitting the uninitialized
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003031 * extent into multiple /initialized uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003032 * There are three possibilities:
3033 * a> There is no split required: Entire extent should be uninitialized
3034 * b> Splits in two extents: Write is happening at either end of the extent
3035 * c> Splits in three extents: Somone is writing in middle of the extent
3036 *
3037 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003038 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003039 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003040 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003041 * into three uninitialized extent(at most). After IO complete, the part
3042 * being filled will be convert to initialized by the end_io callback function
3043 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003044 *
3045 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003046 */
3047static int ext4_split_unwritten_extents(handle_t *handle,
3048 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003049 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003050 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003051 int flags)
3052{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003053 ext4_lblk_t eof_block;
3054 ext4_lblk_t ee_block;
3055 struct ext4_extent *ex;
3056 unsigned int ee_len;
3057 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003058
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003059 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3060 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003061 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003062
3063 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3064 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003065 if (eof_block < map->m_lblk + map->m_len)
3066 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003067 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003068 * It is safe to convert extent to initialized via explicit
3069 * zeroout only if extent is fully insde i_size or new_size.
3070 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003071 depth = ext_depth(inode);
3072 ex = path[depth].p_ext;
3073 ee_block = le32_to_cpu(ex->ee_block);
3074 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003075
Yongqiang Yang667eff32011-05-03 12:25:07 -04003076 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3077 split_flag |= EXT4_EXT_MARK_UNINIT2;
Mingming Cao00314622009-09-28 15:49:08 -04003078
Yongqiang Yang667eff32011-05-03 12:25:07 -04003079 flags |= EXT4_GET_BLOCKS_PRE_IO;
3080 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003081}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003082
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003083static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Mingming Cao00314622009-09-28 15:49:08 -04003084 struct inode *inode,
3085 struct ext4_ext_path *path)
3086{
3087 struct ext4_extent *ex;
Mingming Cao00314622009-09-28 15:49:08 -04003088 int depth;
3089 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003090
3091 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003092 ex = path[depth].p_ext;
3093
Yongqiang Yang197217a2011-05-03 11:45:29 -04003094 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3095 "block %llu, max_blocks %u\n", inode->i_ino,
3096 (unsigned long long)le32_to_cpu(ex->ee_block),
3097 ext4_ext_get_actual_len(ex));
3098
Mingming Cao00314622009-09-28 15:49:08 -04003099 err = ext4_ext_get_access(handle, inode, path + depth);
3100 if (err)
3101 goto out;
3102 /* first mark the extent as initialized */
3103 ext4_ext_mark_initialized(ex);
3104
Yongqiang Yang197217a2011-05-03 11:45:29 -04003105 /* note: ext4_ext_correct_indexes() isn't needed here because
3106 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003107 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04003108 ext4_ext_try_to_merge(inode, path, ex);
3109
Mingming Cao00314622009-09-28 15:49:08 -04003110 /* Mark modified extent as dirty */
3111 err = ext4_ext_dirty(handle, inode, path + depth);
3112out:
3113 ext4_ext_show_leaf(inode, path);
3114 return err;
3115}
3116
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003117static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3118 sector_t block, int count)
3119{
3120 int i;
3121 for (i = 0; i < count; i++)
3122 unmap_underlying_metadata(bdev, block + i);
3123}
3124
Theodore Ts'o58590b02010-10-27 21:23:12 -04003125/*
3126 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3127 */
3128static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003129 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003130 struct ext4_ext_path *path,
3131 unsigned int len)
3132{
3133 int i, depth;
3134 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003135 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003136
3137 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3138 return 0;
3139
3140 depth = ext_depth(inode);
3141 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003142
3143 if (unlikely(!eh->eh_entries)) {
3144 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3145 "EOFBLOCKS_FL set");
3146 return -EIO;
3147 }
3148 last_ex = EXT_LAST_EXTENT(eh);
3149 /*
3150 * We should clear the EOFBLOCKS_FL flag if we are writing the
3151 * last block in the last extent in the file. We test this by
3152 * first checking to see if the caller to
3153 * ext4_ext_get_blocks() was interested in the last block (or
3154 * a block beyond the last block) in the current extent. If
3155 * this turns out to be false, we can bail out from this
3156 * function immediately.
3157 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003158 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003159 ext4_ext_get_actual_len(last_ex))
3160 return 0;
3161 /*
3162 * If the caller does appear to be planning to write at or
3163 * beyond the end of the current extent, we then test to see
3164 * if the current extent is the last extent in the file, by
3165 * checking to make sure it was reached via the rightmost node
3166 * at each level of the tree.
3167 */
3168 for (i = depth-1; i >= 0; i--)
3169 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3170 return 0;
3171 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3172 return ext4_mark_inode_dirty(handle, inode);
3173}
3174
Mingming Cao00314622009-09-28 15:49:08 -04003175static int
3176ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003177 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003178 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003179 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003180{
3181 int ret = 0;
3182 int err = 0;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003183 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
Mingming Cao00314622009-09-28 15:49:08 -04003184
3185 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3186 "block %llu, max_blocks %u, flags %d, allocated %u",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003187 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003188 flags, allocated);
3189 ext4_ext_show_leaf(inode, path);
3190
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003191 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003192 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003193 ret = ext4_split_unwritten_extents(handle, inode, map,
3194 path, flags);
Mingming5f5249502009-11-10 10:48:04 -05003195 /*
3196 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003197 * that this IO needs to conversion to written when IO is
Mingming5f5249502009-11-10 10:48:04 -05003198 * completed
3199 */
Eric Sandeene9e3bce2011-02-12 08:17:34 -05003200 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
Theodore Ts'obd2d0212010-10-27 21:30:10 -04003201 io->flag = EXT4_IO_END_UNWRITTEN;
Eric Sandeene9e3bce2011-02-12 08:17:34 -05003202 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3203 } else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003204 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003205 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003206 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003207 goto out;
3208 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003209 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003210 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003211 ret = ext4_convert_unwritten_extents_endio(handle, inode,
Mingming Cao00314622009-09-28 15:49:08 -04003212 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003213 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003214 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003215 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3216 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003217 } else
3218 err = ret;
Mingming Cao00314622009-09-28 15:49:08 -04003219 goto out2;
3220 }
3221 /* buffered IO case */
3222 /*
3223 * repeat fallocate creation request
3224 * we already have an unwritten extent
3225 */
3226 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3227 goto map_out;
3228
3229 /* buffered READ or buffered write_begin() lookup */
3230 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3231 /*
3232 * We have blocks reserved already. We
3233 * return allocated blocks so that delalloc
3234 * won't do block reservation for us. But
3235 * the buffer head will be unmapped so that
3236 * a read from the block returns 0s.
