blob: 74bc2d549c58bd57d60ebcb0a143a3fdf4e95997 [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 Guan5946d082013-12-03 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 Guan5946d082013-12-03 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 Guan5946d082013-12-03 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 Guan5946d082013-12-03 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 Guan5946d082013-12-03 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,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400427 int depth, ext4_fsblk_t pblk)
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'oc3491792013-08-16 21:21:41 -0400467 "pblk %llu bad header/extent: %s - magic %x, "
468 "entries %u, max %u(%u), depth %u(%u)",
469 (unsigned long long) pblk, error_msg,
470 le16_to_cpu(eh->eh_magic),
471 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
472 max, le16_to_cpu(eh->eh_depth), depth);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400473 return -EIO;
474}
475
Theodore Ts'oc3491792013-08-16 21:21:41 -0400476#define ext4_ext_check(inode, eh, depth, pblk) \
477 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
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{
Theodore Ts'oc3491792013-08-16 21:21:41 -0400481 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400482}
483
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400484static struct buffer_head *
485__read_extent_tree_block(const char *function, unsigned int line,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400486 struct inode *inode, ext4_fsblk_t pblk, int depth,
487 int flags)
Darrick J. Wongf8489122012-04-29 18:21:10 -0400488{
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400489 struct buffer_head *bh;
490 int err;
Darrick J. Wongf8489122012-04-29 18:21:10 -0400491
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400492 bh = sb_getblk(inode->i_sb, pblk);
493 if (unlikely(!bh))
494 return ERR_PTR(-ENOMEM);
495
496 if (!bh_uptodate_or_lock(bh)) {
497 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
498 err = bh_submit_read(bh);
499 if (err < 0)
500 goto errout;
501 }
Theodore Ts'o7869a4a2013-08-16 22:05:14 -0400502 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400503 return bh;
504 err = __ext4_ext_check(function, line, inode,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400505 ext_block_hdr(bh), depth, pblk);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400506 if (err)
507 goto errout;
Darrick J. Wongf8489122012-04-29 18:21:10 -0400508 set_buffer_verified(bh);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400509 /*
510 * If this is a leaf block, cache all of its entries
511 */
512 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
513 struct ext4_extent_header *eh = ext_block_hdr(bh);
514 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
515 ext4_lblk_t prev = 0;
516 int i;
517
518 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
519 unsigned int status = EXTENT_STATUS_WRITTEN;
520 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
521 int len = ext4_ext_get_actual_len(ex);
522
523 if (prev && (prev != lblk))
524 ext4_es_cache_extent(inode, prev,
525 lblk - prev, ~0,
526 EXTENT_STATUS_HOLE);
527
528 if (ext4_ext_is_uninitialized(ex))
529 status = EXTENT_STATUS_UNWRITTEN;
530 ext4_es_cache_extent(inode, lblk, len,
531 ext4_ext_pblock(ex), status);
532 prev = lblk + len;
533 }
534 }
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400535 return bh;
536errout:
537 put_bh(bh);
538 return ERR_PTR(err);
539
Darrick J. Wongf8489122012-04-29 18:21:10 -0400540}
541
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400542#define read_extent_tree_block(inode, pblk, depth, flags) \
543 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
544 (depth), (flags))
Darrick J. Wongf8489122012-04-29 18:21:10 -0400545
Theodore Ts'o7869a4a2013-08-16 22:05:14 -0400546/*
547 * This function is called to cache a file's extent information in the
548 * extent status tree
549 */
550int ext4_ext_precache(struct inode *inode)
551{
552 struct ext4_inode_info *ei = EXT4_I(inode);
553 struct ext4_ext_path *path = NULL;
554 struct buffer_head *bh;
555 int i = 0, depth, ret = 0;
556
557 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
558 return 0; /* not an extent-mapped inode */
559
560 down_read(&ei->i_data_sem);
561 depth = ext_depth(inode);
562
563 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
564 GFP_NOFS);
565 if (path == NULL) {
566 up_read(&ei->i_data_sem);
567 return -ENOMEM;
568 }
569
570 /* Don't cache anything if there are no external extent blocks */
571 if (depth == 0)
572 goto out;
573 path[0].p_hdr = ext_inode_hdr(inode);
574 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
575 if (ret)
576 goto out;
577 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
578 while (i >= 0) {
579 /*
580 * If this is a leaf block or we've reached the end of
581 * the index block, go up
582 */
583 if ((i == depth) ||
584 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
585 brelse(path[i].p_bh);
586 path[i].p_bh = NULL;
587 i--;
588 continue;
589 }
590 bh = read_extent_tree_block(inode,
591 ext4_idx_pblock(path[i].p_idx++),
592 depth - i - 1,
593 EXT4_EX_FORCE_CACHE);
594 if (IS_ERR(bh)) {
595 ret = PTR_ERR(bh);
596 break;
597 }
598 i++;
599 path[i].p_bh = bh;
600 path[i].p_hdr = ext_block_hdr(bh);
601 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
602 }
603 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
604out:
605 up_read(&ei->i_data_sem);
606 ext4_ext_drop_refs(path);
607 kfree(path);
608 return ret;
609}
610
Alex Tomasa86c6182006-10-11 01:21:03 -0700611#ifdef EXT_DEBUG
612static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
613{
614 int k, l = path->p_depth;
615
616 ext_debug("path:");
617 for (k = 0; k <= l; k++, path++) {
618 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700619 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400620 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700621 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400622 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700623 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400624 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400625 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400626 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700627 } else
628 ext_debug(" []");
629 }
630 ext_debug("\n");
631}
632
633static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
634{
635 int depth = ext_depth(inode);
636 struct ext4_extent_header *eh;
637 struct ext4_extent *ex;
638 int i;
639
640 if (!path)
641 return;
642
643 eh = path[depth].p_hdr;
644 ex = EXT_FIRST_EXTENT(eh);
645
Mingming553f9002009-09-18 13:34:55 -0400646 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
647
Alex Tomasa86c6182006-10-11 01:21:03 -0700648 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400649 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
650 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400651 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700652 }
653 ext_debug("\n");
654}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400655
656static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
657 ext4_fsblk_t newblock, int level)
658{
659 int depth = ext_depth(inode);
660 struct ext4_extent *ex;
661
662 if (depth != level) {
663 struct ext4_extent_idx *idx;
664 idx = path[level].p_idx;
665 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
666 ext_debug("%d: move %d:%llu in new index %llu\n", level,
667 le32_to_cpu(idx->ei_block),
668 ext4_idx_pblock(idx),
669 newblock);
670 idx++;
671 }
672
673 return;
674 }
675
676 ex = path[depth].p_ext;
677 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
678 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
679 le32_to_cpu(ex->ee_block),
680 ext4_ext_pblock(ex),
681 ext4_ext_is_uninitialized(ex),
682 ext4_ext_get_actual_len(ex),
683 newblock);
684 ex++;
685 }
686}
687
Alex Tomasa86c6182006-10-11 01:21:03 -0700688#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400689#define ext4_ext_show_path(inode, path)
690#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400691#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700692#endif
693
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500694void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700695{
696 int depth = path->p_depth;
697 int i;
698
699 for (i = 0; i <= depth; i++, path++)
700 if (path->p_bh) {
701 brelse(path->p_bh);
702 path->p_bh = NULL;
703 }
704}
705
706/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700707 * ext4_ext_binsearch_idx:
708 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400709 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700710 */
711static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500712ext4_ext_binsearch_idx(struct inode *inode,
713 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700714{
715 struct ext4_extent_header *eh = path->p_hdr;
716 struct ext4_extent_idx *r, *l, *m;
717
Alex Tomasa86c6182006-10-11 01:21:03 -0700718
Eric Sandeenbba90742008-01-28 23:58:27 -0500719 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700720
721 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400722 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700723 while (l <= r) {
724 m = l + (r - l) / 2;
725 if (block < le32_to_cpu(m->ei_block))
726 r = m - 1;
727 else
728 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400729 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
730 m, le32_to_cpu(m->ei_block),
731 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700732 }
733
734 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400735 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400736 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700737
738#ifdef CHECK_BINSEARCH
739 {
740 struct ext4_extent_idx *chix, *ix;
741 int k;
742
743 chix = ix = EXT_FIRST_INDEX(eh);
744 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
745 if (k != 0 &&
746 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400747 printk(KERN_DEBUG "k=%d, ix=0x%p, "
748 "first=0x%p\n", k,
749 ix, EXT_FIRST_INDEX(eh));
750 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700751 le32_to_cpu(ix->ei_block),
752 le32_to_cpu(ix[-1].ei_block));
753 }
754 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400755 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700756 if (block < le32_to_cpu(ix->ei_block))
757 break;
758 chix = ix;
759 }
760 BUG_ON(chix != path->p_idx);
761 }
762#endif
763
764}
765
766/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700767 * ext4_ext_binsearch:
768 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400769 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700770 */
771static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500772ext4_ext_binsearch(struct inode *inode,
773 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700774{
775 struct ext4_extent_header *eh = path->p_hdr;
776 struct ext4_extent *r, *l, *m;
777
Alex Tomasa86c6182006-10-11 01:21:03 -0700778 if (eh->eh_entries == 0) {
779 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700780 * this leaf is empty:
781 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700782 */
783 return;
784 }
785
Eric Sandeenbba90742008-01-28 23:58:27 -0500786 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700787
788 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400789 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700790
791 while (l <= r) {
792 m = l + (r - l) / 2;
793 if (block < le32_to_cpu(m->ee_block))
794 r = m - 1;
795 else
796 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400797 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
798 m, le32_to_cpu(m->ee_block),
799 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700800 }
801
802 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400803 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400804 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400805 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400806 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400807 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700808
809#ifdef CHECK_BINSEARCH
810 {
811 struct ext4_extent *chex, *ex;
812 int k;
813
814 chex = ex = EXT_FIRST_EXTENT(eh);
815 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
816 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400817 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700818 if (block < le32_to_cpu(ex->ee_block))
819 break;
820 chex = ex;
821 }
822 BUG_ON(chex != path->p_ext);
823 }
824#endif
825
826}
827
828int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
829{
830 struct ext4_extent_header *eh;
831
832 eh = ext_inode_hdr(inode);
833 eh->eh_depth = 0;
834 eh->eh_entries = 0;
835 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400836 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700837 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700838 return 0;
839}
840
841struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500842ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400843 struct ext4_ext_path *path, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700844{
845 struct ext4_extent_header *eh;
846 struct buffer_head *bh;
847 short int depth, i, ppos = 0, alloc = 0;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500848 int ret;
Alex Tomasa86c6182006-10-11 01:21:03 -0700849
850 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400851 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700852
853 /* account possible depth increase */
854 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800855 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700856 GFP_NOFS);
857 if (!path)
858 return ERR_PTR(-ENOMEM);
859 alloc = 1;
860 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700861 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400862 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700863
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400864 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700865 /* walk through the tree */
866 while (i) {
867 ext_debug("depth %d: num %d, max %d\n",
868 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400869
Alex Tomasa86c6182006-10-11 01:21:03 -0700870 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400871 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700872 path[ppos].p_depth = i;
873 path[ppos].p_ext = NULL;
874
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400875 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
876 flags);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400877 if (IS_ERR(bh)) {
878 ret = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700879 goto err;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500880 }
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400881
Alex Tomasa86c6182006-10-11 01:21:03 -0700882 eh = ext_block_hdr(bh);
883 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500884 if (unlikely(ppos > depth)) {
885 put_bh(bh);
886 EXT4_ERROR_INODE(inode,
887 "ppos %d > depth %d", ppos, depth);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500888 ret = -EIO;
Frank Mayhar273df552010-03-02 11:46:09 -0500889 goto err;
890 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700891 path[ppos].p_bh = bh;
892 path[ppos].p_hdr = eh;
Alex Tomasa86c6182006-10-11 01:21:03 -0700893 }
894
895 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700896 path[ppos].p_ext = NULL;
897 path[ppos].p_idx = NULL;
898
Alex Tomasa86c6182006-10-11 01:21:03 -0700899 /* find extent */
900 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400901 /* if not an empty leaf */
902 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400903 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700904
905 ext4_ext_show_path(inode, path);
906
907 return path;
908
909err:
910 ext4_ext_drop_refs(path);
911 if (alloc)
912 kfree(path);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500913 return ERR_PTR(ret);
Alex Tomasa86c6182006-10-11 01:21:03 -0700914}
915
916/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700917 * ext4_ext_insert_index:
918 * insert new index [@logical;@ptr] into the block at @curp;
919 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700920 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400921static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
922 struct ext4_ext_path *curp,
923 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700924{
925 struct ext4_extent_idx *ix;
926 int len, err;
927
Avantika Mathur7e028972006-12-06 20:41:33 -0800928 err = ext4_ext_get_access(handle, inode, curp);
929 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700930 return err;
931
Frank Mayhar273df552010-03-02 11:46:09 -0500932 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
933 EXT4_ERROR_INODE(inode,
934 "logical %d == ei_block %d!",
935 logical, le32_to_cpu(curp->p_idx->ei_block));
936 return -EIO;
937 }
Robin Dongd4620312011-07-17 23:43:42 -0400938
939 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
940 >= le16_to_cpu(curp->p_hdr->eh_max))) {
941 EXT4_ERROR_INODE(inode,
942 "eh_entries %d >= eh_max %d!",
943 le16_to_cpu(curp->p_hdr->eh_entries),
944 le16_to_cpu(curp->p_hdr->eh_max));
945 return -EIO;
946 }
947
Alex Tomasa86c6182006-10-11 01:21:03 -0700948 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
949 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400950 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700951 ix = curp->p_idx + 1;
952 } else {
953 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400954 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700955 ix = curp->p_idx;
956 }
957
Eric Gouriou80e675f2011-10-27 11:52:18 -0400958 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
959 BUG_ON(len < 0);
960 if (len > 0) {
961 ext_debug("insert new index %d: "
962 "move %d indices from 0x%p to 0x%p\n",
963 logical, len, ix, ix + 1);
964 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
965 }
966
Tao Maf472e022011-10-17 10:13:46 -0400967 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
968 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
969 return -EIO;
970 }
971
Alex Tomasa86c6182006-10-11 01:21:03 -0700972 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700973 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400974 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700975
Frank Mayhar273df552010-03-02 11:46:09 -0500976 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
977 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
978 return -EIO;
979 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700980
981 err = ext4_ext_dirty(handle, inode, curp);
982 ext4_std_error(inode->i_sb, err);
983
984 return err;
985}
986
987/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700988 * ext4_ext_split:
989 * inserts new subtree into the path, using free index entry
990 * at depth @at:
991 * - allocates all needed blocks (new leaf and all intermediate index blocks)
992 * - makes decision where to split
993 * - moves remaining extents and index entries (right to the split point)
994 * into the newly allocated blocks
995 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700996 */
997static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400998 unsigned int flags,
999 struct ext4_ext_path *path,
1000 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -07001001{
1002 struct buffer_head *bh = NULL;
1003 int depth = ext_depth(inode);
1004 struct ext4_extent_header *neh;
1005 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -07001006 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001007 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001008 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001009 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -07001010 int err = 0;
1011
1012 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001013 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07001014
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001015 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -07001016 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -05001017 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1018 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1019 return -EIO;
1020 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001021 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1022 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001023 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -07001024 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001025 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -07001026 } else {
1027 border = newext->ee_block;
1028 ext_debug("leaf will be added."
1029 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001030 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -07001031 }
1032
1033 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001034 * If error occurs, then we break processing
1035 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -07001036 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001037 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -07001038 */
1039
1040 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001041 * Get array to track all allocated blocks.
