blob: 74f23c292e1b3000bb7bd9cf0e953df27488b698 [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
Alex Tomasa86c6182006-10-11 01:21:03 -070032#include <linux/fs.h>
33#include <linux/time.h>
Mingming Caocd02ff02007-10-16 18:38:25 -040034#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070035#include <linux/highuid.h>
36#include <linux/pagemap.h>
37#include <linux/quotaops.h>
38#include <linux/string.h>
39#include <linux/slab.h>
Amit Aroraa2df2a62007-07-17 21:42:41 -040040#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070041#include <asm/uaccess.h>
Eric Sandeen6873fa02008-10-07 00:46:36 -040042#include <linux/fiemap.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040043#include "ext4_jbd2.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070044
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040045#include <trace/events/ext4.h>
46
Allison Hendersond583fb82011-05-25 07:41:43 -040047static int ext4_split_extent(handle_t *handle,
48 struct inode *inode,
49 struct ext4_ext_path *path,
50 struct ext4_map_blocks *map,
51 int split_flag,
52 int flags);
53
Jan Kara487caee2009-08-17 22:17:20 -040054static int ext4_ext_truncate_extend_restart(handle_t *handle,
55 struct inode *inode,
56 int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -070057{
58 int err;
59
Frank Mayhar03901312009-01-07 00:06:22 -050060 if (!ext4_handle_valid(handle))
61 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -070062 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -040063 return 0;
64 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -040065 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -040066 return err;
Jan Kara487caee2009-08-17 22:17:20 -040067 err = ext4_truncate_restart_trans(handle, inode, needed);
Dmitry Monakhov0617b832010-05-17 01:00:00 -040068 if (err == 0)
69 err = -EAGAIN;
Jan Kara487caee2009-08-17 22:17:20 -040070
71 return err;
Alex Tomasa86c6182006-10-11 01:21:03 -070072}
73
74/*
75 * could return:
76 * - EROFS
77 * - ENOMEM
78 */
79static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
80 struct ext4_ext_path *path)
81{
82 if (path->p_bh) {
83 /* path points to block */
84 return ext4_journal_get_write_access(handle, path->p_bh);
85 }
86 /* path points to leaf/index in inode body */
87 /* we use in-core data, no need to protect them */
88 return 0;
89}
90
91/*
92 * could return:
93 * - EROFS
94 * - ENOMEM
95 * - EIO
96 */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -040097#define ext4_ext_dirty(handle, inode, path) \
98 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
99static int __ext4_ext_dirty(const char *where, unsigned int line,
100 handle_t *handle, struct inode *inode,
101 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700102{
103 int err;
104 if (path->p_bh) {
105 /* path points to block */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400106 err = __ext4_handle_dirty_metadata(where, line, handle,
107 inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700108 } else {
109 /* path points to leaf/index in inode body */
110 err = ext4_mark_inode_dirty(handle, inode);
111 }
112 return err;
113}
114
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700115static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700116 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500117 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700118{
Alex Tomasa86c6182006-10-11 01:21:03 -0700119 if (path) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400120 int depth = path->p_depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700121 struct ext4_extent *ex;
Alex Tomasa86c6182006-10-11 01:21:03 -0700122
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500123 /*
124 * Try to predict block placement assuming that we are
125 * filling in a file which will eventually be
126 * non-sparse --- i.e., in the case of libbfd writing
127 * an ELF object sections out-of-order but in a way
128 * the eventually results in a contiguous object or
129 * executable file, or some database extending a table
130 * space file. However, this is actually somewhat
131 * non-ideal if we are writing a sparse file such as
132 * qemu or KVM writing a raw image file that is going
133 * to stay fairly sparse, since it will end up
134 * fragmenting the file system's free space. Maybe we
135 * should have some hueristics or some way to allow
136 * userspace to pass a hint to file system,
Tao Mab8d65682011-01-21 23:21:31 +0800137 * especially if the latter case turns out to be
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500138 * common.
139 */
Avantika Mathur7e028972006-12-06 20:41:33 -0800140 ex = path[depth].p_ext;
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500141 if (ex) {
142 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
143 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
144
145 if (block > ext_block)
146 return ext_pblk + (block - ext_block);
147 else
148 return ext_pblk - (ext_block - block);
149 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700150
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700151 /* it looks like index is empty;
152 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700153 if (path[depth].p_bh)
154 return path[depth].p_bh->b_blocknr;
155 }
156
157 /* OK. use inode's group */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400158 return ext4_inode_to_goal_block(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700159}
160
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400161/*
162 * Allocation for a meta data block
163 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700164static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400165ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700166 struct ext4_ext_path *path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400167 struct ext4_extent *ex, int *err, unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700168{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700169 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700170
171 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Allison Henderson55f020d2011-05-25 07:41:26 -0400172 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
173 NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700174 return newblock;
175}
176
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400177static inline int ext4_ext_space_block(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700178{
179 int size;
180
181 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
182 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100183#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400184 if (!check && size > 6)
185 size = 6;
Alex Tomasa86c6182006-10-11 01:21:03 -0700186#endif
187 return size;
188}
189
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400190static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700191{
192 int size;
193
194 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
195 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100196#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400197 if (!check && size > 5)
198 size = 5;
Alex Tomasa86c6182006-10-11 01:21:03 -0700199#endif
200 return size;
201}
202
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400203static inline int ext4_ext_space_root(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700204{
205 int size;
206
207 size = sizeof(EXT4_I(inode)->i_data);
208 size -= sizeof(struct ext4_extent_header);
209 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100210#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400211 if (!check && size > 3)
212 size = 3;
Alex Tomasa86c6182006-10-11 01:21:03 -0700213#endif
214 return size;
215}
216
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400217static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700218{
219 int size;
220
221 size = sizeof(EXT4_I(inode)->i_data);
222 size -= sizeof(struct ext4_extent_header);
223 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100224#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400225 if (!check && size > 4)
226 size = 4;
Alex Tomasa86c6182006-10-11 01:21:03 -0700227#endif
228 return size;
229}
230
Mingming Caod2a17632008-07-14 17:52:37 -0400231/*
232 * Calculate the number of metadata blocks needed
233 * to allocate @blocks
234 * Worse case is one block per extent
235 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500236int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400237{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500238 struct ext4_inode_info *ei = EXT4_I(inode);
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400239 int idxs;
Mingming Caod2a17632008-07-14 17:52:37 -0400240
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500241 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
242 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400243
244 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500245 * If the new delayed allocation block is contiguous with the
246 * previous da block, it can share index blocks with the
247 * previous block, so we only need to allocate a new index
248 * block every idxs leaf blocks. At ldxs**2 blocks, we need
249 * an additional index block, and at ldxs**3 blocks, yet
250 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400251 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500252 if (ei->i_da_metadata_calc_len &&
253 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400254 int num = 0;
255
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500256 if ((ei->i_da_metadata_calc_len % idxs) == 0)
257 num++;
258 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
259 num++;
260 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
261 num++;
262 ei->i_da_metadata_calc_len = 0;
263 } else
264 ei->i_da_metadata_calc_len++;
265 ei->i_da_metadata_calc_last_lblock++;
266 return num;
267 }
Mingming Caod2a17632008-07-14 17:52:37 -0400268
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500269 /*
270 * In the worst case we need a new set of index blocks at
271 * every level of the inode's extent tree.
272 */
273 ei->i_da_metadata_calc_len = 1;
274 ei->i_da_metadata_calc_last_lblock = lblock;
275 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400276}
277
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400278static int
279ext4_ext_max_entries(struct inode *inode, int depth)
280{
281 int max;
282
283 if (depth == ext_depth(inode)) {
284 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400285 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400286 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400287 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400288 } else {
289 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400290 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400291 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400292 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400293 }
294
295 return max;
296}
297
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400298static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
299{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400300 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400301 int len = ext4_ext_get_actual_len(ext);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400302
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400303 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400304}
305
306static int ext4_valid_extent_idx(struct inode *inode,
307 struct ext4_extent_idx *ext_idx)
308{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400309 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400310
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400311 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400312}
313
314static int ext4_valid_extent_entries(struct inode *inode,
315 struct ext4_extent_header *eh,
316 int depth)
317{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400318 unsigned short entries;
319 if (eh->eh_entries == 0)
320 return 1;
321
322 entries = le16_to_cpu(eh->eh_entries);
323
324 if (depth == 0) {
325 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400326 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400327 while (entries) {
328 if (!ext4_valid_extent(inode, ext))
329 return 0;
330 ext++;
331 entries--;
332 }
333 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400334 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400335 while (entries) {
336 if (!ext4_valid_extent_idx(inode, ext_idx))
337 return 0;
338 ext_idx++;
339 entries--;
340 }
341 }
342 return 1;
343}
344
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400345static int __ext4_ext_check(const char *function, unsigned int line,
346 struct inode *inode, struct ext4_extent_header *eh,
347 int depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400348{
349 const char *error_msg;
350 int max = 0;
351
352 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
353 error_msg = "invalid magic";
354 goto corrupted;
355 }
356 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
357 error_msg = "unexpected eh_depth";
358 goto corrupted;
359 }
360 if (unlikely(eh->eh_max == 0)) {
361 error_msg = "invalid eh_max";
362 goto corrupted;
363 }
364 max = ext4_ext_max_entries(inode, depth);
365 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
366 error_msg = "too large eh_max";
367 goto corrupted;
368 }
369 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
370 error_msg = "invalid eh_entries";
371 goto corrupted;
372 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400373 if (!ext4_valid_extent_entries(inode, eh, depth)) {
374 error_msg = "invalid extent entries";
375 goto corrupted;
376 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400377 return 0;
378
379corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400380 ext4_error_inode(inode, function, line, 0,
Theodore Ts'o24676da2010-05-16 21:00:00 -0400381 "bad header/extent: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400382 "entries %u, max %u(%u), depth %u(%u)",
Theodore Ts'o24676da2010-05-16 21:00:00 -0400383 error_msg, le16_to_cpu(eh->eh_magic),
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400384 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
385 max, le16_to_cpu(eh->eh_depth), depth);
386
387 return -EIO;
388}
389
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400390#define ext4_ext_check(inode, eh, depth) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400391 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400392
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400393int ext4_ext_check_inode(struct inode *inode)
394{
395 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
396}
397
Alex Tomasa86c6182006-10-11 01:21:03 -0700398#ifdef EXT_DEBUG
399static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
400{
401 int k, l = path->p_depth;
402
403 ext_debug("path:");
404 for (k = 0; k <= l; k++, path++) {
405 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700406 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400407 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700408 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400409 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700410 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400411 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400412 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400413 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700414 } else
415 ext_debug(" []");
416 }
417 ext_debug("\n");
418}
419
420static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
421{
422 int depth = ext_depth(inode);
423 struct ext4_extent_header *eh;
424 struct ext4_extent *ex;
425 int i;
426
427 if (!path)
428 return;
429
430 eh = path[depth].p_hdr;
431 ex = EXT_FIRST_EXTENT(eh);
432
Mingming553f9002009-09-18 13:34:55 -0400433 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
434
Alex Tomasa86c6182006-10-11 01:21:03 -0700435 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400436 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
437 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400438 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700439 }
440 ext_debug("\n");
441}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400442
443static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
444 ext4_fsblk_t newblock, int level)
445{
446 int depth = ext_depth(inode);
447 struct ext4_extent *ex;
448
449 if (depth != level) {
450 struct ext4_extent_idx *idx;
451 idx = path[level].p_idx;
452 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
453 ext_debug("%d: move %d:%llu in new index %llu\n", level,
454 le32_to_cpu(idx->ei_block),
455 ext4_idx_pblock(idx),
456 newblock);
457 idx++;
458 }
459
460 return;
461 }
462
463 ex = path[depth].p_ext;
464 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
465 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
466 le32_to_cpu(ex->ee_block),
467 ext4_ext_pblock(ex),
468 ext4_ext_is_uninitialized(ex),
469 ext4_ext_get_actual_len(ex),
470 newblock);
471 ex++;
472 }
473}
474
Alex Tomasa86c6182006-10-11 01:21:03 -0700475#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400476#define ext4_ext_show_path(inode, path)
477#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400478#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700479#endif
480
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500481void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700482{
483 int depth = path->p_depth;
484 int i;
485
486 for (i = 0; i <= depth; i++, path++)
487 if (path->p_bh) {
488 brelse(path->p_bh);
489 path->p_bh = NULL;
490 }
491}
492
493/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700494 * ext4_ext_binsearch_idx:
495 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400496 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700497 */
498static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500499ext4_ext_binsearch_idx(struct inode *inode,
500 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700501{
502 struct ext4_extent_header *eh = path->p_hdr;
503 struct ext4_extent_idx *r, *l, *m;
504
Alex Tomasa86c6182006-10-11 01:21:03 -0700505
Eric Sandeenbba90742008-01-28 23:58:27 -0500506 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700507
508 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400509 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700510 while (l <= r) {
511 m = l + (r - l) / 2;
512 if (block < le32_to_cpu(m->ei_block))
513 r = m - 1;
514 else
515 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400516 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
517 m, le32_to_cpu(m->ei_block),
518 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700519 }
520
521 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700522 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400523 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700524
525#ifdef CHECK_BINSEARCH
526 {
527 struct ext4_extent_idx *chix, *ix;
528 int k;
529
530 chix = ix = EXT_FIRST_INDEX(eh);
531 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
532 if (k != 0 &&
533 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o47760042008-09-08 23:00:52 -0400534 printk(KERN_DEBUG "k=%d, ix=0x%p, "
535 "first=0x%p\n", k,
536 ix, EXT_FIRST_INDEX(eh));
537 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700538 le32_to_cpu(ix->ei_block),
539 le32_to_cpu(ix[-1].ei_block));
540 }
541 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400542 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700543 if (block < le32_to_cpu(ix->ei_block))
544 break;
545 chix = ix;
546 }
547 BUG_ON(chix != path->p_idx);
548 }
549#endif
550
551}
552
553/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700554 * ext4_ext_binsearch:
555 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400556 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700557 */
558static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500559ext4_ext_binsearch(struct inode *inode,
560 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700561{
562 struct ext4_extent_header *eh = path->p_hdr;
563 struct ext4_extent *r, *l, *m;
564
Alex Tomasa86c6182006-10-11 01:21:03 -0700565 if (eh->eh_entries == 0) {
566 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700567 * this leaf is empty:
568 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700569 */
570 return;
571 }
572
Eric Sandeenbba90742008-01-28 23:58:27 -0500573 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700574
575 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400576 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700577
578 while (l <= r) {
579 m = l + (r - l) / 2;
580 if (block < le32_to_cpu(m->ee_block))
581 r = m - 1;
582 else
583 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400584 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
585 m, le32_to_cpu(m->ee_block),
586 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700587 }
588
589 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400590 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400591 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400592 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400593 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400594 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700595
596#ifdef CHECK_BINSEARCH
597 {
598 struct ext4_extent *chex, *ex;
599 int k;
600
601 chex = ex = EXT_FIRST_EXTENT(eh);
602 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
603 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400604 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700605 if (block < le32_to_cpu(ex->ee_block))
606 break;
607 chex = ex;
608 }
609 BUG_ON(chex != path->p_ext);
610 }
611#endif
612
613}
614
615int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
616{
617 struct ext4_extent_header *eh;
618
619 eh = ext_inode_hdr(inode);
620 eh->eh_depth = 0;
621 eh->eh_entries = 0;
622 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400623 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700624 ext4_mark_inode_dirty(handle, inode);
625 ext4_ext_invalidate_cache(inode);
626 return 0;
627}
628
629struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500630ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
631 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700632{
633 struct ext4_extent_header *eh;
634 struct buffer_head *bh;
635 short int depth, i, ppos = 0, alloc = 0;
636
637 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400638 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700639
640 /* account possible depth increase */
641 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800642 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700643 GFP_NOFS);
644 if (!path)
645 return ERR_PTR(-ENOMEM);
646 alloc = 1;
647 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700648 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400649 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700650
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400651 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700652 /* walk through the tree */
653 while (i) {
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400654 int need_to_validate = 0;
655
Alex Tomasa86c6182006-10-11 01:21:03 -0700656 ext_debug("depth %d: num %d, max %d\n",
657 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400658
Alex Tomasa86c6182006-10-11 01:21:03 -0700659 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400660 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700661 path[ppos].p_depth = i;
662 path[ppos].p_ext = NULL;
663
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400664 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
665 if (unlikely(!bh))
Alex Tomasa86c6182006-10-11 01:21:03 -0700666 goto err;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400667 if (!bh_uptodate_or_lock(bh)) {
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400668 trace_ext4_ext_load_extent(inode, block,
669 path[ppos].p_block);
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400670 if (bh_submit_read(bh) < 0) {
671 put_bh(bh);
672 goto err;
673 }
674 /* validate the extent entries */
675 need_to_validate = 1;
676 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700677 eh = ext_block_hdr(bh);
678 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500679 if (unlikely(ppos > depth)) {
680 put_bh(bh);
681 EXT4_ERROR_INODE(inode,
682 "ppos %d > depth %d", ppos, depth);
683 goto err;
684 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700685 path[ppos].p_bh = bh;
686 path[ppos].p_hdr = eh;
687 i--;
688
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400689 if (need_to_validate && ext4_ext_check(inode, eh, i))
Alex Tomasa86c6182006-10-11 01:21:03 -0700690 goto err;
691 }
692
693 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700694 path[ppos].p_ext = NULL;
695 path[ppos].p_idx = NULL;
696
Alex Tomasa86c6182006-10-11 01:21:03 -0700697 /* find extent */
698 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400699 /* if not an empty leaf */
700 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400701 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700702
703 ext4_ext_show_path(inode, path);
704
705 return path;
706
707err:
708 ext4_ext_drop_refs(path);
709 if (alloc)
710 kfree(path);
711 return ERR_PTR(-EIO);
712}
713
714/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700715 * ext4_ext_insert_index:
716 * insert new index [@logical;@ptr] into the block at @curp;
717 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700718 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400719static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
720 struct ext4_ext_path *curp,
721 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700722{
723 struct ext4_extent_idx *ix;
724 int len, err;
725
Avantika Mathur7e028972006-12-06 20:41:33 -0800726 err = ext4_ext_get_access(handle, inode, curp);
727 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700728 return err;
729
Frank Mayhar273df552010-03-02 11:46:09 -0500730 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
731 EXT4_ERROR_INODE(inode,
732 "logical %d == ei_block %d!",
733 logical, le32_to_cpu(curp->p_idx->ei_block));
734 return -EIO;
735 }
Robin Dongd4620312011-07-17 23:43:42 -0400736
737 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
738 >= le16_to_cpu(curp->p_hdr->eh_max))) {
739 EXT4_ERROR_INODE(inode,
740 "eh_entries %d >= eh_max %d!",
741 le16_to_cpu(curp->p_hdr->eh_entries),
742 le16_to_cpu(curp->p_hdr->eh_max));
743 return -EIO;
744 }
745
Alex Tomasa86c6182006-10-11 01:21:03 -0700746 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
747 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400748 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700749 ix = curp->p_idx + 1;
750 } else {
751 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400752 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700753 ix = curp->p_idx;
754 }
755
Eric Gouriou80e675f2011-10-27 11:52:18 -0400756 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
757 BUG_ON(len < 0);
758 if (len > 0) {
759 ext_debug("insert new index %d: "
760 "move %d indices from 0x%p to 0x%p\n",
761 logical, len, ix, ix + 1);
762 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
763 }
764
Tao Maf472e022011-10-17 10:13:46 -0400765 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
766 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
767 return -EIO;
768 }
769
Alex Tomasa86c6182006-10-11 01:21:03 -0700770 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700771 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400772 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700773
Frank Mayhar273df552010-03-02 11:46:09 -0500774 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
775 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
776 return -EIO;
777 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700778
779 err = ext4_ext_dirty(handle, inode, curp);
780 ext4_std_error(inode->i_sb, err);
781
782 return err;
783}
784
785/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700786 * ext4_ext_split:
787 * inserts new subtree into the path, using free index entry
788 * at depth @at:
789 * - allocates all needed blocks (new leaf and all intermediate index blocks)
790 * - makes decision where to split
791 * - moves remaining extents and index entries (right to the split point)
792 * into the newly allocated blocks
793 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700794 */
795static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400796 unsigned int flags,
797 struct ext4_ext_path *path,
798 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -0700799{
800 struct buffer_head *bh = NULL;
801 int depth = ext_depth(inode);
802 struct ext4_extent_header *neh;
803 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -0700804 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700805 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700806 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700807 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700808 int err = 0;
809
810 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700811 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700812
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700813 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700814 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -0500815 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
816 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
817 return -EIO;
818 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700819 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
820 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700821 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700822 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400823 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700824 } else {
825 border = newext->ee_block;
826 ext_debug("leaf will be added."