3237 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003238 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003239 goto out1;
3240 }
3241
3242 /* buffered write, writepage time, convert*/
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003243 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003244 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003245 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003246 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3247 map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003248 if (err < 0)
3249 goto out2;
3250 }
3251
Mingming Cao00314622009-09-28 15:49:08 -04003252out:
3253 if (ret <= 0) {
3254 err = ret;
3255 goto out2;
3256 } else
3257 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003258 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003259 /*
3260 * if we allocated more blocks than requested
3261 * we need to make sure we unmap the extra block
3262 * allocated. The actual needed block will get
3263 * unmapped later when we find the buffer_head marked
3264 * new.
3265 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003266 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003267 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003268 newblock + map->m_len,
3269 allocated - map->m_len);
3270 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003271 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003272
3273 /*
3274 * If we have done fallocate with the offset that is already
3275 * delayed allocated, we would have block reservation
3276 * and quota reservation done in the delayed write path.
3277 * But fallocate would have already updated quota and block
3278 * count for this offset. So cancel these reservation
3279 */
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05003280 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003281 ext4_da_update_reserve_space(inode, allocated, 0);
3282
Mingming Cao00314622009-09-28 15:49:08 -04003283map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003284 map->m_flags |= EXT4_MAP_MAPPED;
Mingming Cao00314622009-09-28 15:49:08 -04003285out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003286 if (allocated > map->m_len)
3287 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003288 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003289 map->m_pblk = newblock;
3290 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003291out2:
3292 if (path) {
3293 ext4_ext_drop_refs(path);
3294 kfree(path);
3295 }
3296 return err ? err : allocated;
3297}
Theodore Ts'o58590b02010-10-27 21:23:12 -04003298
Mingming Cao00314622009-09-28 15:49:08 -04003299/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003300 * get_implied_cluster_alloc - check to see if the requested
3301 * allocation (in the map structure) overlaps with a cluster already
3302 * allocated in an extent.
3303 * @sbi The ext4-specific superblock structure
3304 * @map The requested lblk->pblk mapping
3305 * @ex The extent structure which might contain an implied
3306 * cluster allocation
3307 *
3308 * This function is called by ext4_ext_map_blocks() after we failed to
3309 * find blocks that were already in the inode's extent tree. Hence,
3310 * we know that the beginning of the requested region cannot overlap
3311 * the extent from the inode's extent tree. There are three cases we
3312 * want to catch. The first is this case:
3313 *
3314 * |--- cluster # N--|
3315 * |--- extent ---| |---- requested region ---|
3316 * |==========|
3317 *
3318 * The second case that we need to test for is this one:
3319 *
3320 * |--------- cluster # N ----------------|
3321 * |--- requested region --| |------- extent ----|
3322 * |=======================|
3323 *
3324 * The third case is when the requested region lies between two extents
3325 * within the same cluster:
3326 * |------------- cluster # N-------------|
3327 * |----- ex -----| |---- ex_right ----|
3328 * |------ requested region ------|
3329 * |================|
3330 *
3331 * In each of the above cases, we need to set the map->m_pblk and
3332 * map->m_len so it corresponds to the return the extent labelled as
3333 * "|====|" from cluster #N, since it is already in use for data in
3334 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3335 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3336 * as a new "allocated" block region. Otherwise, we will return 0 and
3337 * ext4_ext_map_blocks() will then allocate one or more new clusters
3338 * by calling ext4_mb_new_blocks().
3339 */
3340static int get_implied_cluster_alloc(struct ext4_sb_info *sbi,
3341 struct ext4_map_blocks *map,
3342 struct ext4_extent *ex,
3343 struct ext4_ext_path *path)
3344{
3345 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3346 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3347 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3348 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3349 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3350 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3351
3352 /* The extent passed in that we are trying to match */
3353 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3354 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3355
3356 /* The requested region passed into ext4_map_blocks() */
3357 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3358 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3359
3360 if ((rr_cluster_start == ex_cluster_end) ||
3361 (rr_cluster_start == ex_cluster_start)) {
3362 if (rr_cluster_start == ex_cluster_end)
3363 ee_start += ee_len - 1;
3364 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3365 c_offset;
3366 map->m_len = min(map->m_len,
3367 (unsigned) sbi->s_cluster_ratio - c_offset);
3368 /*
3369 * Check for and handle this case:
3370 *
3371 * |--------- cluster # N-------------|
3372 * |------- extent ----|
3373 * |--- requested region ---|
3374 * |===========|
3375 */
3376
3377 if (map->m_lblk < ee_block)
3378 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3379
3380 /*
3381 * Check for the case where there is already another allocated
3382 * block to the right of 'ex' but before the end of the cluster.