1042 * We need this to handle errors and free blocks
1043 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -07001044 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -08001045 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07001046 if (!ablocks)
1047 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001048
1049 /* allocate all needed blocks */
1050 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1051 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -04001052 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -04001053 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001054 if (newblock == 0)
1055 goto cleanup;
1056 ablocks[a] = newblock;
1057 }
1058
1059 /* initialize new leaf */
1060 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -05001061 if (unlikely(newblock == 0)) {
1062 EXT4_ERROR_INODE(inode, "newblock == 0!");
1063 err = -EIO;
1064 goto cleanup;
1065 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001066 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001067 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001068 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001069 goto cleanup;
1070 }
1071 lock_buffer(bh);
1072
Avantika Mathur7e028972006-12-06 20:41:33 -08001073 err = ext4_journal_get_create_access(handle, bh);
1074 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001075 goto cleanup;
1076
1077 neh = ext_block_hdr(bh);
1078 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001079 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001080 neh->eh_magic = EXT4_EXT_MAGIC;
1081 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001082
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001083 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -05001084 if (unlikely(path[depth].p_hdr->eh_entries !=
1085 path[depth].p_hdr->eh_max)) {
1086 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1087 path[depth].p_hdr->eh_entries,
1088 path[depth].p_hdr->eh_max);
1089 err = -EIO;
1090 goto cleanup;
1091 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001092 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001093 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1094 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07001095 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001096 struct ext4_extent *ex;
1097 ex = EXT_FIRST_EXTENT(neh);
1098 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001099 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001100 }
1101
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001102 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001103 set_buffer_uptodate(bh);
1104 unlock_buffer(bh);
1105
Frank Mayhar03901312009-01-07 00:06:22 -05001106 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001107 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001108 goto cleanup;
1109 brelse(bh);
1110 bh = NULL;
1111
1112 /* correct old leaf */
1113 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001114 err = ext4_ext_get_access(handle, inode, path + depth);
1115 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001116 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001117 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001118 err = ext4_ext_dirty(handle, inode, path + depth);
1119 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001120 goto cleanup;
1121
1122 }
1123
1124 /* create intermediate indexes */
1125 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001126 if (unlikely(k < 0)) {
1127 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1128 err = -EIO;
1129 goto cleanup;
1130 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001131 if (k)
1132 ext_debug("create %d intermediate indices\n", k);
1133 /* insert new index into current index block */
1134 /* current depth stored in i var */
1135 i = depth - 1;
1136 while (k--) {
1137 oldblock = newblock;
1138 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001139 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001140 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001141 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001142 goto cleanup;
1143 }
1144 lock_buffer(bh);
1145
Avantika Mathur7e028972006-12-06 20:41:33 -08001146 err = ext4_journal_get_create_access(handle, bh);
1147 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001148 goto cleanup;
1149
1150 neh = ext_block_hdr(bh);
1151 neh->eh_entries = cpu_to_le16(1);
1152 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001153 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001154 neh->eh_depth = cpu_to_le16(depth - i);
1155 fidx = EXT_FIRST_INDEX(neh);
1156 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001157 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001158
Eric Sandeenbba90742008-01-28 23:58:27 -05001159 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1160 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001161
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001162 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001163 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1164 EXT_LAST_INDEX(path[i].p_hdr))) {
1165 EXT4_ERROR_INODE(inode,
1166 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1167 le32_to_cpu(path[i].p_ext->ee_block));
1168 err = -EIO;
1169 goto cleanup;
1170 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001171 /* start copy indexes */
1172 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1173 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1174 EXT_MAX_INDEX(path[i].p_hdr));
1175 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001176 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001177 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001178 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001179 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001180 }
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 cleanup;
1188 brelse(bh);
1189 bh = NULL;
1190
1191 /* correct old index */
1192 if (m) {
1193 err = ext4_ext_get_access(handle, inode, path + i);
1194 if (err)
1195 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001196 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001197 err = ext4_ext_dirty(handle, inode, path + i);
1198 if (err)
1199 goto cleanup;
1200 }
1201
1202 i--;
1203 }
1204
1205 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001206 err = ext4_ext_insert_index(handle, inode, path + at,
1207 le32_to_cpu(border), newblock);
1208
1209cleanup:
1210 if (bh) {
1211 if (buffer_locked(bh))
1212 unlock_buffer(bh);
1213 brelse(bh);
1214 }
1215
1216 if (err) {
1217 /* free all allocated blocks in error case */
1218 for (i = 0; i < depth; i++) {
1219 if (!ablocks[i])
1220 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001221 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001222 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001223 }
1224 }
1225 kfree(ablocks);
1226
1227 return err;
1228}
1229
1230/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001231 * ext4_ext_grow_indepth:
1232 * implements tree growing procedure:
1233 * - allocates new block
1234 * - moves top-level data (index block or leaf) into the new block
1235 * - initializes new top-level, creating index that points to the
1236 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001237 */
1238static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001239 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001240 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001241{
Alex Tomasa86c6182006-10-11 01:21:03 -07001242 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001243 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001244 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001245 int err = 0;
1246
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001247 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001248 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001249 if (newblock == 0)
1250 return err;
1251
1252 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001253 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001254 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001255 lock_buffer(bh);
1256
Avantika Mathur7e028972006-12-06 20:41:33 -08001257 err = ext4_journal_get_create_access(handle, bh);
1258 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001259 unlock_buffer(bh);
1260 goto out;
1261 }
1262
1263 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001264 memmove(bh->b_data, EXT4_I(inode)->i_data,
1265 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001266
1267 /* set size of new block */
1268 neh = ext_block_hdr(bh);
1269 /* old root could have indexes or leaves
1270 * so calculate e_max right way */
1271 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001272 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001273 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001274 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001275 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001276 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001277 set_buffer_uptodate(bh);
1278 unlock_buffer(bh);
1279
Frank Mayhar03901312009-01-07 00:06:22 -05001280 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001281 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001282 goto out;
1283
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001284 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001285 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001286 neh->eh_entries = cpu_to_le16(1);
1287 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1288 if (neh->eh_depth == 0) {
1289 /* Root extent block becomes index block */
1290 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1291 EXT_FIRST_INDEX(neh)->ei_block =
1292 EXT_FIRST_EXTENT(neh)->ee_block;
1293 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001294 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001295 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001296 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001297 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001298
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001299 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001300 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001301out:
1302 brelse(bh);
1303
1304 return err;
1305}
1306
1307/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001308 * ext4_ext_create_new_leaf:
1309 * finds empty index and adds new leaf.
1310 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001311 */
1312static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001313 unsigned int mb_flags,
1314 unsigned int gb_flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001315 struct ext4_ext_path *path,
1316 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001317{
1318 struct ext4_ext_path *curp;
1319 int depth, i, err = 0;
1320
1321repeat:
1322 i = depth = ext_depth(inode);
1323
1324 /* walk up to the tree and look for free index entry */
1325 curp = path + depth;
1326 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1327 i--;
1328 curp--;
1329 }
1330
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001331 /* we use already allocated block for index block,
1332 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001333 if (EXT_HAS_FREE_INDEX(curp)) {
1334 /* if we found index with free entry, then use that
1335 * entry: create all needed subtree and add new leaf */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001336 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001337 if (err)
1338 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001339
1340 /* refill path */
1341 ext4_ext_drop_refs(path);
1342 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001343 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001344 path, gb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001345 if (IS_ERR(path))
1346 err = PTR_ERR(path);
1347 } else {
1348 /* tree is full, time to grow in depth */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001349 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001350 if (err)
1351 goto out;
1352
1353 /* refill path */
1354 ext4_ext_drop_refs(path);
1355 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001356 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001357 path, gb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001358 if (IS_ERR(path)) {
1359 err = PTR_ERR(path);
1360 goto out;
1361 }
1362
1363 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001364 * only first (depth 0 -> 1) produces free space;
1365 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001366 */
1367 depth = ext_depth(inode);
1368 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001369 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001370 goto repeat;
1371 }
1372 }
1373
1374out:
1375 return err;
1376}
1377
1378/*
Alex Tomas1988b512008-01-28 23:58:27 -05001379 * search the closest allocated block to the left for *logical
1380 * and returns it at @logical + it's physical address at @phys
1381 * if *logical is the smallest allocated block, the function
1382 * returns 0 at @phys
1383 * return value contains 0 (success) or error code
1384 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001385static int ext4_ext_search_left(struct inode *inode,
1386 struct ext4_ext_path *path,
1387 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001388{
1389 struct ext4_extent_idx *ix;
1390 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001391 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001392
Frank Mayhar273df552010-03-02 11:46:09 -05001393 if (unlikely(path == NULL)) {
1394 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1395 return -EIO;
1396 }
Alex Tomas1988b512008-01-28 23:58:27 -05001397 depth = path->p_depth;
1398 *phys = 0;
1399
1400 if (depth == 0 && path->p_ext == NULL)
1401 return 0;
1402
1403 /* usually extent in the path covers blocks smaller
1404 * then *logical, but it can be that extent is the
1405 * first one in the file */
1406
1407 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001408 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001409 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001410 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1411 EXT4_ERROR_INODE(inode,
1412 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1413 *logical, le32_to_cpu(ex->ee_block));
1414 return -EIO;
1415 }
Alex Tomas1988b512008-01-28 23:58:27 -05001416 while (--depth >= 0) {
1417 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001418 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1419 EXT4_ERROR_INODE(inode,
1420 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001421 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001422 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001423 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001424 depth);
1425 return -EIO;
1426 }
Alex Tomas1988b512008-01-28 23:58:27 -05001427 }
1428 return 0;
1429 }
1430
Frank Mayhar273df552010-03-02 11:46:09 -05001431 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1432 EXT4_ERROR_INODE(inode,
1433 "logical %d < ee_block %d + ee_len %d!",
1434 *logical, le32_to_cpu(ex->ee_block), ee_len);
1435 return -EIO;
1436 }
Alex Tomas1988b512008-01-28 23:58:27 -05001437
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001438 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001439 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001440 return 0;
1441}
1442
1443/*
1444 * search the closest allocated block to the right for *logical
1445 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001446 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001447 * returns 0 at @phys
1448 * return value contains 0 (success) or error code
1449 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001450static int ext4_ext_search_right(struct inode *inode,
1451 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001452 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1453 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001454{
1455 struct buffer_head *bh = NULL;
1456 struct ext4_extent_header *eh;
1457 struct ext4_extent_idx *ix;
1458 struct ext4_extent *ex;
1459 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001460 int depth; /* Note, NOT eh_depth; depth from top of tree */
1461 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001462
Frank Mayhar273df552010-03-02 11:46:09 -05001463 if (unlikely(path == NULL)) {
1464 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1465 return -EIO;
1466 }
Alex Tomas1988b512008-01-28 23:58:27 -05001467 depth = path->p_depth;
1468 *phys = 0;
1469
1470 if (depth == 0 && path->p_ext == NULL)
1471 return 0;
1472
1473 /* usually extent in the path covers blocks smaller
1474 * then *logical, but it can be that extent is the
1475 * first one in the file */
1476
1477 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001478 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001479 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001480 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1481 EXT4_ERROR_INODE(inode,
1482 "first_extent(path[%d].p_hdr) != ex",
1483 depth);
1484 return -EIO;
1485 }
Alex Tomas1988b512008-01-28 23:58:27 -05001486 while (--depth >= 0) {
1487 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001488 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1489 EXT4_ERROR_INODE(inode,
1490 "ix != EXT_FIRST_INDEX *logical %d!",
1491 *logical);
1492 return -EIO;
1493 }
Alex Tomas1988b512008-01-28 23:58:27 -05001494 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001495 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001496 }
1497
Frank Mayhar273df552010-03-02 11:46:09 -05001498 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1499 EXT4_ERROR_INODE(inode,
1500 "logical %d < ee_block %d + ee_len %d!",
1501 *logical, le32_to_cpu(ex->ee_block), ee_len);
1502 return -EIO;
1503 }
Alex Tomas1988b512008-01-28 23:58:27 -05001504
1505 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1506 /* next allocated block in this leaf */
1507 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001508 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001509 }
1510
1511 /* go up and search for index to the right */
1512 while (--depth >= 0) {
1513 ix = path[depth].p_idx;
1514 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001515 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001516 }
1517
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001518 /* we've gone up to the root and found no index to the right */
1519 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001520
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001521got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001522 /* we've found index to the right, let's
1523 * follow it and find the closest allocated
1524 * block to the right */
1525 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001526 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001527 while (++depth < path->p_depth) {
Eric Sandeen395a87b2009-03-10 18:18:47 -04001528 /* subtract from p_depth to get proper eh_depth */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001529 bh = read_extent_tree_block(inode, block,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001530 path->p_depth - depth, 0);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001531 if (IS_ERR(bh))
1532 return PTR_ERR(bh);
1533 eh = ext_block_hdr(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001534 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001535 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001536 put_bh(bh);
1537 }
1538
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001539 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001540 if (IS_ERR(bh))
1541 return PTR_ERR(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001542 eh = ext_block_hdr(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001543 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001544found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001545 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001546 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001547 *ret_ex = ex;
1548 if (bh)
1549 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001550 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001551}
1552
1553/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001554 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001555 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001556 * NOTE: it considers block number from index entry as
1557 * allocated block. Thus, index entries have to be consistent
1558 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001559 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001560static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001561ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1562{
1563 int depth;
1564
1565 BUG_ON(path == NULL);
1566 depth = path->p_depth;
1567
1568 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001569 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001570
1571 while (depth >= 0) {
1572 if (depth == path->p_depth) {
1573 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001574 if (path[depth].p_ext &&
1575 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001576 EXT_LAST_EXTENT(path[depth].p_hdr))
1577 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1578 } else {
1579 /* index */
1580 if (path[depth].p_idx !=
1581 EXT_LAST_INDEX(path[depth].p_hdr))
1582 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1583 }
1584 depth--;
1585 }
1586
Lukas Czernerf17722f2011-06-06 00:05:17 -04001587 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001588}
1589
1590/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001591 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001592 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001593 */
Robin Dong57187892011-07-23 21:49:07 -04001594static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001595{
1596 int depth;
1597
1598 BUG_ON(path == NULL);
1599 depth = path->p_depth;
1600
1601 /* zero-tree has no leaf blocks at all */
1602 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001603 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001604
1605 /* go to index block */
1606 depth--;
1607
1608 while (depth >= 0) {
1609 if (path[depth].p_idx !=
1610 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001611 return (ext4_lblk_t)
1612 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001613 depth--;
1614 }
1615
Lukas Czernerf17722f2011-06-06 00:05:17 -04001616 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001617}
1618
1619/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001620 * ext4_ext_correct_indexes:
1621 * if leaf gets modified and modified extent is first in the leaf,
1622 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001623 * TODO: do we need to correct tree in all cases?
1624 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001625static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001626 struct ext4_ext_path *path)
1627{
1628 struct ext4_extent_header *eh;
1629 int depth = ext_depth(inode);
1630 struct ext4_extent *ex;
1631 __le32 border;
1632 int k, err = 0;
1633
1634 eh = path[depth].p_hdr;
1635 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001636
1637 if (unlikely(ex == NULL || eh == NULL)) {
1638 EXT4_ERROR_INODE(inode,
1639 "ex %p == NULL or eh %p == NULL", ex, eh);
1640 return -EIO;
1641 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001642
1643 if (depth == 0) {
1644 /* there is no tree at all */
1645 return 0;
1646 }
1647
1648 if (ex != EXT_FIRST_EXTENT(eh)) {
1649 /* we correct tree if first leaf got modified only */
1650 return 0;
1651 }
1652
1653 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001654 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001655 */
1656 k = depth - 1;
1657 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001658 err = ext4_ext_get_access(handle, inode, path + k);
1659 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001660 return err;
1661 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001662 err = ext4_ext_dirty(handle, inode, path + k);
1663 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001664 return err;
1665
1666 while (k--) {
1667 /* change all left-side indexes */
1668 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1669 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001670 err = ext4_ext_get_access(handle, inode, path + k);
1671 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001672 break;
1673 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001674 err = ext4_ext_dirty(handle, inode, path + k);
1675 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001676 break;
1677 }
1678
1679 return err;
1680}
1681
Akira Fujita748de672009-06-17 19:24:03 -04001682int
Alex Tomasa86c6182006-10-11 01:21:03 -07001683ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1684 struct ext4_extent *ex2)
1685{
Eric Sandeenda0169b2013-11-04 09:58:26 -05001686 unsigned short ext1_ee_len, ext2_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001687
1688 /*
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001689 * Make sure that both extents are initialized. We don't merge
1690 * uninitialized extents so that we can be sure that end_io code has
1691 * the extent that was written properly split out and conversion to
1692 * initialized is trivial.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001693 */
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001694 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001695 return 0;
1696
1697 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1698 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1699
1700 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001701 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001702 return 0;
1703
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001704 /*
1705 * To allow future support for preallocated extents to be added
1706 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001707 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001708 */
Eric Sandeenda0169b2013-11-04 09:58:26 -05001709 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001710 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001711#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001712 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001713 return 0;
1714#endif
1715
Theodore Ts'obf89d162010-10-27 21:30:14 -04001716 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001717 return 1;
1718 return 0;
1719}
1720
1721/*
Amit Arora56055d32007-07-17 21:42:38 -04001722 * This function tries to merge the "ex" extent to the next extent in the tree.
