blob: d480d49d2696e121ef5c4aac63dfb40f3e1a1b0b [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"
Theodore Ts'o4a092d72012-11-28 13:03:30 -050044#include "ext4_extents.h"
Tao Maf19d5872012-12-10 14:05:51 -050045#include "xattr.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070046
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040047#include <trace/events/ext4.h>
48
Lukas Czerner5f95d212012-03-19 23:03:19 -040049/*
50 * used by extent splitting.
51 */
52#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 due to ENOSPC */
54#define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
55#define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
56
Dmitry Monakhovdee1f972012-10-10 01:04:58 -040057#define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58#define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59
Darrick J. Wong7ac59902012-04-29 18:37:10 -040060static __le32 ext4_extent_block_csum(struct inode *inode,
61 struct ext4_extent_header *eh)
62{
63 struct ext4_inode_info *ei = EXT4_I(inode);
64 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
65 __u32 csum;
66
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
68 EXT4_EXTENT_TAIL_OFFSET(eh));
69 return cpu_to_le32(csum);
70}
71
72static int ext4_extent_block_csum_verify(struct inode *inode,
73 struct ext4_extent_header *eh)
74{
75 struct ext4_extent_tail *et;
76
77 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
78 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
79 return 1;
80
81 et = find_ext4_extent_tail(eh);
82 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
83 return 0;
84 return 1;
85}
86
87static void ext4_extent_block_csum_set(struct inode *inode,
88 struct ext4_extent_header *eh)
89{
90 struct ext4_extent_tail *et;
91
92 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
93 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
94 return;
95
96 et = find_ext4_extent_tail(eh);
97 et->et_checksum = ext4_extent_block_csum(inode, eh);
98}
99
Allison Hendersond583fb82011-05-25 07:41:43 -0400100static int ext4_split_extent(handle_t *handle,
101 struct inode *inode,
102 struct ext4_ext_path *path,
103 struct ext4_map_blocks *map,
104 int split_flag,
105 int flags);
106
Lukas Czerner5f95d212012-03-19 23:03:19 -0400107static int ext4_split_extent_at(handle_t *handle,
108 struct inode *inode,
109 struct ext4_ext_path *path,
110 ext4_lblk_t split,
111 int split_flag,
112 int flags);
113
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500114static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -0500115 struct extent_status *newes);
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500116
Jan Kara487caee2009-08-17 22:17:20 -0400117static int ext4_ext_truncate_extend_restart(handle_t *handle,
118 struct inode *inode,
119 int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -0700120{
121 int err;
122
Frank Mayhar03901312009-01-07 00:06:22 -0500123 if (!ext4_handle_valid(handle))
124 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700125 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -0400126 return 0;
127 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -0400128 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -0400129 return err;
Jan Kara487caee2009-08-17 22:17:20 -0400130 err = ext4_truncate_restart_trans(handle, inode, needed);
Dmitry Monakhov0617b832010-05-17 01:00:00 -0400131 if (err == 0)
132 err = -EAGAIN;
Jan Kara487caee2009-08-17 22:17:20 -0400133
134 return err;
Alex Tomasa86c6182006-10-11 01:21:03 -0700135}
136
137/*
138 * could return:
139 * - EROFS
140 * - ENOMEM
141 */
142static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
143 struct ext4_ext_path *path)
144{
145 if (path->p_bh) {
146 /* path points to block */
147 return ext4_journal_get_write_access(handle, path->p_bh);
148 }
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
151 return 0;
152}
153
154/*
155 * could return:
156 * - EROFS
157 * - ENOMEM
158 * - EIO
159 */
Darrick J. Wong26564972013-04-19 14:04:12 -0400160int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700162{
163 int err;
164 if (path->p_bh) {
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400165 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
Alex Tomasa86c6182006-10-11 01:21:03 -0700166 /* path points to block */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400167 err = __ext4_handle_dirty_metadata(where, line, handle,
168 inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700169 } else {
170 /* path points to leaf/index in inode body */
171 err = ext4_mark_inode_dirty(handle, inode);
172 }
173 return err;
174}
175
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700176static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700177 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500178 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700179{
Alex Tomasa86c6182006-10-11 01:21:03 -0700180 if (path) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400181 int depth = path->p_depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700182 struct ext4_extent *ex;
Alex Tomasa86c6182006-10-11 01:21:03 -0700183
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500184 /*
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
Tao Mab8d65682011-01-21 23:21:31 +0800198 * especially if the latter case turns out to be
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500199 * common.
200 */
Avantika Mathur7e028972006-12-06 20:41:33 -0800201 ex = path[depth].p_ext;
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500202 if (ex) {
203 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
204 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
205
206 if (block > ext_block)
207 return ext_pblk + (block - ext_block);
208 else
209 return ext_pblk - (ext_block - block);
210 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700211
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700212 /* it looks like index is empty;
213 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700214 if (path[depth].p_bh)
215 return path[depth].p_bh->b_blocknr;
216 }
217
218 /* OK. use inode's group */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400219 return ext4_inode_to_goal_block(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700220}
221
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400222/*
223 * Allocation for a meta data block
224 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700225static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400226ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700227 struct ext4_ext_path *path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400228 struct ext4_extent *ex, int *err, unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700229{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700230 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700231
232 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Allison Henderson55f020d2011-05-25 07:41:26 -0400233 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
234 NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700235 return newblock;
236}
237
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400238static inline int ext4_ext_space_block(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700239{
240 int size;
241
242 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100244#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400245 if (!check && size > 6)
246 size = 6;
Alex Tomasa86c6182006-10-11 01:21:03 -0700247#endif
248 return size;
249}
250
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400251static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700252{
253 int size;
254
255 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
256 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100257#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400258 if (!check && size > 5)
259 size = 5;
Alex Tomasa86c6182006-10-11 01:21:03 -0700260#endif
261 return size;
262}
263
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400264static inline int ext4_ext_space_root(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700265{
266 int size;
267
268 size = sizeof(EXT4_I(inode)->i_data);
269 size -= sizeof(struct ext4_extent_header);
270 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100271#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400272 if (!check && size > 3)
273 size = 3;
Alex Tomasa86c6182006-10-11 01:21:03 -0700274#endif
275 return size;
276}
277
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400278static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700279{
280 int size;
281
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100285#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400286 if (!check && size > 4)
287 size = 4;
Alex Tomasa86c6182006-10-11 01:21:03 -0700288#endif
289 return size;
290}
291
Mingming Caod2a17632008-07-14 17:52:37 -0400292/*
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
296 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500297int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400298{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500299 struct ext4_inode_info *ei = EXT4_I(inode);
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400300 int idxs;
Mingming Caod2a17632008-07-14 17:52:37 -0400301
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500302 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
303 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400304
305 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400312 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500313 if (ei->i_da_metadata_calc_len &&
314 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400315 int num = 0;
316
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500317 if ((ei->i_da_metadata_calc_len % idxs) == 0)
318 num++;
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
320 num++;
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
322 num++;
323 ei->i_da_metadata_calc_len = 0;
324 } else
325 ei->i_da_metadata_calc_len++;
326 ei->i_da_metadata_calc_last_lblock++;
327 return num;
328 }
Mingming Caod2a17632008-07-14 17:52:37 -0400329
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500330 /*
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
333 */
334 ei->i_da_metadata_calc_len = 1;
335 ei->i_da_metadata_calc_last_lblock = lblock;
336 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400337}
338
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400339static int
340ext4_ext_max_entries(struct inode *inode, int depth)
341{
342 int max;
343
344 if (depth == ext_depth(inode)) {
345 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400346 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400347 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400348 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400349 } else {
350 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400351 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400352 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400353 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400354 }
355
356 return max;
357}
358
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400359static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
360{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400361 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400362 int len = ext4_ext_get_actual_len(ext);
Eryu Guanae21dda2013-12-03 21:22:21 -0500363 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
364 ext4_lblk_t last = lblock + len - 1;
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400365
Eryu Guanae21dda2013-12-03 21:22:21 -0500366 if (lblock > last)
Theodore Ts'o31d4f3a2012-03-11 23:30:16 -0400367 return 0;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400368 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400369}
370
371static int ext4_valid_extent_idx(struct inode *inode,
372 struct ext4_extent_idx *ext_idx)
373{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400374 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400375
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400376 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400377}
378
379static int ext4_valid_extent_entries(struct inode *inode,
380 struct ext4_extent_header *eh,
381 int depth)
382{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400383 unsigned short entries;
384 if (eh->eh_entries == 0)
385 return 1;
386
387 entries = le16_to_cpu(eh->eh_entries);
388
389 if (depth == 0) {
390 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400391 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Eryu Guanae21dda2013-12-03 21:22:21 -0500392 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
393 ext4_fsblk_t pblock = 0;
394 ext4_lblk_t lblock = 0;
395 ext4_lblk_t prev = 0;
396 int len = 0;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400397 while (entries) {
398 if (!ext4_valid_extent(inode, ext))
399 return 0;
Eryu Guanae21dda2013-12-03 21:22:21 -0500400
401 /* Check for overlapping extents */
402 lblock = le32_to_cpu(ext->ee_block);
403 len = ext4_ext_get_actual_len(ext);
404 if ((lblock <= prev) && prev) {
405 pblock = ext4_ext_pblock(ext);
406 es->s_last_error_block = cpu_to_le64(pblock);
407 return 0;
408 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400409 ext++;
410 entries--;
Eryu Guanae21dda2013-12-03 21:22:21 -0500411 prev = lblock + len - 1;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400412 }
413 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400414 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400415 while (entries) {
416 if (!ext4_valid_extent_idx(inode, ext_idx))
417 return 0;
418 ext_idx++;
419 entries--;
420 }
421 }
422 return 1;
423}
424
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400425static int __ext4_ext_check(const char *function, unsigned int line,
426 struct inode *inode, struct ext4_extent_header *eh,
427 int depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400428{
429 const char *error_msg;
430 int max = 0;
431
432 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
433 error_msg = "invalid magic";
434 goto corrupted;
435 }
436 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
437 error_msg = "unexpected eh_depth";
438 goto corrupted;
439 }
440 if (unlikely(eh->eh_max == 0)) {
441 error_msg = "invalid eh_max";
442 goto corrupted;
443 }
444 max = ext4_ext_max_entries(inode, depth);
445 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
446 error_msg = "too large eh_max";
447 goto corrupted;
448 }
449 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
450 error_msg = "invalid eh_entries";
451 goto corrupted;
452 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400453 if (!ext4_valid_extent_entries(inode, eh, depth)) {
454 error_msg = "invalid extent entries";
455 goto corrupted;
456 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400457 /* Verify checksum on non-root extent tree nodes */
458 if (ext_depth(inode) != depth &&
459 !ext4_extent_block_csum_verify(inode, eh)) {
460 error_msg = "extent tree corrupted";
461 goto corrupted;
462 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400463 return 0;
464
465corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400466 ext4_error_inode(inode, function, line, 0,
Theodore Ts'o24676da2010-05-16 21:00:00 -0400467 "bad header/extent: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400468 "entries %u, max %u(%u), depth %u(%u)",
Theodore Ts'o24676da2010-05-16 21:00:00 -0400469 error_msg, le16_to_cpu(eh->eh_magic),
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400470 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
471 max, le16_to_cpu(eh->eh_depth), depth);
472
473 return -EIO;
474}
475
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400476#define ext4_ext_check(inode, eh, depth) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400477 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400478
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400479int ext4_ext_check_inode(struct inode *inode)
480{
481 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
482}
483
Darrick J. Wongf8489122012-04-29 18:21:10 -0400484static int __ext4_ext_check_block(const char *function, unsigned int line,
485 struct inode *inode,
486 struct ext4_extent_header *eh,
487 int depth,
488 struct buffer_head *bh)
489{
490 int ret;
491
492 if (buffer_verified(bh))
493 return 0;
494 ret = ext4_ext_check(inode, eh, depth);
495 if (ret)
496 return ret;
497 set_buffer_verified(bh);
498 return ret;
499}
500
501#define ext4_ext_check_block(inode, eh, depth, bh) \
502 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
503
Alex Tomasa86c6182006-10-11 01:21:03 -0700504#ifdef EXT_DEBUG
505static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
506{
507 int k, l = path->p_depth;
508
509 ext_debug("path:");
510 for (k = 0; k <= l; k++, path++) {
511 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700512 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400513 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700514 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400515 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700516 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400517 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400518 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400519 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700520 } else
521 ext_debug(" []");
522 }
523 ext_debug("\n");
524}
525
526static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
527{
528 int depth = ext_depth(inode);
529 struct ext4_extent_header *eh;
530 struct ext4_extent *ex;
531 int i;
532
533 if (!path)
534 return;
535
536 eh = path[depth].p_hdr;
537 ex = EXT_FIRST_EXTENT(eh);
538
Mingming553f9002009-09-18 13:34:55 -0400539 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
540
Alex Tomasa86c6182006-10-11 01:21:03 -0700541 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400542 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
543 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400544 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700545 }
546 ext_debug("\n");
547}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400548
549static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
550 ext4_fsblk_t newblock, int level)
551{
552 int depth = ext_depth(inode);
553 struct ext4_extent *ex;
554
555 if (depth != level) {
556 struct ext4_extent_idx *idx;
557 idx = path[level].p_idx;
558 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
559 ext_debug("%d: move %d:%llu in new index %llu\n", level,
560 le32_to_cpu(idx->ei_block),
561 ext4_idx_pblock(idx),
562 newblock);
563 idx++;
564 }
565
566 return;
567 }
568
569 ex = path[depth].p_ext;
570 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
571 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
572 le32_to_cpu(ex->ee_block),
573 ext4_ext_pblock(ex),
574 ext4_ext_is_uninitialized(ex),
575 ext4_ext_get_actual_len(ex),
576 newblock);
577 ex++;
578 }
579}
580
Alex Tomasa86c6182006-10-11 01:21:03 -0700581#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400582#define ext4_ext_show_path(inode, path)
583#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400584#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700585#endif
586
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500587void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700588{
589 int depth = path->p_depth;
590 int i;
591
592 for (i = 0; i <= depth; i++, path++)
593 if (path->p_bh) {
594 brelse(path->p_bh);
595 path->p_bh = NULL;
596 }
597}
598
599/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700600 * ext4_ext_binsearch_idx:
601 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400602 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700603 */
604static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500605ext4_ext_binsearch_idx(struct inode *inode,
606 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700607{
608 struct ext4_extent_header *eh = path->p_hdr;
609 struct ext4_extent_idx *r, *l, *m;
610
Alex Tomasa86c6182006-10-11 01:21:03 -0700611
Eric Sandeenbba90742008-01-28 23:58:27 -0500612 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700613
614 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400615 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700616 while (l <= r) {
617 m = l + (r - l) / 2;
618 if (block < le32_to_cpu(m->ei_block))
619 r = m - 1;
620 else
621 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400622 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
623 m, le32_to_cpu(m->ei_block),
624 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700625 }
626
627 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400628 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400629 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700630
631#ifdef CHECK_BINSEARCH
632 {
633 struct ext4_extent_idx *chix, *ix;
634 int k;
635
636 chix = ix = EXT_FIRST_INDEX(eh);
637 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
638 if (k != 0 &&
639 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400640 printk(KERN_DEBUG "k=%d, ix=0x%p, "
641 "first=0x%p\n", k,
642 ix, EXT_FIRST_INDEX(eh));
643 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700644 le32_to_cpu(ix->ei_block),
645 le32_to_cpu(ix[-1].ei_block));
646 }
647 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400648 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700649 if (block < le32_to_cpu(ix->ei_block))
650 break;
651 chix = ix;
652 }
653 BUG_ON(chix != path->p_idx);
654 }
655#endif
656
657}
658
659/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700660 * ext4_ext_binsearch:
661 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400662 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700663 */
664static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500665ext4_ext_binsearch(struct inode *inode,
666 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700667{
668 struct ext4_extent_header *eh = path->p_hdr;
669 struct ext4_extent *r, *l, *m;
670
Alex Tomasa86c6182006-10-11 01:21:03 -0700671 if (eh->eh_entries == 0) {
672 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700673 * this leaf is empty:
674 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700675 */
676 return;
677 }
678
Eric Sandeenbba90742008-01-28 23:58:27 -0500679 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700680
681 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400682 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700683
684 while (l <= r) {
685 m = l + (r - l) / 2;
686 if (block < le32_to_cpu(m->ee_block))
687 r = m - 1;
688 else
689 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400690 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
691 m, le32_to_cpu(m->ee_block),
692 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700693 }
694
695 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400696 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400697 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400698 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400699 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400700 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700701
702#ifdef CHECK_BINSEARCH
703 {
704 struct ext4_extent *chex, *ex;
705 int k;
706
707 chex = ex = EXT_FIRST_EXTENT(eh);
708 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
709 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400710 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700711 if (block < le32_to_cpu(ex->ee_block))
712 break;
713 chex = ex;
714 }
715 BUG_ON(chex != path->p_ext);
716 }
717#endif
718
719}
720
721int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
722{
723 struct ext4_extent_header *eh;
724
725 eh = ext_inode_hdr(inode);
726 eh->eh_depth = 0;
727 eh->eh_entries = 0;
728 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400729 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700730 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700731 return 0;
732}
733
734struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500735ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
736 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700737{
738 struct ext4_extent_header *eh;
739 struct buffer_head *bh;
740 short int depth, i, ppos = 0, alloc = 0;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500741 int ret;
Alex Tomasa86c6182006-10-11 01:21:03 -0700742
743 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400744 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700745
746 /* account possible depth increase */
747 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800748 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700749 GFP_NOFS);
750 if (!path)
751 return ERR_PTR(-ENOMEM);
752 alloc = 1;
753 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700754 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400755 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700756
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400757 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700758 /* walk through the tree */
759 while (i) {
760 ext_debug("depth %d: num %d, max %d\n",
761 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400762
Alex Tomasa86c6182006-10-11 01:21:03 -0700763 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400764 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700765 path[ppos].p_depth = i;
766 path[ppos].p_ext = NULL;
767
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400768 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500769 if (unlikely(!bh)) {
770 ret = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700771 goto err;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500772 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400773 if (!bh_uptodate_or_lock(bh)) {
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400774 trace_ext4_ext_load_extent(inode, block,
775 path[ppos].p_block);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500776 ret = bh_submit_read(bh);
777 if (ret < 0) {
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400778 put_bh(bh);
779 goto err;
780 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400781 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700782 eh = ext_block_hdr(bh);
783 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500784 if (unlikely(ppos > depth)) {
785 put_bh(bh);
786 EXT4_ERROR_INODE(inode,
787 "ppos %d > depth %d", ppos, depth);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500788 ret = -EIO;
Frank Mayhar273df552010-03-02 11:46:09 -0500789 goto err;
790 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700791 path[ppos].p_bh = bh;
792 path[ppos].p_hdr = eh;
793 i--;
794
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500795 ret = ext4_ext_check_block(inode, eh, i, bh);
796 if (ret < 0)
Alex Tomasa86c6182006-10-11 01:21:03 -0700797 goto err;
798 }
799
800 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700801 path[ppos].p_ext = NULL;
802 path[ppos].p_idx = NULL;
803
Alex Tomasa86c6182006-10-11 01:21:03 -0700804 /* find extent */
805 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400806 /* if not an empty leaf */
807 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400808 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700809
810 ext4_ext_show_path(inode, path);
811
812 return path;
813
814err:
815 ext4_ext_drop_refs(path);
816 if (alloc)
817 kfree(path);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500818 return ERR_PTR(ret);
Alex Tomasa86c6182006-10-11 01:21:03 -0700819}
820
821/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700822 * ext4_ext_insert_index:
823 * insert new index [@logical;@ptr] into the block at @curp;
824 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700825 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400826static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
827 struct ext4_ext_path *curp,
828 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700829{
830 struct ext4_extent_idx *ix;
831 int len, err;
832
Avantika Mathur7e028972006-12-06 20:41:33 -0800833 err = ext4_ext_get_access(handle, inode, curp);
834 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700835 return err;
836
Frank Mayhar273df552010-03-02 11:46:09 -0500837 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
838 EXT4_ERROR_INODE(inode,
839 "logical %d == ei_block %d!",
840 logical, le32_to_cpu(curp->p_idx->ei_block));
841 return -EIO;
842 }
Robin Dongd4620312011-07-17 23:43:42 -0400843
844 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
845 >= le16_to_cpu(curp->p_hdr->eh_max))) {
846 EXT4_ERROR_INODE(inode,
847 "eh_entries %d >= eh_max %d!",
848 le16_to_cpu(curp->p_hdr->eh_entries),
849 le16_to_cpu(curp->p_hdr->eh_max));
850 return -EIO;
851 }
852
Alex Tomasa86c6182006-10-11 01:21:03 -0700853 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
854 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400855 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700856 ix = curp->p_idx + 1;
857 } else {
858 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400859 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700860 ix = curp->p_idx;
861 }
862
Eric Gouriou80e675f2011-10-27 11:52:18 -0400863 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
864 BUG_ON(len < 0);
865 if (len > 0) {
866 ext_debug("insert new index %d: "
867 "move %d indices from 0x%p to 0x%p\n",
868 logical, len, ix, ix + 1);
869 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
870 }
871
Tao Maf472e022011-10-17 10:13:46 -0400872 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
873 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
874 return -EIO;
875 }
876
Alex Tomasa86c6182006-10-11 01:21:03 -0700877 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700878 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400879 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700880
Frank Mayhar273df552010-03-02 11:46:09 -0500881 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
882 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
883 return -EIO;
884 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700885
886 err = ext4_ext_dirty(handle, inode, curp);
887 ext4_std_error(inode->i_sb, err);
888
889 return err;
890}
891
892/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700893 * ext4_ext_split:
894 * inserts new subtree into the path, using free index entry
895 * at depth @at:
896 * - allocates all needed blocks (new leaf and all intermediate index blocks)
897 * - makes decision where to split
898 * - moves remaining extents and index entries (right to the split point)
899 * into the newly allocated blocks
900 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700901 */
902static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400903 unsigned int flags,
904 struct ext4_ext_path *path,
905 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -0700906{
907 struct buffer_head *bh = NULL;
908 int depth = ext_depth(inode);
909 struct ext4_extent_header *neh;
910 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -0700911 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700912 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700913 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700914 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700915 int err = 0;
916
917 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700918 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700919
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700920 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700921 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -0500922 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
923 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
924 return -EIO;
925 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700926 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
927 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700928 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700929 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400930 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700931 } else {
932 border = newext->ee_block;
933 ext_debug("leaf will be added."
934 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400935 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700936 }
937
938 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700939 * If error occurs, then we break processing
940 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700941 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700942 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700943 */
944
945 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700946 * Get array to track all allocated blocks.
947 * We need this to handle errors and free blocks
948 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700949 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800950 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700951 if (!ablocks)
952 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700953
954 /* allocate all needed blocks */
955 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
956 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400957 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400958 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -0700959 if (newblock == 0)
960 goto cleanup;
961 ablocks[a] = newblock;
962 }
963
964 /* initialize new leaf */
965 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -0500966 if (unlikely(newblock == 0)) {
967 EXT4_ERROR_INODE(inode, "newblock == 0!");
968 err = -EIO;
969 goto cleanup;
970 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700971 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -0500972 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500973 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700974 goto cleanup;
975 }
976 lock_buffer(bh);
977
Avantika Mathur7e028972006-12-06 20:41:33 -0800978 err = ext4_journal_get_create_access(handle, bh);
979 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700980 goto cleanup;
981
982 neh = ext_block_hdr(bh);
983 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400984 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700985 neh->eh_magic = EXT4_EXT_MAGIC;
986 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700987
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700988 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -0500989 if (unlikely(path[depth].p_hdr->eh_entries !=
990 path[depth].p_hdr->eh_max)) {
991 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
992 path[depth].p_hdr->eh_entries,
993 path[depth].p_hdr->eh_max);
994 err = -EIO;
995 goto cleanup;
996 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700997 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400998 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
999 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07001000 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001001 struct ext4_extent *ex;
1002 ex = EXT_FIRST_EXTENT(neh);
1003 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001004 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001005 }
1006
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001007 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001008 set_buffer_uptodate(bh);
1009 unlock_buffer(bh);
1010
Frank Mayhar03901312009-01-07 00:06:22 -05001011 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001012 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001013 goto cleanup;
1014 brelse(bh);
1015 bh = NULL;
1016
1017 /* correct old leaf */
1018 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001019 err = ext4_ext_get_access(handle, inode, path + depth);
1020 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001021 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001022 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001023 err = ext4_ext_dirty(handle, inode, path + depth);
1024 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001025 goto cleanup;
1026
1027 }
1028
1029 /* create intermediate indexes */
1030 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001031 if (unlikely(k < 0)) {
1032 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1033 err = -EIO;
1034 goto cleanup;
1035 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001036 if (k)
1037 ext_debug("create %d intermediate indices\n", k);
1038 /* insert new index into current index block */
1039 /* current depth stored in i var */
1040 i = depth - 1;
1041 while (k--) {
1042 oldblock = newblock;
1043 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001044 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001045 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001046 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001047 goto cleanup;
1048 }
1049 lock_buffer(bh);
1050
Avantika Mathur7e028972006-12-06 20:41:33 -08001051 err = ext4_journal_get_create_access(handle, bh);
1052 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001053 goto cleanup;
1054
1055 neh = ext_block_hdr(bh);
1056 neh->eh_entries = cpu_to_le16(1);
1057 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001058 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001059 neh->eh_depth = cpu_to_le16(depth - i);
1060 fidx = EXT_FIRST_INDEX(neh);
1061 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001062 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001063
Eric Sandeenbba90742008-01-28 23:58:27 -05001064 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1065 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001066
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001067 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001068 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1069 EXT_LAST_INDEX(path[i].p_hdr))) {
1070 EXT4_ERROR_INODE(inode,
1071 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1072 le32_to_cpu(path[i].p_ext->ee_block));
1073 err = -EIO;
1074 goto cleanup;
1075 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001076 /* start copy indexes */
1077 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1078 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1079 EXT_MAX_INDEX(path[i].p_hdr));
1080 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001081 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001082 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001083 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001084 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001085 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001086 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001087 set_buffer_uptodate(bh);
1088 unlock_buffer(bh);
1089
Frank Mayhar03901312009-01-07 00:06:22 -05001090 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001091 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001092 goto cleanup;
1093 brelse(bh);
1094 bh = NULL;
1095
1096 /* correct old index */
1097 if (m) {
1098 err = ext4_ext_get_access(handle, inode, path + i);
1099 if (err)
1100 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001101 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001102 err = ext4_ext_dirty(handle, inode, path + i);
1103 if (err)
1104 goto cleanup;
1105 }
1106
1107 i--;
1108 }
1109
1110 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001111 err = ext4_ext_insert_index(handle, inode, path + at,
1112 le32_to_cpu(border), newblock);
1113
1114cleanup:
1115 if (bh) {
1116 if (buffer_locked(bh))
1117 unlock_buffer(bh);
1118 brelse(bh);
1119 }
1120
1121 if (err) {
1122 /* free all allocated blocks in error case */
1123 for (i = 0; i < depth; i++) {
1124 if (!ablocks[i])
1125 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001126 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001127 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001128 }
1129 }
1130 kfree(ablocks);
1131
1132 return err;
1133}
1134
1135/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001136 * ext4_ext_grow_indepth:
1137 * implements tree growing procedure:
1138 * - allocates new block
1139 * - moves top-level data (index block or leaf) into the new block
1140 * - initializes new top-level, creating index that points to the
1141 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001142 */
1143static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001144 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001145 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001146{
Alex Tomasa86c6182006-10-11 01:21:03 -07001147 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001148 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001149 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001150 int err = 0;
1151
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001152 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001153 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001154 if (newblock == 0)
1155 return err;
1156
1157 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001158 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001159 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001160 lock_buffer(bh);
1161
Avantika Mathur7e028972006-12-06 20:41:33 -08001162 err = ext4_journal_get_create_access(handle, bh);
1163 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001164 unlock_buffer(bh);
1165 goto out;
1166 }
1167
1168 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001169 memmove(bh->b_data, EXT4_I(inode)->i_data,
1170 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001171
1172 /* set size of new block */
1173 neh = ext_block_hdr(bh);
1174 /* old root could have indexes or leaves
1175 * so calculate e_max right way */
1176 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001177 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001178 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001179 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001180 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001181 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001182 set_buffer_uptodate(bh);
1183 unlock_buffer(bh);
1184
Frank Mayhar03901312009-01-07 00:06:22 -05001185 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001186 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001187 goto out;
1188
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001189 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001190 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001191 neh->eh_entries = cpu_to_le16(1);
1192 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1193 if (neh->eh_depth == 0) {
1194 /* Root extent block becomes index block */
1195 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1196 EXT_FIRST_INDEX(neh)->ei_block =
1197 EXT_FIRST_EXTENT(neh)->ee_block;
1198 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001199 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001200 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001201 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001202 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001203
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001204 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001205 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001206out:
1207 brelse(bh);
1208
1209 return err;
1210}
1211
1212/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001213 * ext4_ext_create_new_leaf:
1214 * finds empty index and adds new leaf.