3383 *
3384 * |------------- cluster # N-------------|
3385 * |----- ex -----| |---- ex_right ----|
3386 * |------ requested region ------|
3387 * |================|
3388 */
3389 if (map->m_lblk > ee_block) {
3390 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3391 map->m_len = min(map->m_len, next - map->m_lblk);
3392 }
3393 return 1;
3394 }
3395 return 0;
3396}
3397
3398
3399/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05003400 * Block allocation/map/preallocation routine for extents based files
3401 *
3402 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003403 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05003404 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3405 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05003406 *
3407 * return > 0, number of of blocks already mapped/allocated
3408 * if create == 0 and these are pre-allocated blocks
3409 * buffer head is unmapped
3410 * otherwise blocks are mapped
3411 *
3412 * return = 0, if plain look up failed (blocks have not been allocated)
3413 * buffer head is unmapped
3414 *
3415 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003416 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003417int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3418 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07003419{
3420 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003421 struct ext4_extent newex, *ex, *ex2;
3422 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003423 ext4_fsblk_t newblock = 0;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003424 int free_on_err = 0, err = 0, depth, ret;
3425 unsigned int allocated = 0, offset = 0;
Allison Hendersone8613042011-05-25 07:41:46 -04003426 unsigned int punched_out = 0;
3427 unsigned int result = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05003428 struct ext4_allocation_request ar;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003429 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003430 ext4_lblk_t cluster_offset;
Allison Hendersone8613042011-05-25 07:41:46 -04003431 struct ext4_map_blocks punch_map;
Alex Tomasa86c6182006-10-11 01:21:03 -07003432
Mingming84fe3be2009-09-01 08:44:37 -04003433 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003434 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003435 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07003436
3437 /* check in cache */
Robin Dong015861b2011-07-17 23:27:43 -04003438 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3439 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003440 if (!newex.ee_start_lo && !newex.ee_start_hi) {
Theodore Ts'oc2177052009-05-14 00:58:52 -04003441 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04003442 /*
3443 * block isn't allocated yet and
3444 * user doesn't want to allocate it
3445 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003446 goto out2;
3447 }
3448 /* we should allocate requested block */
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003449 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07003450 /* block is already allocated */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003451 newblock = map->m_lblk
Dave Kleikamp8c55e202007-05-24 13:04:54 -04003452 - le32_to_cpu(newex.ee_block)
Theodore Ts'obf89d162010-10-27 21:30:14 -04003453 + ext4_ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003454 /* number of remaining blocks in the extent */
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05003455 allocated = ext4_ext_get_actual_len(&newex) -
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003456 (map->m_lblk - le32_to_cpu(newex.ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -07003457 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07003458 }
3459 }
3460
3461 /* find extent for this block */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003462 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07003463 if (IS_ERR(path)) {
3464 err = PTR_ERR(path);
3465 path = NULL;
3466 goto out2;
3467 }
3468
3469 depth = ext_depth(inode);
3470
3471 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003472 * consistent leaf must not be empty;
3473 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07003474 * this is why assert can't be put in ext4_ext_find_extent()
3475 */
Frank Mayhar273df552010-03-02 11:46:09 -05003476 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3477 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04003478 "lblock: %lu, depth: %d pblock %lld",
3479 (unsigned long) map->m_lblk, depth,
3480 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05003481 err = -EIO;
3482 goto out2;
3483 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003484
Avantika Mathur7e028972006-12-06 20:41:33 -08003485 ex = path[depth].p_ext;
3486 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003487 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003488 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04003489 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003490
3491 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003492 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04003493 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003494 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04003495 ee_len = ext4_ext_get_actual_len(ex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003496 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003497 if (in_range(map->m_lblk, ee_block, ee_len)) {
3498 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003499 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003500 allocated = ee_len - (map->m_lblk - ee_block);
3501 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3502 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04003503
Allison Hendersone8613042011-05-25 07:41:46 -04003504 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3505 /*
3506 * Do not put uninitialized extent
3507 * in the cache
3508 */
3509 if (!ext4_ext_is_uninitialized(ex)) {
3510 ext4_ext_put_in_cache(inode, ee_block,
3511 ee_len, ee_start);
3512 goto out;
3513 }
3514 ret = ext4_ext_handle_uninitialized_extents(
3515 handle, inode, map, path, flags,
3516 allocated, newblock);
3517 return ret;
Amit Arora56055d32007-07-17 21:42:38 -04003518 }
Allison Hendersone8613042011-05-25 07:41:46 -04003519
3520 /*
3521 * Punch out the map length, but only to the
3522 * end of the extent
3523 */
3524 punched_out = allocated < map->m_len ?
3525 allocated : map->m_len;
3526
3527 /*
3528 * Sense extents need to be converted to
3529 * uninitialized, they must fit in an
3530 * uninitialized extent
3531 */
3532 if (punched_out > EXT_UNINIT_MAX_LEN)
3533 punched_out = EXT_UNINIT_MAX_LEN;
3534
3535 punch_map.m_lblk = map->m_lblk;
3536 punch_map.m_pblk = newblock;
3537 punch_map.m_len = punched_out;
3538 punch_map.m_flags = 0;
3539
3540 /* Check to see if the extent needs to be split */
3541 if (punch_map.m_len != ee_len ||
3542 punch_map.m_lblk != ee_block) {
3543
3544 ret = ext4_split_extent(handle, inode,
3545 path, &punch_map, 0,
3546 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3547 EXT4_GET_BLOCKS_PRE_IO);
3548
3549 if (ret < 0) {
3550 err = ret;
3551 goto out2;
3552 }
3553 /*
3554 * find extent for the block at
3555 * the start of the hole
3556 */
3557 ext4_ext_drop_refs(path);
3558 kfree(path);
3559
3560 path = ext4_ext_find_extent(inode,
3561 map->m_lblk, NULL);
3562 if (IS_ERR(path)) {
3563 err = PTR_ERR(path);
3564 path = NULL;
3565 goto out2;
3566 }
3567
3568 depth = ext_depth(inode);
3569 ex = path[depth].p_ext;
3570 ee_len = ext4_ext_get_actual_len(ex);
3571 ee_block = le32_to_cpu(ex->ee_block);
3572 ee_start = ext4_ext_pblock(ex);
3573
3574 }
3575
3576 ext4_ext_mark_uninitialized(ex);
3577
Allison Hendersonf7d0d372011-07-17 23:17:02 -04003578 ext4_ext_invalidate_cache(inode);
3579
3580 err = ext4_ext_rm_leaf(handle, inode, path,
3581 map->m_lblk, map->m_lblk + punched_out);
3582
3583 if (!err && path->p_hdr->eh_entries == 0) {
3584 /*
3585 * Punch hole freed all of this sub tree,
3586 * so we need to correct eh_depth
3587 */
3588 err = ext4_ext_get_access(handle, inode, path);
3589 if (err == 0) {
3590 ext_inode_hdr(inode)->eh_depth = 0;
3591 ext_inode_hdr(inode)->eh_max =
3592 cpu_to_le16(ext4_ext_space_root(
3593 inode, 0));
3594
3595 err = ext4_ext_dirty(
3596 handle, inode, path);
3597 }
3598 }
Allison Hendersone8613042011-05-25 07:41:46 -04003599
3600 goto out2;
Alex Tomasa86c6182006-10-11 01:21:03 -07003601 }
3602 }
3603
3604 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003605 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07003606 * we couldn't try to create block if create flag is zero
3607 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04003608 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04003609 /*
3610 * put just found gap into cache to speed up
3611 * subsequent requests
3612 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003613 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07003614 goto out2;
3615 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003616
Alex Tomasa86c6182006-10-11 01:21:03 -07003617 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04003618 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07003619 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003620 newex.ee_block = cpu_to_le32(map->m_lblk);
3621 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3622
3623 /*
3624 * If we are doing bigalloc, check to see if the extent returned
3625 * by ext4_ext_find_extent() implies a cluster we can use.