1723 * It always tries to merge towards right. If you want to merge towards
1724 * left, pass "ex - 1" as argument instead of "ex".
1725 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1726 * 1 if they got merged.
1727 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001728static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001729 struct ext4_ext_path *path,
1730 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001731{
1732 struct ext4_extent_header *eh;
1733 unsigned int depth, len;
1734 int merge_done = 0;
Amit Arora56055d32007-07-17 21:42:38 -04001735
1736 depth = ext_depth(inode);
1737 BUG_ON(path[depth].p_hdr == NULL);
1738 eh = path[depth].p_hdr;
1739
1740 while (ex < EXT_LAST_EXTENT(eh)) {
1741 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1742 break;
1743 /* merge with next extent! */
Amit Arora56055d32007-07-17 21:42:38 -04001744 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1745 + ext4_ext_get_actual_len(ex + 1));
Amit Arora56055d32007-07-17 21:42:38 -04001746
1747 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1748 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1749 * sizeof(struct ext4_extent);
1750 memmove(ex + 1, ex + 2, len);
1751 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001752 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001753 merge_done = 1;
1754 WARN_ON(eh->eh_entries == 0);
1755 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001756 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001757 }
1758
1759 return merge_done;
1760}
1761
1762/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001763 * This function does a very simple check to see if we can collapse
1764 * an extent tree with a single extent tree leaf block into the inode.
1765 */
1766static void ext4_ext_try_to_merge_up(handle_t *handle,
1767 struct inode *inode,
1768 struct ext4_ext_path *path)
1769{
1770 size_t s;
1771 unsigned max_root = ext4_ext_space_root(inode, 0);
1772 ext4_fsblk_t blk;
1773
1774 if ((path[0].p_depth != 1) ||
1775 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1776 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1777 return;
1778
1779 /*
1780 * We need to modify the block allocation bitmap and the block
1781 * group descriptor to release the extent tree block. If we
1782 * can't get the journal credits, give up.
1783 */
1784 if (ext4_journal_extend(handle, 2))
1785 return;
1786
1787 /*
1788 * Copy the extent data up to the inode
1789 */
1790 blk = ext4_idx_pblock(path[0].p_idx);
1791 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1792 sizeof(struct ext4_extent_idx);
1793 s += sizeof(struct ext4_extent_header);
1794
1795 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1796 path[0].p_depth = 0;
1797 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1798 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1799 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1800
1801 brelse(path[1].p_bh);
1802 ext4_free_blocks(handle, inode, NULL, blk, 1,
Jan Kara7d734532013-08-17 09:36:54 -04001803 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET |
1804 EXT4_FREE_BLOCKS_RESERVE);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001805}
1806
1807/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001808 * This function tries to merge the @ex extent to neighbours in the tree.
1809 * return 1 if merge left else 0.
1810 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001811static void ext4_ext_try_to_merge(handle_t *handle,
1812 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001813 struct ext4_ext_path *path,
1814 struct ext4_extent *ex) {
1815 struct ext4_extent_header *eh;
1816 unsigned int depth;
1817 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001818
1819 depth = ext_depth(inode);
1820 BUG_ON(path[depth].p_hdr == NULL);
1821 eh = path[depth].p_hdr;
1822
1823 if (ex > EXT_FIRST_EXTENT(eh))
1824 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1825
1826 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001827 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001828
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001829 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001830}
1831
1832/*
Amit Arora25d14f92007-05-24 13:04:13 -04001833 * check if a portion of the "newext" extent overlaps with an
1834 * existing extent.
1835 *
1836 * If there is an overlap discovered, it updates the length of the newext
1837 * such that there will be no overlap, and then returns 1.
1838 * If there is no overlap found, it returns 0.
1839 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001840static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1841 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001842 struct ext4_extent *newext,
1843 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001844{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001845 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001846 unsigned int depth, len1;
1847 unsigned int ret = 0;
1848
1849 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001850 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001851 depth = ext_depth(inode);
1852 if (!path[depth].p_ext)
1853 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001854 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
Amit Arora25d14f92007-05-24 13:04:13 -04001855
1856 /*
1857 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001858 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001859 */
1860 if (b2 < b1) {
1861 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001862 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001863 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001864 b2 = EXT4_LBLK_CMASK(sbi, b2);
Amit Arora25d14f92007-05-24 13:04:13 -04001865 }
1866
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001867 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001868 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001869 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001870 newext->ee_len = cpu_to_le16(len1);
1871 ret = 1;
1872 }
1873
1874 /* check for overlap */
1875 if (b1 + len1 > b2) {
1876 newext->ee_len = cpu_to_le16(b2 - b1);
1877 ret = 1;
1878 }
1879out:
1880 return ret;
1881}
1882
1883/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001884 * ext4_ext_insert_extent:
1885 * tries to merge requsted extent into the existing extent or
1886 * inserts requested extent as new one into the tree,
1887 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001888 */
1889int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1890 struct ext4_ext_path *path,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001891 struct ext4_extent *newext, int gb_flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07001892{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001893 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001894 struct ext4_extent *ex, *fex;
1895 struct ext4_extent *nearex; /* nearest extent */
1896 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001897 int depth, len, err;
1898 ext4_lblk_t next;
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001899 int mb_flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001900
Frank Mayhar273df552010-03-02 11:46:09 -05001901 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1902 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1903 return -EIO;
1904 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001905 depth = ext_depth(inode);
1906 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001907 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001908 if (unlikely(path[depth].p_hdr == NULL)) {
1909 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1910 return -EIO;
1911 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001912
1913 /* try to insert block into found extent and return */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001914 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001915
1916 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001917 * Try to see whether we should rather test the extent on
1918 * right from ex, or from the left of ex. This is because
1919 * ext4_ext_find_extent() can return either extent on the
1920 * left, or on the right from the searched position. This
1921 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001922 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001923 if (ex < EXT_LAST_EXTENT(eh) &&
1924 (le32_to_cpu(ex->ee_block) +
1925 ext4_ext_get_actual_len(ex) <
1926 le32_to_cpu(newext->ee_block))) {
1927 ex += 1;
1928 goto prepend;
1929 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1930 (le32_to_cpu(newext->ee_block) +
1931 ext4_ext_get_actual_len(newext) <
1932 le32_to_cpu(ex->ee_block)))
1933 ex -= 1;
1934
1935 /* Try to append newex to the ex */
1936 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1937 ext_debug("append [%d]%d block to %u:[%d]%d"
1938 "(from %llu)\n",
1939 ext4_ext_is_uninitialized(newext),
1940 ext4_ext_get_actual_len(newext),
1941 le32_to_cpu(ex->ee_block),
1942 ext4_ext_is_uninitialized(ex),
1943 ext4_ext_get_actual_len(ex),
1944 ext4_ext_pblock(ex));
1945 err = ext4_ext_get_access(handle, inode,
1946 path + depth);
1947 if (err)
1948 return err;
1949
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001950 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001951 + ext4_ext_get_actual_len(newext));
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001952 eh = path[depth].p_hdr;
1953 nearex = ex;
1954 goto merge;
1955 }
1956
1957prepend:
1958 /* Try to prepend newex to the ex */
1959 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1960 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1961 "(from %llu)\n",
1962 le32_to_cpu(newext->ee_block),
1963 ext4_ext_is_uninitialized(newext),
1964 ext4_ext_get_actual_len(newext),
1965 le32_to_cpu(ex->ee_block),
1966 ext4_ext_is_uninitialized(ex),
1967 ext4_ext_get_actual_len(ex),
1968 ext4_ext_pblock(ex));
1969 err = ext4_ext_get_access(handle, inode,
1970 path + depth);
1971 if (err)
1972 return err;
1973
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001974 ex->ee_block = newext->ee_block;
1975 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1976 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1977 + ext4_ext_get_actual_len(newext));
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001978 eh = path[depth].p_hdr;
1979 nearex = ex;
1980 goto merge;
1981 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001982 }
1983
Alex Tomasa86c6182006-10-11 01:21:03 -07001984 depth = ext_depth(inode);
1985 eh = path[depth].p_hdr;
1986 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1987 goto has_space;
1988
1989 /* probably next leaf has space for us? */
1990 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001991 next = EXT_MAX_BLOCKS;
1992 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001993 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001994 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001995 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07001996 BUG_ON(npath != NULL);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001997 npath = ext4_ext_find_extent(inode, next, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001998 if (IS_ERR(npath))
1999 return PTR_ERR(npath);
2000 BUG_ON(npath->p_depth != path->p_depth);
2001 eh = npath[depth].p_hdr;
2002 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002003 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002004 le16_to_cpu(eh->eh_entries));
2005 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04002006 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07002007 }
2008 ext_debug("next leaf has no free space(%d,%d)\n",
2009 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2010 }
2011
2012 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002013 * There is no free space in the found leaf.
2014 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07002015 */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002016 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2017 mb_flags = EXT4_MB_USE_RESERVED;
2018 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2019 path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07002020 if (err)
2021 goto cleanup;
2022 depth = ext_depth(inode);
2023 eh = path[depth].p_hdr;
2024
2025has_space:
2026 nearex = path[depth].p_ext;
2027
Avantika Mathur7e028972006-12-06 20:41:33 -08002028 err = ext4_ext_get_access(handle, inode, path + depth);
2029 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002030 goto cleanup;
2031
2032 if (!nearex) {
2033 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002034 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002035 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002036 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04002037 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002038 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04002039 nearex = EXT_FIRST_EXTENT(eh);
2040 } else {
2041 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002042 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04002043 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002044 ext_debug("insert %u:%llu:[%d]%d before: "
2045 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002046 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002047 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04002048 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002049 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002050 nearex);
2051 nearex++;
2052 } else {
2053 /* Insert before */
2054 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04002055 ext_debug("insert %u:%llu:[%d]%d after: "
2056 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04002057 le32_to_cpu(newext->ee_block),
2058 ext4_ext_pblock(newext),
2059 ext4_ext_is_uninitialized(newext),
2060 ext4_ext_get_actual_len(newext),
2061 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002062 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04002063 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2064 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002065 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04002066 "move %d extents from 0x%p to 0x%p\n",
2067 le32_to_cpu(newext->ee_block),
2068 ext4_ext_pblock(newext),
2069 ext4_ext_is_uninitialized(newext),
2070 ext4_ext_get_actual_len(newext),
2071 len, nearex, nearex + 1);
2072 memmove(nearex + 1, nearex,
2073 len * sizeof(struct ext4_extent));
2074 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002075 }
2076
Marcin Slusarze8546d02008-04-17 10:38:59 -04002077 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002078 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002079 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002080 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002081 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002082
2083merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002084 /* try to merge extents */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002085 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002086 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002087
Alex Tomasa86c6182006-10-11 01:21:03 -07002088
2089 /* time to correct all indexes above */
2090 err = ext4_ext_correct_indexes(handle, inode, path);
2091 if (err)
2092 goto cleanup;
2093
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002094 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002095
2096cleanup:
2097 if (npath) {
2098 ext4_ext_drop_refs(npath);
2099 kfree(npath);
2100 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002101 return err;
2102}
2103
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002104static int ext4_fill_fiemap_extents(struct inode *inode,
2105 ext4_lblk_t block, ext4_lblk_t num,
2106 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002107{
2108 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002109 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002110 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002111 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002112 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002113 int exists, depth = 0, err = 0;
2114 unsigned int flags = 0;
2115 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002116
Lukas Czernerf17722f2011-06-06 00:05:17 -04002117 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002118 num = last - block;
2119 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002120 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002121
2122 if (path && ext_depth(inode) != depth) {
2123 /* depth was changed. we have to realloc path */
2124 kfree(path);
2125 path = NULL;
2126 }
2127
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002128 path = ext4_ext_find_extent(inode, block, path, 0);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002129 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002130 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002131 err = PTR_ERR(path);
2132 path = NULL;
2133 break;
2134 }
2135
2136 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002137 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002138 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002139 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2140 err = -EIO;
2141 break;
2142 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002143 ex = path[depth].p_ext;
2144 next = ext4_ext_next_allocated_block(path);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002145 ext4_ext_drop_refs(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002146
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002147 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002148 exists = 0;
2149 if (!ex) {
2150 /* there is no extent yet, so try to allocate
2151 * all requested space */
2152 start = block;
2153 end = block + num;
2154 } else if (le32_to_cpu(ex->ee_block) > block) {
2155 /* need to allocate space before found extent */
2156 start = block;
2157 end = le32_to_cpu(ex->ee_block);
2158 if (block + num < end)
2159 end = block + num;
2160 } else if (block >= le32_to_cpu(ex->ee_block)
2161 + ext4_ext_get_actual_len(ex)) {
2162 /* need to allocate space after found extent */
2163 start = block;
2164 end = block + num;
2165 if (end >= next)
2166 end = next;
2167 } else if (block >= le32_to_cpu(ex->ee_block)) {
2168 /*
2169 * some part of requested space is covered
2170 * by found extent
2171 */
2172 start = block;
2173 end = le32_to_cpu(ex->ee_block)
2174 + ext4_ext_get_actual_len(ex);
2175 if (block + num < end)
2176 end = block + num;
2177 exists = 1;
2178 } else {
2179 BUG();
2180 }
2181 BUG_ON(end <= start);
2182
2183 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002184 es.es_lblk = start;
2185 es.es_len = end - start;
2186 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002187 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002188 es.es_lblk = le32_to_cpu(ex->ee_block);
2189 es.es_len = ext4_ext_get_actual_len(ex);
2190 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002191 if (ext4_ext_is_uninitialized(ex))
2192 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002193 }
2194
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002195 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002196 * Find delayed extent and update es accordingly. We call
2197 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002198 * last existing extent or not.
2199 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002200 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002201 if (!exists && next_del) {
2202 exists = 1;
Jie Liu72dac952013-06-12 23:13:59 -04002203 flags |= (FIEMAP_EXTENT_DELALLOC |
2204 FIEMAP_EXTENT_UNKNOWN);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002205 }
2206 up_read(&EXT4_I(inode)->i_data_sem);
2207
Zheng Liu69eb33d2013-02-18 00:31:07 -05002208 if (unlikely(es.es_len == 0)) {
2209 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002210 err = -EIO;
2211 break;
2212 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002213
Zheng Liuf7fec032013-02-18 00:28:47 -05002214 /*
2215 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2216 * we need to check next == EXT_MAX_BLOCKS because it is
2217 * possible that an extent is with unwritten and delayed
2218 * status due to when an extent is delayed allocated and
2219 * is allocated by fallocate status tree will track both of
2220 * them in a extent.
2221 *
2222 * So we could return a unwritten and delayed extent, and
2223 * its block is equal to 'next'.
