blob: bc0f1910b9cfa7dd5aa93b9c5d889c6fbacd037d [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);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400363
Theodore Ts'o31d4f3a2012-03-11 23:30:16 -0400364 if (len == 0)
365 return 0;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400366 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400367}
368
369static int ext4_valid_extent_idx(struct inode *inode,
370 struct ext4_extent_idx *ext_idx)
371{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400372 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400373
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400374 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400375}
376
377static int ext4_valid_extent_entries(struct inode *inode,
378 struct ext4_extent_header *eh,
379 int depth)
380{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400381 unsigned short entries;
382 if (eh->eh_entries == 0)
383 return 1;
384
385 entries = le16_to_cpu(eh->eh_entries);
386
387 if (depth == 0) {
388 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400389 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400390 while (entries) {
391 if (!ext4_valid_extent(inode, ext))
392 return 0;
393 ext++;
394 entries--;
395 }
396 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400397 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400398 while (entries) {
399 if (!ext4_valid_extent_idx(inode, ext_idx))
400 return 0;
401 ext_idx++;
402 entries--;
403 }
404 }
405 return 1;
406}
407
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400408static int __ext4_ext_check(const char *function, unsigned int line,
409 struct inode *inode, struct ext4_extent_header *eh,
410 int depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400411{
412 const char *error_msg;
413 int max = 0;
414
415 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
416 error_msg = "invalid magic";
417 goto corrupted;
418 }
419 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
420 error_msg = "unexpected eh_depth";
421 goto corrupted;
422 }
423 if (unlikely(eh->eh_max == 0)) {
424 error_msg = "invalid eh_max";
425 goto corrupted;
426 }
427 max = ext4_ext_max_entries(inode, depth);
428 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
429 error_msg = "too large eh_max";
430 goto corrupted;
431 }
432 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
433 error_msg = "invalid eh_entries";
434 goto corrupted;
435 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400436 if (!ext4_valid_extent_entries(inode, eh, depth)) {
437 error_msg = "invalid extent entries";
438 goto corrupted;
439 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400440 /* Verify checksum on non-root extent tree nodes */
441 if (ext_depth(inode) != depth &&
442 !ext4_extent_block_csum_verify(inode, eh)) {
443 error_msg = "extent tree corrupted";
444 goto corrupted;
445 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400446 return 0;
447
448corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400449 ext4_error_inode(inode, function, line, 0,
Theodore Ts'o24676da2010-05-16 21:00:00 -0400450 "bad header/extent: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400451 "entries %u, max %u(%u), depth %u(%u)",
Theodore Ts'o24676da2010-05-16 21:00:00 -0400452 error_msg, le16_to_cpu(eh->eh_magic),
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400453 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
454 max, le16_to_cpu(eh->eh_depth), depth);
455
456 return -EIO;
457}
458
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400459#define ext4_ext_check(inode, eh, depth) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400460 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400461
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400462int ext4_ext_check_inode(struct inode *inode)
463{
464 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
465}
466
Darrick J. Wongf8489122012-04-29 18:21:10 -0400467static int __ext4_ext_check_block(const char *function, unsigned int line,
468 struct inode *inode,
469 struct ext4_extent_header *eh,
470 int depth,
471 struct buffer_head *bh)
472{
473 int ret;
474
475 if (buffer_verified(bh))
476 return 0;
477 ret = ext4_ext_check(inode, eh, depth);
478 if (ret)
479 return ret;
480 set_buffer_verified(bh);
481 return ret;
482}
483
484#define ext4_ext_check_block(inode, eh, depth, bh) \
485 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
486
Alex Tomasa86c6182006-10-11 01:21:03 -0700487#ifdef EXT_DEBUG
488static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
489{
490 int k, l = path->p_depth;
491
492 ext_debug("path:");
493 for (k = 0; k <= l; k++, path++) {
494 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700495 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400496 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700497 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400498 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700499 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400500 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400501 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400502 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700503 } else
504 ext_debug(" []");
505 }
506 ext_debug("\n");
507}
508
509static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
510{
511 int depth = ext_depth(inode);
512 struct ext4_extent_header *eh;
513 struct ext4_extent *ex;
514 int i;
515
516 if (!path)
517 return;
518
519 eh = path[depth].p_hdr;
520 ex = EXT_FIRST_EXTENT(eh);
521
Mingming553f9002009-09-18 13:34:55 -0400522 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
523
Alex Tomasa86c6182006-10-11 01:21:03 -0700524 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400525 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
526 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400527 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700528 }
529 ext_debug("\n");
530}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400531
532static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
533 ext4_fsblk_t newblock, int level)
534{
535 int depth = ext_depth(inode);
536 struct ext4_extent *ex;
537
538 if (depth != level) {
539 struct ext4_extent_idx *idx;
540 idx = path[level].p_idx;
541 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
542 ext_debug("%d: move %d:%llu in new index %llu\n", level,
543 le32_to_cpu(idx->ei_block),
544 ext4_idx_pblock(idx),
545 newblock);
546 idx++;
547 }
548
549 return;
550 }
551
552 ex = path[depth].p_ext;
553 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
554 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
555 le32_to_cpu(ex->ee_block),
556 ext4_ext_pblock(ex),
557 ext4_ext_is_uninitialized(ex),
558 ext4_ext_get_actual_len(ex),
559 newblock);
560 ex++;
561 }
562}
563