3626 */
3627 if (cluster_offset && ex &&
3628 get_implied_cluster_alloc(sbi, map, ex, path)) {
3629 ar.len = allocated = map->m_len;
3630 newblock = map->m_pblk;
3631 goto got_allocated_blocks;
3632 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003633
Alex Tomasc9de5602008-01-29 00:19:52 -05003634 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003635 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05003636 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3637 if (err)
3638 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003639 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003640 ex2 = NULL;
3641 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05003642 if (err)
3643 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04003644
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003645 /* Check if the extent after searching to the right implies a
3646 * cluster we can use. */
3647 if ((sbi->s_cluster_ratio > 1) && ex2 &&
3648 get_implied_cluster_alloc(sbi, map, ex2, path)) {
3649 ar.len = allocated = map->m_len;
3650 newblock = map->m_pblk;
3651 goto got_allocated_blocks;
3652 }
3653
Amit Arora749269f2007-07-18 09:02:56 -04003654 /*
3655 * See if request is beyond maximum number of blocks we can have in
3656 * a single extent. For an initialized extent this limit is
3657 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3658 * EXT_UNINIT_MAX_LEN.
3659 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003660 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04003661 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003662 map->m_len = EXT_INIT_MAX_LEN;
3663 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04003664 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003665 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04003666
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003667 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003668 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003669 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04003670 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05003671 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04003672 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003673 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05003674
3675 /* allocate new block */
3676 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003677 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3678 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003679 /*
3680 * We calculate the offset from the beginning of the cluster
3681 * for the logical block number, since when we allocate a
3682 * physical cluster, the physical block should start at the
3683 * same offset from the beginning of the cluster. This is
3684 * needed so that future calls to get_implied_cluster_alloc()
3685 * work correctly.
3686 */
3687 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3688 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3689 ar.goal -= offset;
3690 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05003691 if (S_ISREG(inode->i_mode))
3692 ar.flags = EXT4_MB_HINT_DATA;
3693 else
3694 /* disable in-core preallocation for non-regular files */
3695 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04003696 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
3697 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05003698 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07003699 if (!newblock)
3700 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04003701 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05003702 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003703 free_on_err = 1;
3704 ar.len = EXT4_C2B(sbi, ar.len) - offset;
3705 if (ar.len > allocated)
3706 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07003707
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003708got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07003709 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003710 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05003711 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003712 /* Mark uninitialized */
3713 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04003714 ext4_ext_mark_uninitialized(&newex);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003715 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05003716 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003717 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05003718 * here we flag the IO that really needs the conversion.
Mingming5f5249502009-11-10 10:48:04 -05003719 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003720 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003721 */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003722 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Eric Sandeene9e3bce2011-02-12 08:17:34 -05003723 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
Theodore Ts'obd2d0212010-10-27 21:30:10 -04003724 io->flag = EXT4_IO_END_UNWRITTEN;
Eric Sandeene9e3bce2011-02-12 08:17:34 -05003725 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3726 } else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003727 ext4_set_inode_state(inode,
3728 EXT4_STATE_DIO_UNWRITTEN);
Mingming5f5249502009-11-10 10:48:04 -05003729 }
Jiaying Zhang744692d2010-03-04 16:14:02 -05003730 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003731 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003732 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003733
Eric Sandeend002ebf2011-01-10 13:03:35 -05003734 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04003735 if (!err)
3736 err = ext4_ext_insert_extent(handle, inode, path,
3737 &newex, flags);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003738 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04003739 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3740 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04003741 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05003742 /* not a good idea to call discard here directly,
3743 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04003744 ext4_discard_preallocations(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05003745 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
Maxim Patlasov7132de72011-07-10 19:37:48 -04003746 ext4_ext_get_actual_len(&newex), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07003747 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04003748 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003749
Alex Tomasa86c6182006-10-11 01:21:03 -07003750 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04003751 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05003752 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003753 if (allocated > map->m_len)
3754 allocated = map->m_len;
3755 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07003756
Jan Karab436b9b2009-12-08 23:51:10 -05003757 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003758 * Update reserved blocks/metadata blocks after successful
3759 * block allocation which had been deferred till now.
3760 */
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05003761 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003762 ext4_da_update_reserve_space(inode, allocated, 1);
3763
3764 /*
Jan Karab436b9b2009-12-08 23:51:10 -05003765 * Cache the extent and update transaction to commit on fdatasync only
3766 * when it is _not_ an uninitialized extent.
3767 */
3768 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003769 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
Jan Karab436b9b2009-12-08 23:51:10 -05003770 ext4_update_inode_fsync_trans(handle, inode, 1);
3771 } else
3772 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07003773out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003774 if (allocated > map->m_len)
3775 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07003776 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003777 map->m_flags |= EXT4_MAP_MAPPED;
3778 map->m_pblk = newblock;
3779 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07003780out2:
3781 if (path) {
3782 ext4_ext_drop_refs(path);
3783 kfree(path);
3784 }
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003785 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3786 newblock, map->m_len, err ? err : allocated);
Allison Hendersone8613042011-05-25 07:41:46 -04003787
3788 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3789 punched_out : allocated;
3790
3791 return err ? err : result;
Alex Tomasa86c6182006-10-11 01:21:03 -07003792}
3793
Jan Karacf108bc2008-07-11 19:27:31 -04003794void ext4_ext_truncate(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07003795{
3796 struct address_space *mapping = inode->i_mapping;
3797 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003798 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07003799 handle_t *handle;
Allison Henderson189e8682011-09-06 21:49:44 -04003800 loff_t page_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07003801 int err = 0;
3802
3803 /*
Jiaying Zhang3889fd52011-01-10 12:47:05 -05003804 * finish any pending end_io work so we won't run the risk of
3805 * converting any truncated blocks to initialized later
3806 */
3807 ext4_flush_completed_IO(inode);
3808
3809 /*
Alex Tomasa86c6182006-10-11 01:21:03 -07003810 * probably first extent we're gonna free will be last in block
3811 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04003812 err = ext4_writepage_trans_blocks(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07003813 handle = ext4_journal_start(inode, err);
Jan Karacf108bc2008-07-11 19:27:31 -04003814 if (IS_ERR(handle))
Alex Tomasa86c6182006-10-11 01:21:03 -07003815 return;
Alex Tomasa86c6182006-10-11 01:21:03 -07003816
Allison Henderson189e8682011-09-06 21:49:44 -04003817 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3818 page_len = PAGE_CACHE_SIZE -
3819 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3820
3821 err = ext4_discard_partial_page_buffers(handle,
3822 mapping, inode->i_size, page_len, 0);
3823
3824 if (err)
3825 goto out_stop;
3826 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003827
Jan Kara9ddfc3d2008-07-11 19:27:31 -04003828 if (ext4_orphan_add(handle, inode))
3829 goto out_stop;
3830
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05003831 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07003832 ext4_ext_invalidate_cache(inode);
3833
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04003834 ext4_discard_preallocations(inode);
Alex Tomasc9de5602008-01-29 00:19:52 -05003835
Alex Tomasa86c6182006-10-11 01:21:03 -07003836 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003837 * TODO: optimization is possible here.