827 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400828 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700829 }
830
831 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700832 * If error occurs, then we break processing
833 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700834 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700835 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700836 */
837
838 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700839 * Get array to track all allocated blocks.
840 * We need this to handle errors and free blocks
841 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700842 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800843 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700844 if (!ablocks)
845 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700846
847 /* allocate all needed blocks */
848 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
849 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400850 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400851 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -0700852 if (newblock == 0)
853 goto cleanup;
854 ablocks[a] = newblock;
855 }
856
857 /* initialize new leaf */
858 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -0500859 if (unlikely(newblock == 0)) {
860 EXT4_ERROR_INODE(inode, "newblock == 0!");
861 err = -EIO;
862 goto cleanup;
863 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700864 bh = sb_getblk(inode->i_sb, newblock);
865 if (!bh) {
866 err = -EIO;
867 goto cleanup;
868 }
869 lock_buffer(bh);
870
Avantika Mathur7e028972006-12-06 20:41:33 -0800871 err = ext4_journal_get_create_access(handle, bh);
872 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700873 goto cleanup;
874
875 neh = ext_block_hdr(bh);
876 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400877 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700878 neh->eh_magic = EXT4_EXT_MAGIC;
879 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700880
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700881 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -0500882 if (unlikely(path[depth].p_hdr->eh_entries !=
883 path[depth].p_hdr->eh_max)) {
884 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
885 path[depth].p_hdr->eh_entries,
886 path[depth].p_hdr->eh_max);
887 err = -EIO;
888 goto cleanup;
889 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700890 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400891 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
892 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -0700893 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400894 struct ext4_extent *ex;
895 ex = EXT_FIRST_EXTENT(neh);
896 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400897 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700898 }
899
900 set_buffer_uptodate(bh);
901 unlock_buffer(bh);
902
Frank Mayhar03901312009-01-07 00:06:22 -0500903 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800904 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700905 goto cleanup;
906 brelse(bh);
907 bh = NULL;
908
909 /* correct old leaf */
910 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -0800911 err = ext4_ext_get_access(handle, inode, path + depth);
912 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700913 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400914 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -0800915 err = ext4_ext_dirty(handle, inode, path + depth);
916 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700917 goto cleanup;
918
919 }
920
921 /* create intermediate indexes */
922 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -0500923 if (unlikely(k < 0)) {
924 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
925 err = -EIO;
926 goto cleanup;
927 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700928 if (k)
929 ext_debug("create %d intermediate indices\n", k);
930 /* insert new index into current index block */
931 /* current depth stored in i var */
932 i = depth - 1;
933 while (k--) {
934 oldblock = newblock;
935 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -0500936 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700937 if (!bh) {
938 err = -EIO;
939 goto cleanup;
940 }
941 lock_buffer(bh);
942
Avantika Mathur7e028972006-12-06 20:41:33 -0800943 err = ext4_journal_get_create_access(handle, bh);
944 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700945 goto cleanup;
946
947 neh = ext_block_hdr(bh);
948 neh->eh_entries = cpu_to_le16(1);
949 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400950 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700951 neh->eh_depth = cpu_to_le16(depth - i);
952 fidx = EXT_FIRST_INDEX(neh);
953 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700954 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700955
Eric Sandeenbba90742008-01-28 23:58:27 -0500956 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
957 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700958
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400959 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -0500960 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
961 EXT_LAST_INDEX(path[i].p_hdr))) {
962 EXT4_ERROR_INODE(inode,
963 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
964 le32_to_cpu(path[i].p_ext->ee_block));
965 err = -EIO;
966 goto cleanup;
967 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400968 /* start copy indexes */
969 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
970 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
971 EXT_MAX_INDEX(path[i].p_hdr));
972 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -0700973 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400974 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -0700975 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400976 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700977 }
978 set_buffer_uptodate(bh);
979 unlock_buffer(bh);
980
Frank Mayhar03901312009-01-07 00:06:22 -0500981 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800982 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700983 goto cleanup;
984 brelse(bh);
985 bh = NULL;
986
987 /* correct old index */
988 if (m) {
989 err = ext4_ext_get_access(handle, inode, path + i);
990 if (err)
991 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400992 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700993 err = ext4_ext_dirty(handle, inode, path + i);
994 if (err)
995 goto cleanup;
996 }
997
998 i--;
999 }
1000
1001 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001002 err = ext4_ext_insert_index(handle, inode, path + at,
1003 le32_to_cpu(border), newblock);
1004
1005cleanup:
1006 if (bh) {
1007 if (buffer_locked(bh))
1008 unlock_buffer(bh);
1009 brelse(bh);
1010 }
1011
1012 if (err) {
1013 /* free all allocated blocks in error case */
1014 for (i = 0; i < depth; i++) {
1015 if (!ablocks[i])
1016 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001017 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001018 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001019 }
1020 }
1021 kfree(ablocks);
1022
1023 return err;
1024}
1025
1026/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001027 * ext4_ext_grow_indepth:
1028 * implements tree growing procedure:
1029 * - allocates new block
1030 * - moves top-level data (index block or leaf) into the new block
1031 * - initializes new top-level, creating index that points to the
1032 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001033 */
1034static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001035 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001036 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001037{
Alex Tomasa86c6182006-10-11 01:21:03 -07001038 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001039 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001040 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001041 int err = 0;
1042
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001043 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001044 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001045 if (newblock == 0)
1046 return err;
1047
1048 bh = sb_getblk(inode->i_sb, newblock);
1049 if (!bh) {
1050 err = -EIO;
1051 ext4_std_error(inode->i_sb, err);
1052 return err;
1053 }
1054 lock_buffer(bh);
1055
Avantika Mathur7e028972006-12-06 20:41:33 -08001056 err = ext4_journal_get_create_access(handle, bh);
1057 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001058 unlock_buffer(bh);
1059 goto out;
1060 }
1061
1062 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001063 memmove(bh->b_data, EXT4_I(inode)->i_data,
1064 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001065
1066 /* set size of new block */
1067 neh = ext_block_hdr(bh);
1068 /* old root could have indexes or leaves
1069 * so calculate e_max right way */
1070 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001071 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001072 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001073 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001074 neh->eh_magic = EXT4_EXT_MAGIC;
1075 set_buffer_uptodate(bh);
1076 unlock_buffer(bh);
1077
Frank Mayhar03901312009-01-07 00:06:22 -05001078 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001079 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001080 goto out;
1081
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001082 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001083 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001084 neh->eh_entries = cpu_to_le16(1);
1085 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1086 if (neh->eh_depth == 0) {
1087 /* Root extent block becomes index block */
1088 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1089 EXT_FIRST_INDEX(neh)->ei_block =
1090 EXT_FIRST_EXTENT(neh)->ee_block;
1091 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001092 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001093 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001094 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001095 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001096
Paul Mackerrasb4611ab2011-12-12 11:00:56 -05001097 neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001098 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001099out:
1100 brelse(bh);
1101
1102 return err;
1103}
1104
1105/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001106 * ext4_ext_create_new_leaf:
1107 * finds empty index and adds new leaf.
1108 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001109 */
1110static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001111 unsigned int flags,
1112 struct ext4_ext_path *path,
1113 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001114{
1115 struct ext4_ext_path *curp;
1116 int depth, i, err = 0;
1117
1118repeat:
1119 i = depth = ext_depth(inode);
1120
1121 /* walk up to the tree and look for free index entry */
1122 curp = path + depth;
1123 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1124 i--;
1125 curp--;
1126 }
1127
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001128 /* we use already allocated block for index block,
1129 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001130 if (EXT_HAS_FREE_INDEX(curp)) {
1131 /* if we found index with free entry, then use that
1132 * entry: create all needed subtree and add new leaf */
Allison Henderson55f020d2011-05-25 07:41:26 -04001133 err = ext4_ext_split(handle, inode, flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001134 if (err)
1135 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001136
1137 /* refill path */
1138 ext4_ext_drop_refs(path);
1139 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001140 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1141 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001142 if (IS_ERR(path))
1143 err = PTR_ERR(path);
1144 } else {
1145 /* tree is full, time to grow in depth */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001146 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001147 if (err)
1148 goto out;
1149
1150 /* refill path */
1151 ext4_ext_drop_refs(path);
1152 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001153 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1154 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001155 if (IS_ERR(path)) {
1156 err = PTR_ERR(path);
1157 goto out;
1158 }
1159
1160 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001161 * only first (depth 0 -> 1) produces free space;
1162 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001163 */
1164 depth = ext_depth(inode);
1165 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001166 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001167 goto repeat;
1168 }
1169 }
1170
1171out:
1172 return err;
1173}
1174
1175/*
Alex Tomas1988b512008-01-28 23:58:27 -05001176 * search the closest allocated block to the left for *logical
1177 * and returns it at @logical + it's physical address at @phys
1178 * if *logical is the smallest allocated block, the function
1179 * returns 0 at @phys
1180 * return value contains 0 (success) or error code
1181 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001182static int ext4_ext_search_left(struct inode *inode,
1183 struct ext4_ext_path *path,
1184 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001185{
1186 struct ext4_extent_idx *ix;
1187 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001188 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001189
Frank Mayhar273df552010-03-02 11:46:09 -05001190 if (unlikely(path == NULL)) {
1191 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1192 return -EIO;
1193 }
Alex Tomas1988b512008-01-28 23:58:27 -05001194 depth = path->p_depth;
1195 *phys = 0;
1196
1197 if (depth == 0 && path->p_ext == NULL)
1198 return 0;
1199
1200 /* usually extent in the path covers blocks smaller
1201 * then *logical, but it can be that extent is the
1202 * first one in the file */
1203
1204 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001205 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001206 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001207 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1208 EXT4_ERROR_INODE(inode,
1209 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1210 *logical, le32_to_cpu(ex->ee_block));
1211 return -EIO;
1212 }
Alex Tomas1988b512008-01-28 23:58:27 -05001213 while (--depth >= 0) {
1214 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001215 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1216 EXT4_ERROR_INODE(inode,
1217 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001218 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001219 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001220 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001221 depth);
1222 return -EIO;
1223 }
Alex Tomas1988b512008-01-28 23:58:27 -05001224 }
1225 return 0;
1226 }
1227
Frank Mayhar273df552010-03-02 11:46:09 -05001228 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1229 EXT4_ERROR_INODE(inode,
1230 "logical %d < ee_block %d + ee_len %d!",
1231 *logical, le32_to_cpu(ex->ee_block), ee_len);
1232 return -EIO;
1233 }
Alex Tomas1988b512008-01-28 23:58:27 -05001234
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001235 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001236 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001237 return 0;
1238}
1239
1240/*
1241 * search the closest allocated block to the right for *logical
1242 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001243 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001244 * returns 0 at @phys
1245 * return value contains 0 (success) or error code
1246 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001247static int ext4_ext_search_right(struct inode *inode,
1248 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001249 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1250 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001251{
1252 struct buffer_head *bh = NULL;
1253 struct ext4_extent_header *eh;
1254 struct ext4_extent_idx *ix;
1255 struct ext4_extent *ex;
1256 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001257 int depth; /* Note, NOT eh_depth; depth from top of tree */
1258 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001259
Frank Mayhar273df552010-03-02 11:46:09 -05001260 if (unlikely(path == NULL)) {
1261 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1262 return -EIO;
1263 }
Alex Tomas1988b512008-01-28 23:58:27 -05001264 depth = path->p_depth;
1265 *phys = 0;
1266
1267 if (depth == 0 && path->p_ext == NULL)
1268 return 0;
1269
1270 /* usually extent in the path covers blocks smaller
1271 * then *logical, but it can be that extent is the
1272 * first one in the file */
1273
1274 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001275 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001276 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001277 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1278 EXT4_ERROR_INODE(inode,
1279 "first_extent(path[%d].p_hdr) != ex",
1280 depth);
1281 return -EIO;
1282 }
Alex Tomas1988b512008-01-28 23:58:27 -05001283 while (--depth >= 0) {
1284 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001285 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1286 EXT4_ERROR_INODE(inode,
1287 "ix != EXT_FIRST_INDEX *logical %d!",
1288 *logical);
1289 return -EIO;
1290 }
Alex Tomas1988b512008-01-28 23:58:27 -05001291 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001292 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001293 }
1294
Frank Mayhar273df552010-03-02 11:46:09 -05001295 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1296 EXT4_ERROR_INODE(inode,
1297 "logical %d < ee_block %d + ee_len %d!",
1298 *logical, le32_to_cpu(ex->ee_block), ee_len);
1299 return -EIO;
1300 }
Alex Tomas1988b512008-01-28 23:58:27 -05001301
1302 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1303 /* next allocated block in this leaf */
1304 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001305 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001306 }
1307
1308 /* go up and search for index to the right */
1309 while (--depth >= 0) {
1310 ix = path[depth].p_idx;
1311 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001312 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001313 }
1314
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001315 /* we've gone up to the root and found no index to the right */
1316 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001317
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001318got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001319 /* we've found index to the right, let's
1320 * follow it and find the closest allocated
1321 * block to the right */
1322 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001323 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001324 while (++depth < path->p_depth) {
1325 bh = sb_bread(inode->i_sb, block);
1326 if (bh == NULL)
1327 return -EIO;
1328 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001329 /* subtract from p_depth to get proper eh_depth */
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04001330 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001331 put_bh(bh);
1332 return -EIO;
1333 }
1334 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001335 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001336 put_bh(bh);
1337 }
1338
1339 bh = sb_bread(inode->i_sb, block);
1340 if (bh == NULL)
1341 return -EIO;
1342 eh = ext_block_hdr(bh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04001343 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001344 put_bh(bh);
1345 return -EIO;
1346 }
1347 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001348found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001349 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001350 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001351 *ret_ex = ex;
1352 if (bh)
1353 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001354 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001355}
1356
1357/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001358 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001359 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001360 * NOTE: it considers block number from index entry as
1361 * allocated block. Thus, index entries have to be consistent
1362 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001363 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001364static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001365ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1366{
1367 int depth;
1368
1369 BUG_ON(path == NULL);
1370 depth = path->p_depth;
1371
1372 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001373 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001374
1375 while (depth >= 0) {
1376 if (depth == path->p_depth) {
1377 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001378 if (path[depth].p_ext &&
1379 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001380 EXT_LAST_EXTENT(path[depth].p_hdr))
1381 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1382 } else {
1383 /* index */
1384 if (path[depth].p_idx !=
1385 EXT_LAST_INDEX(path[depth].p_hdr))
1386 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1387 }
1388 depth--;
1389 }
1390
Lukas Czernerf17722f2011-06-06 00:05:17 -04001391 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001392}
1393
1394/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001395 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001396 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001397 */
Robin Dong57187892011-07-23 21:49:07 -04001398static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001399{
1400 int depth;
1401
1402 BUG_ON(path == NULL);
1403 depth = path->p_depth;
1404
1405 /* zero-tree has no leaf blocks at all */
1406 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001407 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001408
1409 /* go to index block */
1410 depth--;
1411
1412 while (depth >= 0) {
1413 if (path[depth].p_idx !=
1414 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001415 return (ext4_lblk_t)
1416 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001417 depth--;
1418 }
1419
Lukas Czernerf17722f2011-06-06 00:05:17 -04001420 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001421}
1422
1423/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001424 * ext4_ext_correct_indexes:
1425 * if leaf gets modified and modified extent is first in the leaf,
1426 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001427 * TODO: do we need to correct tree in all cases?