1215 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001216 */
1217static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001218 unsigned int flags,
1219 struct ext4_ext_path *path,
1220 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001221{
1222 struct ext4_ext_path *curp;
1223 int depth, i, err = 0;
1224
1225repeat:
1226 i = depth = ext_depth(inode);
1227
1228 /* walk up to the tree and look for free index entry */
1229 curp = path + depth;
1230 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1231 i--;
1232 curp--;
1233 }
1234
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001235 /* we use already allocated block for index block,
1236 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001237 if (EXT_HAS_FREE_INDEX(curp)) {
1238 /* if we found index with free entry, then use that
1239 * entry: create all needed subtree and add new leaf */
Allison Henderson55f020d2011-05-25 07:41:26 -04001240 err = ext4_ext_split(handle, inode, flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001241 if (err)
1242 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001243
1244 /* refill path */
1245 ext4_ext_drop_refs(path);
1246 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001247 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1248 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001249 if (IS_ERR(path))
1250 err = PTR_ERR(path);
1251 } else {
1252 /* tree is full, time to grow in depth */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001253 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001254 if (err)
1255 goto out;
1256
1257 /* refill path */
1258 ext4_ext_drop_refs(path);
1259 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001260 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1261 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001262 if (IS_ERR(path)) {
1263 err = PTR_ERR(path);
1264 goto out;
1265 }
1266
1267 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001268 * only first (depth 0 -> 1) produces free space;
1269 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001270 */
1271 depth = ext_depth(inode);
1272 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001273 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001274 goto repeat;
1275 }
1276 }
1277
1278out:
1279 return err;
1280}
1281
1282/*
Alex Tomas1988b512008-01-28 23:58:27 -05001283 * search the closest allocated block to the left for *logical
1284 * and returns it at @logical + it's physical address at @phys
1285 * if *logical is the smallest allocated block, the function
1286 * returns 0 at @phys
1287 * return value contains 0 (success) or error code
1288 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001289static int ext4_ext_search_left(struct inode *inode,
1290 struct ext4_ext_path *path,
1291 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001292{
1293 struct ext4_extent_idx *ix;
1294 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001295 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001296
Frank Mayhar273df552010-03-02 11:46:09 -05001297 if (unlikely(path == NULL)) {
1298 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1299 return -EIO;
1300 }
Alex Tomas1988b512008-01-28 23:58:27 -05001301 depth = path->p_depth;
1302 *phys = 0;
1303
1304 if (depth == 0 && path->p_ext == NULL)
1305 return 0;
1306
1307 /* usually extent in the path covers blocks smaller
1308 * then *logical, but it can be that extent is the
1309 * first one in the file */
1310
1311 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001312 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001313 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001314 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1315 EXT4_ERROR_INODE(inode,
1316 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1317 *logical, le32_to_cpu(ex->ee_block));
1318 return -EIO;
1319 }
Alex Tomas1988b512008-01-28 23:58:27 -05001320 while (--depth >= 0) {
1321 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001322 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1323 EXT4_ERROR_INODE(inode,
1324 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001325 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001326 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001327 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001328 depth);
1329 return -EIO;
1330 }
Alex Tomas1988b512008-01-28 23:58:27 -05001331 }
1332 return 0;
1333 }
1334
Frank Mayhar273df552010-03-02 11:46:09 -05001335 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1336 EXT4_ERROR_INODE(inode,
1337 "logical %d < ee_block %d + ee_len %d!",
1338 *logical, le32_to_cpu(ex->ee_block), ee_len);
1339 return -EIO;
1340 }
Alex Tomas1988b512008-01-28 23:58:27 -05001341
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001342 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001343 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001344 return 0;
1345}
1346
1347/*
1348 * search the closest allocated block to the right for *logical
1349 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001350 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001351 * returns 0 at @phys
1352 * return value contains 0 (success) or error code
1353 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001354static int ext4_ext_search_right(struct inode *inode,
1355 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001356 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1357 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001358{
1359 struct buffer_head *bh = NULL;
1360 struct ext4_extent_header *eh;
1361 struct ext4_extent_idx *ix;
1362 struct ext4_extent *ex;
1363 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001364 int depth; /* Note, NOT eh_depth; depth from top of tree */
1365 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001366
Frank Mayhar273df552010-03-02 11:46:09 -05001367 if (unlikely(path == NULL)) {
1368 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1369 return -EIO;
1370 }
Alex Tomas1988b512008-01-28 23:58:27 -05001371 depth = path->p_depth;
1372 *phys = 0;
1373
1374 if (depth == 0 && path->p_ext == NULL)
1375 return 0;
1376
1377 /* usually extent in the path covers blocks smaller
1378 * then *logical, but it can be that extent is the
1379 * first one in the file */
1380
1381 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001382 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001383 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001384 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1385 EXT4_ERROR_INODE(inode,
1386 "first_extent(path[%d].p_hdr) != ex",
1387 depth);
1388 return -EIO;
1389 }
Alex Tomas1988b512008-01-28 23:58:27 -05001390 while (--depth >= 0) {
1391 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001392 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1393 EXT4_ERROR_INODE(inode,
1394 "ix != EXT_FIRST_INDEX *logical %d!",
1395 *logical);
1396 return -EIO;
1397 }
Alex Tomas1988b512008-01-28 23:58:27 -05001398 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001399 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001400 }
1401
Frank Mayhar273df552010-03-02 11:46:09 -05001402 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1403 EXT4_ERROR_INODE(inode,
1404 "logical %d < ee_block %d + ee_len %d!",
1405 *logical, le32_to_cpu(ex->ee_block), ee_len);
1406 return -EIO;
1407 }
Alex Tomas1988b512008-01-28 23:58:27 -05001408
1409 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1410 /* next allocated block in this leaf */
1411 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001412 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001413 }
1414
1415 /* go up and search for index to the right */
1416 while (--depth >= 0) {
1417 ix = path[depth].p_idx;
1418 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001419 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001420 }
1421
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001422 /* we've gone up to the root and found no index to the right */
1423 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001424
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001425got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001426 /* we've found index to the right, let's
1427 * follow it and find the closest allocated
1428 * block to the right */
1429 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001430 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001431 while (++depth < path->p_depth) {
1432 bh = sb_bread(inode->i_sb, block);
1433 if (bh == NULL)
1434 return -EIO;
1435 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001436 /* subtract from p_depth to get proper eh_depth */
Darrick J. Wongf8489122012-04-29 18:21:10 -04001437 if (ext4_ext_check_block(inode, eh,
1438 path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001439 put_bh(bh);
1440 return -EIO;
1441 }
1442 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001443 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001444 put_bh(bh);
1445 }
1446
1447 bh = sb_bread(inode->i_sb, block);
1448 if (bh == NULL)
1449 return -EIO;
1450 eh = ext_block_hdr(bh);
Darrick J. Wongf8489122012-04-29 18:21:10 -04001451 if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001452 put_bh(bh);
1453 return -EIO;
1454 }
1455 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001456found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001457 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001458 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001459 *ret_ex = ex;
1460 if (bh)
1461 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001462 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001463}
1464
1465/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001466 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001467 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001468 * NOTE: it considers block number from index entry as
1469 * allocated block. Thus, index entries have to be consistent
1470 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001471 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001472static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001473ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1474{
1475 int depth;
1476
1477 BUG_ON(path == NULL);
1478 depth = path->p_depth;
1479
1480 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001481 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001482
1483 while (depth >= 0) {
1484 if (depth == path->p_depth) {
1485 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001486 if (path[depth].p_ext &&
1487 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001488 EXT_LAST_EXTENT(path[depth].p_hdr))
1489 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1490 } else {
1491 /* index */
1492 if (path[depth].p_idx !=
1493 EXT_LAST_INDEX(path[depth].p_hdr))
1494 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1495 }
1496 depth--;
1497 }
1498
Lukas Czernerf17722f2011-06-06 00:05:17 -04001499 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001500}
1501
1502/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001503 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001504 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001505 */
Robin Dong57187892011-07-23 21:49:07 -04001506static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001507{
1508 int depth;
1509
1510 BUG_ON(path == NULL);
1511 depth = path->p_depth;
1512
1513 /* zero-tree has no leaf blocks at all */
1514 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001515 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001516
1517 /* go to index block */
1518 depth--;
1519
1520 while (depth >= 0) {
1521 if (path[depth].p_idx !=
1522 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001523 return (ext4_lblk_t)
1524 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001525 depth--;
1526 }
1527
Lukas Czernerf17722f2011-06-06 00:05:17 -04001528 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001529}
1530
1531/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001532 * ext4_ext_correct_indexes:
1533 * if leaf gets modified and modified extent is first in the leaf,
1534 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001535 * TODO: do we need to correct tree in all cases?
1536 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001537static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001538 struct ext4_ext_path *path)
1539{
1540 struct ext4_extent_header *eh;
1541 int depth = ext_depth(inode);
1542 struct ext4_extent *ex;
1543 __le32 border;
1544 int k, err = 0;
1545
1546 eh = path[depth].p_hdr;
1547 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001548
1549 if (unlikely(ex == NULL || eh == NULL)) {
1550 EXT4_ERROR_INODE(inode,
1551 "ex %p == NULL or eh %p == NULL", ex, eh);
1552 return -EIO;
1553 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001554
1555 if (depth == 0) {
1556 /* there is no tree at all */
1557 return 0;
1558 }
1559
1560 if (ex != EXT_FIRST_EXTENT(eh)) {
1561 /* we correct tree if first leaf got modified only */
1562 return 0;
1563 }
1564
1565 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001566 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001567 */
1568 k = depth - 1;
1569 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001570 err = ext4_ext_get_access(handle, inode, path + k);
1571 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001572 return err;
1573 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001574 err = ext4_ext_dirty(handle, inode, path + k);
1575 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001576 return err;
1577
1578 while (k--) {
1579 /* change all left-side indexes */
1580 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1581 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001582 err = ext4_ext_get_access(handle, inode, path + k);
1583 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001584 break;
1585 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001586 err = ext4_ext_dirty(handle, inode, path + k);
1587 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001588 break;
1589 }
1590
1591 return err;
1592}
1593
Akira Fujita748de672009-06-17 19:24:03 -04001594int
Alex Tomasa86c6182006-10-11 01:21:03 -07001595ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1596 struct ext4_extent *ex2)
1597{
Amit Arora749269f2007-07-18 09:02:56 -04001598 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001599
1600 /*
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001601 * Make sure that both extents are initialized. We don't merge
1602 * uninitialized extents so that we can be sure that end_io code has
1603 * the extent that was written properly split out and conversion to
1604 * initialized is trivial.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001605 */
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001606 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001607 return 0;
1608
Amit Arora749269f2007-07-18 09:02:56 -04001609 if (ext4_ext_is_uninitialized(ex1))
1610 max_len = EXT_UNINIT_MAX_LEN;
1611 else
1612 max_len = EXT_INIT_MAX_LEN;
1613
Amit Aroraa2df2a62007-07-17 21:42:41 -04001614 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1615 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1616
1617 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001618 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001619 return 0;
1620
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001621 /*
1622 * To allow future support for preallocated extents to be added
1623 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001624 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001625 */
Amit Arora749269f2007-07-18 09:02:56 -04001626 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001627 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001628#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001629 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001630 return 0;
1631#endif
1632
Theodore Ts'obf89d162010-10-27 21:30:14 -04001633 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001634 return 1;
1635 return 0;
1636}
1637
1638/*
Amit Arora56055d32007-07-17 21:42:38 -04001639 * This function tries to merge the "ex" extent to the next extent in the tree.
1640 * It always tries to merge towards right. If you want to merge towards
1641 * left, pass "ex - 1" as argument instead of "ex".
1642 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1643 * 1 if they got merged.
1644 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001645static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001646 struct ext4_ext_path *path,
1647 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001648{
1649 struct ext4_extent_header *eh;
1650 unsigned int depth, len;
1651 int merge_done = 0;
1652 int uninitialized = 0;
1653
1654 depth = ext_depth(inode);
1655 BUG_ON(path[depth].p_hdr == NULL);
1656 eh = path[depth].p_hdr;
1657
1658 while (ex < EXT_LAST_EXTENT(eh)) {
1659 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1660 break;
1661 /* merge with next extent! */
1662 if (ext4_ext_is_uninitialized(ex))
1663 uninitialized = 1;
1664 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1665 + ext4_ext_get_actual_len(ex + 1));
1666 if (uninitialized)
1667 ext4_ext_mark_uninitialized(ex);
1668
1669 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1670 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1671 * sizeof(struct ext4_extent);
1672 memmove(ex + 1, ex + 2, len);
1673 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001674 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001675 merge_done = 1;
1676 WARN_ON(eh->eh_entries == 0);
1677 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001678 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001679 }
1680
1681 return merge_done;
1682}
1683
1684/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001685 * This function does a very simple check to see if we can collapse
1686 * an extent tree with a single extent tree leaf block into the inode.
1687 */
1688static void ext4_ext_try_to_merge_up(handle_t *handle,
1689 struct inode *inode,
1690 struct ext4_ext_path *path)
1691{
1692 size_t s;
1693 unsigned max_root = ext4_ext_space_root(inode, 0);
1694 ext4_fsblk_t blk;
1695
1696 if ((path[0].p_depth != 1) ||
1697 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1698 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1699 return;
1700
1701 /*
1702 * We need to modify the block allocation bitmap and the block
1703 * group descriptor to release the extent tree block. If we
1704 * can't get the journal credits, give up.
1705 */
1706 if (ext4_journal_extend(handle, 2))
1707 return;
1708
1709 /*
1710 * Copy the extent data up to the inode
1711 */
1712 blk = ext4_idx_pblock(path[0].p_idx);
1713 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1714 sizeof(struct ext4_extent_idx);
1715 s += sizeof(struct ext4_extent_header);
1716
1717 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1718 path[0].p_depth = 0;
1719 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1720 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1721 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1722
1723 brelse(path[1].p_bh);
1724 ext4_free_blocks(handle, inode, NULL, blk, 1,
1725 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1726}
1727
1728/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001729 * This function tries to merge the @ex extent to neighbours in the tree.
1730 * return 1 if merge left else 0.
1731 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001732static void ext4_ext_try_to_merge(handle_t *handle,
1733 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001734 struct ext4_ext_path *path,
1735 struct ext4_extent *ex) {
1736 struct ext4_extent_header *eh;
1737 unsigned int depth;
1738 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001739
1740 depth = ext_depth(inode);
1741 BUG_ON(path[depth].p_hdr == NULL);
1742 eh = path[depth].p_hdr;
1743
1744 if (ex > EXT_FIRST_EXTENT(eh))
1745 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1746
1747 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001748 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001749
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001750 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001751}
1752
1753/*
Amit Arora25d14f92007-05-24 13:04:13 -04001754 * check if a portion of the "newext" extent overlaps with an
1755 * existing extent.
1756 *
1757 * If there is an overlap discovered, it updates the length of the newext
1758 * such that there will be no overlap, and then returns 1.
1759 * If there is no overlap found, it returns 0.
1760 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001761static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1762 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001763 struct ext4_extent *newext,
1764 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001765{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001766 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001767 unsigned int depth, len1;
1768 unsigned int ret = 0;
1769
1770 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001771 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001772 depth = ext_depth(inode);
1773 if (!path[depth].p_ext)
1774 goto out;
1775 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001776 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001777
1778 /*
1779 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001780 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001781 */
1782 if (b2 < b1) {
1783 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001784 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001785 goto out;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001786 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001787 }
1788
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001789 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001790 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001791 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001792 newext->ee_len = cpu_to_le16(len1);
1793 ret = 1;
1794 }
1795
1796 /* check for overlap */
1797 if (b1 + len1 > b2) {
1798 newext->ee_len = cpu_to_le16(b2 - b1);
1799 ret = 1;
1800 }
1801out:
1802 return ret;
1803}
1804
1805/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001806 * ext4_ext_insert_extent:
1807 * tries to merge requsted extent into the existing extent or
1808 * inserts requested extent as new one into the tree,
1809 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001810 */
1811int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1812 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04001813 struct ext4_extent *newext, int flag)
Alex Tomasa86c6182006-10-11 01:21:03 -07001814{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001815 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001816 struct ext4_extent *ex, *fex;
1817 struct ext4_extent *nearex; /* nearest extent */
1818 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001819 int depth, len, err;
1820 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001821 unsigned uninitialized = 0;
Allison Henderson55f020d2011-05-25 07:41:26 -04001822 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001823
Frank Mayhar273df552010-03-02 11:46:09 -05001824 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1825 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1826 return -EIO;
1827 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001828 depth = ext_depth(inode);
1829 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001830 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001831 if (unlikely(path[depth].p_hdr == NULL)) {
1832 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1833 return -EIO;
1834 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001835
1836 /* try to insert block into found extent and return */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001837 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001838
1839 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001840 * Try to see whether we should rather test the extent on
1841 * right from ex, or from the left of ex. This is because
1842 * ext4_ext_find_extent() can return either extent on the
1843 * left, or on the right from the searched position. This
1844 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001845 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001846 if (ex < EXT_LAST_EXTENT(eh) &&
1847 (le32_to_cpu(ex->ee_block) +
1848 ext4_ext_get_actual_len(ex) <
1849 le32_to_cpu(newext->ee_block))) {
1850 ex += 1;
1851 goto prepend;
1852 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1853 (le32_to_cpu(newext->ee_block) +
1854 ext4_ext_get_actual_len(newext) <
1855 le32_to_cpu(ex->ee_block)))
1856 ex -= 1;
1857
1858 /* Try to append newex to the ex */
1859 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1860 ext_debug("append [%d]%d block to %u:[%d]%d"
1861 "(from %llu)\n",
1862 ext4_ext_is_uninitialized(newext),
1863 ext4_ext_get_actual_len(newext),
1864 le32_to_cpu(ex->ee_block),
1865 ext4_ext_is_uninitialized(ex),
1866 ext4_ext_get_actual_len(ex),
1867 ext4_ext_pblock(ex));
1868 err = ext4_ext_get_access(handle, inode,
1869 path + depth);
1870 if (err)
1871 return err;
1872
1873 /*
1874 * ext4_can_extents_be_merged should have checked
1875 * that either both extents are uninitialized, or
1876 * both aren't. Thus we need to check only one of
1877 * them here.
1878 */
1879 if (ext4_ext_is_uninitialized(ex))
1880 uninitialized = 1;
1881 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001882 + ext4_ext_get_actual_len(newext));
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001883 if (uninitialized)
1884 ext4_ext_mark_uninitialized(ex);
1885 eh = path[depth].p_hdr;
1886 nearex = ex;
1887 goto merge;
1888 }
1889
1890prepend:
1891 /* Try to prepend newex to the ex */
1892 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1893 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1894 "(from %llu)\n",
1895 le32_to_cpu(newext->ee_block),
1896 ext4_ext_is_uninitialized(newext),
1897 ext4_ext_get_actual_len(newext),
1898 le32_to_cpu(ex->ee_block),
1899 ext4_ext_is_uninitialized(ex),
1900 ext4_ext_get_actual_len(ex),
1901 ext4_ext_pblock(ex));
1902 err = ext4_ext_get_access(handle, inode,
1903 path + depth);
1904 if (err)
1905 return err;
1906
1907 /*
1908 * ext4_can_extents_be_merged should have checked
1909 * that either both extents are uninitialized, or
1910 * both aren't. Thus we need to check only one of
1911 * them here.