2224 */
2225 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002226 flags |= FIEMAP_EXTENT_LAST;
2227 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2228 next != EXT_MAX_BLOCKS)) {
2229 EXT4_ERROR_INODE(inode,
2230 "next extent == %u, next "
2231 "delalloc extent = %u",
2232 next, next_del);
2233 err = -EIO;
2234 break;
2235 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002236 }
2237
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002238 if (exists) {
2239 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002240 (__u64)es.es_lblk << blksize_bits,
2241 (__u64)es.es_pblk << blksize_bits,
2242 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002243 flags);
2244 if (err < 0)
2245 break;
2246 if (err == 1) {
2247 err = 0;
2248 break;
2249 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002250 }
2251
Zheng Liu69eb33d2013-02-18 00:31:07 -05002252 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002253 }
2254
2255 if (path) {
2256 ext4_ext_drop_refs(path);
2257 kfree(path);
2258 }
2259
2260 return err;
2261}
2262
Alex Tomasa86c6182006-10-11 01:21:03 -07002263/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002264 * ext4_ext_put_gap_in_cache:
2265 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002266 * and cache this gap
2267 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002268static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002269ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002270 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002271{
2272 int depth = ext_depth(inode);
Andi Shyti27b1b222013-08-28 14:00:00 -04002273 unsigned long len = 0;
2274 ext4_lblk_t lblock = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002275 struct ext4_extent *ex;
2276
2277 ex = path[depth].p_ext;
2278 if (ex == NULL) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002279 /*
2280 * there is no extent yet, so gap is [0;-] and we
2281 * don't cache it
2282 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002283 ext_debug("cache gap(whole file):");
2284 } else if (block < le32_to_cpu(ex->ee_block)) {
2285 lblock = block;
2286 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002287 ext_debug("cache gap(before): %u [%u:%u]",
2288 block,
2289 le32_to_cpu(ex->ee_block),
2290 ext4_ext_get_actual_len(ex));
Zheng Liud100eef2013-02-18 00:29:59 -05002291 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2292 ext4_es_insert_extent(inode, lblock, len, ~0,
2293 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002294 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002295 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002296 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002297 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002298 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002299
2300 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002301 ext_debug("cache gap(after): [%u:%u] %u",
2302 le32_to_cpu(ex->ee_block),
2303 ext4_ext_get_actual_len(ex),
2304 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002305 BUG_ON(next == lblock);
2306 len = next - lblock;
Zheng Liud100eef2013-02-18 00:29:59 -05002307 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2308 ext4_es_insert_extent(inode, lblock, len, ~0,
2309 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002310 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07002311 BUG();
2312 }
2313
Eric Sandeenbba90742008-01-28 23:58:27 -05002314 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002315}
2316
2317/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002318 * ext4_ext_rm_idx:
2319 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002320 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002321static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002322 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002323{
Alex Tomasa86c6182006-10-11 01:21:03 -07002324 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002325 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002326
2327 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002328 depth--;
2329 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002330 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002331 if (unlikely(path->p_hdr->eh_entries == 0)) {
2332 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2333 return -EIO;
2334 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002335 err = ext4_ext_get_access(handle, inode, path);
2336 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002337 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002338
2339 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2340 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2341 len *= sizeof(struct ext4_extent_idx);
2342 memmove(path->p_idx, path->p_idx + 1, len);
2343 }
2344
Marcin Slusarze8546d02008-04-17 10:38:59 -04002345 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002346 err = ext4_ext_dirty(handle, inode, path);
2347 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002348 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002349 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002350 trace_ext4_ext_rm_idx(inode, leaf);
2351
Peter Huewe7dc57612011-02-21 21:01:42 -05002352 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002353 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002354
2355 while (--depth >= 0) {
2356 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2357 break;
2358 path--;
2359 err = ext4_ext_get_access(handle, inode, path);
2360 if (err)
2361 break;
2362 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2363 err = ext4_ext_dirty(handle, inode, path);
2364 if (err)
2365 break;
2366 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002367 return err;
2368}
2369
2370/*
Mingming Caoee12b632008-08-19 22:16:05 -04002371 * ext4_ext_calc_credits_for_single_extent:
2372 * This routine returns max. credits that needed to insert an extent
2373 * to the extent tree.
2374 * When pass the actual path, the caller should calculate credits
2375 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002376 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002377int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002378 struct ext4_ext_path *path)
2379{
Alex Tomasa86c6182006-10-11 01:21:03 -07002380 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002381 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002382 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002383
Alex Tomasa86c6182006-10-11 01:21:03 -07002384 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002385 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002386 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2387
2388 /*
2389 * There are some space in the leaf tree, no
2390 * need to account for leaf block credit
2391 *
2392 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002393 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002394 * accounted.
2395 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002396 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002397 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002398 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002399 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002400 }
2401
Mingming Cao525f4ed2008-08-19 22:15:58 -04002402 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002403}
Alex Tomasa86c6182006-10-11 01:21:03 -07002404
Mingming Caoee12b632008-08-19 22:16:05 -04002405/*
Jan Karafffb2732013-06-04 13:01:11 -04002406 * How many index/leaf blocks need to change/allocate to add @extents extents?
Mingming Caoee12b632008-08-19 22:16:05 -04002407 *
Jan Karafffb2732013-06-04 13:01:11 -04002408 * If we add a single extent, then in the worse case, each tree level
2409 * index/leaf need to be changed in case of the tree split.
Mingming Caoee12b632008-08-19 22:16:05 -04002410 *
Jan Karafffb2732013-06-04 13:01:11 -04002411 * If more extents are inserted, they could cause the whole tree split more
2412 * than once, but this is really rare.
Mingming Caoee12b632008-08-19 22:16:05 -04002413 */
Jan Karafffb2732013-06-04 13:01:11 -04002414int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
Mingming Caoee12b632008-08-19 22:16:05 -04002415{
2416 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002417 int depth;
2418
2419 /* If we are converting the inline data, only one is needed here. */
2420 if (ext4_has_inline_data(inode))
2421 return 1;
2422
2423 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002424
Jan Karafffb2732013-06-04 13:01:11 -04002425 if (extents <= 1)
Mingming Caoee12b632008-08-19 22:16:05 -04002426 index = depth * 2;
2427 else
2428 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002429
Mingming Caoee12b632008-08-19 22:16:05 -04002430 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002431}
2432
Theodore Ts'o981250c2013-06-12 11:48:29 -04002433static inline int get_default_free_blocks_flags(struct inode *inode)
2434{
2435 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2436 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2437 else if (ext4_should_journal_data(inode))
2438 return EXT4_FREE_BLOCKS_FORGET;
2439 return 0;
2440}
2441
Alex Tomasa86c6182006-10-11 01:21:03 -07002442static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002443 struct ext4_extent *ex,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002444 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002445 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002446{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002447 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002448 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002449 ext4_fsblk_t pblk;
Theodore Ts'o981250c2013-06-12 11:48:29 -04002450 int flags = get_default_free_blocks_flags(inode);
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002451
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002452 /*
2453 * For bigalloc file systems, we never free a partial cluster
2454 * at the beginning of the extent. Instead, we make a note
2455 * that we tried freeing the cluster, and check to see if we
2456 * need to free it on a subsequent call to ext4_remove_blocks,
2457 * or at the end of the ext4_truncate() operation.
2458 */
2459 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2460
Aditya Kalid8990242011-09-09 19:18:51 -04002461 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002462 /*
2463 * If we have a partial cluster, and it's different from the
2464 * cluster of the last block, we need to explicitly free the
2465 * partial cluster here.
2466 */
2467 pblk = ext4_ext_pblock(ex) + ee_len - 1;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002468 if ((*partial_cluster > 0) &&
2469 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002470 ext4_free_blocks(handle, inode, NULL,
2471 EXT4_C2B(sbi, *partial_cluster),
2472 sbi->s_cluster_ratio, flags);
2473 *partial_cluster = 0;
2474 }
2475
Alex Tomasa86c6182006-10-11 01:21:03 -07002476#ifdef EXTENTS_STATS
2477 {
2478 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002479 spin_lock(&sbi->s_ext_stats_lock);
2480 sbi->s_ext_blocks += ee_len;
2481 sbi->s_ext_extents++;
2482 if (ee_len < sbi->s_ext_min)
2483 sbi->s_ext_min = ee_len;
2484 if (ee_len > sbi->s_ext_max)
2485 sbi->s_ext_max = ee_len;
2486 if (ext_depth(inode) > sbi->s_depth_max)
2487 sbi->s_depth_max = ext_depth(inode);
2488 spin_unlock(&sbi->s_ext_stats_lock);
2489 }
2490#endif
2491 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002492 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002493 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002494 ext4_lblk_t num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002495 unsigned int unaligned;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002496
Amit Aroraa2df2a62007-07-17 21:42:41 -04002497 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002498 pblk = ext4_ext_pblock(ex) + ee_len - num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002499 /*
2500 * Usually we want to free partial cluster at the end of the
2501 * extent, except for the situation when the cluster is still
2502 * used by any other extent (partial_cluster is negative).
2503 */
2504 if (*partial_cluster < 0 &&
2505 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2506 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2507
2508 ext_debug("free last %u blocks starting %llu partial %lld\n",
2509 num, pblk, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002510 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2511 /*
2512 * If the block range to be freed didn't start at the
2513 * beginning of a cluster, and we removed the entire
Lukas Czernerd23142c2013-05-27 23:33:35 -04002514 * extent and the cluster is not used by any other extent,
2515 * save the partial cluster here, since we might need to
2516 * delete if we determine that the truncate operation has
2517 * removed all of the blocks in the cluster.
2518 *
2519 * On the other hand, if we did not manage to free the whole
2520 * extent, we have to mark the cluster as used (store negative
2521 * cluster number in partial_cluster).
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002522 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05002523 unaligned = EXT4_PBLK_COFF(sbi, pblk);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002524 if (unaligned && (ee_len == num) &&
2525 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002526 *partial_cluster = EXT4_B2C(sbi, pblk);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002527 else if (unaligned)
2528 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2529 else if (*partial_cluster > 0)
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002530 *partial_cluster = 0;
Lukas Czerner78fb9cd2013-05-27 23:32:35 -04002531 } else
2532 ext4_error(sbi->s_sb, "strange request: removal(2) "
2533 "%u-%u from %u:%u\n",
2534 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002535 return 0;
2536}
2537
Allison Hendersond583fb82011-05-25 07:41:43 -04002538
2539/*
2540 * ext4_ext_rm_leaf() Removes the extents associated with the
2541 * blocks appearing between "start" and "end", and splits the extents
2542 * if "start" and "end" appear in the same extent
2543 *
2544 * @handle: The journal handle
2545 * @inode: The files inode
2546 * @path: The path to the leaf
Lukas Czernerd23142c2013-05-27 23:33:35 -04002547 * @partial_cluster: The cluster which we'll have to free if all extents
2548 * has been released from it. It gets negative in case
2549 * that the cluster is still used.
Allison Hendersond583fb82011-05-25 07:41:43 -04002550 * @start: The first block to remove
2551 * @end: The last block to remove
2552 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002553static int
2554ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002555 struct ext4_ext_path *path,
2556 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002557 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002558{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002559 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002560 int err = 0, correct_index = 0;
2561 int depth = ext_depth(inode), credits;
2562 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002563 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002564 unsigned num;
2565 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002566 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002567 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002568 struct ext4_extent *ex;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002569 ext4_fsblk_t pblk;
Alex Tomasa86c6182006-10-11 01:21:03 -07002570
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002571 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002572 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002573 if (!path[depth].p_hdr)
2574 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2575 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002576 if (unlikely(path[depth].p_hdr == NULL)) {
2577 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2578 return -EIO;
2579 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002580 /* find where to start removing */
Ashish Sangwan6ae06ff2013-07-01 08:12:41 -04002581 ex = path[depth].p_ext;
2582 if (!ex)
2583 ex = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002584
2585 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002586 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002587
Aditya Kalid8990242011-09-09 19:18:51 -04002588 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2589
Alex Tomasa86c6182006-10-11 01:21:03 -07002590 while (ex >= EXT_FIRST_EXTENT(eh) &&
2591 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002592
2593 if (ext4_ext_is_uninitialized(ex))
2594 uninitialized = 1;
2595 else
2596 uninitialized = 0;
2597
Mingming553f9002009-09-18 13:34:55 -04002598 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2599 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002600 path[depth].p_ext = ex;
2601
2602 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002603 b = ex_ee_block+ex_ee_len - 1 < end ?
2604 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002605
2606 ext_debug(" border %u:%u\n", a, b);
2607
Allison Hendersond583fb82011-05-25 07:41:43 -04002608 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002609 if (end < ex_ee_block) {
Lukas Czernerd23142c2013-05-27 23:33:35 -04002610 /*
2611 * We're going to skip this extent and move to another,
2612 * so if this extent is not cluster aligned we have
2613 * to mark the current cluster as used to avoid
2614 * accidentally freeing it later on
2615 */
2616 pblk = ext4_ext_pblock(ex);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05002617 if (EXT4_PBLK_COFF(sbi, pblk))
Lukas Czernerd23142c2013-05-27 23:33:35 -04002618 *partial_cluster =
2619 -((long long)EXT4_B2C(sbi, pblk));
Allison Hendersond583fb82011-05-25 07:41:43 -04002620 ex--;
2621 ex_ee_block = le32_to_cpu(ex->ee_block);
2622 ex_ee_len = ext4_ext_get_actual_len(ex);
2623 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002624 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002625 EXT4_ERROR_INODE(inode,
2626 "can not handle truncate %u:%u "
2627 "on extent %u:%u",
2628 start, end, ex_ee_block,
2629 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002630 err = -EIO;
2631 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002632 } else if (a != ex_ee_block) {
2633 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002634 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002635 } else {
2636 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002637 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002638 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002639 /*
2640 * 3 for leaf, sb, and inode plus 2 (bmap and group
2641 * descriptor) for each block group; assume two block
2642 * groups plus ex_ee_len/blocks_per_block_group for
2643 * the worst case
2644 */
2645 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002646 if (ex == EXT_FIRST_EXTENT(eh)) {
2647 correct_index = 1;
2648 credits += (ext_depth(inode)) + 1;
2649 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002650 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002651
Jan Kara487caee2009-08-17 22:17:20 -04002652 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002653 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002654 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002655
2656 err = ext4_ext_get_access(handle, inode, path + depth);
2657 if (err)
2658 goto out;
2659
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002660 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2661 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002662 if (err)
2663 goto out;
2664
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002665 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002666 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002667 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002668
Alex Tomasa86c6182006-10-11 01:21:03 -07002669 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002670 /*
2671 * Do not mark uninitialized if all the blocks in the
2672 * extent have been removed.
2673 */
2674 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002675 ext4_ext_mark_uninitialized(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002676 /*
2677 * If the extent was completely released,
2678 * we need to remove it from the leaf
2679 */
2680 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002681 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002682 /*
2683 * For hole punching, we need to scoot all the
2684 * extents up when an extent is removed so that
2685 * we dont have blank extents in the middle
2686 */
2687 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2688 sizeof(struct ext4_extent));
2689
2690 /* Now get rid of the one at the end */
2691 memset(EXT_LAST_EXTENT(eh), 0,
2692 sizeof(struct ext4_extent));
2693 }
2694 le16_add_cpu(&eh->eh_entries, -1);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002695 } else if (*partial_cluster > 0)
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002696 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002697
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002698 err = ext4_ext_dirty(handle, inode, path + depth);
2699 if (err)
2700 goto out;
2701
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002702 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002703 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002704 ex--;
2705 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002706 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002707 }
2708
2709 if (correct_index && eh->eh_entries)
2710 err = ext4_ext_correct_indexes(handle, inode, path);
2711
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002712 /*
Lukas Czernerd23142c2013-05-27 23:33:35 -04002713 * Free the partial cluster only if the current extent does not
2714 * reference it. Otherwise we might free used cluster.