1428 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001429static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001430 struct ext4_ext_path *path)
1431{
1432 struct ext4_extent_header *eh;
1433 int depth = ext_depth(inode);
1434 struct ext4_extent *ex;
1435 __le32 border;
1436 int k, err = 0;
1437
1438 eh = path[depth].p_hdr;
1439 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001440
1441 if (unlikely(ex == NULL || eh == NULL)) {
1442 EXT4_ERROR_INODE(inode,
1443 "ex %p == NULL or eh %p == NULL", ex, eh);
1444 return -EIO;
1445 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001446
1447 if (depth == 0) {
1448 /* there is no tree at all */
1449 return 0;
1450 }
1451
1452 if (ex != EXT_FIRST_EXTENT(eh)) {
1453 /* we correct tree if first leaf got modified only */
1454 return 0;
1455 }
1456
1457 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001458 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001459 */
1460 k = depth - 1;
1461 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001462 err = ext4_ext_get_access(handle, inode, path + k);
1463 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001464 return err;
1465 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001466 err = ext4_ext_dirty(handle, inode, path + k);
1467 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001468 return err;
1469
1470 while (k--) {
1471 /* change all left-side indexes */
1472 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1473 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001474 err = ext4_ext_get_access(handle, inode, path + k);
1475 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001476 break;
1477 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001478 err = ext4_ext_dirty(handle, inode, path + k);
1479 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001480 break;
1481 }
1482
1483 return err;
1484}
1485
Akira Fujita748de672009-06-17 19:24:03 -04001486int
Alex Tomasa86c6182006-10-11 01:21:03 -07001487ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1488 struct ext4_extent *ex2)
1489{
Amit Arora749269f2007-07-18 09:02:56 -04001490 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001491
1492 /*
1493 * Make sure that either both extents are uninitialized, or
1494 * both are _not_.
1495 */
1496 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1497 return 0;
1498
Amit Arora749269f2007-07-18 09:02:56 -04001499 if (ext4_ext_is_uninitialized(ex1))
1500 max_len = EXT_UNINIT_MAX_LEN;
1501 else
1502 max_len = EXT_INIT_MAX_LEN;
1503
Amit Aroraa2df2a62007-07-17 21:42:41 -04001504 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1505 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1506
1507 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001508 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001509 return 0;
1510
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001511 /*
1512 * To allow future support for preallocated extents to be added
1513 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001514 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001515 */
Amit Arora749269f2007-07-18 09:02:56 -04001516 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001517 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001518#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001519 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001520 return 0;
1521#endif
1522
Theodore Ts'obf89d162010-10-27 21:30:14 -04001523 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001524 return 1;
1525 return 0;
1526}
1527
1528/*
Amit Arora56055d32007-07-17 21:42:38 -04001529 * This function tries to merge the "ex" extent to the next extent in the tree.
1530 * It always tries to merge towards right. If you want to merge towards
1531 * left, pass "ex - 1" as argument instead of "ex".
1532 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1533 * 1 if they got merged.
1534 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001535static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001536 struct ext4_ext_path *path,
1537 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001538{
1539 struct ext4_extent_header *eh;
1540 unsigned int depth, len;
1541 int merge_done = 0;
1542 int uninitialized = 0;
1543
1544 depth = ext_depth(inode);
1545 BUG_ON(path[depth].p_hdr == NULL);
1546 eh = path[depth].p_hdr;
1547
1548 while (ex < EXT_LAST_EXTENT(eh)) {
1549 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1550 break;
1551 /* merge with next extent! */
1552 if (ext4_ext_is_uninitialized(ex))
1553 uninitialized = 1;
1554 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1555 + ext4_ext_get_actual_len(ex + 1));
1556 if (uninitialized)
1557 ext4_ext_mark_uninitialized(ex);
1558
1559 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1560 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1561 * sizeof(struct ext4_extent);
1562 memmove(ex + 1, ex + 2, len);
1563 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001564 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001565 merge_done = 1;
1566 WARN_ON(eh->eh_entries == 0);
1567 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001568 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001569 }
1570
1571 return merge_done;
1572}
1573
1574/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001575 * This function tries to merge the @ex extent to neighbours in the tree.
1576 * return 1 if merge left else 0.
1577 */
1578static int ext4_ext_try_to_merge(struct inode *inode,
1579 struct ext4_ext_path *path,
1580 struct ext4_extent *ex) {
1581 struct ext4_extent_header *eh;
1582 unsigned int depth;
1583 int merge_done = 0;
1584 int ret = 0;
1585
1586 depth = ext_depth(inode);
1587 BUG_ON(path[depth].p_hdr == NULL);
1588 eh = path[depth].p_hdr;
1589
1590 if (ex > EXT_FIRST_EXTENT(eh))
1591 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1592
1593 if (!merge_done)
1594 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1595
1596 return ret;
1597}
1598
1599/*
Amit Arora25d14f92007-05-24 13:04:13 -04001600 * check if a portion of the "newext" extent overlaps with an
1601 * existing extent.
1602 *
1603 * If there is an overlap discovered, it updates the length of the newext
1604 * such that there will be no overlap, and then returns 1.
1605 * If there is no overlap found, it returns 0.
1606 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001607static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1608 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001609 struct ext4_extent *newext,
1610 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001611{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001612 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001613 unsigned int depth, len1;
1614 unsigned int ret = 0;
1615
1616 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001617 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001618 depth = ext_depth(inode);
1619 if (!path[depth].p_ext)
1620 goto out;
1621 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001622 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001623
1624 /*
1625 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001626 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001627 */
1628 if (b2 < b1) {
1629 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001630 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001631 goto out;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001632 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001633 }
1634
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001635 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001636 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001637 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001638 newext->ee_len = cpu_to_le16(len1);
1639 ret = 1;
1640 }
1641
1642 /* check for overlap */
1643 if (b1 + len1 > b2) {
1644 newext->ee_len = cpu_to_le16(b2 - b1);
1645 ret = 1;
1646 }
1647out:
1648 return ret;
1649}
1650
1651/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001652 * ext4_ext_insert_extent:
1653 * tries to merge requsted extent into the existing extent or
1654 * inserts requested extent as new one into the tree,
1655 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001656 */
1657int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1658 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04001659 struct ext4_extent *newext, int flag)
Alex Tomasa86c6182006-10-11 01:21:03 -07001660{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001661 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001662 struct ext4_extent *ex, *fex;
1663 struct ext4_extent *nearex; /* nearest extent */
1664 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001665 int depth, len, err;
1666 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001667 unsigned uninitialized = 0;
Allison Henderson55f020d2011-05-25 07:41:26 -04001668 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001669
Frank Mayhar273df552010-03-02 11:46:09 -05001670 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1671 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1672 return -EIO;
1673 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001674 depth = ext_depth(inode);
1675 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001676 if (unlikely(path[depth].p_hdr == NULL)) {
1677 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1678 return -EIO;
1679 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001680
1681 /* try to insert block into found extent and return */
Jiaying Zhang744692d2010-03-04 16:14:02 -05001682 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
Mingming Cao00314622009-09-28 15:49:08 -04001683 && ext4_can_extents_be_merged(inode, ex, newext)) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001684 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04001685 ext4_ext_is_uninitialized(newext),
1686 ext4_ext_get_actual_len(newext),
1687 le32_to_cpu(ex->ee_block),
1688 ext4_ext_is_uninitialized(ex),
1689 ext4_ext_get_actual_len(ex),
1690 ext4_ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001691 err = ext4_ext_get_access(handle, inode, path + depth);
1692 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001693 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001694
1695 /*
1696 * ext4_can_extents_be_merged should have checked that either
1697 * both extents are uninitialized, or both aren't. Thus we
1698 * need to check only one of them here.
1699 */
1700 if (ext4_ext_is_uninitialized(ex))
1701 uninitialized = 1;
1702 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1703 + ext4_ext_get_actual_len(newext));
1704 if (uninitialized)
1705 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001706 eh = path[depth].p_hdr;
1707 nearex = ex;
1708 goto merge;
1709 }
1710
Alex Tomasa86c6182006-10-11 01:21:03 -07001711 depth = ext_depth(inode);
1712 eh = path[depth].p_hdr;
1713 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1714 goto has_space;
1715
1716 /* probably next leaf has space for us? */
1717 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001718 next = EXT_MAX_BLOCKS;
1719 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001720 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001721 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001722 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07001723 BUG_ON(npath != NULL);
1724 npath = ext4_ext_find_extent(inode, next, NULL);
1725 if (IS_ERR(npath))
1726 return PTR_ERR(npath);
1727 BUG_ON(npath->p_depth != path->p_depth);
1728 eh = npath[depth].p_hdr;
1729 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001730 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001731 le16_to_cpu(eh->eh_entries));
1732 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04001733 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07001734 }
1735 ext_debug("next leaf has no free space(%d,%d)\n",
1736 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1737 }
1738
1739 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001740 * There is no free space in the found leaf.
1741 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001742 */
Allison Henderson55f020d2011-05-25 07:41:26 -04001743 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1744 flags = EXT4_MB_USE_ROOT_BLOCKS;
1745 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001746 if (err)
1747 goto cleanup;
1748 depth = ext_depth(inode);
1749 eh = path[depth].p_hdr;
1750
1751has_space:
1752 nearex = path[depth].p_ext;
1753
Avantika Mathur7e028972006-12-06 20:41:33 -08001754 err = ext4_ext_get_access(handle, inode, path + depth);
1755 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001756 goto cleanup;
1757
1758 if (!nearex) {
1759 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001760 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001761 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001762 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001763 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001764 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04001765 nearex = EXT_FIRST_EXTENT(eh);
1766 } else {
1767 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001768 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04001769 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001770 ext_debug("insert %u:%llu:[%d]%d before: "
1771 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001772 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001773 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001774 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001775 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04001776 nearex);
1777 nearex++;
1778 } else {
1779 /* Insert before */
1780 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04001781 ext_debug("insert %u:%llu:[%d]%d after: "
1782 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04001783 le32_to_cpu(newext->ee_block),
1784 ext4_ext_pblock(newext),
1785 ext4_ext_is_uninitialized(newext),
1786 ext4_ext_get_actual_len(newext),
1787 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001788 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04001789 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1790 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001791 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04001792 "move %d extents from 0x%p to 0x%p\n",
1793 le32_to_cpu(newext->ee_block),
1794 ext4_ext_pblock(newext),
1795 ext4_ext_is_uninitialized(newext),
1796 ext4_ext_get_actual_len(newext),
1797 len, nearex, nearex + 1);
1798 memmove(nearex + 1, nearex,
1799 len * sizeof(struct ext4_extent));
1800 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001801 }
1802
Marcin Slusarze8546d02008-04-17 10:38:59 -04001803 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04001804 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07001805 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001806 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001807 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001808
1809merge:
1810 /* try to merge extents to the right */
Jiaying Zhang744692d2010-03-04 16:14:02 -05001811 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
Mingming Cao00314622009-09-28 15:49:08 -04001812 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001813
1814 /* try to merge extents to the left */
1815
1816 /* time to correct all indexes above */
1817 err = ext4_ext_correct_indexes(handle, inode, path);
1818 if (err)
1819 goto cleanup;
1820
1821 err = ext4_ext_dirty(handle, inode, path + depth);
1822
1823cleanup:
1824 if (npath) {
1825 ext4_ext_drop_refs(npath);
1826 kfree(npath);
1827 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001828 ext4_ext_invalidate_cache(inode);
1829 return err;
1830}
1831
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001832static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1833 ext4_lblk_t num, ext_prepare_callback func,
1834 void *cbdata)
Eric Sandeen6873fa02008-10-07 00:46:36 -04001835{
1836 struct ext4_ext_path *path = NULL;
1837 struct ext4_ext_cache cbex;
1838 struct ext4_extent *ex;
1839 ext4_lblk_t next, start = 0, end = 0;
1840 ext4_lblk_t last = block + num;
1841 int depth, exists, err = 0;
1842
1843 BUG_ON(func == NULL);
1844 BUG_ON(inode == NULL);
1845
Lukas Czernerf17722f2011-06-06 00:05:17 -04001846 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04001847 num = last - block;
1848 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05001849 down_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001850 path = ext4_ext_find_extent(inode, block, path);
Theodore Ts'ofab3a542009-12-09 21:30:02 -05001851 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001852 if (IS_ERR(path)) {
1853 err = PTR_ERR(path);
1854 path = NULL;
1855 break;
1856 }
1857
1858 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05001859 if (unlikely(path[depth].p_hdr == NULL)) {
1860 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1861 err = -EIO;
1862 break;
1863 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04001864 ex = path[depth].p_ext;
1865 next = ext4_ext_next_allocated_block(path);
1866
1867 exists = 0;
1868 if (!ex) {
1869 /* there is no extent yet, so try to allocate
1870 * all requested space */
1871 start = block;
1872 end = block + num;
1873 } else if (le32_to_cpu(ex->ee_block) > block) {
1874 /* need to allocate space before found extent */
1875 start = block;
1876 end = le32_to_cpu(ex->ee_block);
1877 if (block + num < end)
1878 end = block + num;
1879 } else if (block >= le32_to_cpu(ex->ee_block)
1880 + ext4_ext_get_actual_len(ex)) {
1881 /* need to allocate space after found extent */
1882 start = block;
1883 end = block + num;
1884 if (end >= next)
1885 end = next;
1886 } else if (block >= le32_to_cpu(ex->ee_block)) {
1887 /*
1888 * some part of requested space is covered
1889 * by found extent
1890 */
1891 start = block;
1892 end = le32_to_cpu(ex->ee_block)
1893 + ext4_ext_get_actual_len(ex);
1894 if (block + num < end)
1895 end = block + num;
1896 exists = 1;
1897 } else {
1898 BUG();
1899 }
1900 BUG_ON(end <= start);
1901
1902 if (!exists) {
1903 cbex.ec_block = start;
1904 cbex.ec_len = end - start;
1905 cbex.ec_start = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04001906 } else {
1907 cbex.ec_block = le32_to_cpu(ex->ee_block);
1908 cbex.ec_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001909 cbex.ec_start = ext4_ext_pblock(ex);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001910 }
1911
Frank Mayhar273df552010-03-02 11:46:09 -05001912 if (unlikely(cbex.ec_len == 0)) {
1913 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1914 err = -EIO;
1915 break;
1916 }
Lukas Czernerc03f8aa2011-06-06 00:06:52 -04001917 err = func(inode, next, &cbex, ex, cbdata);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001918 ext4_ext_drop_refs(path);
1919
1920 if (err < 0)
1921 break;
1922
1923 if (err == EXT_REPEAT)
1924 continue;
1925 else if (err == EXT_BREAK) {
1926 err = 0;
1927 break;
1928 }
1929
1930 if (ext_depth(inode) != depth) {
1931 /* depth was changed. we have to realloc path */
1932 kfree(path);
1933 path = NULL;
1934 }
1935
1936 block = cbex.ec_block + cbex.ec_len;
1937 }
1938
1939 if (path) {
1940 ext4_ext_drop_refs(path);
1941 kfree(path);
1942 }
1943
1944 return err;
1945}
1946
Avantika Mathur09b88252006-12-06 20:41:36 -08001947static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001948ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05001949 __u32 len, ext4_fsblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001950{
1951 struct ext4_ext_cache *cex;
1952 BUG_ON(len == 0);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001953 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Aditya Kalid8990242011-09-09 19:18:51 -04001954 trace_ext4_ext_put_in_cache(inode, block, len, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001955 cex = &EXT4_I(inode)->i_cached_extent;
Alex Tomasa86c6182006-10-11 01:21:03 -07001956 cex->ec_block = block;
1957 cex->ec_len = len;
1958 cex->ec_start = start;
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001959 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001960}
1961
1962/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001963 * ext4_ext_put_gap_in_cache:
1964 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001965 * and cache this gap
1966 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001967static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001968ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001969 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001970{
1971 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001972 unsigned long len;
1973 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001974 struct ext4_extent *ex;
1975
1976 ex = path[depth].p_ext;
1977 if (ex == NULL) {
1978 /* there is no extent yet, so gap is [0;-] */
1979 lblock = 0;
Lukas Czernerf17722f2011-06-06 00:05:17 -04001980 len = EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001981 ext_debug("cache gap(whole file):");
1982 } else if (block < le32_to_cpu(ex->ee_block)) {
1983 lblock = block;
1984 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05001985 ext_debug("cache gap(before): %u [%u:%u]",
1986 block,
1987 le32_to_cpu(ex->ee_block),
1988 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001989 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001990 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001991 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001992 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001993 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001994
1995 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05001996 ext_debug("cache gap(after): [%u:%u] %u",
1997 le32_to_cpu(ex->ee_block),
1998 ext4_ext_get_actual_len(ex),
1999 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002000 BUG_ON(next == lblock);
2001 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002002 } else {
2003 lblock = len = 0;
2004 BUG();
2005 }
2006
Eric Sandeenbba90742008-01-28 23:58:27 -05002007 ext_debug(" -> %u:%lu\n", lblock, len);
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002008 ext4_ext_put_in_cache(inode, lblock, len, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002009}
2010
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002011/*
Robin Dongb7ca1e82011-07-23 21:53:25 -04002012 * ext4_ext_check_cache()
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002013 * Checks to see if the given block is in the cache.
2014 * If it is, the cached extent is stored in the given
2015 * cache extent pointer. If the cached extent is a hole,
2016 * this routine should be used instead of
2017 * ext4_ext_in_cache if the calling function needs to
2018 * know the size of the hole.
2019 *
2020 * @inode: The files inode
2021 * @block: The block to look for in the cache
2022 * @ex: Pointer where the cached extent will be stored
2023 * if it contains block
2024 *
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002025 * Return 0 if cache is invalid; 1 if the cache is valid
2026 */
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002027static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2028 struct ext4_ext_cache *ex){
Alex Tomasa86c6182006-10-11 01:21:03 -07002029 struct ext4_ext_cache *cex;
Vivek Haldar77f41352011-05-22 21:24:16 -04002030 struct ext4_sb_info *sbi;
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002031 int ret = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002032
Theodore Ts'o60e66792010-05-17 07:00:00 -04002033 /*
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002034 * We borrow i_block_reservation_lock to protect i_cached_extent
2035 */
2036 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07002037 cex = &EXT4_I(inode)->i_cached_extent;
Vivek Haldar77f41352011-05-22 21:24:16 -04002038 sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002039
2040 /* has cache valid data? */
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002041 if (cex->ec_len == 0)
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002042 goto errout;
Alex Tomasa86c6182006-10-11 01:21:03 -07002043
Akinobu Mita731eb1a2010-03-03 23:55:01 -05002044 if (in_range(block, cex->ec_block, cex->ec_len)) {
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002045 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
Eric Sandeenbba90742008-01-28 23:58:27 -05002046 ext_debug("%u cached by %u:%u:%llu\n",
2047 block,
2048 cex->ec_block, cex->ec_len, cex->ec_start);
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002049 ret = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07002050 }
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002051errout:
Vivek Haldar77f41352011-05-22 21:24:16 -04002052 if (!ret)
2053 sbi->extent_cache_misses++;
2054 else
2055 sbi->extent_cache_hits++;
Aditya Kalid8990242011-09-09 19:18:51 -04002056 trace_ext4_ext_in_cache(inode, block, ret);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002057 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2058 return ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002059}
2060
2061/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002062 * ext4_ext_in_cache()
2063 * Checks to see if the given block is in the cache.