Alex Tomasa86c6182006-10-11 01:21:03 -0700564#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400565#define ext4_ext_show_path(inode, path)
566#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400567#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700568#endif
569
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500570void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700571{
572 int depth = path->p_depth;
573 int i;
574
575 for (i = 0; i <= depth; i++, path++)
576 if (path->p_bh) {
577 brelse(path->p_bh);
578 path->p_bh = NULL;
579 }
580}
581
582/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700583 * ext4_ext_binsearch_idx:
584 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400585 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700586 */
587static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500588ext4_ext_binsearch_idx(struct inode *inode,
589 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700590{
591 struct ext4_extent_header *eh = path->p_hdr;
592 struct ext4_extent_idx *r, *l, *m;
593
Alex Tomasa86c6182006-10-11 01:21:03 -0700594
Eric Sandeenbba90742008-01-28 23:58:27 -0500595 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700596
597 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400598 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700599 while (l <= r) {
600 m = l + (r - l) / 2;
601 if (block < le32_to_cpu(m->ei_block))
602 r = m - 1;
603 else
604 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400605 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
606 m, le32_to_cpu(m->ei_block),
607 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700608 }
609
610 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400611 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400612 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700613
614#ifdef CHECK_BINSEARCH
615 {
616 struct ext4_extent_idx *chix, *ix;
617 int k;
618
619 chix = ix = EXT_FIRST_INDEX(eh);
620 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
621 if (k != 0 &&
622 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400623 printk(KERN_DEBUG "k=%d, ix=0x%p, "
624 "first=0x%p\n", k,
625 ix, EXT_FIRST_INDEX(eh));
626 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700627 le32_to_cpu(ix->ei_block),
628 le32_to_cpu(ix[-1].ei_block));
629 }
630 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400631 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700632 if (block < le32_to_cpu(ix->ei_block))
633 break;
634 chix = ix;
635 }
636 BUG_ON(chix != path->p_idx);
637 }
638#endif
639
640}
641
642/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700643 * ext4_ext_binsearch:
644 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400645 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700646 */
647static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500648ext4_ext_binsearch(struct inode *inode,
649 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700650{
651 struct ext4_extent_header *eh = path->p_hdr;
652 struct ext4_extent *r, *l, *m;
653
Alex Tomasa86c6182006-10-11 01:21:03 -0700654 if (eh->eh_entries == 0) {
655 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700656 * this leaf is empty:
657 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700658 */
659 return;
660 }
661
Eric Sandeenbba90742008-01-28 23:58:27 -0500662 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700663
664 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400665 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700666
667 while (l <= r) {
668 m = l + (r - l) / 2;
669 if (block < le32_to_cpu(m->ee_block))
670 r = m - 1;
671 else
672 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400673 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
674 m, le32_to_cpu(m->ee_block),
675 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700676 }
677
678 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400679 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400680 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400681 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400682 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400683 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700684
685#ifdef CHECK_BINSEARCH
686 {
687 struct ext4_extent *chex, *ex;
688 int k;
689
690 chex = ex = EXT_FIRST_EXTENT(eh);
691 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
692 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400693 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700694 if (block < le32_to_cpu(ex->ee_block))
695 break;
696 chex = ex;
697 }
698 BUG_ON(chex != path->p_ext);
699 }
700#endif
701
702}
703
704int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
705{
706 struct ext4_extent_header *eh;
707
708 eh = ext_inode_hdr(inode);
709 eh->eh_depth = 0;
710 eh->eh_entries = 0;
711 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400712 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700713 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700714 return 0;
715}
716
717struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500718ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
719 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700720{
721 struct ext4_extent_header *eh;
722 struct buffer_head *bh;
723 short int depth, i, ppos = 0, alloc = 0;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500724 int ret;
Alex Tomasa86c6182006-10-11 01:21:03 -0700725
726 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400727 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700728
729 /* account possible depth increase */
730 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800731 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700732 GFP_NOFS);
733 if (!path)
734 return ERR_PTR(-ENOMEM);
735 alloc = 1;
736 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700737 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400738 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700739
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400740 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700741 /* walk through the tree */
742 while (i) {
743 ext_debug("depth %d: num %d, max %d\n",
744 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400745
Alex Tomasa86c6182006-10-11 01:21:03 -0700746 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400747 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700748 path[ppos].p_depth = i;
749 path[ppos].p_ext = NULL;
750
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400751 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500752 if (unlikely(!bh)) {
753 ret = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700754 goto err;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500755 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400756 if (!bh_uptodate_or_lock(bh)) {
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400757 trace_ext4_ext_load_extent(inode, block,