3838 * Probably we need not scan at all,
3839 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07003840 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003841
3842 /* we have to know where to truncate from in crash case */
3843 EXT4_I(inode)->i_disksize = inode->i_size;
3844 ext4_mark_inode_dirty(handle, inode);
3845
3846 last_block = (inode->i_size + sb->s_blocksize - 1)
3847 >> EXT4_BLOCK_SIZE_BITS(sb);
Allison Hendersonc6a03712011-07-17 23:21:03 -04003848 err = ext4_ext_remove_space(inode, last_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07003849
3850 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04003851 * transaction synchronous.
3852 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003853 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05003854 ext4_handle_sync(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07003855
Jan Kara9ddfc3d2008-07-11 19:27:31 -04003856 up_write(&EXT4_I(inode)->i_data_sem);
Eric Gouriouf6d2f6b2011-05-22 21:33:00 -04003857
3858out_stop:
Alex Tomasa86c6182006-10-11 01:21:03 -07003859 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003860 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07003861 * then we need to clear up the orphan record which we created above.
3862 * However, if this was a real unlink then we were called by
3863 * ext4_delete_inode(), and we allow that function to clean up the
3864 * orphan info for us.
3865 */
3866 if (inode->i_nlink)
3867 ext4_orphan_del(handle, inode);
3868
Solofo Ramangalahyef737722008-04-29 22:00:41 -04003869 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3870 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07003871 ext4_journal_stop(handle);
3872}
3873
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003874static void ext4_falloc_update_inode(struct inode *inode,
3875 int mode, loff_t new_size, int update_ctime)
3876{
3877 struct timespec now;
3878
3879 if (update_ctime) {
3880 now = current_fs_time(inode->i_sb);
3881 if (!timespec_equal(&inode->i_ctime, &now))
3882 inode->i_ctime = now;
3883 }
3884 /*
3885 * Update only when preallocation was requested beyond
3886 * the file size.
3887 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04003888 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3889 if (new_size > i_size_read(inode))
3890 i_size_write(inode, new_size);
3891 if (new_size > EXT4_I(inode)->i_disksize)
3892 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003893 } else {
3894 /*
3895 * Mark that we allocate beyond EOF so the subsequent truncate
3896 * can proceed even if the new size is the same as i_size.
3897 */
3898 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003899 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003900 }
3901
3902}
3903
Amit Aroraa2df2a62007-07-17 21:42:41 -04003904/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01003905 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04003906 * operation, which gets called from sys_fallocate system call.
3907 * For block-mapped files, posix_fallocate should fall back to the method
3908 * of writing zeroes to the required new blocks (the same behavior which is
3909 * expected for file systems which do not support fallocate() system call).
3910 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01003911long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04003912{
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01003913 struct inode *inode = file->f_path.dentry->d_inode;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003914 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003915 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05003916 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003917 int ret = 0;
3918 int ret2 = 0;
3919 int retries = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003920 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003921 unsigned int credits, blkbits = inode->i_blkbits;
3922
3923 /*
3924 * currently supporting (pre)allocate mode for extent-based
3925 * files _only_
3926 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003927 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Amit Aroraa2df2a62007-07-17 21:42:41 -04003928 return -EOPNOTSUPP;
3929
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003930 /* Return error if mode is not supported */
3931 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3932 return -EOPNOTSUPP;
3933
3934 if (mode & FALLOC_FL_PUNCH_HOLE)
3935 return ext4_punch_hole(file, offset, len);
3936
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003937 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003938 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003939 /*
3940 * We can't just convert len to max_blocks because
3941 * If blocksize = 4096 offset = 3072 and len = 2048
3942 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04003943 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003944 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003945 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04003946 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04003947 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04003948 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05003949 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04003950 ret = inode_newsize_ok(inode, (len + offset));
3951 if (ret) {
3952 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003953 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04003954 return ret;
3955 }
Amit Aroraa2df2a62007-07-17 21:42:41 -04003956retry:
3957 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003958 map.m_lblk = map.m_lblk + ret;
3959 map.m_len = max_blocks = max_blocks - ret;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003960 handle = ext4_journal_start(inode, credits);
3961 if (IS_ERR(handle)) {
3962 ret = PTR_ERR(handle);
3963 break;
3964 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003965 ret = ext4_map_blocks(handle, inode, &map,
Vivek Haldar556b27a2011-05-25 07:41:54 -04003966 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
3967 EXT4_GET_BLOCKS_NO_NORMALIZE);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05003968 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05003969#ifdef EXT4FS_DEBUG
3970 WARN_ON(ret <= 0);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003971 printk(KERN_ERR "%s: ext4_ext_map_blocks "
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05003972 "returned error inode#%lu, block=%u, "
Thadeu Lima de Souza Cascardo9fd97842009-01-26 19:26:26 -05003973 "max_blocks=%u", __func__,
Kazuya Mioa6371b62010-10-27 21:30:15 -04003974 inode->i_ino, map.m_lblk, max_blocks);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05003975#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04003976 ext4_mark_inode_dirty(handle, inode);
3977 ret2 = ext4_journal_stop(handle);
3978 break;
3979 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003980 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003981 blkbits) >> blkbits))
3982 new_size = offset + len;
3983 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04003984 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003985
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003986 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003987 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04003988 ext4_mark_inode_dirty(handle, inode);
3989 ret2 = ext4_journal_stop(handle);
3990 if (ret2)
3991 break;
3992 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003993 if (ret == -ENOSPC &&
3994 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3995 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003996 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003997 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05003998 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003999 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4000 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004001 return ret > 0 ? ret2 : ret;
4002}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004003
4004/*
Mingming Cao00314622009-09-28 15:49:08 -04004005 * This function convert a range of blocks to written extents
4006 * The caller of this function will pass the start offset and the size.