2064 * If it is, the cached extent is stored in the given
2065 * extent pointer.
2066 *
2067 * @inode: The files inode
2068 * @block: The block to look for in the cache
2069 * @ex: Pointer where the cached extent will be stored
2070 * if it contains block
2071 *
2072 * Return 0 if cache is invalid; 1 if the cache is valid
2073 */
2074static int
2075ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2076 struct ext4_extent *ex)
2077{
2078 struct ext4_ext_cache cex;
2079 int ret = 0;
2080
2081 if (ext4_ext_check_cache(inode, block, &cex)) {
2082 ex->ee_block = cpu_to_le32(cex.ec_block);
2083 ext4_ext_store_pblock(ex, cex.ec_start);
2084 ex->ee_len = cpu_to_le16(cex.ec_len);
2085 ret = 1;
2086 }
2087
2088 return ret;
2089}
2090
2091
2092/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002093 * ext4_ext_rm_idx:
2094 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002095 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002096static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07002097 struct ext4_ext_path *path)
2098{
Alex Tomasa86c6182006-10-11 01:21:03 -07002099 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002100 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002101
2102 /* free index block */
2103 path--;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002104 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002105 if (unlikely(path->p_hdr->eh_entries == 0)) {
2106 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2107 return -EIO;
2108 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002109 err = ext4_ext_get_access(handle, inode, path);
2110 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002111 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002112
2113 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2114 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2115 len *= sizeof(struct ext4_extent_idx);
2116 memmove(path->p_idx, path->p_idx + 1, len);
2117 }
2118
Marcin Slusarze8546d02008-04-17 10:38:59 -04002119 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002120 err = ext4_ext_dirty(handle, inode, path);
2121 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002122 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002123 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002124 trace_ext4_ext_rm_idx(inode, leaf);
2125
Peter Huewe7dc57612011-02-21 21:01:42 -05002126 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002127 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Alex Tomasa86c6182006-10-11 01:21:03 -07002128 return err;
2129}
2130
2131/*
Mingming Caoee12b632008-08-19 22:16:05 -04002132 * ext4_ext_calc_credits_for_single_extent:
2133 * This routine returns max. credits that needed to insert an extent
2134 * to the extent tree.
2135 * When pass the actual path, the caller should calculate credits
2136 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002137 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002138int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002139 struct ext4_ext_path *path)
2140{
Alex Tomasa86c6182006-10-11 01:21:03 -07002141 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002142 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002143 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002144
Alex Tomasa86c6182006-10-11 01:21:03 -07002145 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002146 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002147 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2148
2149 /*
2150 * There are some space in the leaf tree, no
2151 * need to account for leaf block credit
2152 *
2153 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002154 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002155 * accounted.
2156 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002157 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002158 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002159 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002160 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002161 }
2162
Mingming Cao525f4ed2008-08-19 22:15:58 -04002163 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002164}
Alex Tomasa86c6182006-10-11 01:21:03 -07002165
Mingming Caoee12b632008-08-19 22:16:05 -04002166/*
2167 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2168 *
2169 * if nrblocks are fit in a single extent (chunk flag is 1), then
2170 * in the worse case, each tree level index/leaf need to be changed
2171 * if the tree split due to insert a new extent, then the old tree
2172 * index/leaf need to be updated too
2173 *
2174 * If the nrblocks are discontiguous, they could cause
2175 * the whole tree split more than once, but this is really rare.
2176 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002177int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04002178{
2179 int index;
2180 int depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002181
Mingming Caoee12b632008-08-19 22:16:05 -04002182 if (chunk)
2183 index = depth * 2;
2184 else
2185 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002186
Mingming Caoee12b632008-08-19 22:16:05 -04002187 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002188}
2189
2190static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002191 struct ext4_extent *ex,
2192 ext4_fsblk_t *partial_cluster,
2193 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002194{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002195 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002196 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002197 ext4_fsblk_t pblk;
Theodore Ts'oe6362602009-11-23 07:17:05 -05002198 int flags = EXT4_FREE_BLOCKS_FORGET;
Alex Tomasa86c6182006-10-11 01:21:03 -07002199
Alex Tomasc9de5602008-01-29 00:19:52 -05002200 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Theodore Ts'oe6362602009-11-23 07:17:05 -05002201 flags |= EXT4_FREE_BLOCKS_METADATA;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002202 /*
2203 * For bigalloc file systems, we never free a partial cluster
2204 * at the beginning of the extent. Instead, we make a note
2205 * that we tried freeing the cluster, and check to see if we
2206 * need to free it on a subsequent call to ext4_remove_blocks,
2207 * or at the end of the ext4_truncate() operation.
2208 */
2209 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2210
Aditya Kalid8990242011-09-09 19:18:51 -04002211 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002212 /*
2213 * If we have a partial cluster, and it's different from the
2214 * cluster of the last block, we need to explicitly free the
2215 * partial cluster here.
2216 */
2217 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2218 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2219 ext4_free_blocks(handle, inode, NULL,
2220 EXT4_C2B(sbi, *partial_cluster),
2221 sbi->s_cluster_ratio, flags);
2222 *partial_cluster = 0;
2223 }
2224
Alex Tomasa86c6182006-10-11 01:21:03 -07002225#ifdef EXTENTS_STATS
2226 {
2227 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002228 spin_lock(&sbi->s_ext_stats_lock);
2229 sbi->s_ext_blocks += ee_len;
2230 sbi->s_ext_extents++;
2231 if (ee_len < sbi->s_ext_min)
2232 sbi->s_ext_min = ee_len;
2233 if (ee_len > sbi->s_ext_max)
2234 sbi->s_ext_max = ee_len;
2235 if (ext_depth(inode) > sbi->s_depth_max)
2236 sbi->s_depth_max = ext_depth(inode);
2237 spin_unlock(&sbi->s_ext_stats_lock);
2238 }
2239#endif
2240 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002241 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002242 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002243 ext4_lblk_t num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002244
Amit Aroraa2df2a62007-07-17 21:42:41 -04002245 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002246 pblk = ext4_ext_pblock(ex) + ee_len - num;
2247 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2248 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2249 /*
2250 * If the block range to be freed didn't start at the
2251 * beginning of a cluster, and we removed the entire
2252 * extent, save the partial cluster here, since we
2253 * might need to delete if we determine that the
2254 * truncate operation has removed all of the blocks in
2255 * the cluster.
2256 */
2257 if (pblk & (sbi->s_cluster_ratio - 1) &&
2258 (ee_len == num))
2259 *partial_cluster = EXT4_B2C(sbi, pblk);
2260 else
2261 *partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002262 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002263 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002264 /* head removal */
2265 ext4_lblk_t num;
2266 ext4_fsblk_t start;
2267
2268 num = to - from;
2269 start = ext4_ext_pblock(ex);
2270
2271 ext_debug("free first %u blocks starting %llu\n", num, start);
H Hartley Sweetenee90d572011-10-18 11:01:51 -04002272 ext4_free_blocks(handle, inode, NULL, start, num, flags);
Allison Hendersond583fb82011-05-25 07:41:43 -04002273
Alex Tomasa86c6182006-10-11 01:21:03 -07002274 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002275 printk(KERN_INFO "strange request: removal(2) "
2276 "%u-%u from %u:%u\n",
2277 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002278 }
2279 return 0;
2280}
2281
Allison Hendersond583fb82011-05-25 07:41:43 -04002282
2283/*
2284 * ext4_ext_rm_leaf() Removes the extents associated with the
2285 * blocks appearing between "start" and "end", and splits the extents
2286 * if "start" and "end" appear in the same extent
2287 *
2288 * @handle: The journal handle
2289 * @inode: The files inode
2290 * @path: The path to the leaf
2291 * @start: The first block to remove
2292 * @end: The last block to remove
2293 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002294static int
2295ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002296 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2297 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002298{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002299 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002300 int err = 0, correct_index = 0;
2301 int depth = ext_depth(inode), credits;
2302 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002303 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002304 unsigned num;
2305 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002306 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002307 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002308 struct ext4_extent *ex;
2309
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002310 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002311 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002312 if (!path[depth].p_hdr)
2313 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2314 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002315 if (unlikely(path[depth].p_hdr == NULL)) {
2316 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2317 return -EIO;
2318 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002319 /* find where to start removing */
2320 ex = EXT_LAST_EXTENT(eh);
2321
2322 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002323 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002324
Aditya Kalid8990242011-09-09 19:18:51 -04002325 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2326
Alex Tomasa86c6182006-10-11 01:21:03 -07002327 while (ex >= EXT_FIRST_EXTENT(eh) &&
2328 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002329
2330 if (ext4_ext_is_uninitialized(ex))
2331 uninitialized = 1;
2332 else
2333 uninitialized = 0;
2334
Mingming553f9002009-09-18 13:34:55 -04002335 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2336 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002337 path[depth].p_ext = ex;
2338
2339 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002340 b = ex_ee_block+ex_ee_len - 1 < end ?
2341 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002342
2343 ext_debug(" border %u:%u\n", a, b);
2344
Allison Hendersond583fb82011-05-25 07:41:43 -04002345 /* If this extent is beyond the end of the hole, skip it */
2346 if (end <= ex_ee_block) {
2347 ex--;
2348 ex_ee_block = le32_to_cpu(ex->ee_block);
2349 ex_ee_len = ext4_ext_get_actual_len(ex);
2350 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002351 } else if (b != ex_ee_block + ex_ee_len - 1) {
2352 EXT4_ERROR_INODE(inode," bad truncate %u:%u\n",
2353 start, end);
2354 err = -EIO;
2355 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002356 } else if (a != ex_ee_block) {
2357 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002358 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002359 } else {
2360 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002361 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002362 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002363 /*
2364 * 3 for leaf, sb, and inode plus 2 (bmap and group
2365 * descriptor) for each block group; assume two block
2366 * groups plus ex_ee_len/blocks_per_block_group for
2367 * the worst case
2368 */
2369 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002370 if (ex == EXT_FIRST_EXTENT(eh)) {
2371 correct_index = 1;
2372 credits += (ext_depth(inode)) + 1;
2373 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002374 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002375
Jan Kara487caee2009-08-17 22:17:20 -04002376 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002377 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002378 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002379
2380 err = ext4_ext_get_access(handle, inode, path + depth);
2381 if (err)
2382 goto out;
2383
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002384 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2385 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002386 if (err)
2387 goto out;
2388
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002389 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002390 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002391 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002392
Alex Tomasa86c6182006-10-11 01:21:03 -07002393 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002394 /*
2395 * Do not mark uninitialized if all the blocks in the
2396 * extent have been removed.
2397 */
2398 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002399 ext4_ext_mark_uninitialized(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002400 /*
2401 * If the extent was completely released,
2402 * we need to remove it from the leaf
2403 */
2404 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002405 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002406 /*
2407 * For hole punching, we need to scoot all the
2408 * extents up when an extent is removed so that
2409 * we dont have blank extents in the middle
2410 */
2411 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2412 sizeof(struct ext4_extent));
2413
2414 /* Now get rid of the one at the end */
2415 memset(EXT_LAST_EXTENT(eh), 0,
2416 sizeof(struct ext4_extent));
2417 }
2418 le16_add_cpu(&eh->eh_entries, -1);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002419 } else
2420 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002421
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002422 err = ext4_ext_dirty(handle, inode, path + depth);
2423 if (err)
2424 goto out;
2425
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002426 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002427 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002428 ex--;
2429 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002430 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002431 }
2432
2433 if (correct_index && eh->eh_entries)
2434 err = ext4_ext_correct_indexes(handle, inode, path);
2435
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002436 /*
2437 * If there is still a entry in the leaf node, check to see if
2438 * it references the partial cluster. This is the only place
2439 * where it could; if it doesn't, we can free the cluster.
2440 */
2441 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2442 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2443 *partial_cluster)) {
2444 int flags = EXT4_FREE_BLOCKS_FORGET;
2445
2446 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2447 flags |= EXT4_FREE_BLOCKS_METADATA;
2448
2449 ext4_free_blocks(handle, inode, NULL,
2450 EXT4_C2B(sbi, *partial_cluster),
2451 sbi->s_cluster_ratio, flags);
2452 *partial_cluster = 0;
2453 }
2454
Alex Tomasa86c6182006-10-11 01:21:03 -07002455 /* if this leaf is free, then we should
2456 * remove it from index block above */
2457 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2458 err = ext4_ext_rm_idx(handle, inode, path + depth);
2459
2460out:
2461 return err;
2462}
2463
2464/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002465 * ext4_ext_more_to_rm:
2466 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002467 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002468static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002469ext4_ext_more_to_rm(struct ext4_ext_path *path)
2470{
2471 BUG_ON(path->p_idx == NULL);
2472
2473 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2474 return 0;
2475
2476 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002477 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002478 * so we have to consider current index for truncation
2479 */
2480 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2481 return 0;
2482 return 1;
2483}
2484
Allison Hendersonc6a03712011-07-17 23:21:03 -04002485static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07002486{
2487 struct super_block *sb = inode->i_sb;
2488 int depth = ext_depth(inode);
2489 struct ext4_ext_path *path;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002490 ext4_fsblk_t partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002491 handle_t *handle;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002492 int i, err;
Alex Tomasa86c6182006-10-11 01:21:03 -07002493
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002494 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002495
2496 /* probably first extent we're gonna free will be last in block */
2497 handle = ext4_journal_start(inode, depth + 1);
2498 if (IS_ERR(handle))
2499 return PTR_ERR(handle);
2500
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002501again:
Alex Tomasa86c6182006-10-11 01:21:03 -07002502 ext4_ext_invalidate_cache(inode);
2503
Aditya Kalid8990242011-09-09 19:18:51 -04002504 trace_ext4_ext_remove_space(inode, start, depth);
2505
Alex Tomasa86c6182006-10-11 01:21:03 -07002506 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002507 * We start scanning from right side, freeing all the blocks
2508 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002509 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002510 depth = ext_depth(inode);
Josef Bacik216553c2008-04-29 22:02:02 -04002511 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07002512 if (path == NULL) {
2513 ext4_journal_stop(handle);
2514 return -ENOMEM;
2515 }
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002516 path[0].p_depth = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -07002517 path[0].p_hdr = ext_inode_hdr(inode);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04002518 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002519 err = -EIO;
2520 goto out;
2521 }
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002522 i = err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002523
2524 while (i >= 0 && err == 0) {
2525 if (i == depth) {
2526 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002527 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002528 &partial_cluster, start,
2529 EXT_MAX_BLOCKS - 1);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002530 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002531 brelse(path[i].p_bh);
2532 path[i].p_bh = NULL;
2533 i--;
2534 continue;
2535 }
2536
2537 /* this is index block */
2538 if (!path[i].p_hdr) {
2539 ext_debug("initialize header\n");
2540 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002541 }
2542
Alex Tomasa86c6182006-10-11 01:21:03 -07002543 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002544 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002545 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2546 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2547 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2548 path[i].p_hdr,
2549 le16_to_cpu(path[i].p_hdr->eh_entries));
2550 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002551 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002552 path[i].p_idx--;
2553 }
2554
2555 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2556 i, EXT_FIRST_INDEX(path[i].p_hdr),
2557 path[i].p_idx);
2558 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002559 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002560 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002561 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002562 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002563 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'obf89d162010-10-27 21:30:14 -04002564 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002565 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002566 /* should we reset i_size? */
2567 err = -EIO;
2568 break;
2569 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002570 if (WARN_ON(i + 1 > depth)) {
2571 err = -EIO;
2572 break;
2573 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04002574 if (ext4_ext_check(inode, ext_block_hdr(bh),
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002575 depth - i - 1)) {
2576 err = -EIO;
2577 break;
2578 }
2579 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002580
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002581 /* save actual number of indexes since this
2582 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002583 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2584 i++;
2585 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002586 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002587 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002588 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002589 * handle must be already prepared by the
2590 * truncatei_leaf() */
2591 err = ext4_ext_rm_idx(handle, inode, path + i);
2592 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002593 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002594 brelse(path[i].p_bh);
2595 path[i].p_bh = NULL;
2596 i--;
2597 ext_debug("return to level %d\n", i);
2598 }
2599 }
2600
Aditya Kalid8990242011-09-09 19:18:51 -04002601 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2602 path->p_hdr->eh_entries);
2603
Aditya Kali7b415bf2011-09-09 19:04:51 -04002604 /* If we still have something in the partial cluster and we have removed
2605 * even the first extent, then we should free the blocks in the partial
2606 * cluster as well. */
2607 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2608 int flags = EXT4_FREE_BLOCKS_FORGET;
2609
2610 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2611 flags |= EXT4_FREE_BLOCKS_METADATA;
2612
2613 ext4_free_blocks(handle, inode, NULL,
2614 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2615 EXT4_SB(sb)->s_cluster_ratio, flags);
2616 partial_cluster = 0;
2617 }
2618
Alex Tomasa86c6182006-10-11 01:21:03 -07002619 /* TODO: flexible tree reduction should be here */
2620 if (path->p_hdr->eh_entries == 0) {
2621 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002622 * truncate to zero freed all the tree,
2623 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002624 */
2625 err = ext4_ext_get_access(handle, inode, path);
2626 if (err == 0) {
2627 ext_inode_hdr(inode)->eh_depth = 0;
2628 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002629 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002630 err = ext4_ext_dirty(handle, inode, path);
2631 }
2632 }
2633out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002634 ext4_ext_drop_refs(path);
2635 kfree(path);
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002636 if (err == -EAGAIN)
2637 goto again;
Alex Tomasa86c6182006-10-11 01:21:03 -07002638 ext4_journal_stop(handle);
2639
2640 return err;
2641}
2642
2643/*
2644 * called at mount time
2645 */
2646void ext4_ext_init(struct super_block *sb)
2647{
2648 /*
2649 * possible initialization would be here
2650 */
2651
Theodore Ts'o83982b62009-01-06 14:53:16 -05002652 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002653#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o47760042008-09-08 23:00:52 -04002654 printk(KERN_INFO "EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002655#ifdef AGGRESSIVE_TEST
2656 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07002657#endif
2658#ifdef CHECK_BINSEARCH
2659 printk(", check binsearch");
2660#endif
2661#ifdef EXTENTS_STATS
2662 printk(", stats");
2663#endif
2664 printk("\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04002665#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07002666#ifdef EXTENTS_STATS
2667 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2668 EXT4_SB(sb)->s_ext_min = 1 << 30;
2669 EXT4_SB(sb)->s_ext_max = 0;
2670#endif
2671 }
2672}
2673
2674/*
2675 * called at umount time
2676 */
2677void ext4_ext_release(struct super_block *sb)
2678{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002679 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002680 return;
2681
2682#ifdef EXTENTS_STATS
2683 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2684 struct ext4_sb_info *sbi = EXT4_SB(sb);
2685 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2686 sbi->s_ext_blocks, sbi->s_ext_extents,
2687 sbi->s_ext_blocks / sbi->s_ext_extents);
2688 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2689 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2690 }
2691#endif
2692}
2693
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002694/* FIXME!! we need to try to merge to left or right after zero-out */
2695static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2696{
Lukas Czerner24075182010-10-27 21:30:06 -04002697 ext4_fsblk_t ee_pblock;
2698 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04002699 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002700
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002701 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04002702 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002703
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002704 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04002705 if (ret > 0)
2706 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002707
Lukas Czerner24075182010-10-27 21:30:06 -04002708 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002709}
2710
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002711/*
2712 * used by extent splitting.