1912 */
1913 if (ext4_ext_is_uninitialized(ex))
1914 uninitialized = 1;
1915 ex->ee_block = newext->ee_block;
1916 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1917 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1918 + ext4_ext_get_actual_len(newext));
1919 if (uninitialized)
1920 ext4_ext_mark_uninitialized(ex);
1921 eh = path[depth].p_hdr;
1922 nearex = ex;
1923 goto merge;
1924 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001925 }
1926
Alex Tomasa86c6182006-10-11 01:21:03 -07001927 depth = ext_depth(inode);
1928 eh = path[depth].p_hdr;
1929 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1930 goto has_space;
1931
1932 /* probably next leaf has space for us? */
1933 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001934 next = EXT_MAX_BLOCKS;
1935 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001936 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001937 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001938 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07001939 BUG_ON(npath != NULL);
1940 npath = ext4_ext_find_extent(inode, next, NULL);
1941 if (IS_ERR(npath))
1942 return PTR_ERR(npath);
1943 BUG_ON(npath->p_depth != path->p_depth);
1944 eh = npath[depth].p_hdr;
1945 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001946 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001947 le16_to_cpu(eh->eh_entries));
1948 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04001949 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07001950 }
1951 ext_debug("next leaf has no free space(%d,%d)\n",
1952 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1953 }
1954
1955 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001956 * There is no free space in the found leaf.
1957 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001958 */
Lukas Czerner27dd4382013-04-09 22:11:22 -04001959 if (flag & EXT4_GET_BLOCKS_METADATA_NOFAIL)
1960 flags = EXT4_MB_USE_RESERVED;
Allison Henderson55f020d2011-05-25 07:41:26 -04001961 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001962 if (err)
1963 goto cleanup;
1964 depth = ext_depth(inode);
1965 eh = path[depth].p_hdr;
1966
1967has_space:
1968 nearex = path[depth].p_ext;
1969
Avantika Mathur7e028972006-12-06 20:41:33 -08001970 err = ext4_ext_get_access(handle, inode, path + depth);
1971 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001972 goto cleanup;
1973
1974 if (!nearex) {
1975 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001976 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001977 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001978 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001979 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001980 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04001981 nearex = EXT_FIRST_EXTENT(eh);
1982 } else {
1983 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001984 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04001985 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001986 ext_debug("insert %u:%llu:[%d]%d before: "
1987 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001988 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001989 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001990 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001991 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04001992 nearex);
1993 nearex++;
1994 } else {
1995 /* Insert before */
1996 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04001997 ext_debug("insert %u:%llu:[%d]%d after: "
1998 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04001999 le32_to_cpu(newext->ee_block),
2000 ext4_ext_pblock(newext),
2001 ext4_ext_is_uninitialized(newext),
2002 ext4_ext_get_actual_len(newext),
2003 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002004 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04002005 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2006 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002007 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04002008 "move %d extents from 0x%p to 0x%p\n",
2009 le32_to_cpu(newext->ee_block),
2010 ext4_ext_pblock(newext),
2011 ext4_ext_is_uninitialized(newext),
2012 ext4_ext_get_actual_len(newext),
2013 len, nearex, nearex + 1);
2014 memmove(nearex + 1, nearex,
2015 len * sizeof(struct ext4_extent));
2016 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002017 }
2018
Marcin Slusarze8546d02008-04-17 10:38:59 -04002019 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002020 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002021 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002022 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002023 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002024
2025merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002026 /* try to merge extents */
Jiaying Zhang744692d2010-03-04 16:14:02 -05002027 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002028 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002029
Alex Tomasa86c6182006-10-11 01:21:03 -07002030
2031 /* time to correct all indexes above */
2032 err = ext4_ext_correct_indexes(handle, inode, path);
2033 if (err)
2034 goto cleanup;
2035
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002036 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002037
2038cleanup:
2039 if (npath) {
2040 ext4_ext_drop_refs(npath);
2041 kfree(npath);
2042 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002043 return err;
2044}
2045
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002046static int ext4_fill_fiemap_extents(struct inode *inode,
2047 ext4_lblk_t block, ext4_lblk_t num,
2048 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002049{
2050 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002051 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002052 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002053 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002054 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002055 int exists, depth = 0, err = 0;
2056 unsigned int flags = 0;
2057 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002058
Lukas Czernerf17722f2011-06-06 00:05:17 -04002059 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002060 num = last - block;
2061 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002062 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002063
2064 if (path && ext_depth(inode) != depth) {
2065 /* depth was changed. we have to realloc path */
2066 kfree(path);
2067 path = NULL;
2068 }
2069
Eric Sandeen6873fa02008-10-07 00:46:36 -04002070 path = ext4_ext_find_extent(inode, block, path);
2071 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002072 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002073 err = PTR_ERR(path);
2074 path = NULL;
2075 break;
2076 }
2077
2078 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002079 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002080 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002081 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2082 err = -EIO;
2083 break;
2084 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002085 ex = path[depth].p_ext;
2086 next = ext4_ext_next_allocated_block(path);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002087 ext4_ext_drop_refs(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002088
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002089 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002090 exists = 0;
2091 if (!ex) {
2092 /* there is no extent yet, so try to allocate
2093 * all requested space */
2094 start = block;
2095 end = block + num;
2096 } else if (le32_to_cpu(ex->ee_block) > block) {
2097 /* need to allocate space before found extent */
2098 start = block;
2099 end = le32_to_cpu(ex->ee_block);
2100 if (block + num < end)
2101 end = block + num;
2102 } else if (block >= le32_to_cpu(ex->ee_block)
2103 + ext4_ext_get_actual_len(ex)) {
2104 /* need to allocate space after found extent */
2105 start = block;
2106 end = block + num;
2107 if (end >= next)
2108 end = next;
2109 } else if (block >= le32_to_cpu(ex->ee_block)) {
2110 /*
2111 * some part of requested space is covered
2112 * by found extent
2113 */
2114 start = block;
2115 end = le32_to_cpu(ex->ee_block)
2116 + ext4_ext_get_actual_len(ex);
2117 if (block + num < end)
2118 end = block + num;
2119 exists = 1;
2120 } else {
2121 BUG();
2122 }
2123 BUG_ON(end <= start);
2124
2125 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002126 es.es_lblk = start;
2127 es.es_len = end - start;
2128 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002129 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002130 es.es_lblk = le32_to_cpu(ex->ee_block);
2131 es.es_len = ext4_ext_get_actual_len(ex);
2132 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002133 if (ext4_ext_is_uninitialized(ex))
2134 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002135 }
2136
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002137 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002138 * Find delayed extent and update es accordingly. We call
2139 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002140 * last existing extent or not.
2141 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002142 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002143 if (!exists && next_del) {
2144 exists = 1;
2145 flags |= FIEMAP_EXTENT_DELALLOC;
2146 }
2147 up_read(&EXT4_I(inode)->i_data_sem);
2148
Zheng Liu69eb33d2013-02-18 00:31:07 -05002149 if (unlikely(es.es_len == 0)) {
2150 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002151 err = -EIO;
2152 break;
2153 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002154
Zheng Liuf7fec032013-02-18 00:28:47 -05002155 /*
2156 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2157 * we need to check next == EXT_MAX_BLOCKS because it is
2158 * possible that an extent is with unwritten and delayed
2159 * status due to when an extent is delayed allocated and
2160 * is allocated by fallocate status tree will track both of
2161 * them in a extent.
2162 *
2163 * So we could return a unwritten and delayed extent, and
2164 * its block is equal to 'next'.
2165 */
2166 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002167 flags |= FIEMAP_EXTENT_LAST;
2168 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2169 next != EXT_MAX_BLOCKS)) {
2170 EXT4_ERROR_INODE(inode,
2171 "next extent == %u, next "
2172 "delalloc extent = %u",
2173 next, next_del);
2174 err = -EIO;
2175 break;
2176 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002177 }
2178
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002179 if (exists) {
2180 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002181 (__u64)es.es_lblk << blksize_bits,
2182 (__u64)es.es_pblk << blksize_bits,
2183 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002184 flags);
2185 if (err < 0)
2186 break;
2187 if (err == 1) {
2188 err = 0;
2189 break;
2190 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002191 }
2192
Zheng Liu69eb33d2013-02-18 00:31:07 -05002193 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002194 }
2195
2196 if (path) {
2197 ext4_ext_drop_refs(path);
2198 kfree(path);
2199 }
2200
2201 return err;
2202}
2203
Alex Tomasa86c6182006-10-11 01:21:03 -07002204/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002205 * ext4_ext_put_gap_in_cache:
2206 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002207 * and cache this gap
2208 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002209static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002210ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002211 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002212{
2213 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002214 unsigned long len;
2215 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002216 struct ext4_extent *ex;
2217
2218 ex = path[depth].p_ext;
2219 if (ex == NULL) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002220 /*
2221 * there is no extent yet, so gap is [0;-] and we
2222 * don't cache it
2223 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002224 ext_debug("cache gap(whole file):");
2225 } else if (block < le32_to_cpu(ex->ee_block)) {
2226 lblock = block;
2227 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002228 ext_debug("cache gap(before): %u [%u:%u]",
2229 block,
2230 le32_to_cpu(ex->ee_block),
2231 ext4_ext_get_actual_len(ex));
Zheng Liud100eef2013-02-18 00:29:59 -05002232 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2233 ext4_es_insert_extent(inode, lblock, len, ~0,
2234 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002235 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002236 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002237 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002238 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002239 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002240
2241 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002242 ext_debug("cache gap(after): [%u:%u] %u",
2243 le32_to_cpu(ex->ee_block),
2244 ext4_ext_get_actual_len(ex),
2245 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002246 BUG_ON(next == lblock);
2247 len = next - lblock;
Zheng Liud100eef2013-02-18 00:29:59 -05002248 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2249 ext4_es_insert_extent(inode, lblock, len, ~0,
2250 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002251 } else {
2252 lblock = len = 0;
2253 BUG();
2254 }
2255
Eric Sandeenbba90742008-01-28 23:58:27 -05002256 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002257}
2258
2259/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002260 * ext4_ext_rm_idx:
2261 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002262 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002263static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002264 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002265{
Alex Tomasa86c6182006-10-11 01:21:03 -07002266 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002267 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002268
2269 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002270 depth--;
2271 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002272 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002273 if (unlikely(path->p_hdr->eh_entries == 0)) {
2274 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2275 return -EIO;
2276 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002277 err = ext4_ext_get_access(handle, inode, path);
2278 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002279 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002280
2281 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2282 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2283 len *= sizeof(struct ext4_extent_idx);
2284 memmove(path->p_idx, path->p_idx + 1, len);
2285 }
2286
Marcin Slusarze8546d02008-04-17 10:38:59 -04002287 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002288 err = ext4_ext_dirty(handle, inode, path);
2289 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002290 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002291 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002292 trace_ext4_ext_rm_idx(inode, leaf);
2293
Peter Huewe7dc57612011-02-21 21:01:42 -05002294 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002295 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002296
2297 while (--depth >= 0) {
2298 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2299 break;
2300 path--;
2301 err = ext4_ext_get_access(handle, inode, path);
2302 if (err)
2303 break;
2304 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2305 err = ext4_ext_dirty(handle, inode, path);
2306 if (err)
2307 break;
2308 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002309 return err;
2310}
2311
2312/*
Mingming Caoee12b632008-08-19 22:16:05 -04002313 * ext4_ext_calc_credits_for_single_extent:
2314 * This routine returns max. credits that needed to insert an extent
2315 * to the extent tree.
2316 * When pass the actual path, the caller should calculate credits
2317 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002318 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002319int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002320 struct ext4_ext_path *path)
2321{
Alex Tomasa86c6182006-10-11 01:21:03 -07002322 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002323 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002324 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002325
Alex Tomasa86c6182006-10-11 01:21:03 -07002326 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002327 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002328 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2329
2330 /*
2331 * There are some space in the leaf tree, no
2332 * need to account for leaf block credit
2333 *
2334 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002335 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002336 * accounted.
2337 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002338 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002339 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002340 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002341 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002342 }
2343
Mingming Cao525f4ed2008-08-19 22:15:58 -04002344 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002345}
Alex Tomasa86c6182006-10-11 01:21:03 -07002346
Mingming Caoee12b632008-08-19 22:16:05 -04002347/*
2348 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2349 *
2350 * if nrblocks are fit in a single extent (chunk flag is 1), then
2351 * in the worse case, each tree level index/leaf need to be changed
2352 * if the tree split due to insert a new extent, then the old tree
2353 * index/leaf need to be updated too
2354 *
2355 * If the nrblocks are discontiguous, they could cause
2356 * the whole tree split more than once, but this is really rare.
2357 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002358int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04002359{
2360 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002361 int depth;
2362
2363 /* If we are converting the inline data, only one is needed here. */
2364 if (ext4_has_inline_data(inode))
2365 return 1;
2366
2367 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002368
Mingming Caoee12b632008-08-19 22:16:05 -04002369 if (chunk)
2370 index = depth * 2;
2371 else
2372 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002373
Mingming Caoee12b632008-08-19 22:16:05 -04002374 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002375}
2376
2377static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002378 struct ext4_extent *ex,
2379 ext4_fsblk_t *partial_cluster,
2380 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002381{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002382 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002383 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002384 ext4_fsblk_t pblk;
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002385 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002386
Alex Tomasc9de5602008-01-29 00:19:52 -05002387 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002388 flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2389 else if (ext4_should_journal_data(inode))
2390 flags |= EXT4_FREE_BLOCKS_FORGET;
2391
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002392 /*
2393 * For bigalloc file systems, we never free a partial cluster
2394 * at the beginning of the extent. Instead, we make a note
2395 * that we tried freeing the cluster, and check to see if we
2396 * need to free it on a subsequent call to ext4_remove_blocks,
2397 * or at the end of the ext4_truncate() operation.
2398 */
2399 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2400
Aditya Kalid8990242011-09-09 19:18:51 -04002401 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002402 /*
2403 * If we have a partial cluster, and it's different from the
2404 * cluster of the last block, we need to explicitly free the
2405 * partial cluster here.
2406 */
2407 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2408 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2409 ext4_free_blocks(handle, inode, NULL,
2410 EXT4_C2B(sbi, *partial_cluster),
2411 sbi->s_cluster_ratio, flags);
2412 *partial_cluster = 0;
2413 }
2414
Alex Tomasa86c6182006-10-11 01:21:03 -07002415#ifdef EXTENTS_STATS
2416 {
2417 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002418 spin_lock(&sbi->s_ext_stats_lock);
2419 sbi->s_ext_blocks += ee_len;
2420 sbi->s_ext_extents++;
2421 if (ee_len < sbi->s_ext_min)
2422 sbi->s_ext_min = ee_len;
2423 if (ee_len > sbi->s_ext_max)
2424 sbi->s_ext_max = ee_len;
2425 if (ext_depth(inode) > sbi->s_depth_max)
2426 sbi->s_depth_max = ext_depth(inode);
2427 spin_unlock(&sbi->s_ext_stats_lock);
2428 }
2429#endif
2430 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002431 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002432 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002433 ext4_lblk_t num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002434
Amit Aroraa2df2a62007-07-17 21:42:41 -04002435 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002436 pblk = ext4_ext_pblock(ex) + ee_len - num;
2437 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2438 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2439 /*
2440 * If the block range to be freed didn't start at the
2441 * beginning of a cluster, and we removed the entire
2442 * extent, save the partial cluster here, since we
2443 * might need to delete if we determine that the
2444 * truncate operation has removed all of the blocks in
2445 * the cluster.