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002715 */
Lukas Czernerd23142c2013-05-27 23:33:35 -04002716 if (*partial_cluster > 0 &&
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002717 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2718 *partial_cluster)) {
Theodore Ts'o981250c2013-06-12 11:48:29 -04002719 int flags = get_default_free_blocks_flags(inode);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002720
2721 ext4_free_blocks(handle, inode, NULL,
2722 EXT4_C2B(sbi, *partial_cluster),
2723 sbi->s_cluster_ratio, flags);
2724 *partial_cluster = 0;
2725 }
2726
Alex Tomasa86c6182006-10-11 01:21:03 -07002727 /* if this leaf is free, then we should
2728 * remove it from index block above */
2729 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002730 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002731
2732out:
2733 return err;
2734}
2735
2736/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002737 * ext4_ext_more_to_rm:
2738 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002739 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002740static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002741ext4_ext_more_to_rm(struct ext4_ext_path *path)
2742{
2743 BUG_ON(path->p_idx == NULL);
2744
2745 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2746 return 0;
2747
2748 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002749 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002750 * so we have to consider current index for truncation
2751 */
2752 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2753 return 0;
2754 return 1;
2755}
2756
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002757int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2758 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002759{
2760 struct super_block *sb = inode->i_sb;
2761 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002762 struct ext4_ext_path *path = NULL;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002763 long long partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002764 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002765 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002766
Lukas Czerner5f95d212012-03-19 23:03:19 -04002767 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002768
2769 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002770 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002771 if (IS_ERR(handle))
2772 return PTR_ERR(handle);
2773
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002774again:
Lukas Czerner61801322013-05-27 23:32:35 -04002775 trace_ext4_ext_remove_space(inode, start, end, depth);
Aditya Kalid8990242011-09-09 19:18:51 -04002776
Alex Tomasa86c6182006-10-11 01:21:03 -07002777 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002778 * Check if we are removing extents inside the extent tree. If that
2779 * is the case, we are going to punch a hole inside the extent tree
2780 * so we have to check whether we need to split the extent covering
2781 * the last block to remove so we can easily remove the part of it
2782 * in ext4_ext_rm_leaf().
2783 */
2784 if (end < EXT_MAX_BLOCKS - 1) {
2785 struct ext4_extent *ex;
2786 ext4_lblk_t ee_block;
2787
2788 /* find extent for this block */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002789 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002790 if (IS_ERR(path)) {
2791 ext4_journal_stop(handle);
2792 return PTR_ERR(path);
2793 }
2794 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002795 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002796 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002797 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002798 if (depth) {
2799 EXT4_ERROR_INODE(inode,
2800 "path[%d].p_hdr == NULL",
2801 depth);
2802 err = -EIO;
2803 }
2804 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002805 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002806
2807 ee_block = le32_to_cpu(ex->ee_block);
2808
2809 /*
2810 * See if the last block is inside the extent, if so split
2811 * the extent at 'end' block so we can easily remove the
2812 * tail of the first part of the split extent in
2813 * ext4_ext_rm_leaf().
2814 */
2815 if (end >= ee_block &&
2816 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2817 int split_flag = 0;
2818
2819 if (ext4_ext_is_uninitialized(ex))
2820 split_flag = EXT4_EXT_MARK_UNINIT1 |
2821 EXT4_EXT_MARK_UNINIT2;
2822
2823 /*
2824 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002825 * block in the first new extent. Also we should not
2826 * fail removing space due to ENOSPC so try to use
2827 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002828 */
2829 err = ext4_split_extent_at(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04002830 end + 1, split_flag,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002831 EXT4_EX_NOCACHE |
Lukas Czerner27dd4382013-04-09 22:11:22 -04002832 EXT4_GET_BLOCKS_PRE_IO |
2833 EXT4_GET_BLOCKS_METADATA_NOFAIL);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002834
2835 if (err < 0)
2836 goto out;
2837 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002838 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002839 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002840 * We start scanning from right side, freeing all the blocks
2841 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002842 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002843 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002844 if (path) {
2845 int k = i = depth;
2846 while (--k > 0)
2847 path[k].p_block =
2848 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2849 } else {
2850 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2851 GFP_NOFS);
2852 if (path == NULL) {
2853 ext4_journal_stop(handle);
2854 return -ENOMEM;
2855 }
2856 path[0].p_depth = depth;
2857 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002858 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002859
Theodore Ts'oc3491792013-08-16 21:21:41 -04002860 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
Ashish Sangwan968dee72012-07-22 22:49:08 -04002861 err = -EIO;
2862 goto out;
2863 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002864 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002865 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002866
2867 while (i >= 0 && err == 0) {
2868 if (i == depth) {
2869 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002870 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002871 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002872 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002873 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002874 brelse(path[i].p_bh);
2875 path[i].p_bh = NULL;
2876 i--;
2877 continue;
2878 }
2879
2880 /* this is index block */
2881 if (!path[i].p_hdr) {
2882 ext_debug("initialize header\n");
2883 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002884 }
2885
Alex Tomasa86c6182006-10-11 01:21:03 -07002886 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002887 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002888 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2889 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2890 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2891 path[i].p_hdr,
2892 le16_to_cpu(path[i].p_hdr->eh_entries));
2893 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002894 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002895 path[i].p_idx--;
2896 }
2897
2898 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2899 i, EXT_FIRST_INDEX(path[i].p_hdr),
2900 path[i].p_idx);
2901 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002902 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002903 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002904 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002905 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002906 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002907 bh = read_extent_tree_block(inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002908 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2909 EXT4_EX_NOCACHE);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002910 if (IS_ERR(bh)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002911 /* should we reset i_size? */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002912 err = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002913 break;
2914 }
Theodore Ts'o76828c82013-07-15 12:27:47 -04002915 /* Yield here to deal with large extent trees.
2916 * Should be a no-op if we did IO above. */
2917 cond_resched();
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002918 if (WARN_ON(i + 1 > depth)) {
2919 err = -EIO;
2920 break;
2921 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002922 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002923
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002924 /* save actual number of indexes since this
2925 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002926 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2927 i++;
2928 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002929 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002930 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002931 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002932 * handle must be already prepared by the
2933 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002934 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002935 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002936 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002937 brelse(path[i].p_bh);
2938 path[i].p_bh = NULL;
2939 i--;
2940 ext_debug("return to level %d\n", i);
2941 }
2942 }
2943
Lukas Czerner61801322013-05-27 23:32:35 -04002944 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2945 partial_cluster, path->p_hdr->eh_entries);
Aditya Kalid8990242011-09-09 19:18:51 -04002946
Aditya Kali7b415bf2011-09-09 19:04:51 -04002947 /* If we still have something in the partial cluster and we have removed
2948 * even the first extent, then we should free the blocks in the partial
2949 * cluster as well. */
Lukas Czernerd23142c2013-05-27 23:33:35 -04002950 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
Theodore Ts'o981250c2013-06-12 11:48:29 -04002951 int flags = get_default_free_blocks_flags(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04002952
2953 ext4_free_blocks(handle, inode, NULL,
2954 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2955 EXT4_SB(sb)->s_cluster_ratio, flags);
2956 partial_cluster = 0;
2957 }
2958
Alex Tomasa86c6182006-10-11 01:21:03 -07002959 /* TODO: flexible tree reduction should be here */
2960 if (path->p_hdr->eh_entries == 0) {
2961 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002962 * truncate to zero freed all the tree,
2963 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002964 */
2965 err = ext4_ext_get_access(handle, inode, path);
2966 if (err == 0) {
2967 ext_inode_hdr(inode)->eh_depth = 0;
2968 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002969 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002970 err = ext4_ext_dirty(handle, inode, path);
2971 }
2972 }
2973out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002974 ext4_ext_drop_refs(path);
2975 kfree(path);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002976 if (err == -EAGAIN) {
2977 path = NULL;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002978 goto again;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002979 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002980 ext4_journal_stop(handle);
2981
2982 return err;
2983}
2984
2985/*
2986 * called at mount time
2987 */
2988void ext4_ext_init(struct super_block *sb)
2989{
2990 /*
2991 * possible initialization would be here
2992 */
2993
Theodore Ts'o83982b62009-01-06 14:53:16 -05002994 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002995#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04002996 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002997#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04002998 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07002999#endif
3000#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04003001 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07003002#endif
3003#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04003004 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07003005#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04003006 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04003007#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07003008#ifdef EXTENTS_STATS
3009 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3010 EXT4_SB(sb)->s_ext_min = 1 << 30;
3011 EXT4_SB(sb)->s_ext_max = 0;
3012#endif
3013 }
3014}
3015
3016/*
3017 * called at umount time
3018 */
3019void ext4_ext_release(struct super_block *sb)
3020{
Theodore Ts'o83982b62009-01-06 14:53:16 -05003021 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07003022 return;
3023
3024#ifdef EXTENTS_STATS
3025 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3026 struct ext4_sb_info *sbi = EXT4_SB(sb);
3027 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3028 sbi->s_ext_blocks, sbi->s_ext_extents,
3029 sbi->s_ext_blocks / sbi->s_ext_extents);
3030 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3031 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3032 }
3033#endif
3034}
3035
Zheng Liud7b2a002013-08-28 14:47:06 -04003036static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3037{
3038 ext4_lblk_t ee_block;
3039 ext4_fsblk_t ee_pblock;
3040 unsigned int ee_len;
3041
3042 ee_block = le32_to_cpu(ex->ee_block);
3043 ee_len = ext4_ext_get_actual_len(ex);
3044 ee_pblock = ext4_ext_pblock(ex);
3045
3046 if (ee_len == 0)
3047 return 0;
3048
3049 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3050 EXTENT_STATUS_WRITTEN);
3051}
3052
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003053/* FIXME!! we need to try to merge to left or right after zero-out */
3054static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3055{
Lukas Czerner24075182010-10-27 21:30:06 -04003056 ext4_fsblk_t ee_pblock;
3057 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04003058 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003059
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003060 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003061 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003062
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04003063 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04003064 if (ret > 0)
3065 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003066
Lukas Czerner24075182010-10-27 21:30:06 -04003067 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003068}
3069
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003070/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003071 * ext4_split_extent_at() splits an extent at given block.
3072 *
3073 * @handle: the journal handle
3074 * @inode: the file inode
3075 * @path: the path to the extent
3076 * @split: the logical block where the extent is splitted.
3077 * @split_flags: indicates if the extent could be zeroout if split fails, and
3078 * the states(init or uninit) of new extents.
3079 * @flags: flags used to insert new extent to extent tree.
3080 *
3081 *
3082 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3083 * of which are deterimined by split_flag.
3084 *
3085 * There are two cases:
3086 * a> the extent are splitted into two extent.
3087 * b> split is not needed, and just mark the extent.
3088 *
3089 * return 0 on success.
3090 */
3091static int ext4_split_extent_at(handle_t *handle,
3092 struct inode *inode,
3093 struct ext4_ext_path *path,
3094 ext4_lblk_t split,
3095 int split_flag,
3096 int flags)
3097{
3098 ext4_fsblk_t newblock;
3099 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04003100 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003101 struct ext4_extent *ex2 = NULL;
3102 unsigned int ee_len, depth;
3103 int err = 0;
3104
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003105 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3106 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3107
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003108 ext_debug("ext4_split_extents_at: inode %lu, logical"
3109 "block %llu\n", inode->i_ino, (unsigned long long)split);
3110
3111 ext4_ext_show_leaf(inode, path);
3112
3113 depth = ext_depth(inode);
3114 ex = path[depth].p_ext;
3115 ee_block = le32_to_cpu(ex->ee_block);
3116 ee_len = ext4_ext_get_actual_len(ex);
3117 newblock = split - ee_block + ext4_ext_pblock(ex);
3118
3119 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003120 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3121 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3122 EXT4_EXT_MARK_UNINIT1 |
3123 EXT4_EXT_MARK_UNINIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003124
3125 err = ext4_ext_get_access(handle, inode, path + depth);
3126 if (err)
3127 goto out;
3128
3129 if (split == ee_block) {
3130 /*
3131 * case b: block @split is the block that the extent begins with
3132 * then we just change the state of the extent, and splitting
3133 * is not needed.
3134 */
3135 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3136 ext4_ext_mark_uninitialized(ex);
3137 else
3138 ext4_ext_mark_initialized(ex);
3139
3140 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003141 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003142
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003143 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003144 goto out;
3145 }
3146
3147 /* case a */
3148 memcpy(&orig_ex, ex, sizeof(orig_ex));
3149 ex->ee_len = cpu_to_le16(split - ee_block);
3150 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3151 ext4_ext_mark_uninitialized(ex);
3152
3153 /*
3154 * path may lead to new leaf, not to original leaf any more
3155 * after ext4_ext_insert_extent() returns,
3156 */
3157 err = ext4_ext_dirty(handle, inode, path + depth);
3158 if (err)
3159 goto fix_extent_len;
3160
3161 ex2 = &newex;
3162 ex2->ee_block = cpu_to_le32(split);
3163 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3164 ext4_ext_store_pblock(ex2, newblock);
3165 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3166 ext4_ext_mark_uninitialized(ex2);
3167
3168 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3169 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003170 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003171 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003172 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003173 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003174 zero_ex.ee_len = cpu_to_le16(
3175 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003176 ext4_ext_store_pblock(&zero_ex,
3177 ext4_ext_pblock(ex2));
3178 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003179 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003180 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003181 zero_ex.ee_len = cpu_to_le16(
3182 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003183 ext4_ext_store_pblock(&zero_ex,
3184 ext4_ext_pblock(ex));
3185 }
3186 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003187 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003188 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003189 zero_ex.ee_len = cpu_to_le16(
3190 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003191 ext4_ext_store_pblock(&zero_ex,
3192 ext4_ext_pblock(&orig_ex));
3193 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003194
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003195 if (err)
3196 goto fix_extent_len;
3197 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003198 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003199 ext4_ext_try_to_merge(handle, inode, path, ex);
3200 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003201 if (err)
3202 goto fix_extent_len;
3203
3204 /* update extent status tree */
Zheng Liud7b2a002013-08-28 14:47:06 -04003205 err = ext4_zeroout_es(inode, &zero_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003206
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003207 goto out;
3208 } else if (err)
3209 goto fix_extent_len;
3210
3211out:
3212 ext4_ext_show_leaf(inode, path);
3213 return err;
3214
3215fix_extent_len:
3216 ex->ee_len = orig_ex.ee_len;
3217 ext4_ext_dirty(handle, inode, path + depth);
3218 return err;
3219}
3220
3221/*
3222 * ext4_split_extents() splits an extent and mark extent which is covered
3223 * by @map as split_flags indicates
3224 *
Anatol Pomozov70261f52013-08-28 14:40:12 -04003225 * It may result in splitting the extent into multiple extents (up to three)
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003226 * There are three possibilities:
3227 * a> There is no split required
3228 * b> Splits in two extents: Split is happening at either end of the extent
3229 * c> Splits in three extents: Somone is splitting in middle of the extent
3230 *
3231 */
3232static int ext4_split_extent(handle_t *handle,
3233 struct inode *inode,
3234 struct ext4_ext_path *path,
3235 struct ext4_map_blocks *map,
3236 int split_flag,
3237 int flags)
3238{
3239 ext4_lblk_t ee_block;
3240 struct ext4_extent *ex;
3241 unsigned int ee_len, depth;
3242 int err = 0;
3243 int uninitialized;
3244 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003245 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003246
3247 depth = ext_depth(inode);
3248 ex = path[depth].p_ext;
3249 ee_block = le32_to_cpu(ex->ee_block);
3250 ee_len = ext4_ext_get_actual_len(ex);
3251 uninitialized = ext4_ext_is_uninitialized(ex);
3252
3253 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003254 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003255 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3256 if (uninitialized)
3257 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3258 EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003259 if (split_flag & EXT4_EXT_DATA_VALID2)
3260 split_flag1 |= EXT4_EXT_DATA_VALID1;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003261 err = ext4_split_extent_at(handle, inode, path,
3262 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003263 if (err)
3264 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003265 } else {
3266 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003267 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003268 /*
3269 * Update path is required because previous ext4_split_extent_at() may
3270 * result in split of original leaf or extent zeroout.