2713 */
2714#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2715 due to ENOSPC */
2716#define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2717#define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2718
2719/*
2720 * ext4_split_extent_at() splits an extent at given block.
2721 *
2722 * @handle: the journal handle
2723 * @inode: the file inode
2724 * @path: the path to the extent
2725 * @split: the logical block where the extent is splitted.
2726 * @split_flags: indicates if the extent could be zeroout if split fails, and
2727 * the states(init or uninit) of new extents.
2728 * @flags: flags used to insert new extent to extent tree.
2729 *
2730 *
2731 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2732 * of which are deterimined by split_flag.
2733 *
2734 * There are two cases:
2735 * a> the extent are splitted into two extent.
2736 * b> split is not needed, and just mark the extent.
2737 *
2738 * return 0 on success.
2739 */
2740static int ext4_split_extent_at(handle_t *handle,
2741 struct inode *inode,
2742 struct ext4_ext_path *path,
2743 ext4_lblk_t split,
2744 int split_flag,
2745 int flags)
2746{
2747 ext4_fsblk_t newblock;
2748 ext4_lblk_t ee_block;
2749 struct ext4_extent *ex, newex, orig_ex;
2750 struct ext4_extent *ex2 = NULL;
2751 unsigned int ee_len, depth;
2752 int err = 0;
2753
2754 ext_debug("ext4_split_extents_at: inode %lu, logical"
2755 "block %llu\n", inode->i_ino, (unsigned long long)split);
2756
2757 ext4_ext_show_leaf(inode, path);
2758
2759 depth = ext_depth(inode);
2760 ex = path[depth].p_ext;
2761 ee_block = le32_to_cpu(ex->ee_block);
2762 ee_len = ext4_ext_get_actual_len(ex);
2763 newblock = split - ee_block + ext4_ext_pblock(ex);
2764
2765 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2766
2767 err = ext4_ext_get_access(handle, inode, path + depth);
2768 if (err)
2769 goto out;
2770
2771 if (split == ee_block) {
2772 /*
2773 * case b: block @split is the block that the extent begins with
2774 * then we just change the state of the extent, and splitting
2775 * is not needed.
2776 */
2777 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2778 ext4_ext_mark_uninitialized(ex);
2779 else
2780 ext4_ext_mark_initialized(ex);
2781
2782 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2783 ext4_ext_try_to_merge(inode, path, ex);
2784
2785 err = ext4_ext_dirty(handle, inode, path + depth);
2786 goto out;
2787 }
2788
2789 /* case a */
2790 memcpy(&orig_ex, ex, sizeof(orig_ex));
2791 ex->ee_len = cpu_to_le16(split - ee_block);
2792 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2793 ext4_ext_mark_uninitialized(ex);
2794
2795 /*
2796 * path may lead to new leaf, not to original leaf any more
2797 * after ext4_ext_insert_extent() returns,
2798 */
2799 err = ext4_ext_dirty(handle, inode, path + depth);
2800 if (err)
2801 goto fix_extent_len;
2802
2803 ex2 = &newex;
2804 ex2->ee_block = cpu_to_le32(split);
2805 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2806 ext4_ext_store_pblock(ex2, newblock);
2807 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2808 ext4_ext_mark_uninitialized(ex2);
2809
2810 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2811 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2812 err = ext4_ext_zeroout(inode, &orig_ex);
2813 if (err)
2814 goto fix_extent_len;
2815 /* update the extent length and mark as initialized */
2816 ex->ee_len = cpu_to_le32(ee_len);
2817 ext4_ext_try_to_merge(inode, path, ex);
2818 err = ext4_ext_dirty(handle, inode, path + depth);
2819 goto out;
2820 } else if (err)
2821 goto fix_extent_len;
2822
2823out:
2824 ext4_ext_show_leaf(inode, path);
2825 return err;
2826
2827fix_extent_len:
2828 ex->ee_len = orig_ex.ee_len;
2829 ext4_ext_dirty(handle, inode, path + depth);
2830 return err;
2831}
2832
2833/*
2834 * ext4_split_extents() splits an extent and mark extent which is covered
2835 * by @map as split_flags indicates
2836 *
2837 * It may result in splitting the extent into multiple extents (upto three)
2838 * There are three possibilities:
2839 * a> There is no split required
2840 * b> Splits in two extents: Split is happening at either end of the extent
2841 * c> Splits in three extents: Somone is splitting in middle of the extent
2842 *
2843 */
2844static int ext4_split_extent(handle_t *handle,
2845 struct inode *inode,
2846 struct ext4_ext_path *path,
2847 struct ext4_map_blocks *map,
2848 int split_flag,
2849 int flags)
2850{
2851 ext4_lblk_t ee_block;
2852 struct ext4_extent *ex;
2853 unsigned int ee_len, depth;
2854 int err = 0;
2855 int uninitialized;
2856 int split_flag1, flags1;
2857
2858 depth = ext_depth(inode);
2859 ex = path[depth].p_ext;
2860 ee_block = le32_to_cpu(ex->ee_block);
2861 ee_len = ext4_ext_get_actual_len(ex);
2862 uninitialized = ext4_ext_is_uninitialized(ex);
2863
2864 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2865 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2866 EXT4_EXT_MAY_ZEROOUT : 0;
2867 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2868 if (uninitialized)
2869 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2870 EXT4_EXT_MARK_UNINIT2;
2871 err = ext4_split_extent_at(handle, inode, path,
2872 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04002873 if (err)
2874 goto out;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002875 }
2876
2877 ext4_ext_drop_refs(path);
2878 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2879 if (IS_ERR(path))
2880 return PTR_ERR(path);
2881
2882 if (map->m_lblk >= ee_block) {
2883 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2884 EXT4_EXT_MAY_ZEROOUT : 0;
2885 if (uninitialized)
2886 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2887 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2888 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2889 err = ext4_split_extent_at(handle, inode, path,
2890 map->m_lblk, split_flag1, flags);
2891 if (err)
2892 goto out;
2893 }
2894
2895 ext4_ext_show_leaf(inode, path);
2896out:
2897 return err ? err : map->m_len;
2898}
2899
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002900#define EXT4_EXT_ZERO_LEN 7
Amit Arora56055d32007-07-17 21:42:38 -04002901/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002902 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04002903 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002904 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04002905 * uninitialized).
2906 * There are three possibilities:
2907 * a> There is no split required: Entire extent should be initialized
2908 * b> Splits in two extents: Write is happening at either end of the extent
2909 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04002910 *
2911 * Pre-conditions:
2912 * - The extent pointed to by 'path' is uninitialized.
2913 * - The extent pointed to by 'path' contains a superset
2914 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
2915 *
2916 * Post-conditions on success:
2917 * - the returned value is the number of blocks beyond map->l_lblk
2918 * that are allocated and initialized.
2919 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04002920 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002921static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002922 struct inode *inode,
2923 struct ext4_map_blocks *map,
2924 struct ext4_ext_path *path)
Amit Arora56055d32007-07-17 21:42:38 -04002925{
Eric Gouriou6f91bc52011-10-27 11:43:23 -04002926 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04002927 struct ext4_map_blocks split_map;
2928 struct ext4_extent zero_ex;
2929 struct ext4_extent *ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002930 ext4_lblk_t ee_block, eof_block;
Dan Carpenterf85b2872011-10-26 03:42:36 -04002931 unsigned int ee_len, depth;
2932 int allocated;
Amit Arora56055d32007-07-17 21:42:38 -04002933 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04002934 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002935
2936 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2937 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002938 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002939
2940 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2941 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002942 if (eof_block < map->m_lblk + map->m_len)
2943 eof_block = map->m_lblk + map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04002944
2945 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04002946 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04002947 ex = path[depth].p_ext;
2948 ee_block = le32_to_cpu(ex->ee_block);
2949 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04002950 allocated = ee_len - (map->m_lblk - ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04002951
Eric Gouriou6f91bc52011-10-27 11:43:23 -04002952 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
2953
2954 /* Pre-conditions */
2955 BUG_ON(!ext4_ext_is_uninitialized(ex));
2956 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04002957
2958 /*
2959 * Attempt to transfer newly initialized blocks from the currently
2960 * uninitialized extent to its left neighbor. This is much cheaper
2961 * than an insertion followed by a merge as those involve costly
2962 * memmove() calls. This is the common case in steady state for
2963 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
2964 * writes.
2965 *
2966 * Limitations of the current logic:
2967 * - L1: we only deal with writes at the start of the extent.
2968 * The approach could be extended to writes at the end
2969 * of the extent but this scenario was deemed less common.
2970 * - L2: we do not deal with writes covering the whole extent.
2971 * This would require removing the extent if the transfer
2972 * is possible.
2973 * - L3: we only attempt to merge with an extent stored in the
2974 * same extent tree node.
2975 */
2976 if ((map->m_lblk == ee_block) && /*L1*/
2977 (map->m_len < ee_len) && /*L2*/
2978 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
2979 struct ext4_extent *prev_ex;
2980 ext4_lblk_t prev_lblk;
2981 ext4_fsblk_t prev_pblk, ee_pblk;
2982 unsigned int prev_len, write_len;
2983
2984 prev_ex = ex - 1;
2985 prev_lblk = le32_to_cpu(prev_ex->ee_block);
2986 prev_len = ext4_ext_get_actual_len(prev_ex);
2987 prev_pblk = ext4_ext_pblock(prev_ex);
2988 ee_pblk = ext4_ext_pblock(ex);
2989 write_len = map->m_len;
2990
2991 /*
2992 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
2993 * upon those conditions:
2994 * - C1: prev_ex is initialized,
2995 * - C2: prev_ex is logically abutting ex,
2996 * - C3: prev_ex is physically abutting ex,
2997 * - C4: prev_ex can receive the additional blocks without
2998 * overflowing the (initialized) length limit.
2999 */
3000 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3001 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3002 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3003 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3004 err = ext4_ext_get_access(handle, inode, path + depth);
3005 if (err)
3006 goto out;
3007
3008 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3009 map, ex, prev_ex);
3010
3011 /* Shift the start of ex by 'write_len' blocks */
3012 ex->ee_block = cpu_to_le32(ee_block + write_len);
3013 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3014 ex->ee_len = cpu_to_le16(ee_len - write_len);
3015 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3016
3017 /* Extend prev_ex by 'write_len' blocks */
3018 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3019
3020 /* Mark the block containing both extents as dirty */
3021 ext4_ext_dirty(handle, inode, path + depth);
3022
3023 /* Update path to point to the right extent */
3024 path[depth].p_ext = prev_ex;
3025
3026 /* Result: number of initialized blocks past m_lblk */
3027 allocated = write_len;
3028 goto out;
3029 }
3030 }
3031
Yongqiang Yang667eff32011-05-03 12:25:07 -04003032 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003033 /*
3034 * It is safe to convert extent to initialized via explicit
3035 * zeroout only if extent is fully insde i_size or new_size.
3036 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003037 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003038
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003039 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003040 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3041 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3042 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003043 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003044 goto out;
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003045
3046 err = ext4_ext_get_access(handle, inode, path + depth);
3047 if (err)
3048 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003049 ext4_ext_mark_initialized(ex);
3050 ext4_ext_try_to_merge(inode, path, ex);
3051 err = ext4_ext_dirty(handle, inode, path + depth);
3052 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003053 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003054
Amit Arora56055d32007-07-17 21:42:38 -04003055 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003056 * four cases:
3057 * 1. split the extent into three extents.
3058 * 2. split the extent into two extents, zeroout the first half.
3059 * 3. split the extent into two extents, zeroout the second half.
3060 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003061 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003062 split_map.m_lblk = map->m_lblk;
3063 split_map.m_len = map->m_len;
3064
3065 if (allocated > map->m_len) {
3066 if (allocated <= EXT4_EXT_ZERO_LEN &&
3067 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3068 /* case 3 */
3069 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003070 cpu_to_le32(map->m_lblk);
3071 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003072 ext4_ext_store_pblock(&zero_ex,
3073 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3074 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003075 if (err)
3076 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003077 split_map.m_lblk = map->m_lblk;
3078 split_map.m_len = allocated;
3079 } else if ((map->m_lblk - ee_block + map->m_len <
3080 EXT4_EXT_ZERO_LEN) &&
3081 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3082 /* case 2 */
3083 if (map->m_lblk != ee_block) {
3084 zero_ex.ee_block = ex->ee_block;
3085 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3086 ee_block);
3087 ext4_ext_store_pblock(&zero_ex,
3088 ext4_ext_pblock(ex));
3089 err = ext4_ext_zeroout(inode, &zero_ex);
3090 if (err)
3091 goto out;
3092 }
3093
Yongqiang Yang667eff32011-05-03 12:25:07 -04003094 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003095 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3096 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003097 }
3098 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003099
3100 allocated = ext4_split_extent(handle, inode, path,
3101 &split_map, split_flag, 0);
3102 if (allocated < 0)
3103 err = allocated;
3104
Amit Arora56055d32007-07-17 21:42:38 -04003105out:
3106 return err ? err : allocated;
3107}
3108
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003109/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003110 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003111 * ext4_get_blocks_dio_write() when DIO to write
3112 * to an uninitialized extent.
3113 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003114 * Writing to an uninitialized extent may result in splitting the uninitialized
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003115 * extent into multiple /initialized uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003116 * There are three possibilities:
3117 * a> There is no split required: Entire extent should be uninitialized
3118 * b> Splits in two extents: Write is happening at either end of the extent
3119 * c> Splits in three extents: Somone is writing in middle of the extent
3120 *
3121 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003122 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003123 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003124 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003125 * into three uninitialized extent(at most). After IO complete, the part
3126 * being filled will be convert to initialized by the end_io callback function
3127 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003128 *
3129 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003130 */
3131static int ext4_split_unwritten_extents(handle_t *handle,
3132 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003133 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003134 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003135 int flags)
3136{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003137 ext4_lblk_t eof_block;
3138 ext4_lblk_t ee_block;
3139 struct ext4_extent *ex;
3140 unsigned int ee_len;
3141 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003142
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003143 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3144 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003145 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003146
3147 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3148 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003149 if (eof_block < map->m_lblk + map->m_len)
3150 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003151 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003152 * It is safe to convert extent to initialized via explicit
3153 * zeroout only if extent is fully insde i_size or new_size.
3154 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003155 depth = ext_depth(inode);
3156 ex = path[depth].p_ext;
3157 ee_block = le32_to_cpu(ex->ee_block);
3158 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003159
Yongqiang Yang667eff32011-05-03 12:25:07 -04003160 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3161 split_flag |= EXT4_EXT_MARK_UNINIT2;
Mingming Cao00314622009-09-28 15:49:08 -04003162
Yongqiang Yang667eff32011-05-03 12:25:07 -04003163 flags |= EXT4_GET_BLOCKS_PRE_IO;
3164 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003165}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003166
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003167static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Mingming Cao00314622009-09-28 15:49:08 -04003168 struct inode *inode,
3169 struct ext4_ext_path *path)
3170{
3171 struct ext4_extent *ex;
Mingming Cao00314622009-09-28 15:49:08 -04003172 int depth;
3173 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003174
3175 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003176 ex = path[depth].p_ext;
3177
Yongqiang Yang197217a2011-05-03 11:45:29 -04003178 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3179 "block %llu, max_blocks %u\n", inode->i_ino,
3180 (unsigned long long)le32_to_cpu(ex->ee_block),
3181 ext4_ext_get_actual_len(ex));
3182
Mingming Cao00314622009-09-28 15:49:08 -04003183 err = ext4_ext_get_access(handle, inode, path + depth);
3184 if (err)
3185 goto out;
3186 /* first mark the extent as initialized */
3187 ext4_ext_mark_initialized(ex);
3188
Yongqiang Yang197217a2011-05-03 11:45:29 -04003189 /* note: ext4_ext_correct_indexes() isn't needed here because
3190 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003191 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04003192 ext4_ext_try_to_merge(inode, path, ex);
3193
Mingming Cao00314622009-09-28 15:49:08 -04003194 /* Mark modified extent as dirty */
3195 err = ext4_ext_dirty(handle, inode, path + depth);
3196out:
3197 ext4_ext_show_leaf(inode, path);
3198 return err;
3199}
3200
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003201static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3202 sector_t block, int count)
3203{
3204 int i;
3205 for (i = 0; i < count; i++)
3206 unmap_underlying_metadata(bdev, block + i);
3207}
3208
Theodore Ts'o58590b02010-10-27 21:23:12 -04003209/*
3210 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3211 */
3212static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003213 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003214 struct ext4_ext_path *path,
3215 unsigned int len)
3216{
3217 int i, depth;
3218 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003219 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003220
3221 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3222 return 0;
3223
3224 depth = ext_depth(inode);
3225 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003226
3227 if (unlikely(!eh->eh_entries)) {
3228 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3229 "EOFBLOCKS_FL set");
3230 return -EIO;
3231 }
3232 last_ex = EXT_LAST_EXTENT(eh);
3233 /*
3234 * We should clear the EOFBLOCKS_FL flag if we are writing the
3235 * last block in the last extent in the file. We test this by
3236 * first checking to see if the caller to
3237 * ext4_ext_get_blocks() was interested in the last block (or
3238 * a block beyond the last block) in the current extent. If
3239 * this turns out to be false, we can bail out from this
3240 * function immediately.