2446 */
2447 if (pblk & (sbi->s_cluster_ratio - 1) &&
2448 (ee_len == num))
2449 *partial_cluster = EXT4_B2C(sbi, pblk);
2450 else
2451 *partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002452 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002453 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002454 /* head removal */
2455 ext4_lblk_t num;
2456 ext4_fsblk_t start;
2457
2458 num = to - from;
2459 start = ext4_ext_pblock(ex);
2460
2461 ext_debug("free first %u blocks starting %llu\n", num, start);
H Hartley Sweetenee90d572011-10-18 11:01:51 -04002462 ext4_free_blocks(handle, inode, NULL, start, num, flags);
Allison Hendersond583fb82011-05-25 07:41:43 -04002463
Alex Tomasa86c6182006-10-11 01:21:03 -07002464 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002465 printk(KERN_INFO "strange request: removal(2) "
2466 "%u-%u from %u:%u\n",
2467 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002468 }
2469 return 0;
2470}
2471
Allison Hendersond583fb82011-05-25 07:41:43 -04002472
2473/*
2474 * ext4_ext_rm_leaf() Removes the extents associated with the
2475 * blocks appearing between "start" and "end", and splits the extents
2476 * if "start" and "end" appear in the same extent
2477 *
2478 * @handle: The journal handle
2479 * @inode: The files inode
2480 * @path: The path to the leaf
2481 * @start: The first block to remove
2482 * @end: The last block to remove
2483 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002484static int
2485ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002486 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2487 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002488{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002489 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002490 int err = 0, correct_index = 0;
2491 int depth = ext_depth(inode), credits;
2492 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002493 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002494 unsigned num;
2495 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002496 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002497 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002498 struct ext4_extent *ex;
2499
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002500 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002501 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002502 if (!path[depth].p_hdr)
2503 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2504 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002505 if (unlikely(path[depth].p_hdr == NULL)) {
2506 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2507 return -EIO;
2508 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002509 /* find where to start removing */
2510 ex = EXT_LAST_EXTENT(eh);
2511
2512 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002513 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002514
Aditya Kalid8990242011-09-09 19:18:51 -04002515 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2516
Alex Tomasa86c6182006-10-11 01:21:03 -07002517 while (ex >= EXT_FIRST_EXTENT(eh) &&
2518 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002519
2520 if (ext4_ext_is_uninitialized(ex))
2521 uninitialized = 1;
2522 else
2523 uninitialized = 0;
2524
Mingming553f9002009-09-18 13:34:55 -04002525 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2526 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002527 path[depth].p_ext = ex;
2528
2529 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002530 b = ex_ee_block+ex_ee_len - 1 < end ?
2531 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002532
2533 ext_debug(" border %u:%u\n", a, b);
2534
Allison Hendersond583fb82011-05-25 07:41:43 -04002535 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002536 if (end < ex_ee_block) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002537 ex--;
2538 ex_ee_block = le32_to_cpu(ex->ee_block);
2539 ex_ee_len = ext4_ext_get_actual_len(ex);
2540 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002541 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002542 EXT4_ERROR_INODE(inode,
2543 "can not handle truncate %u:%u "
2544 "on extent %u:%u",
2545 start, end, ex_ee_block,
2546 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002547 err = -EIO;
2548 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002549 } else if (a != ex_ee_block) {
2550 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002551 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002552 } else {
2553 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002554 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002555 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002556 /*
2557 * 3 for leaf, sb, and inode plus 2 (bmap and group
2558 * descriptor) for each block group; assume two block
2559 * groups plus ex_ee_len/blocks_per_block_group for
2560 * the worst case
2561 */
2562 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002563 if (ex == EXT_FIRST_EXTENT(eh)) {
2564 correct_index = 1;
2565 credits += (ext_depth(inode)) + 1;
2566 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002567 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002568
Jan Kara487caee2009-08-17 22:17:20 -04002569 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002570 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002571 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002572
2573 err = ext4_ext_get_access(handle, inode, path + depth);
2574 if (err)
2575 goto out;
2576
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002577 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2578 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002579 if (err)
2580 goto out;
2581
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002582 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002583 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002584 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002585
Alex Tomasa86c6182006-10-11 01:21:03 -07002586 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002587 /*
2588 * Do not mark uninitialized if all the blocks in the
2589 * extent have been removed.
2590 */
2591 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002592 ext4_ext_mark_uninitialized(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002593 /*
2594 * If the extent was completely released,
2595 * we need to remove it from the leaf
2596 */
2597 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002598 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002599 /*
2600 * For hole punching, we need to scoot all the
2601 * extents up when an extent is removed so that
2602 * we dont have blank extents in the middle
2603 */
2604 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2605 sizeof(struct ext4_extent));
2606
2607 /* Now get rid of the one at the end */
2608 memset(EXT_LAST_EXTENT(eh), 0,
2609 sizeof(struct ext4_extent));
2610 }
2611 le16_add_cpu(&eh->eh_entries, -1);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002612 } else
2613 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002614
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002615 err = ext4_ext_dirty(handle, inode, path + depth);
2616 if (err)
2617 goto out;
2618
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002619 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002620 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002621 ex--;
2622 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002623 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002624 }
2625
2626 if (correct_index && eh->eh_entries)
2627 err = ext4_ext_correct_indexes(handle, inode, path);
2628
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002629 /*
2630 * If there is still a entry in the leaf node, check to see if
2631 * it references the partial cluster. This is the only place
2632 * where it could; if it doesn't, we can free the cluster.
2633 */
2634 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2635 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2636 *partial_cluster)) {
2637 int flags = EXT4_FREE_BLOCKS_FORGET;
2638
2639 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2640 flags |= EXT4_FREE_BLOCKS_METADATA;
2641
2642 ext4_free_blocks(handle, inode, NULL,
2643 EXT4_C2B(sbi, *partial_cluster),
2644 sbi->s_cluster_ratio, flags);
2645 *partial_cluster = 0;
2646 }
2647
Alex Tomasa86c6182006-10-11 01:21:03 -07002648 /* if this leaf is free, then we should
2649 * remove it from index block above */
2650 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002651 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002652
2653out:
2654 return err;
2655}
2656
2657/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002658 * ext4_ext_more_to_rm:
2659 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002660 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002661static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002662ext4_ext_more_to_rm(struct ext4_ext_path *path)
2663{
2664 BUG_ON(path->p_idx == NULL);
2665
2666 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2667 return 0;
2668
2669 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002670 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002671 * so we have to consider current index for truncation
2672 */
2673 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2674 return 0;
2675 return 1;
2676}
2677
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002678int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2679 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002680{
2681 struct super_block *sb = inode->i_sb;
2682 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002683 struct ext4_ext_path *path = NULL;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002684 ext4_fsblk_t partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002685 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002686 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002687
Lukas Czerner5f95d212012-03-19 23:03:19 -04002688 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002689
2690 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002691 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002692 if (IS_ERR(handle))
2693 return PTR_ERR(handle);
2694
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002695again:
Aditya Kalid8990242011-09-09 19:18:51 -04002696 trace_ext4_ext_remove_space(inode, start, depth);
2697
Alex Tomasa86c6182006-10-11 01:21:03 -07002698 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002699 * Check if we are removing extents inside the extent tree. If that
2700 * is the case, we are going to punch a hole inside the extent tree
2701 * so we have to check whether we need to split the extent covering
2702 * the last block to remove so we can easily remove the part of it
2703 * in ext4_ext_rm_leaf().
2704 */
2705 if (end < EXT_MAX_BLOCKS - 1) {
2706 struct ext4_extent *ex;
2707 ext4_lblk_t ee_block;
2708
2709 /* find extent for this block */
2710 path = ext4_ext_find_extent(inode, end, NULL);
2711 if (IS_ERR(path)) {
2712 ext4_journal_stop(handle);
2713 return PTR_ERR(path);
2714 }
2715 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002716 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002717 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002718 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002719 if (depth) {
2720 EXT4_ERROR_INODE(inode,
2721 "path[%d].p_hdr == NULL",
2722 depth);
2723 err = -EIO;
2724 }
2725 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002726 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002727
2728 ee_block = le32_to_cpu(ex->ee_block);
2729
2730 /*
2731 * See if the last block is inside the extent, if so split
2732 * the extent at 'end' block so we can easily remove the
2733 * tail of the first part of the split extent in
2734 * ext4_ext_rm_leaf().
2735 */
2736 if (end >= ee_block &&
2737 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2738 int split_flag = 0;
2739
2740 if (ext4_ext_is_uninitialized(ex))
2741 split_flag = EXT4_EXT_MARK_UNINIT1 |
2742 EXT4_EXT_MARK_UNINIT2;
2743
2744 /*
2745 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002746 * block in the first new extent. Also we should not
2747 * fail removing space due to ENOSPC so try to use
2748 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002749 */
2750 err = ext4_split_extent_at(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04002751 end + 1, split_flag,
2752 EXT4_GET_BLOCKS_PRE_IO |
2753 EXT4_GET_BLOCKS_METADATA_NOFAIL);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002754
2755 if (err < 0)
2756 goto out;
2757 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002758 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002759 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002760 * We start scanning from right side, freeing all the blocks
2761 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002762 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002763 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002764 if (path) {
2765 int k = i = depth;
2766 while (--k > 0)
2767 path[k].p_block =
2768 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2769 } else {
2770 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2771 GFP_NOFS);
2772 if (path == NULL) {
2773 ext4_journal_stop(handle);
2774 return -ENOMEM;
2775 }
2776 path[0].p_depth = depth;
2777 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002778 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002779
Ashish Sangwan968dee72012-07-22 22:49:08 -04002780 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2781 err = -EIO;
2782 goto out;
2783 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002784 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002785 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002786
2787 while (i >= 0 && err == 0) {
2788 if (i == depth) {
2789 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002790 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002791 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002792 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002793 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002794 brelse(path[i].p_bh);
2795 path[i].p_bh = NULL;
2796 i--;
2797 continue;
2798 }
2799
2800 /* this is index block */
2801 if (!path[i].p_hdr) {
2802 ext_debug("initialize header\n");
2803 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002804 }
2805
Alex Tomasa86c6182006-10-11 01:21:03 -07002806 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002807 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002808 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2809 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2810 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2811 path[i].p_hdr,
2812 le16_to_cpu(path[i].p_hdr->eh_entries));
2813 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002814 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002815 path[i].p_idx--;
2816 }
2817
2818 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2819 i, EXT_FIRST_INDEX(path[i].p_hdr),
2820 path[i].p_idx);
2821 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002822 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002823 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002824 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002825 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002826 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'obf89d162010-10-27 21:30:14 -04002827 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002828 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002829 /* should we reset i_size? */
2830 err = -EIO;
2831 break;
2832 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002833 if (WARN_ON(i + 1 > depth)) {
2834 err = -EIO;
2835 break;
2836 }
Darrick J. Wongf8489122012-04-29 18:21:10 -04002837 if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2838 depth - i - 1, bh)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002839 err = -EIO;
2840 break;
2841 }
2842 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002843
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002844 /* save actual number of indexes since this
2845 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002846 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2847 i++;
2848 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002849 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002850 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002851 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002852 * handle must be already prepared by the
2853 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002854 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002855 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002856 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002857 brelse(path[i].p_bh);
2858 path[i].p_bh = NULL;
2859 i--;
2860 ext_debug("return to level %d\n", i);
2861 }
2862 }
2863
Aditya Kalid8990242011-09-09 19:18:51 -04002864 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2865 path->p_hdr->eh_entries);
2866
Aditya Kali7b415bf2011-09-09 19:04:51 -04002867 /* If we still have something in the partial cluster and we have removed
2868 * even the first extent, then we should free the blocks in the partial
2869 * cluster as well. */
2870 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2871 int flags = EXT4_FREE_BLOCKS_FORGET;
2872
2873 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2874 flags |= EXT4_FREE_BLOCKS_METADATA;
2875
2876 ext4_free_blocks(handle, inode, NULL,
2877 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2878 EXT4_SB(sb)->s_cluster_ratio, flags);
2879 partial_cluster = 0;
2880 }
2881
Alex Tomasa86c6182006-10-11 01:21:03 -07002882 /* TODO: flexible tree reduction should be here */
2883 if (path->p_hdr->eh_entries == 0) {
2884 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002885 * truncate to zero freed all the tree,
2886 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002887 */
2888 err = ext4_ext_get_access(handle, inode, path);
2889 if (err == 0) {
2890 ext_inode_hdr(inode)->eh_depth = 0;
2891 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002892 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002893 err = ext4_ext_dirty(handle, inode, path);
2894 }
2895 }
2896out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002897 ext4_ext_drop_refs(path);
2898 kfree(path);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002899 if (err == -EAGAIN) {
2900 path = NULL;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002901 goto again;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002902 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002903 ext4_journal_stop(handle);
2904
2905 return err;
2906}
2907
2908/*
2909 * called at mount time
2910 */
2911void ext4_ext_init(struct super_block *sb)
2912{
2913 /*
2914 * possible initialization would be here
2915 */
2916
Theodore Ts'o83982b62009-01-06 14:53:16 -05002917 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002918#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04002919 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002920#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04002921 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07002922#endif
2923#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04002924 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07002925#endif
2926#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04002927 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07002928#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04002929 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04002930#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07002931#ifdef EXTENTS_STATS
2932 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2933 EXT4_SB(sb)->s_ext_min = 1 << 30;
2934 EXT4_SB(sb)->s_ext_max = 0;
2935#endif
2936 }
2937}
2938
2939/*
2940 * called at umount time
2941 */
2942void ext4_ext_release(struct super_block *sb)
2943{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002944 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002945 return;
2946
2947#ifdef EXTENTS_STATS
2948 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2949 struct ext4_sb_info *sbi = EXT4_SB(sb);
2950 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2951 sbi->s_ext_blocks, sbi->s_ext_extents,
2952 sbi->s_ext_blocks / sbi->s_ext_extents);
2953 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2954 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2955 }
2956#endif
2957}
2958
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002959/* FIXME!! we need to try to merge to left or right after zero-out */
2960static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2961{
Lukas Czerner24075182010-10-27 21:30:06 -04002962 ext4_fsblk_t ee_pblock;
2963 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04002964 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002965
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002966 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04002967 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002968
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002969 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04002970 if (ret > 0)
2971 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002972
Lukas Czerner24075182010-10-27 21:30:06 -04002973 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002974}
2975
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002976/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002977 * ext4_split_extent_at() splits an extent at given block.
2978 *
2979 * @handle: the journal handle
2980 * @inode: the file inode
2981 * @path: the path to the extent
2982 * @split: the logical block where the extent is splitted.
2983 * @split_flags: indicates if the extent could be zeroout if split fails, and
2984 * the states(init or uninit) of new extents.
2985 * @flags: flags used to insert new extent to extent tree.
2986 *
2987 *
2988 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2989 * of which are deterimined by split_flag.
2990 *
2991 * There are two cases:
2992 * a> the extent are splitted into two extent.
2993 * b> split is not needed, and just mark the extent.
2994 *
2995 * return 0 on success.
2996 */
2997static int ext4_split_extent_at(handle_t *handle,
2998 struct inode *inode,
2999 struct ext4_ext_path *path,
3000 ext4_lblk_t split,
3001 int split_flag,
3002 int flags)
3003{
3004 ext4_fsblk_t newblock;
3005 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04003006 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003007 struct ext4_extent *ex2 = NULL;
3008 unsigned int ee_len, depth;
3009 int err = 0;
3010
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003011 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3012 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3013
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003014 ext_debug("ext4_split_extents_at: inode %lu, logical"
3015 "block %llu\n", inode->i_ino, (unsigned long long)split);
3016
3017 ext4_ext_show_leaf(inode, path);
3018
3019 depth = ext_depth(inode);
3020 ex = path[depth].p_ext;
3021 ee_block = le32_to_cpu(ex->ee_block);
3022 ee_len = ext4_ext_get_actual_len(ex);
3023 newblock = split - ee_block + ext4_ext_pblock(ex);
3024
3025 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003026 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3027 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3028 EXT4_EXT_MARK_UNINIT1 |
3029 EXT4_EXT_MARK_UNINIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003030
3031 err = ext4_ext_get_access(handle, inode, path + depth);
3032 if (err)
3033 goto out;
3034
3035 if (split == ee_block) {
3036 /*
3037 * case b: block @split is the block that the extent begins with
3038 * then we just change the state of the extent, and splitting
3039 * is not needed.