3271 */
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003272 ext4_ext_drop_refs(path);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04003273 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003274 if (IS_ERR(path))
3275 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003276 depth = ext_depth(inode);
3277 ex = path[depth].p_ext;
3278 uninitialized = ext4_ext_is_uninitialized(ex);
3279 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003280
3281 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003282 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3283 if (uninitialized) {
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003284 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003285 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3286 EXT4_EXT_MARK_UNINIT2);
3287 }
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003288 err = ext4_split_extent_at(handle, inode, path,
3289 map->m_lblk, split_flag1, flags);
3290 if (err)
3291 goto out;
3292 }
3293
3294 ext4_ext_show_leaf(inode, path);
3295out:
Zheng Liu3a225672013-03-10 21:20:23 -04003296 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003297}
3298
Amit Arora56055d32007-07-17 21:42:38 -04003299/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003300 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04003301 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003302 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04003303 * uninitialized).
3304 * There are three possibilities:
3305 * a> There is no split required: Entire extent should be initialized
3306 * b> Splits in two extents: Write is happening at either end of the extent
3307 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003308 *
3309 * Pre-conditions:
3310 * - The extent pointed to by 'path' is uninitialized.
3311 * - The extent pointed to by 'path' contains a superset
3312 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3313 *
3314 * Post-conditions on success:
3315 * - the returned value is the number of blocks beyond map->l_lblk
3316 * that are allocated and initialized.
3317 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003318 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003319static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003320 struct inode *inode,
3321 struct ext4_map_blocks *map,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003322 struct ext4_ext_path *path,
3323 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003324{
Zheng Liu67a5da52012-08-17 09:54:17 -04003325 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003326 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003327 struct ext4_map_blocks split_map;
3328 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003329 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003330 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003331 unsigned int ee_len, depth, map_len = map->m_len;
3332 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003333 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003334 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003335
3336 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3337 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003338 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003339
Zheng Liu67a5da52012-08-17 09:54:17 -04003340 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003341 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3342 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003343 if (eof_block < map->m_lblk + map_len)
3344 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003345
3346 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003347 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003348 ex = path[depth].p_ext;
3349 ee_block = le32_to_cpu(ex->ee_block);
3350 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003351 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003352
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003353 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3354
3355 /* Pre-conditions */
3356 BUG_ON(!ext4_ext_is_uninitialized(ex));
3357 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003358
3359 /*
3360 * Attempt to transfer newly initialized blocks from the currently
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003361 * uninitialized extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003362 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003363 * memmove() calls. Transferring to the left is the common case in
3364 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3365 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003366 *
3367 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003368 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003369 * This would require removing the extent if the transfer
3370 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003371 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003372 * same extent tree node.
3373 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003374 if ((map->m_lblk == ee_block) &&
3375 /* See if we can merge left */
3376 (map_len < ee_len) && /*L1*/
3377 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003378 ext4_lblk_t prev_lblk;
3379 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003380 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003381
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003382 abut_ex = ex - 1;
3383 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3384 prev_len = ext4_ext_get_actual_len(abut_ex);
3385 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003386 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003387
3388 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003389 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003390 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003391 * - C1: abut_ex is initialized,
3392 * - C2: abut_ex is logically abutting ex,
3393 * - C3: abut_ex is physically abutting ex,
3394 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003395 * overflowing the (initialized) length limit.
3396 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003397 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003398 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3399 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003400 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003401 err = ext4_ext_get_access(handle, inode, path + depth);
3402 if (err)
3403 goto out;
3404
3405 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003406 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003407
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003408 /* Shift the start of ex by 'map_len' blocks */
3409 ex->ee_block = cpu_to_le32(ee_block + map_len);
3410 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3411 ex->ee_len = cpu_to_le16(ee_len - map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003412 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3413
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003414 /* Extend abut_ex by 'map_len' blocks */
3415 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003416
3417 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003418 allocated = map_len;
3419 }
3420 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3421 (map_len < ee_len) && /*L1*/
3422 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3423 /* See if we can merge right */
3424 ext4_lblk_t next_lblk;
3425 ext4_fsblk_t next_pblk, ee_pblk;
3426 unsigned int next_len;
3427
3428 abut_ex = ex + 1;
3429 next_lblk = le32_to_cpu(abut_ex->ee_block);
3430 next_len = ext4_ext_get_actual_len(abut_ex);
3431 next_pblk = ext4_ext_pblock(abut_ex);
3432 ee_pblk = ext4_ext_pblock(ex);
3433
3434 /*
3435 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3436 * upon those conditions:
3437 * - C1: abut_ex is initialized,
3438 * - C2: abut_ex is logically abutting ex,
3439 * - C3: abut_ex is physically abutting ex,
3440 * - C4: abut_ex can receive the additional blocks without
3441 * overflowing the (initialized) length limit.
3442 */
3443 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3444 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3445 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3446 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3447 err = ext4_ext_get_access(handle, inode, path + depth);
3448 if (err)
3449 goto out;
3450
3451 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3452 map, ex, abut_ex);
3453
3454 /* Shift the start of abut_ex by 'map_len' blocks */
3455 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3456 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3457 ex->ee_len = cpu_to_le16(ee_len - map_len);
3458 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3459
3460 /* Extend abut_ex by 'map_len' blocks */
3461 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3462
3463 /* Result: number of initialized blocks past m_lblk */
3464 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003465 }
3466 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003467 if (allocated) {
3468 /* Mark the block containing both extents as dirty */
3469 ext4_ext_dirty(handle, inode, path + depth);
3470
3471 /* Update path to point to the right extent */
3472 path[depth].p_ext = abut_ex;
3473 goto out;
3474 } else
3475 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003476
Yongqiang Yang667eff32011-05-03 12:25:07 -04003477 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003478 /*
3479 * It is safe to convert extent to initialized via explicit
Yongqiang Yang9e740562014-01-06 14:05:23 -05003480 * zeroout only if extent is fully inside i_size or new_size.
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003481 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003482 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003483
Zheng Liu67a5da52012-08-17 09:54:17 -04003484 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3485 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003486 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003487
3488 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3489 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003490 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003491 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003492 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003493 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003494 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003495 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003496
3497 err = ext4_ext_get_access(handle, inode, path + depth);
3498 if (err)
3499 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003500 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003501 ext4_ext_try_to_merge(handle, inode, path, ex);
3502 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003503 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003504 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003505
Amit Arora56055d32007-07-17 21:42:38 -04003506 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003507 * four cases:
3508 * 1. split the extent into three extents.
3509 * 2. split the extent into two extents, zeroout the first half.
3510 * 3. split the extent into two extents, zeroout the second half.
3511 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003512 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003513 split_map.m_lblk = map->m_lblk;
3514 split_map.m_len = map->m_len;
3515
Zheng Liu67a5da52012-08-17 09:54:17 -04003516 if (max_zeroout && (allocated > map->m_len)) {
3517 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003518 /* case 3 */
3519 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003520 cpu_to_le32(map->m_lblk);
3521 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003522 ext4_ext_store_pblock(&zero_ex,
3523 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3524 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003525 if (err)
3526 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003527 split_map.m_lblk = map->m_lblk;
3528 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003529 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003530 /* case 2 */
3531 if (map->m_lblk != ee_block) {
3532 zero_ex.ee_block = ex->ee_block;
3533 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3534 ee_block);
3535 ext4_ext_store_pblock(&zero_ex,
3536 ext4_ext_pblock(ex));
3537 err = ext4_ext_zeroout(inode, &zero_ex);
3538 if (err)
3539 goto out;
3540 }
3541
Yongqiang Yang667eff32011-05-03 12:25:07 -04003542 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003543 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3544 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003545 }
3546 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003547
3548 allocated = ext4_split_extent(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003549 &split_map, split_flag, flags);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003550 if (allocated < 0)
3551 err = allocated;
3552
Amit Arora56055d32007-07-17 21:42:38 -04003553out:
Zheng Liuadb23552013-03-10 21:13:05 -04003554 /* If we have gotten a failure, don't zero out status tree */
3555 if (!err)
Zheng Liud7b2a002013-08-28 14:47:06 -04003556 err = ext4_zeroout_es(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003557 return err ? err : allocated;
3558}
3559
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003560/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003561 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003562 * ext4_get_blocks_dio_write() when DIO to write
3563 * to an uninitialized extent.
3564 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003565 * Writing to an uninitialized extent may result in splitting the uninitialized
Wang Sheng-Hui30cb27d2012-08-18 22:38:07 -04003566 * extent into multiple initialized/uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003567 * There are three possibilities:
3568 * a> There is no split required: Entire extent should be uninitialized
3569 * b> Splits in two extents: Write is happening at either end of the extent
3570 * c> Splits in three extents: Somone is writing in middle of the extent
3571 *
3572 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003573 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003574 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003575 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003576 * into three uninitialized extent(at most). After IO complete, the part
3577 * being filled will be convert to initialized by the end_io callback function
3578 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003579 *
3580 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003581 */
3582static int ext4_split_unwritten_extents(handle_t *handle,
3583 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003584 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003585 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003586 int flags)
3587{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003588 ext4_lblk_t eof_block;
3589 ext4_lblk_t ee_block;
3590 struct ext4_extent *ex;
3591 unsigned int ee_len;
3592 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003593
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003594 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3595 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003596 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003597
3598 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3599 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003600 if (eof_block < map->m_lblk + map->m_len)
3601 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003602 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003603 * It is safe to convert extent to initialized via explicit
3604 * zeroout only if extent is fully insde i_size or new_size.
3605 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003606 depth = ext_depth(inode);
3607 ex = path[depth].p_ext;
3608 ee_block = le32_to_cpu(ex->ee_block);
3609 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003610
Yongqiang Yang667eff32011-05-03 12:25:07 -04003611 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3612 split_flag |= EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003613 if (flags & EXT4_GET_BLOCKS_CONVERT)
3614 split_flag |= EXT4_EXT_DATA_VALID2;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003615 flags |= EXT4_GET_BLOCKS_PRE_IO;
3616 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003617}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003618
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003619static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003620 struct inode *inode,
3621 struct ext4_map_blocks *map,
3622 struct ext4_ext_path *path)
Mingming Cao00314622009-09-28 15:49:08 -04003623{
3624 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003625 ext4_lblk_t ee_block;
3626 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003627 int depth;
3628 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003629
3630 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003631 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003632 ee_block = le32_to_cpu(ex->ee_block);
3633 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003634
Yongqiang Yang197217a2011-05-03 11:45:29 -04003635 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3636 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003637 (unsigned long long)ee_block, ee_len);
3638
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003639 /* If extent is larger than requested it is a clear sign that we still
3640 * have some extent state machine issues left. So extent_split is still
3641 * required.
3642 * TODO: Once all related issues will be fixed this situation should be
3643 * illegal.
3644 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003645 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003646#ifdef EXT4_DEBUG
3647 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3648 " len %u; IO logical block %llu, len %u\n",
3649 inode->i_ino, (unsigned long long)ee_block, ee_len,
3650 (unsigned long long)map->m_lblk, map->m_len);
3651#endif
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003652 err = ext4_split_unwritten_extents(handle, inode, map, path,
3653 EXT4_GET_BLOCKS_CONVERT);
3654 if (err < 0)
3655 goto out;
3656 ext4_ext_drop_refs(path);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04003657 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003658 if (IS_ERR(path)) {
3659 err = PTR_ERR(path);
3660 goto out;
3661 }
3662 depth = ext_depth(inode);
3663 ex = path[depth].p_ext;
3664 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003665
Mingming Cao00314622009-09-28 15:49:08 -04003666 err = ext4_ext_get_access(handle, inode, path + depth);
3667 if (err)
3668 goto out;
3669 /* first mark the extent as initialized */
3670 ext4_ext_mark_initialized(ex);
3671
Yongqiang Yang197217a2011-05-03 11:45:29 -04003672 /* note: ext4_ext_correct_indexes() isn't needed here because
3673 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003674 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003675 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003676
Mingming Cao00314622009-09-28 15:49:08 -04003677 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003678 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003679out:
3680 ext4_ext_show_leaf(inode, path);
3681 return err;
3682}
3683
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003684static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3685 sector_t block, int count)
3686{
3687 int i;
3688 for (i = 0; i < count; i++)
3689 unmap_underlying_metadata(bdev, block + i);
3690}
3691
Theodore Ts'o58590b02010-10-27 21:23:12 -04003692/*
3693 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3694 */
3695static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003696 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003697 struct ext4_ext_path *path,
3698 unsigned int len)
3699{
3700 int i, depth;
3701 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003702 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003703
3704 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3705 return 0;
3706
3707 depth = ext_depth(inode);
3708 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003709
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003710 /*
3711 * We're going to remove EOFBLOCKS_FL entirely in future so we
3712 * do not care for this case anymore. Simply remove the flag
3713 * if there are no extents.
3714 */
3715 if (unlikely(!eh->eh_entries))
3716 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003717 last_ex = EXT_LAST_EXTENT(eh);
3718 /*
3719 * We should clear the EOFBLOCKS_FL flag if we are writing the
3720 * last block in the last extent in the file. We test this by
3721 * first checking to see if the caller to
3722 * ext4_ext_get_blocks() was interested in the last block (or
3723 * a block beyond the last block) in the current extent. If
3724 * this turns out to be false, we can bail out from this
3725 * function immediately.
3726 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003727 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003728 ext4_ext_get_actual_len(last_ex))
3729 return 0;
3730 /*
3731 * If the caller does appear to be planning to write at or
3732 * beyond the end of the current extent, we then test to see
3733 * if the current extent is the last extent in the file, by
3734 * checking to make sure it was reached via the rightmost node
3735 * at each level of the tree.
3736 */
3737 for (i = depth-1; i >= 0; i--)
3738 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3739 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003740out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003741 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3742 return ext4_mark_inode_dirty(handle, inode);
3743}
3744
Aditya Kali7b415bf2011-09-09 19:04:51 -04003745/**
3746 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3747 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003748 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003749 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003750int ext4_find_delalloc_range(struct inode *inode,
3751 ext4_lblk_t lblk_start,
3752 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003753{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003754 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003755
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003756 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003757 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003758 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003759 else if (es.es_lblk <= lblk_start &&
3760 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003761 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003762 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003763 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003764 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003765 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003766}
3767
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003768int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003769{
3770 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3771 ext4_lblk_t lblk_start, lblk_end;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003772 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003773 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3774
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003775 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003776}
3777
3778/**
3779 * Determines how many complete clusters (out of those specified by the 'map')
3780 * are under delalloc and were reserved quota for.
3781 * This function is called when we are writing out the blocks that were
3782 * originally written with their allocation delayed, but then the space was
3783 * allocated using fallocate() before the delayed allocation could be resolved.
3784 * The cases to look for are:
3785 * ('=' indicated delayed allocated blocks
3786 * '-' indicates non-delayed allocated blocks)
3787 * (a) partial clusters towards beginning and/or end outside of allocated range
3788 * are not delalloc'ed.
3789 * Ex:
3790 * |----c---=|====c====|====c====|===-c----|
3791 * |++++++ allocated ++++++|
3792 * ==> 4 complete clusters in above example
3793 *
3794 * (b) partial cluster (outside of allocated range) towards either end is
3795 * marked for delayed allocation. In this case, we will exclude that
3796 * cluster.
3797 * Ex:
3798 * |----====c========|========c========|
3799 * |++++++ allocated ++++++|
3800 * ==> 1 complete clusters in above example
3801 *
3802 * Ex:
3803 * |================c================|
3804 * |++++++ allocated ++++++|
3805 * ==> 0 complete clusters in above example
3806 *
3807 * The ext4_da_update_reserve_space will be called only if we
3808 * determine here that there were some "entire" clusters that span
3809 * this 'allocated' range.
3810 * In the non-bigalloc case, this function will just end up returning num_blks
3811 * without ever calling ext4_find_delalloc_range.