3241 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003242 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003243 ext4_ext_get_actual_len(last_ex))
3244 return 0;
3245 /*
3246 * If the caller does appear to be planning to write at or
3247 * beyond the end of the current extent, we then test to see
3248 * if the current extent is the last extent in the file, by
3249 * checking to make sure it was reached via the rightmost node
3250 * at each level of the tree.
3251 */
3252 for (i = depth-1; i >= 0; i--)
3253 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3254 return 0;
3255 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3256 return ext4_mark_inode_dirty(handle, inode);
3257}
3258
Aditya Kali7b415bf2011-09-09 19:04:51 -04003259/**
3260 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3261 *
3262 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3263 * whether there are any buffers marked for delayed allocation. It returns '1'
3264 * on the first delalloc'ed buffer head found. If no buffer head in the given
3265 * range is marked for delalloc, it returns 0.
3266 * lblk_start should always be <= lblk_end.
3267 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3268 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3269 * block sooner). This is useful when blocks are truncated sequentially from
3270 * lblk_start towards lblk_end.
3271 */
3272static int ext4_find_delalloc_range(struct inode *inode,
3273 ext4_lblk_t lblk_start,
3274 ext4_lblk_t lblk_end,
3275 int search_hint_reverse)
3276{
3277 struct address_space *mapping = inode->i_mapping;
3278 struct buffer_head *head, *bh = NULL;
3279 struct page *page;
3280 ext4_lblk_t i, pg_lblk;
3281 pgoff_t index;
3282
Robin Dong8c48f7e2011-12-18 23:05:43 -05003283 if (!test_opt(inode->i_sb, DELALLOC))
3284 return 0;
3285
Aditya Kali7b415bf2011-09-09 19:04:51 -04003286 /* reverse search wont work if fs block size is less than page size */
3287 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3288 search_hint_reverse = 0;
3289
3290 if (search_hint_reverse)
3291 i = lblk_end;
3292 else
3293 i = lblk_start;
3294
3295 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3296
3297 while ((i >= lblk_start) && (i <= lblk_end)) {
3298 page = find_get_page(mapping, index);
Aditya Kali5356f262011-09-09 19:20:51 -04003299 if (!page)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003300 goto nextpage;
3301
Aditya Kali7b415bf2011-09-09 19:04:51 -04003302 if (!page_has_buffers(page))
3303 goto nextpage;
3304
3305 head = page_buffers(page);
3306 if (!head)
3307 goto nextpage;
3308
3309 bh = head;
3310 pg_lblk = index << (PAGE_CACHE_SHIFT -
3311 inode->i_blkbits);
3312 do {
3313 if (unlikely(pg_lblk < lblk_start)) {
3314 /*
3315 * This is possible when fs block size is less
3316 * than page size and our cluster starts/ends in
3317 * middle of the page. So we need to skip the
3318 * initial few blocks till we reach the 'lblk'
3319 */
3320 pg_lblk++;
3321 continue;
3322 }
3323
Aditya Kali5356f262011-09-09 19:20:51 -04003324 /* Check if the buffer is delayed allocated and that it
3325 * is not yet mapped. (when da-buffers are mapped during
3326 * their writeout, their da_mapped bit is set.)
3327 */
3328 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04003329 page_cache_release(page);
Aditya Kalid8990242011-09-09 19:18:51 -04003330 trace_ext4_find_delalloc_range(inode,
3331 lblk_start, lblk_end,
3332 search_hint_reverse,
3333 1, i);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003334 return 1;
3335 }
3336 if (search_hint_reverse)
3337 i--;
3338 else
3339 i++;
3340 } while ((i >= lblk_start) && (i <= lblk_end) &&
3341 ((bh = bh->b_this_page) != head));
3342nextpage:
3343 if (page)
3344 page_cache_release(page);
3345 /*
3346 * Move to next page. 'i' will be the first lblk in the next
3347 * page.
3348 */
3349 if (search_hint_reverse)
3350 index--;
3351 else
3352 index++;
3353 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3354 }
3355
Aditya Kalid8990242011-09-09 19:18:51 -04003356 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3357 search_hint_reverse, 0, 0);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003358 return 0;
3359}
3360
3361int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3362 int search_hint_reverse)
3363{
3364 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3365 ext4_lblk_t lblk_start, lblk_end;
3366 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3367 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3368
3369 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3370 search_hint_reverse);
3371}
3372
3373/**
3374 * Determines how many complete clusters (out of those specified by the 'map')
3375 * are under delalloc and were reserved quota for.
3376 * This function is called when we are writing out the blocks that were
3377 * originally written with their allocation delayed, but then the space was
3378 * allocated using fallocate() before the delayed allocation could be resolved.
3379 * The cases to look for are:
3380 * ('=' indicated delayed allocated blocks
3381 * '-' indicates non-delayed allocated blocks)
3382 * (a) partial clusters towards beginning and/or end outside of allocated range
3383 * are not delalloc'ed.
3384 * Ex:
3385 * |----c---=|====c====|====c====|===-c----|
3386 * |++++++ allocated ++++++|
3387 * ==> 4 complete clusters in above example
3388 *
3389 * (b) partial cluster (outside of allocated range) towards either end is
3390 * marked for delayed allocation. In this case, we will exclude that
3391 * cluster.
3392 * Ex:
3393 * |----====c========|========c========|
3394 * |++++++ allocated ++++++|
3395 * ==> 1 complete clusters in above example
3396 *
3397 * Ex:
3398 * |================c================|
3399 * |++++++ allocated ++++++|
3400 * ==> 0 complete clusters in above example
3401 *
3402 * The ext4_da_update_reserve_space will be called only if we
3403 * determine here that there were some "entire" clusters that span
3404 * this 'allocated' range.
3405 * In the non-bigalloc case, this function will just end up returning num_blks
3406 * without ever calling ext4_find_delalloc_range.
3407 */
3408static unsigned int
3409get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3410 unsigned int num_blks)
3411{
3412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3413 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3414 ext4_lblk_t lblk_from, lblk_to, c_offset;
3415 unsigned int allocated_clusters = 0;
3416
3417 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3418 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3419
3420 /* max possible clusters for this allocation */
3421 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3422
Aditya Kalid8990242011-09-09 19:18:51 -04003423 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3424
Aditya Kali7b415bf2011-09-09 19:04:51 -04003425 /* Check towards left side */
3426 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3427 if (c_offset) {
3428 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3429 lblk_to = lblk_from + c_offset - 1;
3430
3431 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3432 allocated_clusters--;
3433 }
3434
3435 /* Now check towards right. */
3436 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3437 if (allocated_clusters && c_offset) {
3438 lblk_from = lblk_start + num_blks;
3439 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3440
3441 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3442 allocated_clusters--;
3443 }
3444
3445 return allocated_clusters;
3446}
3447
Mingming Cao00314622009-09-28 15:49:08 -04003448static int
3449ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003450 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003451 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003452 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003453{
3454 int ret = 0;
3455 int err = 0;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003456 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
Mingming Cao00314622009-09-28 15:49:08 -04003457
Zheng Liu88635ca2011-12-28 19:00:25 -05003458 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3459 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003460 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003461 flags, allocated);
3462 ext4_ext_show_leaf(inode, path);
3463
Aditya Kalid8990242011-09-09 19:18:51 -04003464 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3465 newblock);
3466
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003467 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003468 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003469 ret = ext4_split_unwritten_extents(handle, inode, map,
3470 path, flags);
Mingming5f524952009-11-10 10:48:04 -05003471 /*
3472 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003473 * that this IO needs to conversion to written when IO is
Mingming5f524952009-11-10 10:48:04 -05003474 * completed
3475 */
Tao Ma0edeb712011-10-31 17:30:44 -04003476 if (io)
3477 ext4_set_io_unwritten_flag(inode, io);
3478 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003479 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003480 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003481 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003482 goto out;
3483 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003484 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003485 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003486 ret = ext4_convert_unwritten_extents_endio(handle, inode,
Mingming Cao00314622009-09-28 15:49:08 -04003487 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003488 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003489 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003490 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3491 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003492 } else
3493 err = ret;
Mingming Cao00314622009-09-28 15:49:08 -04003494 goto out2;
3495 }
3496 /* buffered IO case */
3497 /*
3498 * repeat fallocate creation request
3499 * we already have an unwritten extent
3500 */
3501 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3502 goto map_out;
3503
3504 /* buffered READ or buffered write_begin() lookup */
3505 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3506 /*
3507 * We have blocks reserved already. We
3508 * return allocated blocks so that delalloc
3509 * won't do block reservation for us. But
3510 * the buffer head will be unmapped so that
3511 * a read from the block returns 0s.
3512 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003513 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003514 goto out1;
3515 }
3516
3517 /* buffered write, writepage time, convert*/
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003518 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003519 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05003520 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04003521out:
3522 if (ret <= 0) {
3523 err = ret;
3524 goto out2;
3525 } else
3526 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003527 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003528 /*
3529 * if we allocated more blocks than requested
3530 * we need to make sure we unmap the extra block
3531 * allocated. The actual needed block will get
3532 * unmapped later when we find the buffer_head marked
3533 * new.
3534 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003535 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003536 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003537 newblock + map->m_len,
3538 allocated - map->m_len);
3539 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003540 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003541
3542 /*
3543 * If we have done fallocate with the offset that is already
3544 * delayed allocated, we would have block reservation
3545 * and quota reservation done in the delayed write path.
3546 * But fallocate would have already updated quota and block
3547 * count for this offset. So cancel these reservation
3548 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003549 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3550 unsigned int reserved_clusters;
3551 reserved_clusters = get_reserved_cluster_alloc(inode,
3552 map->m_lblk, map->m_len);
3553 if (reserved_clusters)
3554 ext4_da_update_reserve_space(inode,
3555 reserved_clusters,
3556 0);
3557 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003558
Mingming Cao00314622009-09-28 15:49:08 -04003559map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003560 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003561 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3562 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3563 map->m_len);
3564 if (err < 0)
3565 goto out2;
3566 }
Mingming Cao00314622009-09-28 15:49:08 -04003567out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003568 if (allocated > map->m_len)
3569 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003570 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003571 map->m_pblk = newblock;
3572 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003573out2:
3574 if (path) {
3575 ext4_ext_drop_refs(path);
3576 kfree(path);
3577 }
3578 return err ? err : allocated;
3579}
Theodore Ts'o58590b02010-10-27 21:23:12 -04003580
Mingming Cao00314622009-09-28 15:49:08 -04003581/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003582 * get_implied_cluster_alloc - check to see if the requested
3583 * allocation (in the map structure) overlaps with a cluster already
3584 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04003585 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003586 * @map The requested lblk->pblk mapping
3587 * @ex The extent structure which might contain an implied
3588 * cluster allocation
3589 *
3590 * This function is called by ext4_ext_map_blocks() after we failed to
3591 * find blocks that were already in the inode's extent tree. Hence,
3592 * we know that the beginning of the requested region cannot overlap
3593 * the extent from the inode's extent tree. There are three cases we
3594 * want to catch. The first is this case:
3595 *
3596 * |--- cluster # N--|
3597 * |--- extent ---| |---- requested region ---|
3598 * |==========|
3599 *
3600 * The second case that we need to test for is this one:
3601 *
3602 * |--------- cluster # N ----------------|
3603 * |--- requested region --| |------- extent ----|
3604 * |=======================|
3605 *
3606 * The third case is when the requested region lies between two extents
3607 * within the same cluster:
3608 * |------------- cluster # N-------------|
3609 * |----- ex -----| |---- ex_right ----|
3610 * |------ requested region ------|
3611 * |================|
3612 *
3613 * In each of the above cases, we need to set the map->m_pblk and
3614 * map->m_len so it corresponds to the return the extent labelled as
3615 * "|====|" from cluster #N, since it is already in use for data in
3616 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3617 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3618 * as a new "allocated" block region. Otherwise, we will return 0 and
3619 * ext4_ext_map_blocks() will then allocate one or more new clusters
3620 * by calling ext4_mb_new_blocks().
3621 */
Aditya Kalid8990242011-09-09 19:18:51 -04003622static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003623 struct ext4_map_blocks *map,
3624 struct ext4_extent *ex,
3625 struct ext4_ext_path *path)
3626{
Aditya Kalid8990242011-09-09 19:18:51 -04003627 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003628 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3629 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05003630 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003631 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3632 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3633 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3634
3635 /* The extent passed in that we are trying to match */
3636 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3637 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3638
3639 /* The requested region passed into ext4_map_blocks() */
3640 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003641
3642 if ((rr_cluster_start == ex_cluster_end) ||
3643 (rr_cluster_start == ex_cluster_start)) {
3644 if (rr_cluster_start == ex_cluster_end)
3645 ee_start += ee_len - 1;
3646 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3647 c_offset;
3648 map->m_len = min(map->m_len,
3649 (unsigned) sbi->s_cluster_ratio - c_offset);
3650 /*
3651 * Check for and handle this case:
3652 *
3653 * |--------- cluster # N-------------|
3654 * |------- extent ----|
3655 * |--- requested region ---|
3656 * |===========|
3657 */
3658
3659 if (map->m_lblk < ee_block)
3660 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3661
3662 /*
3663 * Check for the case where there is already another allocated
3664 * block to the right of 'ex' but before the end of the cluster.
3665 *
3666 * |------------- cluster # N-------------|
3667 * |----- ex -----| |---- ex_right ----|
3668 * |------ requested region ------|
3669 * |================|
3670 */
3671 if (map->m_lblk > ee_block) {
3672 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3673 map->m_len = min(map->m_len, next - map->m_lblk);
3674 }
Aditya Kalid8990242011-09-09 19:18:51 -04003675
3676 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003677 return 1;
3678 }
Aditya Kalid8990242011-09-09 19:18:51 -04003679
3680 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003681 return 0;
3682}
3683
3684
3685/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05003686 * Block allocation/map/preallocation routine for extents based files
3687 *
3688 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003689 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05003690 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3691 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05003692 *
3693 * return > 0, number of of blocks already mapped/allocated
3694 * if create == 0 and these are pre-allocated blocks
3695 * buffer head is unmapped
3696 * otherwise blocks are mapped
3697 *
3698 * return = 0, if plain look up failed (blocks have not been allocated)
3699 * buffer head is unmapped
3700 *
3701 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003702 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003703int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3704 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07003705{
3706 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003707 struct ext4_extent newex, *ex, *ex2;
3708 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003709 ext4_fsblk_t newblock = 0;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003710 int free_on_err = 0, err = 0, depth, ret;
3711 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04003712 unsigned int allocated_clusters = 0;
Allison Hendersone8613042011-05-25 07:41:46 -04003713 unsigned int punched_out = 0;
3714 unsigned int result = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05003715 struct ext4_allocation_request ar;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003716 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003717 ext4_lblk_t cluster_offset;
Alex Tomasa86c6182006-10-11 01:21:03 -07003718
Mingming84fe3be2009-09-01 08:44:37 -04003719 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003720 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003721 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07003722
3723 /* check in cache */
Robin Dong015861b2011-07-17 23:27:43 -04003724 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3725 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003726 if (!newex.ee_start_lo && !newex.ee_start_hi) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04003727 if ((sbi->s_cluster_ratio > 1) &&
3728 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3729 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3730
Theodore Ts'oc2177052009-05-14 00:58:52 -04003731 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04003732 /*
3733 * block isn't allocated yet and
3734 * user doesn't want to allocate it
3735 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003736 goto out2;
3737 }
3738 /* we should allocate requested block */
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003739 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07003740 /* block is already allocated */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003741 if (sbi->s_cluster_ratio > 1)
3742 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003743 newblock = map->m_lblk
Dave Kleikamp8c55e202007-05-24 13:04:54 -04003744 - le32_to_cpu(newex.ee_block)
Theodore Ts'obf89d162010-10-27 21:30:14 -04003745 + ext4_ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003746 /* number of remaining blocks in the extent */
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05003747 allocated = ext4_ext_get_actual_len(&newex) -
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003748 (map->m_lblk - le32_to_cpu(newex.ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -07003749 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07003750 }
3751 }
3752
3753 /* find extent for this block */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003754 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07003755 if (IS_ERR(path)) {
3756 err = PTR_ERR(path);
3757 path = NULL;
3758 goto out2;
3759 }
3760
3761 depth = ext_depth(inode);
3762
3763 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003764 * consistent leaf must not be empty;
3765 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07003766 * this is why assert can't be put in ext4_ext_find_extent()
3767 */
Frank Mayhar273df552010-03-02 11:46:09 -05003768 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3769 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04003770 "lblock: %lu, depth: %d pblock %lld",
3771 (unsigned long) map->m_lblk, depth,
3772 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05003773 err = -EIO;
3774 goto out2;
3775 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003776
Avantika Mathur7e028972006-12-06 20:41:33 -08003777 ex = path[depth].p_ext;
3778 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003779 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003780 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04003781 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003782
3783 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003784 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04003785 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003786 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04003787 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04003788
3789 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3790
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003791 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003792 if (in_range(map->m_lblk, ee_block, ee_len)) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04003793 struct ext4_map_blocks punch_map;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04003794 ext4_fsblk_t partial_cluster = 0;
3795
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003796 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003797 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003798 allocated = ee_len - (map->m_lblk - ee_block);
3799 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3800 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04003801
Allison Hendersone8613042011-05-25 07:41:46 -04003802 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3803 /*
3804 * Do not put uninitialized extent
3805 * in the cache
3806 */
3807 if (!ext4_ext_is_uninitialized(ex)) {
3808 ext4_ext_put_in_cache(inode, ee_block,
3809 ee_len, ee_start);
3810 goto out;
3811 }
3812 ret = ext4_ext_handle_uninitialized_extents(
3813 handle, inode, map, path, flags,
3814 allocated, newblock);
3815 return ret;
Amit Arora56055d32007-07-17 21:42:38 -04003816 }
Allison Hendersone8613042011-05-25 07:41:46 -04003817
3818 /*
3819 * Punch out the map length, but only to the
3820 * end of the extent
3821 */
3822 punched_out = allocated < map->m_len ?