758 path[ppos].p_block);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500759 ret = bh_submit_read(bh);
760 if (ret < 0) {
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400761 put_bh(bh);
762 goto err;
763 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400764 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700765 eh = ext_block_hdr(bh);
766 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500767 if (unlikely(ppos > depth)) {
768 put_bh(bh);
769 EXT4_ERROR_INODE(inode,
770 "ppos %d > depth %d", ppos, depth);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500771 ret = -EIO;
Frank Mayhar273df552010-03-02 11:46:09 -0500772 goto err;
773 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700774 path[ppos].p_bh = bh;
775 path[ppos].p_hdr = eh;
776 i--;
777
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500778 ret = ext4_ext_check_block(inode, eh, i, bh);
779 if (ret < 0)
Alex Tomasa86c6182006-10-11 01:21:03 -0700780 goto err;
781 }
782
783 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700784 path[ppos].p_ext = NULL;
785 path[ppos].p_idx = NULL;
786
Alex Tomasa86c6182006-10-11 01:21:03 -0700787 /* find extent */
788 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400789 /* if not an empty leaf */
790 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400791 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700792
793 ext4_ext_show_path(inode, path);
794
795 return path;
796
797err:
798 ext4_ext_drop_refs(path);
799 if (alloc)
800 kfree(path);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500801 return ERR_PTR(ret);
Alex Tomasa86c6182006-10-11 01:21:03 -0700802}
803
804/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700805 * ext4_ext_insert_index:
806 * insert new index [@logical;@ptr] into the block at @curp;
807 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700808 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400809static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
810 struct ext4_ext_path *curp,
811 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700812{
813 struct ext4_extent_idx *ix;
814 int len, err;
815
Avantika Mathur7e028972006-12-06 20:41:33 -0800816 err = ext4_ext_get_access(handle, inode, curp);
817 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700818 return err;
819
Frank Mayhar273df552010-03-02 11:46:09 -0500820 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
821 EXT4_ERROR_INODE(inode,
822 "logical %d == ei_block %d!",
823 logical, le32_to_cpu(curp->p_idx->ei_block));
824 return -EIO;
825 }
Robin Dongd4620312011-07-17 23:43:42 -0400826
827 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
828 >= le16_to_cpu(curp->p_hdr->eh_max))) {
829 EXT4_ERROR_INODE(inode,
830 "eh_entries %d >= eh_max %d!",
831 le16_to_cpu(curp->p_hdr->eh_entries),
832 le16_to_cpu(curp->p_hdr->eh_max));
833 return -EIO;
834 }
835
Alex Tomasa86c6182006-10-11 01:21:03 -0700836 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
837 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400838 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700839 ix = curp->p_idx + 1;
840 } else {
841 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400842 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700843 ix = curp->p_idx;
844 }
845
Eric Gouriou80e675f2011-10-27 11:52:18 -0400846 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
847 BUG_ON(len < 0);
848 if (len > 0) {
849 ext_debug("insert new index %d: "
850 "move %d indices from 0x%p to 0x%p\n",
851 logical, len, ix, ix + 1);
852 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
853 }
854
Tao Maf472e022011-10-17 10:13:46 -0400855 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
856 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
857 return -EIO;
858 }
859
Alex Tomasa86c6182006-10-11 01:21:03 -0700860 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700861 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400862 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700863
Frank Mayhar273df552010-03-02 11:46:09 -0500864 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
865 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
866 return -EIO;
867 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700868
869 err = ext4_ext_dirty(handle, inode, curp);
870 ext4_std_error(inode->i_sb, err);
871
872 return err;
873}
874
875/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700876 * ext4_ext_split:
877 * inserts new subtree into the path, using free index entry
878 * at depth @at:
879 * - allocates all needed blocks (new leaf and all intermediate index blocks)
880 * - makes decision where to split
881 * - moves remaining extents and index entries (right to the split point)
882 * into the newly allocated blocks
883 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700884 */
885static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400886 unsigned int flags,
887 struct ext4_ext_path *path,
888 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -0700889{
890 struct buffer_head *bh = NULL;
891 int depth = ext_depth(inode);
892 struct ext4_extent_header *neh;
893 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -0700894 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700895 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700896 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700897 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700898 int err = 0;
899
900 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700901 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700902
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700903 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700904 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -0500905 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
906 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
907 return -EIO;
908 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700909 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
910 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700911 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700912 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400913 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700914 } else {
915 border = newext->ee_block;
916 ext_debug("leaf will be added."
917 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400918 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700919 }
920
921 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700922 * If error occurs, then we break processing
923 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700924 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700925 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700926 */
927
928 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700929 * Get array to track all allocated blocks.