3040 */
3041 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3042 ext4_ext_mark_uninitialized(ex);
3043 else
3044 ext4_ext_mark_initialized(ex);
3045
3046 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003047 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003048
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003049 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003050 goto out;
3051 }
3052
3053 /* case a */
3054 memcpy(&orig_ex, ex, sizeof(orig_ex));
3055 ex->ee_len = cpu_to_le16(split - ee_block);
3056 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3057 ext4_ext_mark_uninitialized(ex);
3058
3059 /*
3060 * path may lead to new leaf, not to original leaf any more
3061 * after ext4_ext_insert_extent() returns,
3062 */
3063 err = ext4_ext_dirty(handle, inode, path + depth);
3064 if (err)
3065 goto fix_extent_len;
3066
3067 ex2 = &newex;
3068 ex2->ee_block = cpu_to_le32(split);
3069 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3070 ext4_ext_store_pblock(ex2, newblock);
3071 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3072 ext4_ext_mark_uninitialized(ex2);
3073
3074 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3075 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003076 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003077 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003078 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003079 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003080 zero_ex.ee_len = cpu_to_le16(
3081 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003082 ext4_ext_store_pblock(&zero_ex,
3083 ext4_ext_pblock(ex2));
3084 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003085 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003086 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003087 zero_ex.ee_len = cpu_to_le16(
3088 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003089 ext4_ext_store_pblock(&zero_ex,
3090 ext4_ext_pblock(ex));
3091 }
3092 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003093 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003094 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003095 zero_ex.ee_len = cpu_to_le16(
3096 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003097 ext4_ext_store_pblock(&zero_ex,
3098 ext4_ext_pblock(&orig_ex));
3099 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003100
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003101 if (err)
3102 goto fix_extent_len;
3103 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003104 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003105 ext4_ext_try_to_merge(handle, inode, path, ex);
3106 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003107 if (err)
3108 goto fix_extent_len;
3109
3110 /* update extent status tree */
3111 err = ext4_es_zeroout(inode, &zero_ex);
3112
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003113 goto out;
3114 } else if (err)
3115 goto fix_extent_len;
3116
3117out:
3118 ext4_ext_show_leaf(inode, path);
3119 return err;
3120
3121fix_extent_len:
3122 ex->ee_len = orig_ex.ee_len;
3123 ext4_ext_dirty(handle, inode, path + depth);
3124 return err;
3125}
3126
3127/*
3128 * ext4_split_extents() splits an extent and mark extent which is covered
3129 * by @map as split_flags indicates
3130 *
3131 * It may result in splitting the extent into multiple extents (upto three)
3132 * There are three possibilities:
3133 * a> There is no split required
3134 * b> Splits in two extents: Split is happening at either end of the extent
3135 * c> Splits in three extents: Somone is splitting in middle of the extent
3136 *
3137 */
3138static int ext4_split_extent(handle_t *handle,
3139 struct inode *inode,
3140 struct ext4_ext_path *path,
3141 struct ext4_map_blocks *map,
3142 int split_flag,
3143 int flags)
3144{
3145 ext4_lblk_t ee_block;
3146 struct ext4_extent *ex;
3147 unsigned int ee_len, depth;
3148 int err = 0;
3149 int uninitialized;
3150 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003151 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003152
3153 depth = ext_depth(inode);
3154 ex = path[depth].p_ext;
3155 ee_block = le32_to_cpu(ex->ee_block);
3156 ee_len = ext4_ext_get_actual_len(ex);
3157 uninitialized = ext4_ext_is_uninitialized(ex);
3158
3159 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003160 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003161 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3162 if (uninitialized)
3163 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3164 EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003165 if (split_flag & EXT4_EXT_DATA_VALID2)
3166 split_flag1 |= EXT4_EXT_DATA_VALID1;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003167 err = ext4_split_extent_at(handle, inode, path,
3168 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003169 if (err)
3170 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003171 } else {
3172 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003173 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003174 /*
3175 * Update path is required because previous ext4_split_extent_at() may
3176 * result in split of original leaf or extent zeroout.
3177 */
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003178 ext4_ext_drop_refs(path);
3179 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3180 if (IS_ERR(path))
3181 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003182 depth = ext_depth(inode);
3183 ex = path[depth].p_ext;
3184 uninitialized = ext4_ext_is_uninitialized(ex);
3185 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003186
3187 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003188 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3189 if (uninitialized) {
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003190 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003191 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3192 EXT4_EXT_MARK_UNINIT2);
3193 }
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003194 err = ext4_split_extent_at(handle, inode, path,
3195 map->m_lblk, split_flag1, flags);
3196 if (err)
3197 goto out;
3198 }
3199
3200 ext4_ext_show_leaf(inode, path);
3201out:
Zheng Liu3a225672013-03-10 21:20:23 -04003202 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003203}
3204
Amit Arora56055d32007-07-17 21:42:38 -04003205/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003206 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04003207 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003208 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04003209 * uninitialized).
3210 * There are three possibilities:
3211 * a> There is no split required: Entire extent should be initialized
3212 * b> Splits in two extents: Write is happening at either end of the extent
3213 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003214 *
3215 * Pre-conditions:
3216 * - The extent pointed to by 'path' is uninitialized.
3217 * - The extent pointed to by 'path' contains a superset
3218 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3219 *
3220 * Post-conditions on success:
3221 * - the returned value is the number of blocks beyond map->l_lblk
3222 * that are allocated and initialized.
3223 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003224 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003225static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003226 struct inode *inode,
3227 struct ext4_map_blocks *map,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003228 struct ext4_ext_path *path,
3229 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003230{
Zheng Liu67a5da52012-08-17 09:54:17 -04003231 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003232 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003233 struct ext4_map_blocks split_map;
3234 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003235 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003236 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003237 unsigned int ee_len, depth, map_len = map->m_len;
3238 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003239 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003240 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003241
3242 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3243 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003244 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003245
Zheng Liu67a5da52012-08-17 09:54:17 -04003246 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003247 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3248 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003249 if (eof_block < map->m_lblk + map_len)
3250 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003251
3252 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003253 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003254 ex = path[depth].p_ext;
3255 ee_block = le32_to_cpu(ex->ee_block);
3256 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003257 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003258
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003259 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3260
3261 /* Pre-conditions */
3262 BUG_ON(!ext4_ext_is_uninitialized(ex));
3263 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003264
3265 /*
3266 * Attempt to transfer newly initialized blocks from the currently
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003267 * uninitialized extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003268 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003269 * memmove() calls. Transferring to the left is the common case in
3270 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3271 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003272 *
3273 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003274 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003275 * This would require removing the extent if the transfer
3276 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003277 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003278 * same extent tree node.
3279 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003280 if ((map->m_lblk == ee_block) &&
3281 /* See if we can merge left */
3282 (map_len < ee_len) && /*L1*/
3283 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003284 ext4_lblk_t prev_lblk;
3285 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003286 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003287
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003288 abut_ex = ex - 1;
3289 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3290 prev_len = ext4_ext_get_actual_len(abut_ex);
3291 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003292 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003293
3294 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003295 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003296 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003297 * - C1: abut_ex is initialized,
3298 * - C2: abut_ex is logically abutting ex,
3299 * - C3: abut_ex is physically abutting ex,
3300 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003301 * overflowing the (initialized) length limit.
3302 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003303 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003304 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3305 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003306 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003307 err = ext4_ext_get_access(handle, inode, path + depth);
3308 if (err)
3309 goto out;
3310
3311 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003312 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003313
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003314 /* Shift the start of ex by 'map_len' blocks */
3315 ex->ee_block = cpu_to_le32(ee_block + map_len);
3316 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3317 ex->ee_len = cpu_to_le16(ee_len - map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003318 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3319
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003320 /* Extend abut_ex by 'map_len' blocks */
3321 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003322
3323 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003324 allocated = map_len;
3325 }
3326 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3327 (map_len < ee_len) && /*L1*/
3328 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3329 /* See if we can merge right */
3330 ext4_lblk_t next_lblk;
3331 ext4_fsblk_t next_pblk, ee_pblk;
3332 unsigned int next_len;
3333
3334 abut_ex = ex + 1;
3335 next_lblk = le32_to_cpu(abut_ex->ee_block);
3336 next_len = ext4_ext_get_actual_len(abut_ex);
3337 next_pblk = ext4_ext_pblock(abut_ex);
3338 ee_pblk = ext4_ext_pblock(ex);
3339
3340 /*
3341 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3342 * upon those conditions:
3343 * - C1: abut_ex is initialized,
3344 * - C2: abut_ex is logically abutting ex,
3345 * - C3: abut_ex is physically abutting ex,
3346 * - C4: abut_ex can receive the additional blocks without
3347 * overflowing the (initialized) length limit.
3348 */
3349 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3350 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3351 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3352 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3353 err = ext4_ext_get_access(handle, inode, path + depth);
3354 if (err)
3355 goto out;
3356
3357 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3358 map, ex, abut_ex);
3359
3360 /* Shift the start of abut_ex by 'map_len' blocks */
3361 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3362 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3363 ex->ee_len = cpu_to_le16(ee_len - map_len);
3364 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3365
3366 /* Extend abut_ex by 'map_len' blocks */
3367 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3368
3369 /* Result: number of initialized blocks past m_lblk */
3370 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003371 }
3372 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003373 if (allocated) {
3374 /* Mark the block containing both extents as dirty */
3375 ext4_ext_dirty(handle, inode, path + depth);
3376
3377 /* Update path to point to the right extent */
3378 path[depth].p_ext = abut_ex;
3379 goto out;
3380 } else
3381 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003382
Yongqiang Yang667eff32011-05-03 12:25:07 -04003383 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003384 /*
3385 * It is safe to convert extent to initialized via explicit
3386 * zeroout only if extent is fully insde i_size or new_size.
3387 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003388 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003389
Zheng Liu67a5da52012-08-17 09:54:17 -04003390 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3391 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003392 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003393
3394 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3395 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003396 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003397 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003398 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003399 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003400 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003401 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003402
3403 err = ext4_ext_get_access(handle, inode, path + depth);
3404 if (err)
3405 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003406 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003407 ext4_ext_try_to_merge(handle, inode, path, ex);
3408 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003409 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003410 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003411
Amit Arora56055d32007-07-17 21:42:38 -04003412 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003413 * four cases:
3414 * 1. split the extent into three extents.
3415 * 2. split the extent into two extents, zeroout the first half.
3416 * 3. split the extent into two extents, zeroout the second half.
3417 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003418 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003419 split_map.m_lblk = map->m_lblk;
3420 split_map.m_len = map->m_len;
3421
Zheng Liu67a5da52012-08-17 09:54:17 -04003422 if (max_zeroout && (allocated > map->m_len)) {
3423 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003424 /* case 3 */
3425 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003426 cpu_to_le32(map->m_lblk);
3427 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003428 ext4_ext_store_pblock(&zero_ex,
3429 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3430 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003431 if (err)
3432 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003433 split_map.m_lblk = map->m_lblk;
3434 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003435 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003436 /* case 2 */
3437 if (map->m_lblk != ee_block) {
3438 zero_ex.ee_block = ex->ee_block;
3439 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3440 ee_block);
3441 ext4_ext_store_pblock(&zero_ex,
3442 ext4_ext_pblock(ex));
3443 err = ext4_ext_zeroout(inode, &zero_ex);
3444 if (err)
3445 goto out;
3446 }
3447
Yongqiang Yang667eff32011-05-03 12:25:07 -04003448 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003449 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3450 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003451 }
3452 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003453
3454 allocated = ext4_split_extent(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003455 &split_map, split_flag, flags);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003456 if (allocated < 0)
3457 err = allocated;
3458
Amit Arora56055d32007-07-17 21:42:38 -04003459out:
Zheng Liuadb23552013-03-10 21:13:05 -04003460 /* If we have gotten a failure, don't zero out status tree */
3461 if (!err)
3462 err = ext4_es_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003463 return err ? err : allocated;
3464}
3465
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003466/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003467 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003468 * ext4_get_blocks_dio_write() when DIO to write
3469 * to an uninitialized extent.
3470 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003471 * Writing to an uninitialized extent may result in splitting the uninitialized
Wang Sheng-Hui30cb27d2012-08-18 22:38:07 -04003472 * extent into multiple initialized/uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003473 * There are three possibilities:
3474 * a> There is no split required: Entire extent should be uninitialized
3475 * b> Splits in two extents: Write is happening at either end of the extent
3476 * c> Splits in three extents: Somone is writing in middle of the extent
3477 *
3478 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003479 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003480 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003481 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003482 * into three uninitialized extent(at most). After IO complete, the part
3483 * being filled will be convert to initialized by the end_io callback function
3484 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003485 *
3486 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003487 */
3488static int ext4_split_unwritten_extents(handle_t *handle,
3489 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003490 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003491 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003492 int flags)
3493{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003494 ext4_lblk_t eof_block;
3495 ext4_lblk_t ee_block;
3496 struct ext4_extent *ex;
3497 unsigned int ee_len;
3498 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003499
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003500 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3501 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003502 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003503
3504 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3505 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003506 if (eof_block < map->m_lblk + map->m_len)
3507 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003508 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003509 * It is safe to convert extent to initialized via explicit
3510 * zeroout only if extent is fully insde i_size or new_size.
3511 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003512 depth = ext_depth(inode);
3513 ex = path[depth].p_ext;
3514 ee_block = le32_to_cpu(ex->ee_block);
3515 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003516
Yongqiang Yang667eff32011-05-03 12:25:07 -04003517 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3518 split_flag |= EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003519 if (flags & EXT4_GET_BLOCKS_CONVERT)
3520 split_flag |= EXT4_EXT_DATA_VALID2;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003521 flags |= EXT4_GET_BLOCKS_PRE_IO;
3522 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003523}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003524
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003525static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003526 struct inode *inode,
3527 struct ext4_map_blocks *map,
3528 struct ext4_ext_path *path)
Mingming Cao00314622009-09-28 15:49:08 -04003529{
3530 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003531 ext4_lblk_t ee_block;
3532 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003533 int depth;
3534 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003535
3536 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003537 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003538 ee_block = le32_to_cpu(ex->ee_block);
3539 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003540
Yongqiang Yang197217a2011-05-03 11:45:29 -04003541 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3542 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003543 (unsigned long long)ee_block, ee_len);
3544
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003545 /* If extent is larger than requested it is a clear sign that we still
3546 * have some extent state machine issues left. So extent_split is still
3547 * required.
3548 * TODO: Once all related issues will be fixed this situation should be
3549 * illegal.
3550 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003551 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003552#ifdef EXT4_DEBUG
3553 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3554 " len %u; IO logical block %llu, len %u\n",
3555 inode->i_ino, (unsigned long long)ee_block, ee_len,
3556 (unsigned long long)map->m_lblk, map->m_len);
3557#endif
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003558 err = ext4_split_unwritten_extents(handle, inode, map, path,
3559 EXT4_GET_BLOCKS_CONVERT);
3560 if (err < 0)
3561 goto out;
3562 ext4_ext_drop_refs(path);
3563 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3564 if (IS_ERR(path)) {
3565 err = PTR_ERR(path);
3566 goto out;
3567 }
3568 depth = ext_depth(inode);
3569 ex = path[depth].p_ext;
3570 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003571
Mingming Cao00314622009-09-28 15:49:08 -04003572 err = ext4_ext_get_access(handle, inode, path + depth);
3573 if (err)
3574 goto out;
3575 /* first mark the extent as initialized */
3576 ext4_ext_mark_initialized(ex);
3577
Yongqiang Yang197217a2011-05-03 11:45:29 -04003578 /* note: ext4_ext_correct_indexes() isn't needed here because
3579 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003580 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003581 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003582
Mingming Cao00314622009-09-28 15:49:08 -04003583 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003584 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003585out:
3586 ext4_ext_show_leaf(inode, path);
3587 return err;
3588}
3589
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003590static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3591 sector_t block, int count)
3592{
3593 int i;
3594 for (i = 0; i < count; i++)
3595 unmap_underlying_metadata(bdev, block + i);
3596}
3597
Theodore Ts'o58590b02010-10-27 21:23:12 -04003598/*
3599 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3600 */
3601static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003602 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003603 struct ext4_ext_path *path,
3604 unsigned int len)
3605{
3606 int i, depth;
3607 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003608 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003609
3610 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3611 return 0;
3612
3613 depth = ext_depth(inode);
3614 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003615
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003616 /*
3617 * We're going to remove EOFBLOCKS_FL entirely in future so we
3618 * do not care for this case anymore. Simply remove the flag
3619 * if there are no extents.
3620 */
3621 if (unlikely(!eh->eh_entries))
3622 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003623 last_ex = EXT_LAST_EXTENT(eh);
3624 /*
3625 * We should clear the EOFBLOCKS_FL flag if we are writing the
3626 * last block in the last extent in the file. We test this by
3627 * first checking to see if the caller to
3628 * ext4_ext_get_blocks() was interested in the last block (or
3629 * a block beyond the last block) in the current extent. If
3630 * this turns out to be false, we can bail out from this
3631 * function immediately.
3632 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003633 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003634 ext4_ext_get_actual_len(last_ex))
3635 return 0;
3636 /*
3637 * If the caller does appear to be planning to write at or
3638 * beyond the end of the current extent, we then test to see
3639 * if the current extent is the last extent in the file, by
3640 * checking to make sure it was reached via the rightmost node
3641 * at each level of the tree.