4007 * all unwritten extents within this range will be converted to
4008 * written extents.
4009 *
4010 * This function is called from the direct IO end io call back
4011 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004012 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004013 */
4014int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
Eric Sandeena1de02d2010-02-04 23:58:38 -05004015 ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004016{
4017 handle_t *handle;
Mingming Cao00314622009-09-28 15:49:08 -04004018 unsigned int max_blocks;
4019 int ret = 0;
4020 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004021 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004022 unsigned int credits, blkbits = inode->i_blkbits;
4023
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004024 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004025 /*
4026 * We can't just convert len to max_blocks because
4027 * If blocksize = 4096 offset = 3072 and len = 2048
4028 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004029 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4030 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004031 /*
4032 * credits to insert 1 extent into extent tree
4033 */
4034 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4035 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004036 map.m_lblk += ret;
4037 map.m_len = (max_blocks -= ret);
Mingming Cao00314622009-09-28 15:49:08 -04004038 handle = ext4_journal_start(inode, credits);
4039 if (IS_ERR(handle)) {
4040 ret = PTR_ERR(handle);
4041 break;
4042 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004043 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004044 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Mingming Cao00314622009-09-28 15:49:08 -04004045 if (ret <= 0) {
4046 WARN_ON(ret <= 0);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004047 printk(KERN_ERR "%s: ext4_ext_map_blocks "
Mingming Cao00314622009-09-28 15:49:08 -04004048 "returned error inode#%lu, block=%u, "
4049 "max_blocks=%u", __func__,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004050 inode->i_ino, map.m_lblk, map.m_len);
Mingming Cao00314622009-09-28 15:49:08 -04004051 }
4052 ext4_mark_inode_dirty(handle, inode);
4053 ret2 = ext4_journal_stop(handle);
4054 if (ret <= 0 || ret2 )
4055 break;
4056 }
4057 return ret > 0 ? ret2 : ret;
4058}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004059
Mingming Cao00314622009-09-28 15:49:08 -04004060/*
Eric Sandeen6873fa02008-10-07 00:46:36 -04004061 * Callback function called for each extent to gather FIEMAP information.
4062 */
Lukas Czernerc03f8aa2011-06-06 00:06:52 -04004063static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
Eric Sandeen6873fa02008-10-07 00:46:36 -04004064 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4065 void *data)
4066{
Eric Sandeen6873fa02008-10-07 00:46:36 -04004067 __u64 logical;
4068 __u64 physical;
4069 __u64 length;
4070 __u32 flags = 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004071 int ret = 0;
4072 struct fiemap_extent_info *fieinfo = data;
4073 unsigned char blksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004074
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004075 blksize_bits = inode->i_sb->s_blocksize_bits;
4076 logical = (__u64)newex->ec_block << blksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004077
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05004078 if (newex->ec_start == 0) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004079 /*
4080 * No extent in extent-tree contains block @newex->ec_start,
4081 * then the block may stay in 1)a hole or 2)delayed-extent.
4082 *
4083 * Holes or delayed-extents are processed as follows.
4084 * 1. lookup dirty pages with specified range in pagecache.
4085 * If no page is got, then there is no delayed-extent and
4086 * return with EXT_CONTINUE.
4087 * 2. find the 1st mapped buffer,
4088 * 3. check if the mapped buffer is both in the request range
4089 * and a delayed buffer. If not, there is no delayed-extent,
4090 * then return.
4091 * 4. a delayed-extent is found, the extent will be collected.
4092 */
4093 ext4_lblk_t end = 0;
4094 pgoff_t last_offset;
4095 pgoff_t offset;
4096 pgoff_t index;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004097 pgoff_t start_index = 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004098 struct page **pages = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004099 struct buffer_head *bh = NULL;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004100 struct buffer_head *head = NULL;
4101 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4102
4103 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4104 if (pages == NULL)
4105 return -ENOMEM;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004106
4107 offset = logical >> PAGE_SHIFT;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004108repeat:
4109 last_offset = offset;
4110 head = NULL;
4111 ret = find_get_pages_tag(inode->i_mapping, &offset,
4112 PAGECACHE_TAG_DIRTY, nr_pages, pages);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004113
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004114 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4115 /* First time, try to find a mapped buffer. */
4116 if (ret == 0) {
4117out:
4118 for (index = 0; index < ret; index++)
4119 page_cache_release(pages[index]);
4120 /* just a hole. */
4121 kfree(pages);
4122 return EXT_CONTINUE;
4123 }
Yongqiang Yangb2213492011-05-24 11:36:58 -04004124 index = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004125
Yongqiang Yangb2213492011-05-24 11:36:58 -04004126next_page:
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004127 /* Try to find the 1st mapped buffer. */
Yongqiang Yangb2213492011-05-24 11:36:58 -04004128 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004129 blksize_bits;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004130 if (!page_has_buffers(pages[index]))
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004131 goto out;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004132 head = page_buffers(pages[index]);
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004133 if (!head)
4134 goto out;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004135
Yongqiang Yangb2213492011-05-24 11:36:58 -04004136 index++;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004137 bh = head;
4138 do {
Yongqiang Yangb2213492011-05-24 11:36:58 -04004139 if (end >= newex->ec_block +
4140 newex->ec_len)
4141 /* The buffer is out of
4142 * the request range.
4143 */
4144 goto out;
4145
4146 if (buffer_mapped(bh) &&
4147 end >= newex->ec_block) {
4148 start_index = index - 1;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004149 /* get the 1st mapped buffer. */
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004150 goto found_mapped_buffer;
4151 }
Yongqiang Yangb2213492011-05-24 11:36:58 -04004152
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004153 bh = bh->b_this_page;
4154 end++;
4155 } while (bh != head);
4156
Yongqiang Yangb2213492011-05-24 11:36:58 -04004157 /* No mapped buffer in the range found in this page,
4158 * We need to look up next page.