930 * We need this to handle errors and free blocks
931 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700932 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800933 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700934 if (!ablocks)
935 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700936
937 /* allocate all needed blocks */
938 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
939 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400940 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400941 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -0700942 if (newblock == 0)
943 goto cleanup;
944 ablocks[a] = newblock;
945 }
946
947 /* initialize new leaf */
948 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -0500949 if (unlikely(newblock == 0)) {
950 EXT4_ERROR_INODE(inode, "newblock == 0!");
951 err = -EIO;
952 goto cleanup;
953 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700954 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -0500955 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500956 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700957 goto cleanup;
958 }
959 lock_buffer(bh);
960
Avantika Mathur7e028972006-12-06 20:41:33 -0800961 err = ext4_journal_get_create_access(handle, bh);
962 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700963 goto cleanup;
964
965 neh = ext_block_hdr(bh);
966 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400967 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700968 neh->eh_magic = EXT4_EXT_MAGIC;
969 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700970
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700971 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -0500972 if (unlikely(path[depth].p_hdr->eh_entries !=
973 path[depth].p_hdr->eh_max)) {
974 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
975 path[depth].p_hdr->eh_entries,
976 path[depth].p_hdr->eh_max);
977 err = -EIO;
978 goto cleanup;
979 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700980 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400981 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
982 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -0700983 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400984 struct ext4_extent *ex;
985 ex = EXT_FIRST_EXTENT(neh);
986 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400987 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700988 }
989
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400990 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700991 set_buffer_uptodate(bh);
992 unlock_buffer(bh);
993
Frank Mayhar03901312009-01-07 00:06:22 -0500994 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800995 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700996 goto cleanup;
997 brelse(bh);
998 bh = NULL;
999
1000 /* correct old leaf */
1001 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001002 err = ext4_ext_get_access(handle, inode, path + depth);
1003 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001004 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001005 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001006 err = ext4_ext_dirty(handle, inode, path + depth);
1007 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001008 goto cleanup;
1009
1010 }
1011
1012 /* create intermediate indexes */
1013 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001014 if (unlikely(k < 0)) {
1015 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1016 err = -EIO;
1017 goto cleanup;
1018 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001019 if (k)
1020 ext_debug("create %d intermediate indices\n", k);
1021 /* insert new index into current index block */
1022 /* current depth stored in i var */
1023 i = depth - 1;
1024 while (k--) {
1025 oldblock = newblock;
1026 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001027 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001028 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001029 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001030 goto cleanup;
1031 }
1032 lock_buffer(bh);
1033
Avantika Mathur7e028972006-12-06 20:41:33 -08001034 err = ext4_journal_get_create_access(handle, bh);
1035 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001036 goto cleanup;
1037
1038 neh = ext_block_hdr(bh);
1039 neh->eh_entries = cpu_to_le16(1);
1040 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001041 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001042 neh->eh_depth = cpu_to_le16(depth - i);
1043 fidx = EXT_FIRST_INDEX(neh);
1044 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001045 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001046
Eric Sandeenbba90742008-01-28 23:58:27 -05001047 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1048 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001049
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001050 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001051 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1052 EXT_LAST_INDEX(path[i].p_hdr))) {
1053 EXT4_ERROR_INODE(inode,
1054 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1055 le32_to_cpu(path[i].p_ext->ee_block));
1056 err = -EIO;
1057 goto cleanup;
1058 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001059 /* start copy indexes */
1060 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1061 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1062 EXT_MAX_INDEX(path[i].p_hdr));
1063 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001064 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001065 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001066 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001067 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001068 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001069 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001070 set_buffer_uptodate(bh);
1071 unlock_buffer(bh);
1072
Frank Mayhar03901312009-01-07 00:06:22 -05001073 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001074 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001075 goto cleanup;
1076 brelse(bh);
1077 bh = NULL;
1078
1079 /* correct old index */
1080 if (m) {
1081 err = ext4_ext_get_access(handle, inode, path + i);
1082 if (err)
1083 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001084 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001085 err = ext4_ext_dirty(handle, inode, path + i);
1086 if (err)
1087 goto cleanup;
1088 }
1089
1090 i--;
1091 }
1092
1093 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001094 err = ext4_ext_insert_index(handle, inode, path + at,
1095 le32_to_cpu(border), newblock);
1096
1097cleanup:
1098 if (bh) {
1099 if (buffer_locked(bh))
1100 unlock_buffer(bh);
1101 brelse(bh);
1102 }
1103
1104 if (err) {
1105 /* free all allocated blocks in error case */
1106 for (i = 0; i < depth; i++) {
1107 if (!ablocks[i])
1108 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001109 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001110 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001111 }
1112 }
1113 kfree(ablocks);
1114
1115 return err;
1116}
1117
1118/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001119 * ext4_ext_grow_indepth:
1120 * implements tree growing procedure:
1121 * - allocates new block
1122 * - moves top-level data (index block or leaf) into the new block
1123 * - initializes new top-level, creating index that points to the
1124 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001125 */
1126static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001127 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001128 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001129{
Alex Tomasa86c6182006-10-11 01:21:03 -07001130 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001131 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001132 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001133 int err = 0;
1134
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001135 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001136 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001137 if (newblock == 0)
1138 return err;
1139
1140 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001141 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001142 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001143 lock_buffer(bh);
1144
Avantika Mathur7e028972006-12-06 20:41:33 -08001145 err = ext4_journal_get_create_access(handle, bh);
1146 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001147 unlock_buffer(bh);
1148 goto out;
1149 }
1150
1151 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001152 memmove(bh->b_data, EXT4_I(inode)->i_data,
1153 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001154
1155 /* set size of new block */
1156 neh = ext_block_hdr(bh);
1157 /* old root could have indexes or leaves
1158 * so calculate e_max right way */
1159 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001160 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001161 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001162 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001163 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001164 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001165 set_buffer_uptodate(bh);
1166 unlock_buffer(bh);
1167
Frank Mayhar03901312009-01-07 00:06:22 -05001168 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001169 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001170 goto out;
1171
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001172 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001173 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001174 neh->eh_entries = cpu_to_le16(1);
1175 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1176 if (neh->eh_depth == 0) {
1177 /* Root extent block becomes index block */
1178 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1179 EXT_FIRST_INDEX(neh)->ei_block =
1180 EXT_FIRST_EXTENT(neh)->ee_block;
1181 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001182 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001183 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001184 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001185 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001186
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001187 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001188 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001189out:
1190 brelse(bh);
1191
1192 return err;
1193}
1194
1195/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001196 * ext4_ext_create_new_leaf:
1197 * finds empty index and adds new leaf.