3812 */
3813static unsigned int
3814get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3815 unsigned int num_blks)
3816{
3817 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3818 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3819 ext4_lblk_t lblk_from, lblk_to, c_offset;
3820 unsigned int allocated_clusters = 0;
3821
3822 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3823 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3824
3825 /* max possible clusters for this allocation */
3826 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3827
Aditya Kalid8990242011-09-09 19:18:51 -04003828 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3829
Aditya Kali7b415bf2011-09-09 19:04:51 -04003830 /* Check towards left side */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003831 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003832 if (c_offset) {
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003833 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003834 lblk_to = lblk_from + c_offset - 1;
3835
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003836 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003837 allocated_clusters--;
3838 }
3839
3840 /* Now check towards right. */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003841 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003842 if (allocated_clusters && c_offset) {
3843 lblk_from = lblk_start + num_blks;
3844 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3845
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003846 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003847 allocated_clusters--;
3848 }
3849
3850 return allocated_clusters;
3851}
3852
Mingming Cao00314622009-09-28 15:49:08 -04003853static int
3854ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003855 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003856 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003857 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003858{
3859 int ret = 0;
3860 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003861 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003862
Zheng Liu88635ca2011-12-28 19:00:25 -05003863 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3864 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003865 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003866 flags, allocated);
3867 ext4_ext_show_leaf(inode, path);
3868
Lukas Czerner27dd4382013-04-09 22:11:22 -04003869 /*
3870 * When writing into uninitialized space, we should not fail to
3871 * allocate metadata blocks for the new extent block if needed.
3872 */
3873 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3874
Zheng Liub5645532012-11-08 14:33:43 -05003875 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3876 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04003877
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003878 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003879 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003880 ret = ext4_split_unwritten_extents(handle, inode, map,
3881 path, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04003882 if (ret <= 0)
3883 goto out;
Mingming5f524952009-11-10 10:48:04 -05003884 /*
3885 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003886 * that this IO needs to conversion to written when IO is
Mingming5f524952009-11-10 10:48:04 -05003887 * completed
3888 */
Tao Ma0edeb712011-10-31 17:30:44 -04003889 if (io)
3890 ext4_set_io_unwritten_flag(inode, io);
3891 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003892 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05003893 map->m_flags |= EXT4_MAP_UNWRITTEN;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003894 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003895 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003896 goto out;
3897 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003898 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003899 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003900 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Mingming Cao00314622009-09-28 15:49:08 -04003901 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003902 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003903 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003904 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3905 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003906 } else
3907 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04003908 map->m_flags |= EXT4_MAP_MAPPED;
Eric Whitney15cc1762014-02-12 10:42:45 -05003909 map->m_pblk = newblock;
Zheng Liucdee7842013-03-10 21:08:52 -04003910 if (allocated > map->m_len)
3911 allocated = map->m_len;
3912 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003913 goto out2;
3914 }
3915 /* buffered IO case */
3916 /*
3917 * repeat fallocate creation request
3918 * we already have an unwritten extent
3919 */
Zheng Liua25a4e12013-02-18 00:28:04 -05003920 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3921 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003922 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05003923 }
Mingming Cao00314622009-09-28 15:49:08 -04003924
3925 /* buffered READ or buffered write_begin() lookup */
3926 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3927 /*
3928 * We have blocks reserved already. We
3929 * return allocated blocks so that delalloc
3930 * won't do block reservation for us. But
3931 * the buffer head will be unmapped so that
3932 * a read from the block returns 0s.
3933 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003934 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003935 goto out1;
3936 }
3937
3938 /* buffered write, writepage time, convert*/
Lukas Czerner27dd4382013-04-09 22:11:22 -04003939 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003940 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05003941 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04003942out:
3943 if (ret <= 0) {
3944 err = ret;
3945 goto out2;
3946 } else
3947 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003948 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003949 /*
3950 * if we allocated more blocks than requested
3951 * we need to make sure we unmap the extra block
3952 * allocated. The actual needed block will get
3953 * unmapped later when we find the buffer_head marked
3954 * new.
3955 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003956 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003957 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003958 newblock + map->m_len,
3959 allocated - map->m_len);
3960 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003961 }
Zheng Liu3a225672013-03-10 21:20:23 -04003962 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003963
3964 /*
3965 * If we have done fallocate with the offset that is already
3966 * delayed allocated, we would have block reservation
3967 * and quota reservation done in the delayed write path.
3968 * But fallocate would have already updated quota and block
3969 * count for this offset. So cancel these reservation
3970 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003971 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3972 unsigned int reserved_clusters;
3973 reserved_clusters = get_reserved_cluster_alloc(inode,
3974 map->m_lblk, map->m_len);
3975 if (reserved_clusters)
3976 ext4_da_update_reserve_space(inode,
3977 reserved_clusters,
3978 0);
3979 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003980
Mingming Cao00314622009-09-28 15:49:08 -04003981map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003982 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003983 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3984 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3985 map->m_len);
3986 if (err < 0)
3987 goto out2;
3988 }
Mingming Cao00314622009-09-28 15:49:08 -04003989out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003990 if (allocated > map->m_len)
3991 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003992 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003993 map->m_pblk = newblock;
3994 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003995out2:
3996 if (path) {
3997 ext4_ext_drop_refs(path);
3998 kfree(path);
3999 }
4000 return err ? err : allocated;
4001}
Theodore Ts'o58590b02010-10-27 21:23:12 -04004002
Mingming Cao00314622009-09-28 15:49:08 -04004003/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004004 * get_implied_cluster_alloc - check to see if the requested
4005 * allocation (in the map structure) overlaps with a cluster already
4006 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04004007 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004008 * @map The requested lblk->pblk mapping
4009 * @ex The extent structure which might contain an implied
4010 * cluster allocation
4011 *
4012 * This function is called by ext4_ext_map_blocks() after we failed to
4013 * find blocks that were already in the inode's extent tree. Hence,
4014 * we know that the beginning of the requested region cannot overlap
4015 * the extent from the inode's extent tree. There are three cases we
4016 * want to catch. The first is this case:
4017 *
4018 * |--- cluster # N--|
4019 * |--- extent ---| |---- requested region ---|
4020 * |==========|
4021 *
4022 * The second case that we need to test for is this one:
4023 *
4024 * |--------- cluster # N ----------------|
4025 * |--- requested region --| |------- extent ----|
4026 * |=======================|
4027 *
4028 * The third case is when the requested region lies between two extents
4029 * within the same cluster:
4030 * |------------- cluster # N-------------|
4031 * |----- ex -----| |---- ex_right ----|
4032 * |------ requested region ------|
4033 * |================|
4034 *
4035 * In each of the above cases, we need to set the map->m_pblk and
4036 * map->m_len so it corresponds to the return the extent labelled as
4037 * "|====|" from cluster #N, since it is already in use for data in
4038 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4039 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4040 * as a new "allocated" block region. Otherwise, we will return 0 and
4041 * ext4_ext_map_blocks() will then allocate one or more new clusters
4042 * by calling ext4_mb_new_blocks().
4043 */
Aditya Kalid8990242011-09-09 19:18:51 -04004044static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004045 struct ext4_map_blocks *map,
4046 struct ext4_extent *ex,
4047 struct ext4_ext_path *path)
4048{
Aditya Kalid8990242011-09-09 19:18:51 -04004049 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004050 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004051 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05004052 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004053 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4054 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4055 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4056
4057 /* The extent passed in that we are trying to match */
4058 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4059 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4060
4061 /* The requested region passed into ext4_map_blocks() */
4062 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004063
4064 if ((rr_cluster_start == ex_cluster_end) ||
4065 (rr_cluster_start == ex_cluster_start)) {
4066 if (rr_cluster_start == ex_cluster_end)
4067 ee_start += ee_len - 1;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004068 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004069 map->m_len = min(map->m_len,
4070 (unsigned) sbi->s_cluster_ratio - c_offset);
4071 /*
4072 * Check for and handle this case:
4073 *
4074 * |--------- cluster # N-------------|
4075 * |------- extent ----|
4076 * |--- requested region ---|
4077 * |===========|
4078 */
4079
4080 if (map->m_lblk < ee_block)
4081 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4082
4083 /*
4084 * Check for the case where there is already another allocated
4085 * block to the right of 'ex' but before the end of the cluster.
4086 *
4087 * |------------- cluster # N-------------|
4088 * |----- ex -----| |---- ex_right ----|
4089 * |------ requested region ------|
4090 * |================|
4091 */
4092 if (map->m_lblk > ee_block) {
4093 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4094 map->m_len = min(map->m_len, next - map->m_lblk);
4095 }
Aditya Kalid8990242011-09-09 19:18:51 -04004096
4097 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004098 return 1;
4099 }
Aditya Kalid8990242011-09-09 19:18:51 -04004100
4101 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004102 return 0;
4103}
4104
4105
4106/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05004107 * Block allocation/map/preallocation routine for extents based files
4108 *
4109 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004110 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004111 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4112 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004113 *
4114 * return > 0, number of of blocks already mapped/allocated
4115 * if create == 0 and these are pre-allocated blocks
4116 * buffer head is unmapped
4117 * otherwise blocks are mapped
4118 *
4119 * return = 0, if plain look up failed (blocks have not been allocated)
4120 * buffer head is unmapped
4121 *
4122 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004123 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004124int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4125 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004126{
4127 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004128 struct ext4_extent newex, *ex, *ex2;
4129 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004130 ext4_fsblk_t newblock = 0;
Zheng Liu37794732012-11-08 14:47:52 -05004131 int free_on_err = 0, err = 0, depth;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004132 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004133 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004134 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004135 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004136 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004137 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07004138
Mingming84fe3be2009-09-01 08:44:37 -04004139 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004140 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004141 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004142
Alex Tomasa86c6182006-10-11 01:21:03 -07004143 /* find extent for this block */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04004144 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004145 if (IS_ERR(path)) {
4146 err = PTR_ERR(path);
4147 path = NULL;
4148 goto out2;
4149 }
4150
4151 depth = ext_depth(inode);
4152
4153 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004154 * consistent leaf must not be empty;
4155 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07004156 * this is why assert can't be put in ext4_ext_find_extent()
4157 */
Frank Mayhar273df552010-03-02 11:46:09 -05004158 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4159 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004160 "lblock: %lu, depth: %d pblock %lld",
4161 (unsigned long) map->m_lblk, depth,
4162 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004163 err = -EIO;
4164 goto out2;
4165 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004166
Avantika Mathur7e028972006-12-06 20:41:33 -08004167 ex = path[depth].p_ext;
4168 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004169 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004170 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004171 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004172
4173 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004174 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004175 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004176 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004177 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004178
4179 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4180
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004181 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004182 if (in_range(map->m_lblk, ee_block, ee_len)) {
4183 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004184 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004185 allocated = ee_len - (map->m_lblk - ee_block);
4186 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4187 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004188
Zheng Liu69eb33d2013-02-18 00:31:07 -05004189 if (!ext4_ext_is_uninitialized(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004190 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004191
Zheng Liu37794732012-11-08 14:47:52 -05004192 allocated = ext4_ext_handle_uninitialized_extents(
Lukas Czerner78771912012-03-19 23:05:43 -04004193 handle, inode, map, path, flags,
4194 allocated, newblock);
Zheng Liu37794732012-11-08 14:47:52 -05004195 goto out3;
Alex Tomasa86c6182006-10-11 01:21:03 -07004196 }
4197 }
4198
Aditya Kali7b415bf2011-09-09 19:04:51 -04004199 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05004200 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04004201 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4202
Alex Tomasa86c6182006-10-11 01:21:03 -07004203 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004204 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004205 * we couldn't try to create block if create flag is zero
4206 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004207 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004208 /*
4209 * put just found gap into cache to speed up
4210 * subsequent requests
4211 */
Zheng Liud100eef2013-02-18 00:29:59 -05004212 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4213 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004214 goto out2;
4215 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004216
Alex Tomasa86c6182006-10-11 01:21:03 -07004217 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004218 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004219 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004220 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004221 newex.ee_block = cpu_to_le32(map->m_lblk);
Eric Whitneyd0abafa2014-01-06 14:00:23 -05004222 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004223
4224 /*
4225 * If we are doing bigalloc, check to see if the extent returned
4226 * by ext4_ext_find_extent() implies a cluster we can use.
4227 */
4228 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004229 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004230 ar.len = allocated = map->m_len;
4231 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004232 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004233 goto got_allocated_blocks;
4234 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004235
Alex Tomasc9de5602008-01-29 00:19:52 -05004236 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004237 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004238 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4239 if (err)
4240 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004241 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004242 ex2 = NULL;
4243 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004244 if (err)
4245 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004246
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004247 /* Check if the extent after searching to the right implies a
4248 * cluster we can use. */
4249 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004250 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004251 ar.len = allocated = map->m_len;
4252 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004253 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004254 goto got_allocated_blocks;
4255 }
4256
Amit Arora749269f2007-07-18 09:02:56 -04004257 /*
4258 * See if request is beyond maximum number of blocks we can have in
4259 * a single extent. For an initialized extent this limit is
4260 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4261 * EXT_UNINIT_MAX_LEN.
4262 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004263 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004264 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004265 map->m_len = EXT_INIT_MAX_LEN;
4266 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004267 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004268 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004269
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004270 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004271 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004272 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004273 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004274 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004275 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004276 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004277
4278 /* allocate new block */
4279 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004280 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4281 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004282 /*
4283 * We calculate the offset from the beginning of the cluster
4284 * for the logical block number, since when we allocate a
4285 * physical cluster, the physical block should start at the
4286 * same offset from the beginning of the cluster. This is
4287 * needed so that future calls to get_implied_cluster_alloc()
4288 * work correctly.
4289 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004290 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004291 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4292 ar.goal -= offset;
4293 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004294 if (S_ISREG(inode->i_mode))
4295 ar.flags = EXT4_MB_HINT_DATA;
4296 else
4297 /* disable in-core preallocation for non-regular files */
4298 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004299 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4300 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004301 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004302 if (!newblock)
4303 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004304 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004305 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004306 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004307 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004308 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4309 if (ar.len > allocated)
4310 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004311
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004312got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004313 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004314 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004315 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004316 /* Mark uninitialized */
4317 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04004318 ext4_ext_mark_uninitialized(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004319 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004320 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004321 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004322 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004323 * here we flag the IO that really needs the conversion.
Mingming5f524952009-11-10 10:48:04 -05004324 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004325 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004326 */
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004327 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4328 set_unwritten = 1;
Jiaying Zhang744692d2010-03-04 16:14:02 -05004329 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004330 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004331 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004332
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004333 err = 0;
4334 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4335 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4336 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004337 if (!err)
4338 err = ext4_ext_insert_extent(handle, inode, path,
4339 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004340
4341 if (!err && set_unwritten) {
4342 if (io)
4343 ext4_set_io_unwritten_flag(inode, io);
4344 else
4345 ext4_set_inode_state(inode,
4346 EXT4_STATE_DIO_UNWRITTEN);
4347 }
4348
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004349 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004350 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4351 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004352 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004353 /* not a good idea to call discard here directly,
4354 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004355 ext4_discard_preallocations(inode);
Theodore Ts'oc8e15132013-07-15 00:09:37 -04004356 ext4_free_blocks(handle, inode, NULL, newblock,
4357 EXT4_C2B(sbi, allocated_clusters), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004358 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004359 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004360
Alex Tomasa86c6182006-10-11 01:21:03 -07004361 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004362 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004363 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004364 if (allocated > map->m_len)
4365 allocated = map->m_len;
4366 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004367
Jan Karab436b9b2009-12-08 23:51:10 -05004368 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004369 * Update reserved blocks/metadata blocks after successful
4370 * block allocation which had been deferred till now.
4371 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004372 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004373 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004374 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004375 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004376 */
4377 reserved_clusters = get_reserved_cluster_alloc(inode,
4378 map->m_lblk, allocated);
4379 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4380 if (reserved_clusters) {
4381 /*
4382 * We have clusters reserved for this range.
4383 * But since we are not doing actual allocation
4384 * and are simply using blocks from previously
4385 * allocated cluster, we should release the
4386 * reservation and not claim quota.