4159 */
4160 if (index >= ret) {
4161 /* There is no page left, but we need to limit
4162 * newex->ec_len.
4163 */
4164 newex->ec_len = end - newex->ec_block;
4165 goto out;
4166 }
4167 goto next_page;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004168 } else {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004169 /*Find contiguous delayed buffers. */
4170 if (ret > 0 && pages[0]->index == last_offset)
4171 head = page_buffers(pages[0]);
4172 bh = head;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004173 index = 1;
4174 start_index = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004175 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004176
4177found_mapped_buffer:
4178 if (bh != NULL && buffer_delay(bh)) {
4179 /* 1st or contiguous delayed buffer found. */
4180 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4181 /*
4182 * 1st delayed buffer found, record
4183 * the start of extent.
4184 */
4185 flags |= FIEMAP_EXTENT_DELALLOC;
4186 newex->ec_block = end;
4187 logical = (__u64)end << blksize_bits;
4188 }
4189 /* Find contiguous delayed buffers. */
4190 do {
4191 if (!buffer_delay(bh))
4192 goto found_delayed_extent;
4193 bh = bh->b_this_page;
4194 end++;
4195 } while (bh != head);
4196
Yongqiang Yangb2213492011-05-24 11:36:58 -04004197 for (; index < ret; index++) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004198 if (!page_has_buffers(pages[index])) {
4199 bh = NULL;
4200 break;
4201 }
4202 head = page_buffers(pages[index]);
4203 if (!head) {
4204 bh = NULL;
4205 break;
4206 }
Yongqiang Yangb2213492011-05-24 11:36:58 -04004207
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004208 if (pages[index]->index !=
Yongqiang Yangb2213492011-05-24 11:36:58 -04004209 pages[start_index]->index + index
4210 - start_index) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004211 /* Blocks are not contiguous. */
4212 bh = NULL;
4213 break;
4214 }
4215 bh = head;
4216 do {
4217 if (!buffer_delay(bh))
4218 /* Delayed-extent ends. */
4219 goto found_delayed_extent;
4220 bh = bh->b_this_page;
4221 end++;
4222 } while (bh != head);
4223 }
4224 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4225 /* a hole found. */
4226 goto out;
4227
4228found_delayed_extent:
4229 newex->ec_len = min(end - newex->ec_block,
4230 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4231 if (ret == nr_pages && bh != NULL &&
4232 newex->ec_len < EXT_INIT_MAX_LEN &&
4233 buffer_delay(bh)) {
4234 /* Have not collected an extent and continue. */
4235 for (index = 0; index < ret; index++)
4236 page_cache_release(pages[index]);
4237 goto repeat;
4238 }
4239
4240 for (index = 0; index < ret; index++)
4241 page_cache_release(pages[index]);
4242 kfree(pages);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004243 }
4244
4245 physical = (__u64)newex->ec_start << blksize_bits;
4246 length = (__u64)newex->ec_len << blksize_bits;
4247
4248 if (ex && ext4_ext_is_uninitialized(ex))
4249 flags |= FIEMAP_EXTENT_UNWRITTEN;
4250
Lukas Czernerc03f8aa2011-06-06 00:06:52 -04004251 if (next == EXT_MAX_BLOCKS)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004252 flags |= FIEMAP_EXTENT_LAST;
4253
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004254 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
Eric Sandeen6873fa02008-10-07 00:46:36 -04004255 length, flags);
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004256 if (ret < 0)
4257 return ret;
4258 if (ret == 1)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004259 return EXT_BREAK;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004260 return EXT_CONTINUE;
4261}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004262/* fiemap flags we can handle specified here */
4263#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4264
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004265static int ext4_xattr_fiemap(struct inode *inode,
4266 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004267{
4268 __u64 physical = 0;
4269 __u64 length;
4270 __u32 flags = FIEMAP_EXTENT_LAST;
4271 int blockbits = inode->i_sb->s_blocksize_bits;
4272 int error = 0;
4273
4274 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004275 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004276 struct ext4_iloc iloc;
4277 int offset; /* offset of xattr in inode */
4278
4279 error = ext4_get_inode_loc(inode, &iloc);
4280 if (error)
4281 return error;
4282 physical = iloc.bh->b_blocknr << blockbits;
4283 offset = EXT4_GOOD_OLD_INODE_SIZE +
4284 EXT4_I(inode)->i_extra_isize;
4285 physical += offset;
4286 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4287 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004288 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004289 } else { /* external block */
4290 physical = EXT4_I(inode)->i_file_acl << blockbits;
4291 length = inode->i_sb->s_blocksize;
4292 }
4293
4294 if (physical)
4295 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4296 length, flags);
4297 return (error < 0 ? error : 0);
4298}
4299
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004300/*
4301 * ext4_ext_punch_hole
4302 *
4303 * Punches a hole of "length" bytes in a file starting
4304 * at byte "offset"
4305 *
4306 * @inode: The inode of the file to punch a hole in
4307 * @offset: The starting byte offset of the hole
4308 * @length: The length of the hole
4309 *
4310 * Returns the number of blocks removed or negative on err
4311 */
4312int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4313{
4314 struct inode *inode = file->f_path.dentry->d_inode;
4315 struct super_block *sb = inode->i_sb;
4316 struct ext4_ext_cache cache_ex;
4317 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4318 struct address_space *mapping = inode->i_mapping;
4319 struct ext4_map_blocks map;
4320 handle_t *handle;
Allison Hendersonba062082011-09-03 11:55:59 -04004321 loff_t first_page, last_page, page_len;
4322 loff_t first_page_offset, last_page_offset;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004323 int ret, credits, blocks_released, err = 0;
4324
Allison Henderson2be47512011-09-03 11:56:52 -04004325 /* No need to punch hole beyond i_size */
4326 if (offset >= inode->i_size)
4327 return 0;
4328
4329 /*
4330 * If the hole extends beyond i_size, set the hole
4331 * to end after the page that contains i_size
4332 */
4333 if (offset + length > inode->i_size) {
4334 length = inode->i_size +
4335 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4336 offset;
4337 }
4338
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004339 first_block = (offset + sb->s_blocksize - 1) >>
4340 EXT4_BLOCK_SIZE_BITS(sb);
4341 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4342
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004343 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4344 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4345
4346 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4347 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4348
4349 /*
4350 * Write out all dirty pages to avoid race conditions
4351 * Then release them.