1198 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001199 */
1200static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001201 unsigned int flags,
1202 struct ext4_ext_path *path,
1203 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001204{
1205 struct ext4_ext_path *curp;
1206 int depth, i, err = 0;
1207
1208repeat:
1209 i = depth = ext_depth(inode);
1210
1211 /* walk up to the tree and look for free index entry */
1212 curp = path + depth;
1213 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1214 i--;
1215 curp--;
1216 }
1217
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001218 /* we use already allocated block for index block,
1219 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001220 if (EXT_HAS_FREE_INDEX(curp)) {
1221 /* if we found index with free entry, then use that
1222 * entry: create all needed subtree and add new leaf */
Allison Henderson55f020d2011-05-25 07:41:26 -04001223 err = ext4_ext_split(handle, inode, flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001224 if (err)
1225 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001226
1227 /* refill path */
1228 ext4_ext_drop_refs(path);
1229 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001230 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1231 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001232 if (IS_ERR(path))
1233 err = PTR_ERR(path);
1234 } else {
1235 /* tree is full, time to grow in depth */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001236 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001237 if (err)
1238 goto out;
1239
1240 /* refill path */
1241 ext4_ext_drop_refs(path);
1242 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001243 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1244 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001245 if (IS_ERR(path)) {
1246 err = PTR_ERR(path);
1247 goto out;
1248 }
1249
1250 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001251 * only first (depth 0 -> 1) produces free space;
1252 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001253 */
1254 depth = ext_depth(inode);
1255 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001256 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001257 goto repeat;
1258 }
1259 }
1260
1261out:
1262 return err;
1263}
1264
1265/*
Alex Tomas1988b512008-01-28 23:58:27 -05001266 * search the closest allocated block to the left for *logical
1267 * and returns it at @logical + it's physical address at @phys
1268 * if *logical is the smallest allocated block, the function
1269 * returns 0 at @phys
1270 * return value contains 0 (success) or error code
1271 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001272static int ext4_ext_search_left(struct inode *inode,
1273 struct ext4_ext_path *path,
1274 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001275{
1276 struct ext4_extent_idx *ix;
1277 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001278 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001279
Frank Mayhar273df552010-03-02 11:46:09 -05001280 if (unlikely(path == NULL)) {
1281 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1282 return -EIO;
1283 }
Alex Tomas1988b512008-01-28 23:58:27 -05001284 depth = path->p_depth;
1285 *phys = 0;
1286
1287 if (depth == 0 && path->p_ext == NULL)
1288 return 0;
1289
1290 /* usually extent in the path covers blocks smaller
1291 * then *logical, but it can be that extent is the
1292 * first one in the file */
1293
1294 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001295 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001296 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001297 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1298 EXT4_ERROR_INODE(inode,
1299 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1300 *logical, le32_to_cpu(ex->ee_block));
1301 return -EIO;
1302 }
Alex Tomas1988b512008-01-28 23:58:27 -05001303 while (--depth >= 0) {
1304 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001305 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1306 EXT4_ERROR_INODE(inode,
1307 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001308 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001309 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001310 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001311 depth);
1312 return -EIO;
1313 }
Alex Tomas1988b512008-01-28 23:58:27 -05001314 }
1315 return 0;
1316 }
1317
Frank Mayhar273df552010-03-02 11:46:09 -05001318 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1319 EXT4_ERROR_INODE(inode,
1320 "logical %d < ee_block %d + ee_len %d!",
1321 *logical, le32_to_cpu(ex->ee_block), ee_len);
1322 return -EIO;
1323 }
Alex Tomas1988b512008-01-28 23:58:27 -05001324
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001325 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001326 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001327 return 0;
1328}
1329
1330/*
1331 * search the closest allocated block to the right for *logical
1332 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001333 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001334 * returns 0 at @phys
1335 * return value contains 0 (success) or error code
1336 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001337static int ext4_ext_search_right(struct inode *inode,
1338 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001339 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1340 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001341{
1342 struct buffer_head *bh = NULL;
1343 struct ext4_extent_header *eh;
1344 struct ext4_extent_idx *ix;
1345 struct ext4_extent *ex;
1346 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001347 int depth; /* Note, NOT eh_depth; depth from top of tree */
1348 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001349
Frank Mayhar273df552010-03-02 11:46:09 -05001350 if (unlikely(path == NULL)) {
1351 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1352 return -EIO;
1353 }
Alex Tomas1988b512008-01-28 23:58:27 -05001354 depth = path->p_depth;
1355 *phys = 0;
1356
1357 if (depth == 0 && path->p_ext == NULL)
1358 return 0;
1359
1360 /* usually extent in the path covers blocks smaller
1361 * then *logical, but it can be that extent is the
1362 * first one in the file */
1363
1364 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001365 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001366 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001367 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1368 EXT4_ERROR_INODE(inode,
1369 "first_extent(path[%d].p_hdr) != ex",
1370 depth);
1371 return -EIO;
1372 }
Alex Tomas1988b512008-01-28 23:58:27 -05001373 while (--depth >= 0) {
1374 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001375 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1376 EXT4_ERROR_INODE(inode,
1377 "ix != EXT_FIRST_INDEX *logical %d!",
1378 *logical);
1379 return -EIO;
1380 }
Alex Tomas1988b512008-01-28 23:58:27 -05001381 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001382 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001383 }
1384
Frank Mayhar273df552010-03-02 11:46:09 -05001385 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1386 EXT4_ERROR_INODE(inode,
1387 "logical %d < ee_block %d + ee_len %d!",
1388 *logical, le32_to_cpu(ex->ee_block), ee_len);
1389 return -EIO;
1390 }
Alex Tomas1988b512008-01-28 23:58:27 -05001391
1392 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1393 /* next allocated block in this leaf */
1394 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001395 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001396 }
1397
1398 /* go up and search for index to the right */
1399 while (--depth >= 0) {
1400 ix = path[depth].p_idx;
1401 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001402 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001403 }
1404
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001405 /* we've gone up to the root and found no index to the right */
1406 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001407