4387 */
4388 ext4_da_update_reserve_space(inode,
4389 reserved_clusters, 0);
4390 }
4391 } else {
4392 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004393 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f262011-09-09 19:20:51 -04004394 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004395 int reservation = allocated_clusters -
4396 reserved_clusters;
4397 /*
4398 * It seems we claimed few clusters outside of
4399 * the range of this allocation. We should give
4400 * it back to the reservation pool. This can
4401 * happen in the following case:
4402 *
4403 * * Suppose s_cluster_ratio is 4 (i.e., each
4404 * cluster has 4 blocks. Thus, the clusters
4405 * are [0-3],[4-7],[8-11]...
4406 * * First comes delayed allocation write for
4407 * logical blocks 10 & 11. Since there were no
4408 * previous delayed allocated blocks in the
4409 * range [8-11], we would reserve 1 cluster
4410 * for this write.
4411 * * Next comes write for logical blocks 3 to 8.
4412 * In this case, we will reserve 2 clusters
4413 * (for [0-3] and [4-7]; and not for [8-11] as
4414 * that range has a delayed allocated blocks.
4415 * Thus total reserved clusters now becomes 3.
4416 * * Now, during the delayed allocation writeout
4417 * time, we will first write blocks [3-8] and
4418 * allocate 3 clusters for writing these
4419 * blocks. Also, we would claim all these
4420 * three clusters above.
4421 * * Now when we come here to writeout the
4422 * blocks [10-11], we would expect to claim
4423 * the reservation of 1 cluster we had made
4424 * (and we would claim it since there are no
4425 * more delayed allocated blocks in the range
4426 * [8-11]. But our reserved cluster count had
4427 * already gone to 0.
4428 *
4429 * Thus, at the step 4 above when we determine
4430 * that there are still some unwritten delayed
4431 * allocated blocks outside of our current
4432 * block range, we should increment the
4433 * reserved clusters count so that when the
4434 * remaining blocks finally gets written, we
4435 * could claim them.
4436 */
Aditya Kali5356f262011-09-09 19:20:51 -04004437 dquot_reserve_block(inode,
4438 EXT4_C2B(sbi, reservation));
4439 spin_lock(&ei->i_block_reservation_lock);
4440 ei->i_reserved_data_blocks += reservation;
4441 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004442 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004443 /*
4444 * We will claim quota for all newly allocated blocks.
4445 * We're updating the reserved space *after* the
4446 * correction above so we do not accidentally free
4447 * all the metadata reservation because we might
4448 * actually need it later on.
4449 */
4450 ext4_da_update_reserve_space(inode, allocated_clusters,
4451 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004452 }
4453 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004454
4455 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004456 * Cache the extent and update transaction to commit on fdatasync only
4457 * when it is _not_ an uninitialized extent.
4458 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004459 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004460 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004461 else
Jan Karab436b9b2009-12-08 23:51:10 -05004462 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004463out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004464 if (allocated > map->m_len)
4465 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004466 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004467 map->m_flags |= EXT4_MAP_MAPPED;
4468 map->m_pblk = newblock;
4469 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004470out2:
4471 if (path) {
4472 ext4_ext_drop_refs(path);
4473 kfree(path);
4474 }
Allison Hendersone8613042011-05-25 07:41:46 -04004475
Zheng Liu37794732012-11-08 14:47:52 -05004476out3:
Theodore Ts'o63b99962013-07-16 10:28:47 -04004477 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4478 err ? err : allocated);
4479 ext4_es_lru_add(inode);
Lukas Czerner78771912012-03-19 23:05:43 -04004480 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004481}
4482
Theodore Ts'o819c4922013-04-03 12:47:17 -04004483void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004484{
Alex Tomasa86c6182006-10-11 01:21:03 -07004485 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004486 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004487 int err = 0;
4488
4489 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004490 * TODO: optimization is possible here.
4491 * Probably we need not scan at all,
4492 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004493 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004494
4495 /* we have to know where to truncate from in crash case */
4496 EXT4_I(inode)->i_disksize = inode->i_size;
4497 ext4_mark_inode_dirty(handle, inode);
4498
4499 last_block = (inode->i_size + sb->s_blocksize - 1)
4500 >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004501retry:
Zheng Liu51865fd2012-11-08 21:57:32 -05004502 err = ext4_es_remove_extent(inode, last_block,
4503 EXT_MAX_BLOCKS - last_block);
Theodore Ts'o94eec0f2013-07-29 12:12:56 -04004504 if (err == -ENOMEM) {
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004505 cond_resched();
4506 congestion_wait(BLK_RW_ASYNC, HZ/50);
4507 goto retry;
4508 }
4509 if (err) {
4510 ext4_std_error(inode->i_sb, err);
4511 return;
4512 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04004513 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004514 ext4_std_error(inode->i_sb, err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004515}
4516
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004517static void ext4_falloc_update_inode(struct inode *inode,
4518 int mode, loff_t new_size, int update_ctime)
4519{
4520 struct timespec now;
4521
4522 if (update_ctime) {
4523 now = current_fs_time(inode->i_sb);
4524 if (!timespec_equal(&inode->i_ctime, &now))
4525 inode->i_ctime = now;
4526 }
4527 /*
4528 * Update only when preallocation was requested beyond
4529 * the file size.
4530 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04004531 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4532 if (new_size > i_size_read(inode))
4533 i_size_write(inode, new_size);
4534 if (new_size > EXT4_I(inode)->i_disksize)
4535 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004536 } else {
4537 /*
4538 * Mark that we allocate beyond EOF so the subsequent truncate
4539 * can proceed even if the new size is the same as i_size.
4540 */
4541 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004542 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004543 }
4544
4545}
4546
Amit Aroraa2df2a62007-07-17 21:42:41 -04004547/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004548 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04004549 * operation, which gets called from sys_fallocate system call.
4550 * For block-mapped files, posix_fallocate should fall back to the method
4551 * of writing zeroes to the required new blocks (the same behavior which is
4552 * expected for file systems which do not support fallocate() system call).
4553 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004554long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004555{
Al Viro496ad9a2013-01-23 17:07:38 -05004556 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004557 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004558 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004559 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004560 int ret = 0;
4561 int ret2 = 0;
4562 int retries = 0;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004563 int flags;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004564 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004565 unsigned int credits, blkbits = inode->i_blkbits;
4566
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004567 /* Return error if mode is not supported */
4568 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4569 return -EOPNOTSUPP;
4570
4571 if (mode & FALLOC_FL_PUNCH_HOLE)
Ashish Sangwanaeb28172013-07-01 08:12:38 -04004572 return ext4_punch_hole(inode, offset, len);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004573
Tao Ma0c8d4142012-12-10 14:06:03 -05004574 ret = ext4_convert_inline_data(inode);
4575 if (ret)
4576 return ret;
4577
Zheng Liu8bad6fc2013-01-28 09:21:37 -05004578 /*
4579 * currently supporting (pre)allocate mode for extent-based
4580 * files _only_
4581 */
4582 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4583 return -EOPNOTSUPP;
4584
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004585 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004586 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004587 /*
4588 * We can't just convert len to max_blocks because
4589 * If blocksize = 4096 offset = 3072 and len = 2048
4590 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004591 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004592 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004593 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04004594 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04004595 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004596 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004597 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004598 ret = inode_newsize_ok(inode, (len + offset));
4599 if (ret) {
4600 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004601 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004602 return ret;
4603 }
Greg Harm3c6fe772011-10-31 18:41:47 -04004604 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004605 if (mode & FALLOC_FL_KEEP_SIZE)
4606 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
Greg Harm3c6fe772011-10-31 18:41:47 -04004607 /*
4608 * Don't normalize the request if it can fit in one extent so
4609 * that it doesn't get unnecessarily split into multiple
4610 * extents.
4611 */
4612 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4613 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004614
Amit Aroraa2df2a62007-07-17 21:42:41 -04004615retry:
4616 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004617 map.m_lblk = map.m_lblk + ret;
4618 map.m_len = max_blocks = max_blocks - ret;
Theodore Ts'o9924a922013-02-08 21:59:22 -05004619 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4620 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004621 if (IS_ERR(handle)) {
4622 ret = PTR_ERR(handle);
4623 break;
4624 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004625 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004626 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004627#ifdef EXT4FS_DEBUG
Lukas Czernerb06acd32013-01-28 21:21:12 -05004628 ext4_warning(inode->i_sb,
4629 "inode #%lu: block %u: len %u: "
4630 "ext4_ext_map_blocks returned %d",
4631 inode->i_ino, map.m_lblk,
4632 map.m_len, ret);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004633#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04004634 ext4_mark_inode_dirty(handle, inode);
4635 ret2 = ext4_journal_stop(handle);
4636 break;
4637 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004638 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004639 blkbits) >> blkbits))
4640 new_size = offset + len;
4641 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04004642 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004643
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004644 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004645 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04004646 ext4_mark_inode_dirty(handle, inode);
Zheng Liuf4e95b32012-06-30 19:12:57 -04004647 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4648 ext4_handle_sync(handle);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004649 ret2 = ext4_journal_stop(handle);
4650 if (ret2)
4651 break;
4652 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004653 if (ret == -ENOSPC &&
4654 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4655 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004656 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004657 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004658 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004659 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4660 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004661 return ret > 0 ? ret2 : ret;
4662}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004663
4664/*
Mingming Cao00314622009-09-28 15:49:08 -04004665 * This function convert a range of blocks to written extents
4666 * The caller of this function will pass the start offset and the size.
4667 * all unwritten extents within this range will be converted to
4668 * written extents.
4669 *
4670 * This function is called from the direct IO end io call back
4671 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004672 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004673 */
Jan Kara6b523df2013-06-04 13:21:11 -04004674int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4675 loff_t offset, ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004676{
Mingming Cao00314622009-09-28 15:49:08 -04004677 unsigned int max_blocks;
4678 int ret = 0;
4679 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004680 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004681 unsigned int credits, blkbits = inode->i_blkbits;
4682
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004683 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004684 /*
4685 * We can't just convert len to max_blocks because
4686 * If blocksize = 4096 offset = 3072 and len = 2048
4687 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004688 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4689 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004690 /*
Jan Kara6b523df2013-06-04 13:21:11 -04004691 * This is somewhat ugly but the idea is clear: When transaction is
4692 * reserved, everything goes into it. Otherwise we rather start several
4693 * smaller transactions for conversion of each extent separately.
Mingming Cao00314622009-09-28 15:49:08 -04004694 */
Jan Kara6b523df2013-06-04 13:21:11 -04004695 if (handle) {
4696 handle = ext4_journal_start_reserved(handle,
4697 EXT4_HT_EXT_CONVERT);
4698 if (IS_ERR(handle))
4699 return PTR_ERR(handle);
4700 credits = 0;
4701 } else {
4702 /*
4703 * credits to insert 1 extent into extent tree
4704 */
4705 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4706 }
Mingming Cao00314622009-09-28 15:49:08 -04004707 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004708 map.m_lblk += ret;
4709 map.m_len = (max_blocks -= ret);
Jan Kara6b523df2013-06-04 13:21:11 -04004710 if (credits) {
4711 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4712 credits);
4713 if (IS_ERR(handle)) {
4714 ret = PTR_ERR(handle);
4715 break;
4716 }
Mingming Cao00314622009-09-28 15:49:08 -04004717 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004718 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004719 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05004720 if (ret <= 0)
4721 ext4_warning(inode->i_sb,
4722 "inode #%lu: block %u: len %u: "
4723 "ext4_ext_map_blocks returned %d",
4724 inode->i_ino, map.m_lblk,
4725 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04004726 ext4_mark_inode_dirty(handle, inode);
Jan Kara6b523df2013-06-04 13:21:11 -04004727 if (credits)
4728 ret2 = ext4_journal_stop(handle);
4729 if (ret <= 0 || ret2)
Mingming Cao00314622009-09-28 15:49:08 -04004730 break;
4731 }
Jan Kara6b523df2013-06-04 13:21:11 -04004732 if (!credits)
4733 ret2 = ext4_journal_stop(handle);
Mingming Cao00314622009-09-28 15:49:08 -04004734 return ret > 0 ? ret2 : ret;
4735}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004736
Mingming Cao00314622009-09-28 15:49:08 -04004737/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004738 * If newes is not existing extent (newes->ec_pblk equals zero) find
4739 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004740 * return start of the next delayed extent.
4741 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05004742 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004743 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05004744 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004745 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004746static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05004747 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004748{
Zheng Liub3aff3e2012-11-08 21:57:37 -05004749 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05004750 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004751
Zheng Liu69eb33d2013-02-18 00:31:07 -05004752 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004753 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4754 newes->es_lblk + newes->es_len - 1, &es);
4755
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004756 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004757 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004758 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004759 */
Zheng Liu06b0c882013-02-18 00:26:51 -05004760 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05004761 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004762 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004763
Zheng Liu69eb33d2013-02-18 00:31:07 -05004764 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05004765 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004766 newes->es_len = min(es.es_lblk - newes->es_lblk,
4767 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004768 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004769 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004770
Zheng Liu69eb33d2013-02-18 00:31:07 -05004771 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004772 }
4773
Zheng Liu69eb33d2013-02-18 00:31:07 -05004774 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004775 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05004776 if (es.es_len == 0)
4777 next_del = EXT_MAX_BLOCKS;
4778 else
4779 next_del = es.es_lblk;
4780
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004781 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004782}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004783/* fiemap flags we can handle specified here */
4784#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4785
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004786static int ext4_xattr_fiemap(struct inode *inode,
4787 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004788{
4789 __u64 physical = 0;
4790 __u64 length;
4791 __u32 flags = FIEMAP_EXTENT_LAST;
4792 int blockbits = inode->i_sb->s_blocksize_bits;
4793 int error = 0;
4794
4795 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004796 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004797 struct ext4_iloc iloc;
4798 int offset; /* offset of xattr in inode */
4799
4800 error = ext4_get_inode_loc(inode, &iloc);
4801 if (error)
4802 return error;
Jan Karaa60697f2013-05-31 19:38:56 -04004803 physical = (__u64)iloc.bh->b_blocknr << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004804 offset = EXT4_GOOD_OLD_INODE_SIZE +
4805 EXT4_I(inode)->i_extra_isize;
4806 physical += offset;
4807 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4808 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004809 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004810 } else { /* external block */
Jan Karaa60697f2013-05-31 19:38:56 -04004811 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004812 length = inode->i_sb->s_blocksize;
4813 }
4814
4815 if (physical)
4816 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4817 length, flags);
4818 return (error < 0 ? error : 0);
4819}
4820
4821int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4822 __u64 start, __u64 len)
4823{
4824 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004825 int error = 0;
4826
Tao Ma94191982012-12-10 14:06:02 -05004827 if (ext4_has_inline_data(inode)) {
4828 int has_inline = 1;
4829
4830 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4831
4832 if (has_inline)
4833 return error;
4834 }
4835
Theodore Ts'o7869a4a2013-08-16 22:05:14 -04004836 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4837 error = ext4_ext_precache(inode);
4838 if (error)
4839 return error;
4840 }
4841
Eric Sandeen6873fa02008-10-07 00:46:36 -04004842 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004843 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004844 return generic_block_fiemap(inode, fieinfo, start, len,
4845 ext4_get_block);
4846
4847 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4848 return -EBADR;
4849
4850 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4851 error = ext4_xattr_fiemap(inode, fieinfo);
4852 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004853 ext4_lblk_t len_blks;
4854 __u64 last_blk;
4855
Eric Sandeen6873fa02008-10-07 00:46:36 -04004856 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004857 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004858 if (last_blk >= EXT_MAX_BLOCKS)
4859 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004860 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004861
4862 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004863 * Walk the extent tree gathering extent information
4864 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004865 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004866 error = ext4_fill_fiemap_extents(inode, start_blk,
4867 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004868 }
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04004869 ext4_es_lru_add(inode);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004870 return error;
4871}