3642 */
3643 for (i = depth-1; i >= 0; i--)
3644 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3645 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003646out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003647 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3648 return ext4_mark_inode_dirty(handle, inode);
3649}
3650
Aditya Kali7b415bf2011-09-09 19:04:51 -04003651/**
3652 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3653 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003654 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003655 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003656int ext4_find_delalloc_range(struct inode *inode,
3657 ext4_lblk_t lblk_start,
3658 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003659{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003660 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003661
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003662 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003663 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003664 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003665 else if (es.es_lblk <= lblk_start &&
3666 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003667 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003668 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003669 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003670 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003671 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003672}
3673
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003674int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003675{
3676 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3677 ext4_lblk_t lblk_start, lblk_end;
3678 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3679 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3680
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003681 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003682}
3683
3684/**
3685 * Determines how many complete clusters (out of those specified by the 'map')
3686 * are under delalloc and were reserved quota for.
3687 * This function is called when we are writing out the blocks that were
3688 * originally written with their allocation delayed, but then the space was
3689 * allocated using fallocate() before the delayed allocation could be resolved.
3690 * The cases to look for are:
3691 * ('=' indicated delayed allocated blocks
3692 * '-' indicates non-delayed allocated blocks)
3693 * (a) partial clusters towards beginning and/or end outside of allocated range
3694 * are not delalloc'ed.
3695 * Ex:
3696 * |----c---=|====c====|====c====|===-c----|
3697 * |++++++ allocated ++++++|
3698 * ==> 4 complete clusters in above example
3699 *
3700 * (b) partial cluster (outside of allocated range) towards either end is
3701 * marked for delayed allocation. In this case, we will exclude that
3702 * cluster.
3703 * Ex:
3704 * |----====c========|========c========|
3705 * |++++++ allocated ++++++|
3706 * ==> 1 complete clusters in above example
3707 *
3708 * Ex:
3709 * |================c================|
3710 * |++++++ allocated ++++++|
3711 * ==> 0 complete clusters in above example
3712 *
3713 * The ext4_da_update_reserve_space will be called only if we
3714 * determine here that there were some "entire" clusters that span
3715 * this 'allocated' range.
3716 * In the non-bigalloc case, this function will just end up returning num_blks
3717 * without ever calling ext4_find_delalloc_range.
3718 */
3719static unsigned int
3720get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3721 unsigned int num_blks)
3722{
3723 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3724 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3725 ext4_lblk_t lblk_from, lblk_to, c_offset;
3726 unsigned int allocated_clusters = 0;
3727
3728 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3729 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3730
3731 /* max possible clusters for this allocation */
3732 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3733
Aditya Kalid8990242011-09-09 19:18:51 -04003734 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3735
Aditya Kali7b415bf2011-09-09 19:04:51 -04003736 /* Check towards left side */
3737 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3738 if (c_offset) {
3739 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3740 lblk_to = lblk_from + c_offset - 1;
3741
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003742 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003743 allocated_clusters--;
3744 }
3745
3746 /* Now check towards right. */
3747 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3748 if (allocated_clusters && c_offset) {
3749 lblk_from = lblk_start + num_blks;
3750 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3751
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003752 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003753 allocated_clusters--;
3754 }
3755
3756 return allocated_clusters;
3757}
3758
Mingming Cao00314622009-09-28 15:49:08 -04003759static int
3760ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003761 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003762 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003763 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003764{
3765 int ret = 0;
3766 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003767 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003768
Zheng Liu88635ca2011-12-28 19:00:25 -05003769 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3770 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003771 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003772 flags, allocated);
3773 ext4_ext_show_leaf(inode, path);
3774
Lukas Czerner27dd4382013-04-09 22:11:22 -04003775 /*
3776 * When writing into uninitialized space, we should not fail to
3777 * allocate metadata blocks for the new extent block if needed.
3778 */
3779 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3780
Zheng Liub5645532012-11-08 14:33:43 -05003781 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3782 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04003783
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003784 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003785 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003786 ret = ext4_split_unwritten_extents(handle, inode, map,
3787 path, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04003788 if (ret <= 0)
3789 goto out;
Mingming5f5249502009-11-10 10:48:04 -05003790 /*
3791 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003792 * that this IO needs to conversion to written when IO is
Mingming5f5249502009-11-10 10:48:04 -05003793 * completed
3794 */
Tao Ma0edeb712011-10-31 17:30:44 -04003795 if (io)
3796 ext4_set_io_unwritten_flag(inode, io);
3797 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003798 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05003799 map->m_flags |= EXT4_MAP_UNWRITTEN;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003800 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003801 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003802 goto out;
3803 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003804 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003805 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003806 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Mingming Cao00314622009-09-28 15:49:08 -04003807 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003808 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003809 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003810 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3811 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003812 } else
3813 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04003814 map->m_flags |= EXT4_MAP_MAPPED;
3815 if (allocated > map->m_len)
3816 allocated = map->m_len;
3817 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003818 goto out2;
3819 }
3820 /* buffered IO case */
3821 /*
3822 * repeat fallocate creation request
3823 * we already have an unwritten extent
3824 */
Zheng Liua25a4e12013-02-18 00:28:04 -05003825 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3826 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003827 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05003828 }
Mingming Cao00314622009-09-28 15:49:08 -04003829
3830 /* buffered READ or buffered write_begin() lookup */
3831 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3832 /*
3833 * We have blocks reserved already. We
3834 * return allocated blocks so that delalloc
3835 * won't do block reservation for us. But
3836 * the buffer head will be unmapped so that
3837 * a read from the block returns 0s.
3838 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003839 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003840 goto out1;
3841 }
3842
3843 /* buffered write, writepage time, convert*/
Lukas Czerner27dd4382013-04-09 22:11:22 -04003844 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003845 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05003846 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04003847out:
3848 if (ret <= 0) {
3849 err = ret;
3850 goto out2;
3851 } else
3852 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003853 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003854 /*
3855 * if we allocated more blocks than requested
3856 * we need to make sure we unmap the extra block
3857 * allocated. The actual needed block will get
3858 * unmapped later when we find the buffer_head marked
3859 * new.
3860 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003861 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003862 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003863 newblock + map->m_len,
3864 allocated - map->m_len);
3865 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003866 }
Zheng Liu3a225672013-03-10 21:20:23 -04003867 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003868
3869 /*
3870 * If we have done fallocate with the offset that is already
3871 * delayed allocated, we would have block reservation
3872 * and quota reservation done in the delayed write path.
3873 * But fallocate would have already updated quota and block
3874 * count for this offset. So cancel these reservation
3875 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003876 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3877 unsigned int reserved_clusters;
3878 reserved_clusters = get_reserved_cluster_alloc(inode,
3879 map->m_lblk, map->m_len);
3880 if (reserved_clusters)
3881 ext4_da_update_reserve_space(inode,
3882 reserved_clusters,
3883 0);
3884 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003885
Mingming Cao00314622009-09-28 15:49:08 -04003886map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003887 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003888 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3889 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3890 map->m_len);
3891 if (err < 0)
3892 goto out2;
3893 }
Mingming Cao00314622009-09-28 15:49:08 -04003894out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003895 if (allocated > map->m_len)
3896 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003897 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003898 map->m_pblk = newblock;
3899 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003900out2:
3901 if (path) {
3902 ext4_ext_drop_refs(path);
3903 kfree(path);
3904 }
3905 return err ? err : allocated;
3906}
Theodore Ts'o58590b02010-10-27 21:23:12 -04003907
Mingming Cao00314622009-09-28 15:49:08 -04003908/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003909 * get_implied_cluster_alloc - check to see if the requested
3910 * allocation (in the map structure) overlaps with a cluster already
3911 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04003912 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003913 * @map The requested lblk->pblk mapping
3914 * @ex The extent structure which might contain an implied
3915 * cluster allocation
3916 *
3917 * This function is called by ext4_ext_map_blocks() after we failed to
3918 * find blocks that were already in the inode's extent tree. Hence,
3919 * we know that the beginning of the requested region cannot overlap
3920 * the extent from the inode's extent tree. There are three cases we
3921 * want to catch. The first is this case:
3922 *
3923 * |--- cluster # N--|
3924 * |--- extent ---| |---- requested region ---|
3925 * |==========|
3926 *
3927 * The second case that we need to test for is this one:
3928 *
3929 * |--------- cluster # N ----------------|
3930 * |--- requested region --| |------- extent ----|
3931 * |=======================|
3932 *
3933 * The third case is when the requested region lies between two extents
3934 * within the same cluster:
3935 * |------------- cluster # N-------------|
3936 * |----- ex -----| |---- ex_right ----|
3937 * |------ requested region ------|
3938 * |================|
3939 *
3940 * In each of the above cases, we need to set the map->m_pblk and
3941 * map->m_len so it corresponds to the return the extent labelled as
3942 * "|====|" from cluster #N, since it is already in use for data in
3943 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3944 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3945 * as a new "allocated" block region. Otherwise, we will return 0 and
3946 * ext4_ext_map_blocks() will then allocate one or more new clusters
3947 * by calling ext4_mb_new_blocks().
3948 */
Aditya Kalid8990242011-09-09 19:18:51 -04003949static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003950 struct ext4_map_blocks *map,
3951 struct ext4_extent *ex,
3952 struct ext4_ext_path *path)
3953{
Aditya Kalid8990242011-09-09 19:18:51 -04003954 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003955 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3956 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05003957 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003958 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3959 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3960 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3961
3962 /* The extent passed in that we are trying to match */
3963 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3964 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3965
3966 /* The requested region passed into ext4_map_blocks() */
3967 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003968
3969 if ((rr_cluster_start == ex_cluster_end) ||
3970 (rr_cluster_start == ex_cluster_start)) {
3971 if (rr_cluster_start == ex_cluster_end)
3972 ee_start += ee_len - 1;
3973 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3974 c_offset;
3975 map->m_len = min(map->m_len,
3976 (unsigned) sbi->s_cluster_ratio - c_offset);
3977 /*
3978 * Check for and handle this case:
3979 *
3980 * |--------- cluster # N-------------|
3981 * |------- extent ----|
3982 * |--- requested region ---|
3983 * |===========|
3984 */
3985
3986 if (map->m_lblk < ee_block)
3987 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3988
3989 /*
3990 * Check for the case where there is already another allocated
3991 * block to the right of 'ex' but before the end of the cluster.
3992 *
3993 * |------------- cluster # N-------------|
3994 * |----- ex -----| |---- ex_right ----|
3995 * |------ requested region ------|
3996 * |================|
3997 */
3998 if (map->m_lblk > ee_block) {
3999 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4000 map->m_len = min(map->m_len, next - map->m_lblk);
4001 }
Aditya Kalid8990242011-09-09 19:18:51 -04004002
4003 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004004 return 1;
4005 }
Aditya Kalid8990242011-09-09 19:18:51 -04004006
4007 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004008 return 0;
4009}
4010
4011
4012/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05004013 * Block allocation/map/preallocation routine for extents based files
4014 *
4015 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004016 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004017 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4018 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004019 *
4020 * return > 0, number of of blocks already mapped/allocated
4021 * if create == 0 and these are pre-allocated blocks
4022 * buffer head is unmapped
4023 * otherwise blocks are mapped
4024 *
4025 * return = 0, if plain look up failed (blocks have not been allocated)
4026 * buffer head is unmapped
4027 *
4028 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004029 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004030int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4031 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004032{
4033 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004034 struct ext4_extent newex, *ex, *ex2;
4035 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004036 ext4_fsblk_t newblock = 0;
Zheng Liu37794732012-11-08 14:47:52 -05004037 int free_on_err = 0, err = 0, depth;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004038 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004039 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004040 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004041 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004042 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004043 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07004044
Mingming84fe3be2009-09-01 08:44:37 -04004045 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004046 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004047 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004048
Alex Tomasa86c6182006-10-11 01:21:03 -07004049 /* find extent for this block */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004050 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07004051 if (IS_ERR(path)) {
4052 err = PTR_ERR(path);
4053 path = NULL;
4054 goto out2;
4055 }
4056
4057 depth = ext_depth(inode);
4058
4059 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004060 * consistent leaf must not be empty;
4061 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07004062 * this is why assert can't be put in ext4_ext_find_extent()
4063 */
Frank Mayhar273df552010-03-02 11:46:09 -05004064 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4065 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004066 "lblock: %lu, depth: %d pblock %lld",
4067 (unsigned long) map->m_lblk, depth,
4068 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004069 err = -EIO;
4070 goto out2;
4071 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004072
Avantika Mathur7e028972006-12-06 20:41:33 -08004073 ex = path[depth].p_ext;
4074 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004075 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004076 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004077 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004078
4079 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004080 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004081 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004082 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004083 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004084
4085 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4086
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004087 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004088 if (in_range(map->m_lblk, ee_block, ee_len)) {
4089 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004090 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004091 allocated = ee_len - (map->m_lblk - ee_block);
4092 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4093 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004094
Zheng Liu69eb33d2013-02-18 00:31:07 -05004095 if (!ext4_ext_is_uninitialized(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004096 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004097
Zheng Liu37794732012-11-08 14:47:52 -05004098 allocated = ext4_ext_handle_uninitialized_extents(
Lukas Czerner78771912012-03-19 23:05:43 -04004099 handle, inode, map, path, flags,
4100 allocated, newblock);
Zheng Liu37794732012-11-08 14:47:52 -05004101 goto out3;
Alex Tomasa86c6182006-10-11 01:21:03 -07004102 }
4103 }
4104
Aditya Kali7b415bf2011-09-09 19:04:51 -04004105 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05004106 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04004107 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4108
Alex Tomasa86c6182006-10-11 01:21:03 -07004109 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004110 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004111 * we couldn't try to create block if create flag is zero
4112 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004113 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004114 /*
4115 * put just found gap into cache to speed up
4116 * subsequent requests
4117 */
Zheng Liud100eef2013-02-18 00:29:59 -05004118 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4119 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004120 goto out2;
4121 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004122
Alex Tomasa86c6182006-10-11 01:21:03 -07004123 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004124 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004125 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004126 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004127 newex.ee_block = cpu_to_le32(map->m_lblk);
4128 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4129
4130 /*
4131 * If we are doing bigalloc, check to see if the extent returned
4132 * by ext4_ext_find_extent() implies a cluster we can use.
4133 */
4134 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004135 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004136 ar.len = allocated = map->m_len;
4137 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004138 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004139 goto got_allocated_blocks;
4140 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004141
Alex Tomasc9de5602008-01-29 00:19:52 -05004142 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004143 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004144 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4145 if (err)
4146 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004147 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004148 ex2 = NULL;
4149 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004150 if (err)
4151 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004152
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004153 /* Check if the extent after searching to the right implies a
4154 * cluster we can use. */
4155 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004156 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004157 ar.len = allocated = map->m_len;
4158 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004159 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004160 goto got_allocated_blocks;
4161 }
4162
Amit Arora749269f2007-07-18 09:02:56 -04004163 /*
4164 * See if request is beyond maximum number of blocks we can have in
4165 * a single extent. For an initialized extent this limit is
4166 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4167 * EXT_UNINIT_MAX_LEN.
4168 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004169 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004170 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004171 map->m_len = EXT_INIT_MAX_LEN;
4172 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004173 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004174 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004175
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004176 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004177 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004178 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004179 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004180 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004181 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004182 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004183
4184 /* allocate new block */
4185 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004186 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4187 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004188 /*
4189 * We calculate the offset from the beginning of the cluster
4190 * for the logical block number, since when we allocate a
4191 * physical cluster, the physical block should start at the
4192 * same offset from the beginning of the cluster. This is
4193 * needed so that future calls to get_implied_cluster_alloc()
4194 * work correctly.
4195 */
4196 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4197 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4198 ar.goal -= offset;
4199 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004200 if (S_ISREG(inode->i_mode))
4201 ar.flags = EXT4_MB_HINT_DATA;
4202 else
4203 /* disable in-core preallocation for non-regular files */
4204 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004205 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4206 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004207 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004208 if (!newblock)
4209 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004210 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004211 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004212 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004213 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004214 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4215 if (ar.len > allocated)
4216 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004217
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004218got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004219 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004220 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004221 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004222 /* Mark uninitialized */
4223 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04004224 ext4_ext_mark_uninitialized(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004225 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004226 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004227 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004228 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004229 * here we flag the IO that really needs the conversion.