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001408got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001409 /* we've found index to the right, let's
1410 * follow it and find the closest allocated
1411 * block to the right */
1412 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001413 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001414 while (++depth < path->p_depth) {
1415 bh = sb_bread(inode->i_sb, block);
1416 if (bh == NULL)
1417 return -EIO;
1418 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001419 /* subtract from p_depth to get proper eh_depth */
Darrick J. Wongf8489122012-04-29 18:21:10 -04001420 if (ext4_ext_check_block(inode, eh,
1421 path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001422 put_bh(bh);
1423 return -EIO;
1424 }
1425 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001426 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001427 put_bh(bh);
1428 }
1429
1430 bh = sb_bread(inode->i_sb, block);
1431 if (bh == NULL)
1432 return -EIO;
1433 eh = ext_block_hdr(bh);
Darrick J. Wongf8489122012-04-29 18:21:10 -04001434 if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001435 put_bh(bh);
1436 return -EIO;
1437 }
1438 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001439found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001440 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001441 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001442 *ret_ex = ex;
1443 if (bh)
1444 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001445 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001446}
1447
1448/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001449 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001450 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001451 * NOTE: it considers block number from index entry as
1452 * allocated block. Thus, index entries have to be consistent
1453 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001454 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001455static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001456ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1457{
1458 int depth;
1459
1460 BUG_ON(path == NULL);
1461 depth = path->p_depth;
1462
1463 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001464 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001465
1466 while (depth >= 0) {
1467 if (depth == path->p_depth) {
1468 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001469 if (path[depth].p_ext &&
1470 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001471 EXT_LAST_EXTENT(path[depth].p_hdr))
1472 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1473 } else {
1474 /* index */
1475 if (path[depth].p_idx !=
1476 EXT_LAST_INDEX(path[depth].p_hdr))
1477 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1478 }
1479 depth--;
1480 }
1481
Lukas Czernerf17722f2011-06-06 00:05:17 -04001482 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001483}
1484
1485/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001486 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001487 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001488 */
Robin Dong57187892011-07-23 21:49:07 -04001489static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001490{
1491 int depth;
1492
1493 BUG_ON(path == NULL);
1494 depth = path->p_depth;
1495
1496 /* zero-tree has no leaf blocks at all */
1497 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001498 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001499
1500 /* go to index block */
1501 depth--;
1502
1503 while (depth >= 0) {
1504 if (path[depth].p_idx !=
1505 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001506 return (ext4_lblk_t)
1507 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001508 depth--;
1509 }
1510
Lukas Czernerf17722f2011-06-06 00:05:17 -04001511 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001512}
1513
1514/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001515 * ext4_ext_correct_indexes:
1516 * if leaf gets modified and modified extent is first in the leaf,
1517 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001518 * TODO: do we need to correct tree in all cases?
1519 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001520static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001521 struct ext4_ext_path *path)
1522{
1523 struct ext4_extent_header *eh;
1524 int depth = ext_depth(inode);
1525 struct ext4_extent *ex;
1526 __le32 border;
1527 int k, err = 0;
1528
1529 eh = path[depth].p_hdr;
1530 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001531
1532 if (unlikely(ex == NULL || eh == NULL)) {
1533 EXT4_ERROR_INODE(inode,
1534 "ex %p == NULL or eh %p == NULL", ex, eh);
1535 return -EIO;
1536 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001537
1538 if (depth == 0) {
1539 /* there is no tree at all */
1540 return 0;
1541 }
1542
1543 if (ex != EXT_FIRST_EXTENT(eh)) {
1544 /* we correct tree if first leaf got modified only */
1545 return 0;
1546 }
1547
1548 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001549 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001550 */
1551 k = depth - 1;
1552 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001553 err = ext4_ext_get_access(handle, inode, path + k);
1554 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001555 return err;
1556 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001557 err = ext4_ext_dirty(handle, inode, path + k);
1558 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001559 return err;
1560
1561 while (k--) {
1562 /* change all left-side indexes */
1563 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1564 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001565 err = ext4_ext_get_access(handle, inode, path + k);
1566 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001567 break;
1568 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001569 err = ext4_ext_dirty(handle, inode, path + k);
1570 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001571 break;
1572 }
1573
1574 return err;
1575}
1576
Akira Fujita748de672009-06-17 19:24:03 -04001577int
Alex Tomasa86c6182006-10-11 01:21:03 -07001578ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1579 struct ext4_extent *ex2)
1580{
Amit Arora749269f2007-07-18 09:02:56 -04001581 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001582
1583 /*
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001584 * Make sure that both extents are initialized. We don't merge
1585 * uninitialized extents so that we can be sure that end_io code has
1586 * the extent that was written properly split out and conversion to
1587 * initialized is trivial.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001588 */
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001589 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001590 return 0;
1591
Amit Arora749269f2007-07-18 09:02:56 -04001592 if (ext4_ext_is_uninitialized(ex1))
1593 max_len = EXT_UNINIT_MAX_LEN;
1594 else
1595 max_len = EXT_INIT_MAX_LEN;
1596
Amit Aroraa2df2a62007-07-17 21:42:41 -04001597 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1598 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1599
1600 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001601 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001602 return 0;
1603
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001604 /*
1605 * To allow future support for preallocated extents to be added
1606 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001607 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001608 */
Amit Arora749269f2007-07-18 09:02:56 -04001609 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001610 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001611#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001612 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001613 return 0;
1614#endif
1615
Theodore Ts'obf89d162010-10-27 21:30:14 -04001616 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001617 return 1;
1618 return 0;
1619}
1620
1621/*
Amit Arora56055d32007-07-17 21:42:38 -04001622 * This function tries to merge the "ex" extent to the next extent in the tree.