3823 allocated : map->m_len;
3824
3825 /*
3826 * Sense extents need to be converted to
3827 * uninitialized, they must fit in an
3828 * uninitialized extent
3829 */
3830 if (punched_out > EXT_UNINIT_MAX_LEN)
3831 punched_out = EXT_UNINIT_MAX_LEN;
3832
3833 punch_map.m_lblk = map->m_lblk;
3834 punch_map.m_pblk = newblock;
3835 punch_map.m_len = punched_out;
3836 punch_map.m_flags = 0;
3837
3838 /* Check to see if the extent needs to be split */
3839 if (punch_map.m_len != ee_len ||
3840 punch_map.m_lblk != ee_block) {
3841
3842 ret = ext4_split_extent(handle, inode,
3843 path, &punch_map, 0,
3844 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3845 EXT4_GET_BLOCKS_PRE_IO);
3846
3847 if (ret < 0) {
3848 err = ret;
3849 goto out2;
3850 }
3851 /*
3852 * find extent for the block at
3853 * the start of the hole
3854 */
3855 ext4_ext_drop_refs(path);
3856 kfree(path);
3857
3858 path = ext4_ext_find_extent(inode,
3859 map->m_lblk, NULL);
3860 if (IS_ERR(path)) {
3861 err = PTR_ERR(path);
3862 path = NULL;
3863 goto out2;
3864 }
3865
3866 depth = ext_depth(inode);
3867 ex = path[depth].p_ext;
3868 ee_len = ext4_ext_get_actual_len(ex);
3869 ee_block = le32_to_cpu(ex->ee_block);
3870 ee_start = ext4_ext_pblock(ex);
3871
3872 }
3873
3874 ext4_ext_mark_uninitialized(ex);
3875
Allison Hendersonf7d0d372011-07-17 23:17:02 -04003876 ext4_ext_invalidate_cache(inode);
3877
3878 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04003879 &partial_cluster, map->m_lblk,
3880 map->m_lblk + punched_out);
Allison Hendersonf7d0d372011-07-17 23:17:02 -04003881
3882 if (!err && path->p_hdr->eh_entries == 0) {
3883 /*
3884 * Punch hole freed all of this sub tree,
3885 * so we need to correct eh_depth
3886 */
3887 err = ext4_ext_get_access(handle, inode, path);
3888 if (err == 0) {
3889 ext_inode_hdr(inode)->eh_depth = 0;
3890 ext_inode_hdr(inode)->eh_max =
3891 cpu_to_le16(ext4_ext_space_root(
3892 inode, 0));
3893
3894 err = ext4_ext_dirty(
3895 handle, inode, path);
3896 }
3897 }
Allison Hendersone8613042011-05-25 07:41:46 -04003898
3899 goto out2;
Alex Tomasa86c6182006-10-11 01:21:03 -07003900 }
3901 }
3902
Aditya Kali7b415bf2011-09-09 19:04:51 -04003903 if ((sbi->s_cluster_ratio > 1) &&
3904 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3905 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3906
Alex Tomasa86c6182006-10-11 01:21:03 -07003907 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003908 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07003909 * we couldn't try to create block if create flag is zero
3910 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04003911 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04003912 /*
3913 * put just found gap into cache to speed up
3914 * subsequent requests
3915 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003916 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07003917 goto out2;
3918 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003919
Alex Tomasa86c6182006-10-11 01:21:03 -07003920 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04003921 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07003922 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003923 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003924 newex.ee_block = cpu_to_le32(map->m_lblk);
3925 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3926
3927 /*
3928 * If we are doing bigalloc, check to see if the extent returned
3929 * by ext4_ext_find_extent() implies a cluster we can use.
3930 */
3931 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04003932 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003933 ar.len = allocated = map->m_len;
3934 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003935 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003936 goto got_allocated_blocks;
3937 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003938
Alex Tomasc9de5602008-01-29 00:19:52 -05003939 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003940 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05003941 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3942 if (err)
3943 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003944 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003945 ex2 = NULL;
3946 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05003947 if (err)
3948 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04003949
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003950 /* Check if the extent after searching to the right implies a
3951 * cluster we can use. */
3952 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04003953 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003954 ar.len = allocated = map->m_len;
3955 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003956 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003957 goto got_allocated_blocks;
3958 }
3959
Amit Arora749269f2007-07-18 09:02:56 -04003960 /*
3961 * See if request is beyond maximum number of blocks we can have in
3962 * a single extent. For an initialized extent this limit is
3963 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3964 * EXT_UNINIT_MAX_LEN.
3965 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003966 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04003967 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003968 map->m_len = EXT_INIT_MAX_LEN;
3969 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04003970 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003971 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04003972
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003973 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003974 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003975 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04003976 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05003977 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04003978 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003979 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05003980
3981 /* allocate new block */
3982 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003983 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3984 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003985 /*
3986 * We calculate the offset from the beginning of the cluster
3987 * for the logical block number, since when we allocate a
3988 * physical cluster, the physical block should start at the
3989 * same offset from the beginning of the cluster. This is
3990 * needed so that future calls to get_implied_cluster_alloc()
3991 * work correctly.
3992 */
3993 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3994 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3995 ar.goal -= offset;
3996 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05003997 if (S_ISREG(inode->i_mode))
3998 ar.flags = EXT4_MB_HINT_DATA;
3999 else
4000 /* disable in-core preallocation for non-regular files */
4001 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004002 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4003 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004004 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004005 if (!newblock)
4006 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004007 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004008 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004009 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004010 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004011 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4012 if (ar.len > allocated)
4013 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004014
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004015got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004016 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004017 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004018 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004019 /* Mark uninitialized */
4020 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04004021 ext4_ext_mark_uninitialized(&newex);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004022 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004023 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004024 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004025 * here we flag the IO that really needs the conversion.
Mingming5f524952009-11-10 10:48:04 -05004026 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004027 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004028 */
Jiaying Zhang744692d2010-03-04 16:14:02 -05004029 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Tao Ma0edeb712011-10-31 17:30:44 -04004030 if (io)
4031 ext4_set_io_unwritten_flag(inode, io);
4032 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004033 ext4_set_inode_state(inode,
4034 EXT4_STATE_DIO_UNWRITTEN);
Mingming5f524952009-11-10 10:48:04 -05004035 }
Jiaying Zhang744692d2010-03-04 16:14:02 -05004036 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004037 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004038 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004039
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004040 err = 0;
4041 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4042 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4043 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004044 if (!err)
4045 err = ext4_ext_insert_extent(handle, inode, path,
4046 &newex, flags);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004047 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004048 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4049 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004050 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004051 /* not a good idea to call discard here directly,
4052 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004053 ext4_discard_preallocations(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004054 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
Maxim Patlasov7132de72011-07-10 19:37:48 -04004055 ext4_ext_get_actual_len(&newex), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004056 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004057 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004058
Alex Tomasa86c6182006-10-11 01:21:03 -07004059 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004060 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004061 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004062 if (allocated > map->m_len)
4063 allocated = map->m_len;
4064 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004065
Jan Karab436b9b2009-12-08 23:51:10 -05004066 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004067 * Update reserved blocks/metadata blocks after successful
4068 * block allocation which had been deferred till now.
4069 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004070 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004071 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004072 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004073 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004074 */
4075 reserved_clusters = get_reserved_cluster_alloc(inode,
4076 map->m_lblk, allocated);
4077 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4078 if (reserved_clusters) {
4079 /*
4080 * We have clusters reserved for this range.
4081 * But since we are not doing actual allocation
4082 * and are simply using blocks from previously
4083 * allocated cluster, we should release the
4084 * reservation and not claim quota.
4085 */
4086 ext4_da_update_reserve_space(inode,
4087 reserved_clusters, 0);
4088 }
4089 } else {
4090 BUG_ON(allocated_clusters < reserved_clusters);
4091 /* We will claim quota for all newly allocated blocks.*/
4092 ext4_da_update_reserve_space(inode, allocated_clusters,
4093 1);
4094 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f262011-09-09 19:20:51 -04004095 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004096 int reservation = allocated_clusters -
4097 reserved_clusters;
4098 /*
4099 * It seems we claimed few clusters outside of
4100 * the range of this allocation. We should give
4101 * it back to the reservation pool. This can
4102 * happen in the following case:
4103 *
4104 * * Suppose s_cluster_ratio is 4 (i.e., each
4105 * cluster has 4 blocks. Thus, the clusters
4106 * are [0-3],[4-7],[8-11]...
4107 * * First comes delayed allocation write for
4108 * logical blocks 10 & 11. Since there were no
4109 * previous delayed allocated blocks in the
4110 * range [8-11], we would reserve 1 cluster
4111 * for this write.
4112 * * Next comes write for logical blocks 3 to 8.
4113 * In this case, we will reserve 2 clusters
4114 * (for [0-3] and [4-7]; and not for [8-11] as
4115 * that range has a delayed allocated blocks.
4116 * Thus total reserved clusters now becomes 3.
4117 * * Now, during the delayed allocation writeout
4118 * time, we will first write blocks [3-8] and
4119 * allocate 3 clusters for writing these
4120 * blocks. Also, we would claim all these
4121 * three clusters above.
4122 * * Now when we come here to writeout the
4123 * blocks [10-11], we would expect to claim
4124 * the reservation of 1 cluster we had made
4125 * (and we would claim it since there are no
4126 * more delayed allocated blocks in the range
4127 * [8-11]. But our reserved cluster count had
4128 * already gone to 0.
4129 *
4130 * Thus, at the step 4 above when we determine
4131 * that there are still some unwritten delayed
4132 * allocated blocks outside of our current
4133 * block range, we should increment the
4134 * reserved clusters count so that when the
4135 * remaining blocks finally gets written, we
4136 * could claim them.
4137 */
Aditya Kali5356f262011-09-09 19:20:51 -04004138 dquot_reserve_block(inode,
4139 EXT4_C2B(sbi, reservation));
4140 spin_lock(&ei->i_block_reservation_lock);
4141 ei->i_reserved_data_blocks += reservation;
4142 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004143 }
4144 }
4145 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004146
4147 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004148 * Cache the extent and update transaction to commit on fdatasync only
4149 * when it is _not_ an uninitialized extent.
4150 */
4151 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05004152 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
Jan Karab436b9b2009-12-08 23:51:10 -05004153 ext4_update_inode_fsync_trans(handle, inode, 1);
4154 } else
4155 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004156out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004157 if (allocated > map->m_len)
4158 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004159 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004160 map->m_flags |= EXT4_MAP_MAPPED;
4161 map->m_pblk = newblock;
4162 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004163out2:
4164 if (path) {
4165 ext4_ext_drop_refs(path);
4166 kfree(path);
4167 }
Allison Hendersone8613042011-05-25 07:41:46 -04004168 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4169 punched_out : allocated;
4170
Yongqiang Yange7b319e2011-10-29 09:39:51 -04004171 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4172 newblock, map->m_len, err ? err : result);
4173
Allison Hendersone8613042011-05-25 07:41:46 -04004174 return err ? err : result;
Alex Tomasa86c6182006-10-11 01:21:03 -07004175}
4176
Jan Karacf108bc2008-07-11 19:27:31 -04004177void ext4_ext_truncate(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004178{
4179 struct address_space *mapping = inode->i_mapping;
4180 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004181 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004182 handle_t *handle;
Allison Henderson189e8682011-09-06 21:49:44 -04004183 loff_t page_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004184 int err = 0;
4185
4186 /*
Jiaying Zhang3889fd52011-01-10 12:47:05 -05004187 * finish any pending end_io work so we won't run the risk of
4188 * converting any truncated blocks to initialized later
4189 */
4190 ext4_flush_completed_IO(inode);
4191
4192 /*
Alex Tomasa86c6182006-10-11 01:21:03 -07004193 * probably first extent we're gonna free will be last in block
4194 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004195 err = ext4_writepage_trans_blocks(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07004196 handle = ext4_journal_start(inode, err);
Jan Karacf108bc2008-07-11 19:27:31 -04004197 if (IS_ERR(handle))
Alex Tomasa86c6182006-10-11 01:21:03 -07004198 return;
Alex Tomasa86c6182006-10-11 01:21:03 -07004199
Allison Henderson189e8682011-09-06 21:49:44 -04004200 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4201 page_len = PAGE_CACHE_SIZE -
4202 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4203
4204 err = ext4_discard_partial_page_buffers(handle,
4205 mapping, inode->i_size, page_len, 0);
4206
4207 if (err)
4208 goto out_stop;
4209 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004210
Jan Kara9ddfc3d2008-07-11 19:27:31 -04004211 if (ext4_orphan_add(handle, inode))
4212 goto out_stop;
4213
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004214 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07004215 ext4_ext_invalidate_cache(inode);
4216
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004217 ext4_discard_preallocations(inode);
Alex Tomasc9de5602008-01-29 00:19:52 -05004218
Alex Tomasa86c6182006-10-11 01:21:03 -07004219 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004220 * TODO: optimization is possible here.
4221 * Probably we need not scan at all,
4222 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004223 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004224
4225 /* we have to know where to truncate from in crash case */
4226 EXT4_I(inode)->i_disksize = inode->i_size;
4227 ext4_mark_inode_dirty(handle, inode);
4228
4229 last_block = (inode->i_size + sb->s_blocksize - 1)
4230 >> EXT4_BLOCK_SIZE_BITS(sb);
Allison Hendersonc6a03712011-07-17 23:21:03 -04004231 err = ext4_ext_remove_space(inode, last_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07004232
4233 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04004234 * transaction synchronous.
4235 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004236 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05004237 ext4_handle_sync(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07004238
Jan Kara9ddfc3d2008-07-11 19:27:31 -04004239 up_write(&EXT4_I(inode)->i_data_sem);
Eric Gouriouf6d2f6b2011-05-22 21:33:00 -04004240
4241out_stop:
Alex Tomasa86c6182006-10-11 01:21:03 -07004242 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004243 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07004244 * then we need to clear up the orphan record which we created above.
4245 * However, if this was a real unlink then we were called by
4246 * ext4_delete_inode(), and we allow that function to clean up the
4247 * orphan info for us.
4248 */
4249 if (inode->i_nlink)
4250 ext4_orphan_del(handle, inode);
4251
Solofo Ramangalahyef737722008-04-29 22:00:41 -04004252 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4253 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07004254 ext4_journal_stop(handle);
4255}
4256
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004257static void ext4_falloc_update_inode(struct inode *inode,
4258 int mode, loff_t new_size, int update_ctime)
4259{
4260 struct timespec now;
4261
4262 if (update_ctime) {
4263 now = current_fs_time(inode->i_sb);
4264 if (!timespec_equal(&inode->i_ctime, &now))
4265 inode->i_ctime = now;
4266 }
4267 /*
4268 * Update only when preallocation was requested beyond
4269 * the file size.
4270 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04004271 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4272 if (new_size > i_size_read(inode))
4273 i_size_write(inode, new_size);
4274 if (new_size > EXT4_I(inode)->i_disksize)
4275 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004276 } else {
4277 /*
4278 * Mark that we allocate beyond EOF so the subsequent truncate
4279 * can proceed even if the new size is the same as i_size.
4280 */
4281 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004282 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004283 }
4284
4285}
4286
Amit Aroraa2df2a62007-07-17 21:42:41 -04004287/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004288 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04004289 * operation, which gets called from sys_fallocate system call.
4290 * For block-mapped files, posix_fallocate should fall back to the method
4291 * of writing zeroes to the required new blocks (the same behavior which is
4292 * expected for file systems which do not support fallocate() system call).
4293 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004294long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004295{
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004296 struct inode *inode = file->f_path.dentry->d_inode;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004297 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004298 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004299 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004300 int ret = 0;
4301 int ret2 = 0;
4302 int retries = 0;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004303 int flags;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004304 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004305 unsigned int credits, blkbits = inode->i_blkbits;
4306
4307 /*
4308 * currently supporting (pre)allocate mode for extent-based
4309 * files _only_
4310 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004311 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Amit Aroraa2df2a62007-07-17 21:42:41 -04004312 return -EOPNOTSUPP;
4313
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004314 /* Return error if mode is not supported */
4315 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4316 return -EOPNOTSUPP;
4317
4318 if (mode & FALLOC_FL_PUNCH_HOLE)
4319 return ext4_punch_hole(file, offset, len);
4320
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004321 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004322 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004323 /*
4324 * We can't just convert len to max_blocks because
4325 * If blocksize = 4096 offset = 3072 and len = 2048
4326 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004327 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004328 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004329 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04004330 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04004331 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004332 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004333 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004334 ret = inode_newsize_ok(inode, (len + offset));
4335 if (ret) {
4336 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004337 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004338 return ret;
4339 }
Greg Harm3c6fe772011-10-31 18:41:47 -04004340 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004341 if (mode & FALLOC_FL_KEEP_SIZE)
4342 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
Greg Harm3c6fe772011-10-31 18:41:47 -04004343 /*
4344 * Don't normalize the request if it can fit in one extent so
4345 * that it doesn't get unnecessarily split into multiple
4346 * extents.
4347 */
4348 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4349 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004350retry:
4351 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004352 map.m_lblk = map.m_lblk + ret;
4353 map.m_len = max_blocks = max_blocks - ret;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004354 handle = ext4_journal_start(inode, credits);
4355 if (IS_ERR(handle)) {
4356 ret = PTR_ERR(handle);
4357 break;
4358 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004359 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004360 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004361#ifdef EXT4FS_DEBUG
4362 WARN_ON(ret <= 0);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004363 printk(KERN_ERR "%s: ext4_ext_map_blocks "
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004364 "returned error inode#%lu, block=%u, "
Thadeu Lima de Souza Cascardo9fd97842009-01-26 19:26:26 -05004365 "max_blocks=%u", __func__,
Kazuya Mioa6371b62010-10-27 21:30:15 -04004366 inode->i_ino, map.m_lblk, max_blocks);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004367#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04004368 ext4_mark_inode_dirty(handle, inode);
4369 ret2 = ext4_journal_stop(handle);
4370 break;
4371 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004372 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004373 blkbits) >> blkbits))
4374 new_size = offset + len;
4375 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04004376 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004377
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004378 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004379 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04004380 ext4_mark_inode_dirty(handle, inode);
4381 ret2 = ext4_journal_stop(handle);
4382 if (ret2)
4383 break;
4384 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004385 if (ret == -ENOSPC &&
4386 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4387 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004388 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004389 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004390 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004391 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4392 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004393 return ret > 0 ? ret2 : ret;
4394}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004395
4396/*
Mingming Cao00314622009-09-28 15:49:08 -04004397 * This function convert a range of blocks to written extents
4398 * The caller of this function will pass the start offset and the size.