Mingming5f5249502009-11-10 10:48:04 -05004230 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004231 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004232 */
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004233 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4234 set_unwritten = 1;
Jiaying Zhang744692d2010-03-04 16:14:02 -05004235 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004236 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004237 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004238
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004239 err = 0;
4240 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4241 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4242 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004243 if (!err)
4244 err = ext4_ext_insert_extent(handle, inode, path,
4245 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004246
4247 if (!err && set_unwritten) {
4248 if (io)
4249 ext4_set_io_unwritten_flag(inode, io);
4250 else
4251 ext4_set_inode_state(inode,
4252 EXT4_STATE_DIO_UNWRITTEN);
4253 }
4254
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004255 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004256 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4257 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004258 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004259 /* not a good idea to call discard here directly,
4260 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004261 ext4_discard_preallocations(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004262 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
Maxim Patlasov7132de72011-07-10 19:37:48 -04004263 ext4_ext_get_actual_len(&newex), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004264 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004265 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004266
Alex Tomasa86c6182006-10-11 01:21:03 -07004267 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004268 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004269 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004270 if (allocated > map->m_len)
4271 allocated = map->m_len;
4272 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004273
Jan Karab436b9b2009-12-08 23:51:10 -05004274 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004275 * Update reserved blocks/metadata blocks after successful
4276 * block allocation which had been deferred till now.
4277 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004278 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004279 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004280 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004281 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004282 */
4283 reserved_clusters = get_reserved_cluster_alloc(inode,
4284 map->m_lblk, allocated);
4285 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4286 if (reserved_clusters) {
4287 /*
4288 * We have clusters reserved for this range.
4289 * But since we are not doing actual allocation
4290 * and are simply using blocks from previously
4291 * allocated cluster, we should release the
4292 * reservation and not claim quota.
4293 */
4294 ext4_da_update_reserve_space(inode,
4295 reserved_clusters, 0);
4296 }
4297 } else {
4298 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004299 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f2612011-09-09 19:20:51 -04004300 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004301 int reservation = allocated_clusters -
4302 reserved_clusters;
4303 /*
4304 * It seems we claimed few clusters outside of
4305 * the range of this allocation. We should give
4306 * it back to the reservation pool. This can
4307 * happen in the following case:
4308 *
4309 * * Suppose s_cluster_ratio is 4 (i.e., each
4310 * cluster has 4 blocks. Thus, the clusters
4311 * are [0-3],[4-7],[8-11]...
4312 * * First comes delayed allocation write for
4313 * logical blocks 10 & 11. Since there were no
4314 * previous delayed allocated blocks in the
4315 * range [8-11], we would reserve 1 cluster
4316 * for this write.
4317 * * Next comes write for logical blocks 3 to 8.
4318 * In this case, we will reserve 2 clusters
4319 * (for [0-3] and [4-7]; and not for [8-11] as
4320 * that range has a delayed allocated blocks.
4321 * Thus total reserved clusters now becomes 3.
4322 * * Now, during the delayed allocation writeout
4323 * time, we will first write blocks [3-8] and
4324 * allocate 3 clusters for writing these
4325 * blocks. Also, we would claim all these
4326 * three clusters above.
4327 * * Now when we come here to writeout the
4328 * blocks [10-11], we would expect to claim
4329 * the reservation of 1 cluster we had made
4330 * (and we would claim it since there are no
4331 * more delayed allocated blocks in the range
4332 * [8-11]. But our reserved cluster count had
4333 * already gone to 0.
4334 *
4335 * Thus, at the step 4 above when we determine
4336 * that there are still some unwritten delayed
4337 * allocated blocks outside of our current
4338 * block range, we should increment the
4339 * reserved clusters count so that when the
4340 * remaining blocks finally gets written, we
4341 * could claim them.
4342 */
Aditya Kali5356f2612011-09-09 19:20:51 -04004343 dquot_reserve_block(inode,
4344 EXT4_C2B(sbi, reservation));
4345 spin_lock(&ei->i_block_reservation_lock);
4346 ei->i_reserved_data_blocks += reservation;
4347 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004348 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004349 /*
4350 * We will claim quota for all newly allocated blocks.
4351 * We're updating the reserved space *after* the
4352 * correction above so we do not accidentally free
4353 * all the metadata reservation because we might
4354 * actually need it later on.
4355 */
4356 ext4_da_update_reserve_space(inode, allocated_clusters,
4357 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004358 }
4359 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004360
4361 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004362 * Cache the extent and update transaction to commit on fdatasync only
4363 * when it is _not_ an uninitialized extent.
4364 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004365 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004366 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004367 else
Jan Karab436b9b2009-12-08 23:51:10 -05004368 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004369out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004370 if (allocated > map->m_len)
4371 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004372 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004373 map->m_flags |= EXT4_MAP_MAPPED;
4374 map->m_pblk = newblock;
4375 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004376out2:
4377 if (path) {
4378 ext4_ext_drop_refs(path);
4379 kfree(path);
4380 }
Allison Hendersone8613042011-05-25 07:41:46 -04004381
Zheng Liu37794732012-11-08 14:47:52 -05004382out3:
Zheng Liu19b303d2012-11-08 14:34:04 -05004383 trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated);
Yongqiang Yange7b319e2011-10-29 09:39:51 -04004384
Lukas Czerner78771912012-03-19 23:05:43 -04004385 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004386}
4387
Theodore Ts'o819c4922013-04-03 12:47:17 -04004388void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004389{
Alex Tomasa86c6182006-10-11 01:21:03 -07004390 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004391 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004392 int err = 0;
4393
4394 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004395 * TODO: optimization is possible here.
4396 * Probably we need not scan at all,
4397 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004398 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004399
4400 /* we have to know where to truncate from in crash case */
4401 EXT4_I(inode)->i_disksize = inode->i_size;
4402 ext4_mark_inode_dirty(handle, inode);
4403
4404 last_block = (inode->i_size + sb->s_blocksize - 1)
4405 >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'oc07ae682013-07-15 00:09:19 -04004406retry:
Zheng Liu51865fd2012-11-08 21:57:32 -05004407 err = ext4_es_remove_extent(inode, last_block,
4408 EXT_MAX_BLOCKS - last_block);
Theodore Ts'oc7039e92013-07-29 12:12:56 -04004409 if (err == -ENOMEM) {
Theodore Ts'oc07ae682013-07-15 00:09:19 -04004410 cond_resched();
4411 congestion_wait(BLK_RW_ASYNC, HZ/50);
4412 goto retry;
4413 }
4414 if (err) {
4415 ext4_std_error(inode->i_sb, err);
4416 return;
4417 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04004418 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Theodore Ts'oc07ae682013-07-15 00:09:19 -04004419 ext4_std_error(inode->i_sb, err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004420}
4421
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004422static void ext4_falloc_update_inode(struct inode *inode,
4423 int mode, loff_t new_size, int update_ctime)
4424{
4425 struct timespec now;
4426
4427 if (update_ctime) {
4428 now = current_fs_time(inode->i_sb);
4429 if (!timespec_equal(&inode->i_ctime, &now))
4430 inode->i_ctime = now;
4431 }
4432 /*
4433 * Update only when preallocation was requested beyond
4434 * the file size.
4435 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04004436 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4437 if (new_size > i_size_read(inode))
4438 i_size_write(inode, new_size);
4439 if (new_size > EXT4_I(inode)->i_disksize)
4440 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004441 } else {
4442 /*
4443 * Mark that we allocate beyond EOF so the subsequent truncate
4444 * can proceed even if the new size is the same as i_size.
4445 */
4446 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004447 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004448 }
4449
4450}
4451
Amit Aroraa2df2a62007-07-17 21:42:41 -04004452/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004453 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04004454 * operation, which gets called from sys_fallocate system call.
4455 * For block-mapped files, posix_fallocate should fall back to the method
4456 * of writing zeroes to the required new blocks (the same behavior which is
4457 * expected for file systems which do not support fallocate() system call).
4458 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004459long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004460{
Al Viro496ad9a2013-01-23 17:07:38 -05004461 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004462 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004463 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004464 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004465 int ret = 0;
4466 int ret2 = 0;
4467 int retries = 0;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004468 int flags;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004469 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004470 unsigned int credits, blkbits = inode->i_blkbits;
4471
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004472 /* Return error if mode is not supported */
4473 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4474 return -EOPNOTSUPP;
4475
4476 if (mode & FALLOC_FL_PUNCH_HOLE)
4477 return ext4_punch_hole(file, offset, len);
4478
Tao Ma0c8d4142012-12-10 14:06:03 -05004479 ret = ext4_convert_inline_data(inode);
4480 if (ret)
4481 return ret;
4482
Zheng Liu8bad6fc2013-01-28 09:21:37 -05004483 /*
4484 * currently supporting (pre)allocate mode for extent-based
4485 * files _only_
4486 */
4487 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4488 return -EOPNOTSUPP;
4489
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004490 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004491 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004492 /*
4493 * We can't just convert len to max_blocks because
4494 * If blocksize = 4096 offset = 3072 and len = 2048
4495 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004496 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004497 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004498 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04004499 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04004500 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004501 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004502 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004503 ret = inode_newsize_ok(inode, (len + offset));
4504 if (ret) {
4505 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004506 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004507 return ret;
4508 }
Greg Harm3c6fe772011-10-31 18:41:47 -04004509 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004510 if (mode & FALLOC_FL_KEEP_SIZE)
4511 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
Greg Harm3c6fe772011-10-31 18:41:47 -04004512 /*
4513 * Don't normalize the request if it can fit in one extent so
4514 * that it doesn't get unnecessarily split into multiple
4515 * extents.
4516 */
4517 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4518 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004519
Amit Aroraa2df2a62007-07-17 21:42:41 -04004520retry:
4521 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004522 map.m_lblk = map.m_lblk + ret;
4523 map.m_len = max_blocks = max_blocks - ret;
Theodore Ts'o9924a922013-02-08 21:59:22 -05004524 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4525 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004526 if (IS_ERR(handle)) {
4527 ret = PTR_ERR(handle);
4528 break;
4529 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004530 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004531 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004532#ifdef EXT4FS_DEBUG
Lukas Czernerb06acd32013-01-28 21:21:12 -05004533 ext4_warning(inode->i_sb,
4534 "inode #%lu: block %u: len %u: "
4535 "ext4_ext_map_blocks returned %d",
4536 inode->i_ino, map.m_lblk,
4537 map.m_len, ret);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004538#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04004539 ext4_mark_inode_dirty(handle, inode);
4540 ret2 = ext4_journal_stop(handle);
4541 break;
4542 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004543 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004544 blkbits) >> blkbits))
4545 new_size = offset + len;
4546 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04004547 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004548
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004549 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004550 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04004551 ext4_mark_inode_dirty(handle, inode);
Zheng Liuf4e95b32012-06-30 19:12:57 -04004552 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4553 ext4_handle_sync(handle);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004554 ret2 = ext4_journal_stop(handle);
4555 if (ret2)
4556 break;
4557 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004558 if (ret == -ENOSPC &&
4559 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4560 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004561 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004562 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004563 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004564 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4565 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004566 return ret > 0 ? ret2 : ret;
4567}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004568
4569/*
Mingming Cao00314622009-09-28 15:49:08 -04004570 * This function convert a range of blocks to written extents
4571 * The caller of this function will pass the start offset and the size.
4572 * all unwritten extents within this range will be converted to
4573 * written extents.
4574 *
4575 * This function is called from the direct IO end io call back
4576 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004577 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004578 */
4579int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
Eric Sandeena1de02d2010-02-04 23:58:38 -05004580 ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004581{
4582 handle_t *handle;
Mingming Cao00314622009-09-28 15:49:08 -04004583 unsigned int max_blocks;
4584 int ret = 0;
4585 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004586 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004587 unsigned int credits, blkbits = inode->i_blkbits;
4588
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004589 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004590 /*
4591 * We can't just convert len to max_blocks because
4592 * If blocksize = 4096 offset = 3072 and len = 2048
4593 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004594 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4595 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004596 /*
4597 * credits to insert 1 extent into extent tree
4598 */
4599 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4600 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004601 map.m_lblk += ret;
4602 map.m_len = (max_blocks -= ret);
Theodore Ts'o9924a922013-02-08 21:59:22 -05004603 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
Mingming Cao00314622009-09-28 15:49:08 -04004604 if (IS_ERR(handle)) {
4605 ret = PTR_ERR(handle);
4606 break;
4607 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004608 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004609 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05004610 if (ret <= 0)
4611 ext4_warning(inode->i_sb,
4612 "inode #%lu: block %u: len %u: "
4613 "ext4_ext_map_blocks returned %d",
4614 inode->i_ino, map.m_lblk,
4615 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04004616 ext4_mark_inode_dirty(handle, inode);
4617 ret2 = ext4_journal_stop(handle);
4618 if (ret <= 0 || ret2 )
4619 break;
4620 }
4621 return ret > 0 ? ret2 : ret;
4622}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004623
Mingming Cao00314622009-09-28 15:49:08 -04004624/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004625 * If newes is not existing extent (newes->ec_pblk equals zero) find
4626 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004627 * return start of the next delayed extent.
4628 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05004629 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004630 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05004631 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004632 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004633static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05004634 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004635{
Zheng Liub3aff3e2012-11-08 21:57:37 -05004636 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05004637 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004638
Zheng Liu69eb33d2013-02-18 00:31:07 -05004639 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004640 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4641 newes->es_lblk + newes->es_len - 1, &es);
4642
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004643 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004644 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004645 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004646 */
Zheng Liu06b0c882013-02-18 00:26:51 -05004647 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05004648 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004649 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004650
Zheng Liu69eb33d2013-02-18 00:31:07 -05004651 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05004652 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004653 newes->es_len = min(es.es_lblk - newes->es_lblk,
4654 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004655 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004656 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004657
Zheng Liu69eb33d2013-02-18 00:31:07 -05004658 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004659 }
4660
Zheng Liu69eb33d2013-02-18 00:31:07 -05004661 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004662 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05004663 if (es.es_len == 0)
4664 next_del = EXT_MAX_BLOCKS;
4665 else
4666 next_del = es.es_lblk;
4667
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004668 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004669}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004670/* fiemap flags we can handle specified here */
4671#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4672
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004673static int ext4_xattr_fiemap(struct inode *inode,
4674 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004675{
4676 __u64 physical = 0;
4677 __u64 length;
4678 __u32 flags = FIEMAP_EXTENT_LAST;
4679 int blockbits = inode->i_sb->s_blocksize_bits;
4680 int error = 0;
4681
4682 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004683 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004684 struct ext4_iloc iloc;
4685 int offset; /* offset of xattr in inode */
4686
4687 error = ext4_get_inode_loc(inode, &iloc);
4688 if (error)
4689 return error;
Jan Karab9ea8422013-05-31 19:38:56 -04004690 physical = (__u64)iloc.bh->b_blocknr << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004691 offset = EXT4_GOOD_OLD_INODE_SIZE +
4692 EXT4_I(inode)->i_extra_isize;
4693 physical += offset;
4694 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4695 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004696 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004697 } else { /* external block */
Jan Karab9ea8422013-05-31 19:38:56 -04004698 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004699 length = inode->i_sb->s_blocksize;
4700 }
4701
4702 if (physical)
4703 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4704 length, flags);
4705 return (error < 0 ? error : 0);
4706}
4707
4708int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4709 __u64 start, __u64 len)
4710{
4711 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004712 int error = 0;
4713
Tao Ma94191982012-12-10 14:06:02 -05004714 if (ext4_has_inline_data(inode)) {
4715 int has_inline = 1;
4716
4717 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4718
4719 if (has_inline)
4720 return error;
4721 }
4722
Eric Sandeen6873fa02008-10-07 00:46:36 -04004723 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004724 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004725 return generic_block_fiemap(inode, fieinfo, start, len,
4726 ext4_get_block);
4727
4728 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4729 return -EBADR;
4730
4731 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4732 error = ext4_xattr_fiemap(inode, fieinfo);
4733 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004734 ext4_lblk_t len_blks;
4735 __u64 last_blk;
4736
Eric Sandeen6873fa02008-10-07 00:46:36 -04004737 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004738 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004739 if (last_blk >= EXT_MAX_BLOCKS)
4740 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004741 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004742
4743 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004744 * Walk the extent tree gathering extent information
4745 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004746 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004747 error = ext4_fill_fiemap_extents(inode, start_blk,
4748 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004749 }
4750
4751 return error;
4752}