1623 * It always tries to merge towards right. If you want to merge towards
1624 * left, pass "ex - 1" as argument instead of "ex".
1625 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1626 * 1 if they got merged.
1627 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001628static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001629 struct ext4_ext_path *path,
1630 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001631{
1632 struct ext4_extent_header *eh;
1633 unsigned int depth, len;
1634 int merge_done = 0;
1635 int uninitialized = 0;
1636
1637 depth = ext_depth(inode);
1638 BUG_ON(path[depth].p_hdr == NULL);
1639 eh = path[depth].p_hdr;
1640
1641 while (ex < EXT_LAST_EXTENT(eh)) {
1642 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1643 break;
1644 /* merge with next extent! */
1645 if (ext4_ext_is_uninitialized(ex))
1646 uninitialized = 1;
1647 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1648 + ext4_ext_get_actual_len(ex + 1));
1649 if (uninitialized)
1650 ext4_ext_mark_uninitialized(ex);
1651
1652 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1653 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1654 * sizeof(struct ext4_extent);
1655 memmove(ex + 1, ex + 2, len);
1656 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001657 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001658 merge_done = 1;
1659 WARN_ON(eh->eh_entries == 0);
1660 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001661 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001662 }
1663
1664 return merge_done;
1665}
1666
1667/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001668 * This function does a very simple check to see if we can collapse
1669 * an extent tree with a single extent tree leaf block into the inode.
1670 */
1671static void ext4_ext_try_to_merge_up(handle_t *handle,
1672 struct inode *inode,
1673 struct ext4_ext_path *path)
1674{
1675 size_t s;
1676 unsigned max_root = ext4_ext_space_root(inode, 0);
1677 ext4_fsblk_t blk;
1678
1679 if ((path[0].p_depth != 1) ||
1680 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1681 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1682 return;
1683
1684 /*
1685 * We need to modify the block allocation bitmap and the block
1686 * group descriptor to release the extent tree block. If we
1687 * can't get the journal credits, give up.
1688 */
1689 if (ext4_journal_extend(handle, 2))
1690 return;
1691
1692 /*
1693 * Copy the extent data up to the inode
1694 */
1695 blk = ext4_idx_pblock(path[0].p_idx);
1696 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1697 sizeof(struct ext4_extent_idx);
1698 s += sizeof(struct ext4_extent_header);
1699
1700 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1701 path[0].p_depth = 0;
1702 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1703 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1704 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1705
1706 brelse(path[1].p_bh);
1707 ext4_free_blocks(handle, inode, NULL, blk, 1,
1708 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1709}
1710
1711/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001712 * This function tries to merge the @ex extent to neighbours in the tree.
1713 * return 1 if merge left else 0.
1714 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001715static void ext4_ext_try_to_merge(handle_t *handle,
1716 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001717 struct ext4_ext_path *path,
1718 struct ext4_extent *ex) {
1719 struct ext4_extent_header *eh;
1720 unsigned int depth;
1721 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001722
1723 depth = ext_depth(inode);
1724 BUG_ON(path[depth].p_hdr == NULL);
1725 eh = path[depth].p_hdr;
1726
1727 if (ex > EXT_FIRST_EXTENT(eh))
1728 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1729
1730 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001731 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001732
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001733 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001734}
1735
1736/*
Amit Arora25d14f92007-05-24 13:04:13 -04001737 * check if a portion of the "newext" extent overlaps with an
1738 * existing extent.
1739 *
1740 * If there is an overlap discovered, it updates the length of the newext
1741 * such that there will be no overlap, and then returns 1.
1742 * If there is no overlap found, it returns 0.