4399 * all unwritten extents within this range will be converted to
4400 * written extents.
4401 *
4402 * This function is called from the direct IO end io call back
4403 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004404 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004405 */
4406int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
Eric Sandeena1de02d2010-02-04 23:58:38 -05004407 ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004408{
4409 handle_t *handle;
Mingming Cao00314622009-09-28 15:49:08 -04004410 unsigned int max_blocks;
4411 int ret = 0;
4412 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004413 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004414 unsigned int credits, blkbits = inode->i_blkbits;
4415
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004416 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004417 /*
4418 * We can't just convert len to max_blocks because
4419 * If blocksize = 4096 offset = 3072 and len = 2048
4420 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004421 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4422 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004423 /*
4424 * credits to insert 1 extent into extent tree
4425 */
4426 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4427 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004428 map.m_lblk += ret;
4429 map.m_len = (max_blocks -= ret);
Mingming Cao00314622009-09-28 15:49:08 -04004430 handle = ext4_journal_start(inode, credits);
4431 if (IS_ERR(handle)) {
4432 ret = PTR_ERR(handle);
4433 break;
4434 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004435 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004436 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Mingming Cao00314622009-09-28 15:49:08 -04004437 if (ret <= 0) {
4438 WARN_ON(ret <= 0);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004439 printk(KERN_ERR "%s: ext4_ext_map_blocks "
Mingming Cao00314622009-09-28 15:49:08 -04004440 "returned error inode#%lu, block=%u, "
4441 "max_blocks=%u", __func__,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004442 inode->i_ino, map.m_lblk, map.m_len);
Mingming Cao00314622009-09-28 15:49:08 -04004443 }
4444 ext4_mark_inode_dirty(handle, inode);
4445 ret2 = ext4_journal_stop(handle);
4446 if (ret <= 0 || ret2 )
4447 break;
4448 }
4449 return ret > 0 ? ret2 : ret;
4450}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004451
Mingming Cao00314622009-09-28 15:49:08 -04004452/*
Eric Sandeen6873fa02008-10-07 00:46:36 -04004453 * Callback function called for each extent to gather FIEMAP information.
4454 */
Lukas Czernerc03f8aa2011-06-06 00:06:52 -04004455static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
Eric Sandeen6873fa02008-10-07 00:46:36 -04004456 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4457 void *data)
4458{
Eric Sandeen6873fa02008-10-07 00:46:36 -04004459 __u64 logical;
4460 __u64 physical;
4461 __u64 length;
4462 __u32 flags = 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004463 int ret = 0;
4464 struct fiemap_extent_info *fieinfo = data;
4465 unsigned char blksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004466
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004467 blksize_bits = inode->i_sb->s_blocksize_bits;
4468 logical = (__u64)newex->ec_block << blksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004469
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05004470 if (newex->ec_start == 0) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004471 /*
4472 * No extent in extent-tree contains block @newex->ec_start,
4473 * then the block may stay in 1)a hole or 2)delayed-extent.
4474 *
4475 * Holes or delayed-extents are processed as follows.
4476 * 1. lookup dirty pages with specified range in pagecache.
4477 * If no page is got, then there is no delayed-extent and
4478 * return with EXT_CONTINUE.
4479 * 2. find the 1st mapped buffer,
4480 * 3. check if the mapped buffer is both in the request range
4481 * and a delayed buffer. If not, there is no delayed-extent,
4482 * then return.
4483 * 4. a delayed-extent is found, the extent will be collected.
4484 */
4485 ext4_lblk_t end = 0;
4486 pgoff_t last_offset;
4487 pgoff_t offset;
4488 pgoff_t index;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004489 pgoff_t start_index = 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004490 struct page **pages = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004491 struct buffer_head *bh = NULL;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004492 struct buffer_head *head = NULL;
4493 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4494
4495 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4496 if (pages == NULL)
4497 return -ENOMEM;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004498
4499 offset = logical >> PAGE_SHIFT;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004500repeat:
4501 last_offset = offset;
4502 head = NULL;
4503 ret = find_get_pages_tag(inode->i_mapping, &offset,
4504 PAGECACHE_TAG_DIRTY, nr_pages, pages);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004505
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004506 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4507 /* First time, try to find a mapped buffer. */
4508 if (ret == 0) {
4509out:
4510 for (index = 0; index < ret; index++)
4511 page_cache_release(pages[index]);
4512 /* just a hole. */
4513 kfree(pages);
4514 return EXT_CONTINUE;
4515 }
Yongqiang Yangb2213492011-05-24 11:36:58 -04004516 index = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004517
Yongqiang Yangb2213492011-05-24 11:36:58 -04004518next_page:
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004519 /* Try to find the 1st mapped buffer. */
Yongqiang Yangb2213492011-05-24 11:36:58 -04004520 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004521 blksize_bits;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004522 if (!page_has_buffers(pages[index]))
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004523 goto out;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004524 head = page_buffers(pages[index]);
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004525 if (!head)
4526 goto out;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004527
Yongqiang Yangb2213492011-05-24 11:36:58 -04004528 index++;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004529 bh = head;
4530 do {
Yongqiang Yangb2213492011-05-24 11:36:58 -04004531 if (end >= newex->ec_block +
4532 newex->ec_len)
4533 /* The buffer is out of
4534 * the request range.
4535 */
4536 goto out;
4537
4538 if (buffer_mapped(bh) &&
4539 end >= newex->ec_block) {
4540 start_index = index - 1;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004541 /* get the 1st mapped buffer. */
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004542 goto found_mapped_buffer;
4543 }
Yongqiang Yangb2213492011-05-24 11:36:58 -04004544
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004545 bh = bh->b_this_page;
4546 end++;
4547 } while (bh != head);
4548
Yongqiang Yangb2213492011-05-24 11:36:58 -04004549 /* No mapped buffer in the range found in this page,
4550 * We need to look up next page.
4551 */
4552 if (index >= ret) {
4553 /* There is no page left, but we need to limit
4554 * newex->ec_len.
4555 */
4556 newex->ec_len = end - newex->ec_block;
4557 goto out;
4558 }
4559 goto next_page;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004560 } else {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004561 /*Find contiguous delayed buffers. */
4562 if (ret > 0 && pages[0]->index == last_offset)
4563 head = page_buffers(pages[0]);
4564 bh = head;
Yongqiang Yangb2213492011-05-24 11:36:58 -04004565 index = 1;
4566 start_index = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004567 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004568
4569found_mapped_buffer:
4570 if (bh != NULL && buffer_delay(bh)) {
4571 /* 1st or contiguous delayed buffer found. */
4572 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4573 /*
4574 * 1st delayed buffer found, record
4575 * the start of extent.
4576 */
4577 flags |= FIEMAP_EXTENT_DELALLOC;
4578 newex->ec_block = end;
4579 logical = (__u64)end << blksize_bits;
4580 }
4581 /* Find contiguous delayed buffers. */
4582 do {
4583 if (!buffer_delay(bh))
4584 goto found_delayed_extent;
4585 bh = bh->b_this_page;
4586 end++;
4587 } while (bh != head);
4588
Yongqiang Yangb2213492011-05-24 11:36:58 -04004589 for (; index < ret; index++) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004590 if (!page_has_buffers(pages[index])) {
4591 bh = NULL;
4592 break;
4593 }
4594 head = page_buffers(pages[index]);
4595 if (!head) {
4596 bh = NULL;
4597 break;
4598 }
Yongqiang Yangb2213492011-05-24 11:36:58 -04004599
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004600 if (pages[index]->index !=
Yongqiang Yangb2213492011-05-24 11:36:58 -04004601 pages[start_index]->index + index
4602 - start_index) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004603 /* Blocks are not contiguous. */
4604 bh = NULL;
4605 break;
4606 }
4607 bh = head;
4608 do {
4609 if (!buffer_delay(bh))
4610 /* Delayed-extent ends. */
4611 goto found_delayed_extent;
4612 bh = bh->b_this_page;
4613 end++;
4614 } while (bh != head);
4615 }
4616 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4617 /* a hole found. */
4618 goto out;
4619
4620found_delayed_extent:
4621 newex->ec_len = min(end - newex->ec_block,
4622 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4623 if (ret == nr_pages && bh != NULL &&
4624 newex->ec_len < EXT_INIT_MAX_LEN &&
4625 buffer_delay(bh)) {
4626 /* Have not collected an extent and continue. */
4627 for (index = 0; index < ret; index++)
4628 page_cache_release(pages[index]);
4629 goto repeat;
4630 }
4631
4632 for (index = 0; index < ret; index++)
4633 page_cache_release(pages[index]);
4634 kfree(pages);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004635 }
4636
4637 physical = (__u64)newex->ec_start << blksize_bits;
4638 length = (__u64)newex->ec_len << blksize_bits;
4639
4640 if (ex && ext4_ext_is_uninitialized(ex))
4641 flags |= FIEMAP_EXTENT_UNWRITTEN;
4642
Lukas Czernerc03f8aa2011-06-06 00:06:52 -04004643 if (next == EXT_MAX_BLOCKS)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004644 flags |= FIEMAP_EXTENT_LAST;
4645
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004646 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
Eric Sandeen6873fa02008-10-07 00:46:36 -04004647 length, flags);
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004648 if (ret < 0)
4649 return ret;
4650 if (ret == 1)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004651 return EXT_BREAK;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004652 return EXT_CONTINUE;
4653}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004654/* fiemap flags we can handle specified here */
4655#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4656
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004657static int ext4_xattr_fiemap(struct inode *inode,
4658 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004659{
4660 __u64 physical = 0;
4661 __u64 length;
4662 __u32 flags = FIEMAP_EXTENT_LAST;
4663 int blockbits = inode->i_sb->s_blocksize_bits;
4664 int error = 0;
4665
4666 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004667 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004668 struct ext4_iloc iloc;
4669 int offset; /* offset of xattr in inode */
4670
4671 error = ext4_get_inode_loc(inode, &iloc);
4672 if (error)
4673 return error;
4674 physical = iloc.bh->b_blocknr << blockbits;
4675 offset = EXT4_GOOD_OLD_INODE_SIZE +
4676 EXT4_I(inode)->i_extra_isize;
4677 physical += offset;
4678 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4679 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004680 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004681 } else { /* external block */
4682 physical = EXT4_I(inode)->i_file_acl << blockbits;
4683 length = inode->i_sb->s_blocksize;
4684 }
4685
4686 if (physical)
4687 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4688 length, flags);
4689 return (error < 0 ? error : 0);
4690}
4691
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004692/*
4693 * ext4_ext_punch_hole
4694 *
4695 * Punches a hole of "length" bytes in a file starting
4696 * at byte "offset"
4697 *
4698 * @inode: The inode of the file to punch a hole in
4699 * @offset: The starting byte offset of the hole
4700 * @length: The length of the hole
4701 *
4702 * Returns the number of blocks removed or negative on err
4703 */
4704int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4705{
4706 struct inode *inode = file->f_path.dentry->d_inode;
4707 struct super_block *sb = inode->i_sb;
4708 struct ext4_ext_cache cache_ex;
4709 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4710 struct address_space *mapping = inode->i_mapping;
4711 struct ext4_map_blocks map;
4712 handle_t *handle;
Allison Hendersonba062082011-09-03 11:55:59 -04004713 loff_t first_page, last_page, page_len;
4714 loff_t first_page_offset, last_page_offset;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004715 int ret, credits, blocks_released, err = 0;
4716
Allison Henderson2be47512011-09-03 11:56:52 -04004717 /* No need to punch hole beyond i_size */
4718 if (offset >= inode->i_size)
4719 return 0;
4720
4721 /*
4722 * If the hole extends beyond i_size, set the hole
4723 * to end after the page that contains i_size
4724 */
4725 if (offset + length > inode->i_size) {
4726 length = inode->i_size +
4727 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4728 offset;
4729 }
4730
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004731 first_block = (offset + sb->s_blocksize - 1) >>
4732 EXT4_BLOCK_SIZE_BITS(sb);
4733 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4734
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004735 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4736 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4737
4738 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4739 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4740
4741 /*
4742 * Write out all dirty pages to avoid race conditions
4743 * Then release them.
4744 */
4745 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4746 err = filemap_write_and_wait_range(mapping,
Allison Henderson2be47512011-09-03 11:56:52 -04004747 offset, offset + length - 1);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004748
Allison Henderson2be47512011-09-03 11:56:52 -04004749 if (err)
4750 return err;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004751 }
4752
4753 /* Now release the pages */
4754 if (last_page_offset > first_page_offset) {
4755 truncate_inode_pages_range(mapping, first_page_offset,
4756 last_page_offset-1);
4757 }
4758
4759 /* finish any pending end_io work */
4760 ext4_flush_completed_IO(inode);
4761
4762 credits = ext4_writepage_trans_blocks(inode);
4763 handle = ext4_journal_start(inode, credits);
4764 if (IS_ERR(handle))
4765 return PTR_ERR(handle);
4766
4767 err = ext4_orphan_add(handle, inode);
4768 if (err)
4769 goto out;
4770
4771 /*
Allison Hendersonba062082011-09-03 11:55:59 -04004772 * Now we need to zero out the non-page-aligned data in the
4773 * pages at the start and tail of the hole, and unmap the buffer
4774 * heads for the block aligned regions of the page that were
4775 * completely zeroed.
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004776 */
Allison Hendersonba062082011-09-03 11:55:59 -04004777 if (first_page > last_page) {
4778 /*
4779 * If the file space being truncated is contained within a page
4780 * just zero out and unmap the middle of that page
4781 */
4782 err = ext4_discard_partial_page_buffers(handle,
4783 mapping, offset, length, 0);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004784
Allison Hendersonba062082011-09-03 11:55:59 -04004785 if (err)
4786 goto out;
4787 } else {
4788 /*
4789 * zero out and unmap the partial page that contains
4790 * the start of the hole
4791 */
4792 page_len = first_page_offset - offset;
4793 if (page_len > 0) {
4794 err = ext4_discard_partial_page_buffers(handle, mapping,
4795 offset, page_len, 0);
4796 if (err)
4797 goto out;
4798 }
4799
4800 /*
4801 * zero out and unmap the partial page that contains
4802 * the end of the hole
4803 */
4804 page_len = offset + length - last_page_offset;
4805 if (page_len > 0) {
4806 err = ext4_discard_partial_page_buffers(handle, mapping,
4807 last_page_offset, page_len, 0);
4808 if (err)
4809 goto out;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004810 }
4811 }
4812
Allison Henderson2be47512011-09-03 11:56:52 -04004813
4814 /*
4815 * If i_size is contained in the last page, we need to
4816 * unmap and zero the partial page after i_size
4817 */
4818 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4819 inode->i_size % PAGE_CACHE_SIZE != 0) {
4820
4821 page_len = PAGE_CACHE_SIZE -
4822 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4823
4824 if (page_len > 0) {
4825 err = ext4_discard_partial_page_buffers(handle,
4826 mapping, inode->i_size, page_len, 0);
4827
4828 if (err)
4829 goto out;
4830 }
4831 }
4832
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004833 /* If there are no blocks to remove, return now */
4834 if (first_block >= last_block)
4835 goto out;
4836
4837 down_write(&EXT4_I(inode)->i_data_sem);
4838 ext4_ext_invalidate_cache(inode);
4839 ext4_discard_preallocations(inode);
4840
4841 /*
4842 * Loop over all the blocks and identify blocks
4843 * that need to be punched out
4844 */
4845 iblock = first_block;
4846 blocks_released = 0;
4847 while (iblock < last_block) {
4848 max_blocks = last_block - iblock;
4849 num_blocks = 1;
4850 memset(&map, 0, sizeof(map));
4851 map.m_lblk = iblock;
4852 map.m_len = max_blocks;
4853 ret = ext4_ext_map_blocks(handle, inode, &map,
4854 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4855
4856 if (ret > 0) {
4857 blocks_released += ret;
4858 num_blocks = ret;
4859 } else if (ret == 0) {
4860 /*
4861 * If map blocks could not find the block,
4862 * then it is in a hole. If the hole was
4863 * not already cached, then map blocks should
4864 * put it in the cache. So we can get the hole
4865 * out of the cache
4866 */
4867 memset(&cache_ex, 0, sizeof(cache_ex));
4868 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4869 !cache_ex.ec_start) {
4870
4871 /* The hole is cached */
4872 num_blocks = cache_ex.ec_block +
4873 cache_ex.ec_len - iblock;
4874
4875 } else {
4876 /* The block could not be identified */
4877 err = -EIO;
4878 break;
4879 }
4880 } else {
4881 /* Map blocks error */
4882 err = ret;
4883 break;
4884 }
4885
4886 if (num_blocks == 0) {
4887 /* This condition should never happen */
4888 ext_debug("Block lookup failed");
4889 err = -EIO;
4890 break;
4891 }
4892
4893 iblock += num_blocks;
4894 }
4895
4896 if (blocks_released > 0) {
4897 ext4_ext_invalidate_cache(inode);
4898 ext4_discard_preallocations(inode);
4899 }
4900
4901 if (IS_SYNC(inode))
4902 ext4_handle_sync(handle);
4903
4904 up_write(&EXT4_I(inode)->i_data_sem);
4905
4906out:
4907 ext4_orphan_del(handle, inode);
4908 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4909 ext4_mark_inode_dirty(handle, inode);
4910 ext4_journal_stop(handle);
4911 return err;
4912}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004913int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4914 __u64 start, __u64 len)
4915{
4916 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004917 int error = 0;
4918
4919 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004920 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004921 return generic_block_fiemap(inode, fieinfo, start, len,
4922 ext4_get_block);
4923
4924 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4925 return -EBADR;
4926
4927 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4928 error = ext4_xattr_fiemap(inode, fieinfo);
4929 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004930 ext4_lblk_t len_blks;
4931 __u64 last_blk;
4932
Eric Sandeen6873fa02008-10-07 00:46:36 -04004933 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004934 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004935 if (last_blk >= EXT_MAX_BLOCKS)
4936 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004937 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004938
4939 /*
4940 * Walk the extent tree gathering extent information.
4941 * ext4_ext_fiemap_cb will push extents back to user.
4942 */
Eric Sandeen6873fa02008-10-07 00:46:36 -04004943 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4944 ext4_ext_fiemap_cb, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004945 }
4946
4947 return error;
4948}