4352 */
4353 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4354 err = filemap_write_and_wait_range(mapping,
Allison Henderson2be47512011-09-03 11:56:52 -04004355 offset, offset + length - 1);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004356
Allison Henderson2be47512011-09-03 11:56:52 -04004357 if (err)
4358 return err;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004359 }
4360
4361 /* Now release the pages */
4362 if (last_page_offset > first_page_offset) {
4363 truncate_inode_pages_range(mapping, first_page_offset,
4364 last_page_offset-1);
4365 }
4366
4367 /* finish any pending end_io work */
4368 ext4_flush_completed_IO(inode);
4369
4370 credits = ext4_writepage_trans_blocks(inode);
4371 handle = ext4_journal_start(inode, credits);
4372 if (IS_ERR(handle))
4373 return PTR_ERR(handle);
4374
4375 err = ext4_orphan_add(handle, inode);
4376 if (err)
4377 goto out;
4378
4379 /*
Allison Hendersonba062082011-09-03 11:55:59 -04004380 * Now we need to zero out the non-page-aligned data in the
4381 * pages at the start and tail of the hole, and unmap the buffer
4382 * heads for the block aligned regions of the page that were
4383 * completely zeroed.
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004384 */
Allison Hendersonba062082011-09-03 11:55:59 -04004385 if (first_page > last_page) {
4386 /*
4387 * If the file space being truncated is contained within a page
4388 * just zero out and unmap the middle of that page
4389 */
4390 err = ext4_discard_partial_page_buffers(handle,
4391 mapping, offset, length, 0);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004392
Allison Hendersonba062082011-09-03 11:55:59 -04004393 if (err)
4394 goto out;
4395 } else {
4396 /*
4397 * zero out and unmap the partial page that contains
4398 * the start of the hole
4399 */
4400 page_len = first_page_offset - offset;
4401 if (page_len > 0) {
4402 err = ext4_discard_partial_page_buffers(handle, mapping,
4403 offset, page_len, 0);
4404 if (err)
4405 goto out;
4406 }
4407
4408 /*
4409 * zero out and unmap the partial page that contains
4410 * the end of the hole
4411 */
4412 page_len = offset + length - last_page_offset;
4413 if (page_len > 0) {
4414 err = ext4_discard_partial_page_buffers(handle, mapping,
4415 last_page_offset, page_len, 0);
4416 if (err)
4417 goto out;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004418 }
4419 }
4420
Allison Henderson2be47512011-09-03 11:56:52 -04004421
4422 /*
4423 * If i_size is contained in the last page, we need to
4424 * unmap and zero the partial page after i_size
4425 */
4426 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4427 inode->i_size % PAGE_CACHE_SIZE != 0) {
4428
4429 page_len = PAGE_CACHE_SIZE -
4430 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4431
4432 if (page_len > 0) {
4433 err = ext4_discard_partial_page_buffers(handle,
4434 mapping, inode->i_size, page_len, 0);
4435
4436 if (err)
4437 goto out;
4438 }
4439 }
4440
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004441 /* If there are no blocks to remove, return now */
4442 if (first_block >= last_block)
4443 goto out;
4444
4445 down_write(&EXT4_I(inode)->i_data_sem);
4446 ext4_ext_invalidate_cache(inode);
4447 ext4_discard_preallocations(inode);
4448
4449 /*
4450 * Loop over all the blocks and identify blocks
4451 * that need to be punched out
4452 */
4453 iblock = first_block;
4454 blocks_released = 0;
4455 while (iblock < last_block) {
4456 max_blocks = last_block - iblock;
4457 num_blocks = 1;
4458 memset(&map, 0, sizeof(map));
4459 map.m_lblk = iblock;
4460 map.m_len = max_blocks;
4461 ret = ext4_ext_map_blocks(handle, inode, &map,
4462 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4463
4464 if (ret > 0) {
4465 blocks_released += ret;
4466 num_blocks = ret;
4467 } else if (ret == 0) {
4468 /*
4469 * If map blocks could not find the block,
4470 * then it is in a hole. If the hole was
4471 * not already cached, then map blocks should
4472 * put it in the cache. So we can get the hole
4473 * out of the cache
4474 */
4475 memset(&cache_ex, 0, sizeof(cache_ex));
4476 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4477 !cache_ex.ec_start) {
4478
4479 /* The hole is cached */
4480 num_blocks = cache_ex.ec_block +
4481 cache_ex.ec_len - iblock;
4482
4483 } else {
4484 /* The block could not be identified */
4485 err = -EIO;
4486 break;
4487 }
4488 } else {
4489 /* Map blocks error */
4490 err = ret;
4491 break;
4492 }
4493
4494 if (num_blocks == 0) {
4495 /* This condition should never happen */
4496 ext_debug("Block lookup failed");
4497 err = -EIO;
4498 break;
4499 }
4500
4501 iblock += num_blocks;
4502 }
4503
4504 if (blocks_released > 0) {
4505 ext4_ext_invalidate_cache(inode);
4506 ext4_discard_preallocations(inode);
4507 }
4508
4509 if (IS_SYNC(inode))
4510 ext4_handle_sync(handle);
4511
4512 up_write(&EXT4_I(inode)->i_data_sem);
4513
4514out:
4515 ext4_orphan_del(handle, inode);
4516 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4517 ext4_mark_inode_dirty(handle, inode);
4518 ext4_journal_stop(handle);
4519 return err;
4520}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004521int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4522 __u64 start, __u64 len)
4523{
4524 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004525 int error = 0;
4526
4527 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004528 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004529 return generic_block_fiemap(inode, fieinfo, start, len,
4530 ext4_get_block);
4531
4532 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4533 return -EBADR;
4534
4535 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4536 error = ext4_xattr_fiemap(inode, fieinfo);
4537 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004538 ext4_lblk_t len_blks;
4539 __u64 last_blk;
4540
Eric Sandeen6873fa02008-10-07 00:46:36 -04004541 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004542 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004543 if (last_blk >= EXT_MAX_BLOCKS)
4544 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004545 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004546
4547 /*
4548 * Walk the extent tree gathering extent information.
4549 * ext4_ext_fiemap_cb will push extents back to user.
4550 */
Eric Sandeen6873fa02008-10-07 00:46:36 -04004551 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4552 ext4_ext_fiemap_cb, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004553 }
4554
4555 return error;
4556}