1743 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001744static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1745 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001746 struct ext4_extent *newext,
1747 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001748{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001749 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001750 unsigned int depth, len1;
1751 unsigned int ret = 0;
1752
1753 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001754 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001755 depth = ext_depth(inode);
1756 if (!path[depth].p_ext)
1757 goto out;
1758 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001759 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001760
1761 /*
1762 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001763 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001764 */
1765 if (b2 < b1) {
1766 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001767 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001768 goto out;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001769 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001770 }
1771
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001772 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001773 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001774 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001775 newext->ee_len = cpu_to_le16(len1);
1776 ret = 1;
1777 }
1778
1779 /* check for overlap */
1780 if (b1 + len1 > b2) {
1781 newext->ee_len = cpu_to_le16(b2 - b1);
1782 ret = 1;
1783 }
1784out:
1785 return ret;
1786}
1787
1788/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001789 * ext4_ext_insert_extent:
1790 * tries to merge requsted extent into the existing extent or
1791 * inserts requested extent as new one into the tree,
1792 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001793 */
1794int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1795 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04001796 struct ext4_extent *newext, int flag)
Alex Tomasa86c6182006-10-11 01:21:03 -07001797{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001798 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001799 struct ext4_extent *ex, *fex;
1800 struct ext4_extent *nearex; /* nearest extent */
1801 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001802 int depth, len, err;
1803 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001804 unsigned uninitialized = 0;
Allison Henderson55f020d2011-05-25 07:41:26 -04001805 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001806
Frank Mayhar273df552010-03-02 11:46:09 -05001807 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1808 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1809 return -EIO;
1810 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001811 depth = ext_depth(inode);
1812 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001813 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001814 if (unlikely(path[depth].p_hdr == NULL)) {
1815 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1816 return -EIO;
1817 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001818
1819 /* try to insert block into found extent and return */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001820 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001821
1822 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001823 * Try to see whether we should rather test the extent on
1824 * right from ex, or from the left of ex. This is because
1825 * ext4_ext_find_extent() can return either extent on the
1826 * left, or on the right from the searched position. This
1827 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001828 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001829 if (ex < EXT_LAST_EXTENT(eh) &&
1830 (le32_to_cpu(ex->ee_block) +
1831 ext4_ext_get_actual_len(ex) <
1832 le32_to_cpu(newext->ee_block))) {
1833 ex += 1;
1834 goto prepend;
1835 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1836 (le32_to_cpu(newext->ee_block) +
1837 ext4_ext_get_actual_len(newext) <
1838 le32_to_cpu(ex->ee_block)))
1839 ex -= 1;
1840
1841 /* Try to append newex to the ex */
1842 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1843 ext_debug("append [%d]%d block to %u:[%d]%d"
1844 "(from %llu)\n",
1845 ext4_ext_is_uninitialized(newext),
1846 ext4_ext_get_actual_len(newext),
1847 le32_to_cpu(ex->ee_block),
1848 ext4_ext_is_uninitialized(ex),
1849 ext4_ext_get_actual_len(ex),
1850 ext4_ext_pblock(ex));
1851 err = ext4_ext_get_access(handle, inode,
1852 path + depth);
1853 if (err)
1854 return err;
1855
1856 /*
1857 * ext4_can_extents_be_merged should have checked
1858 * that either both extents are uninitialized, or
1859 * both aren't. Thus we need to check only one of
1860 * them here.
1861 */
1862 if (ext4_ext_is_uninitialized(ex))
1863 uninitialized = 1;
1864 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001865 + ext4_ext_get_actual_len(newext));
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001866 if (uninitialized)
1867 ext4_ext_mark_uninitialized(ex);
1868 eh = path[depth].p_hdr;
1869 nearex = ex;
1870 goto merge;
1871 }
1872
1873prepend:
1874 /* Try to prepend newex to the ex */
1875 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1876 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1877 "(from %llu)\n",
1878 le32_to_cpu(newext->ee_block),
1879 ext4_ext_is_uninitialized(newext),
1880 ext4_ext_get_actual_len(newext),
1881 le32_to_cpu(ex->ee_block),
1882 ext4_ext_is_uninitialized(ex),
1883 ext4_ext_get_actual_len(ex),
1884 ext4_ext_pblock(ex));
1885 err = ext4_ext_get_access(handle, inode,
1886 path + depth);
1887 if (err)
1888 return err;
1889
1890 /*
1891 * ext4_can_extents_be_merged should have checked
1892 * that either both extents are uninitialized, or
1893 * both aren't. Thus we need to check only one of
1894 * them here.
1895 */
1896 if (ext4_ext_is_uninitialized(ex))
1897 uninitialized = 1;
1898 ex->ee_block = newext->ee_block;
1899 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1900 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1901 + ext4_ext_get_actual_len(newext));
1902 if (uninitialized)
1903 ext4_ext_mark_uninitialized(ex);
1904 eh = path[depth].p_hdr;
1905 nearex = ex;
1906 goto merge;
1907 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001908 }
1909
Alex Tomasa86c6182006-10-11 01:21:03 -07001910 depth = ext_depth(inode);
1911 eh = path[depth].p_hdr;
1912 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1913 goto has_space;
1914
1915 /* probably next leaf has space for us? */
1916 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001917 next = EXT_MAX_BLOCKS;
1918 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001919 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001920 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001921 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07001922 BUG_ON(npath != NULL);
1923 npath = ext4_ext_find_extent(inode, next, NULL);
1924 if (IS_ERR(npath))
1925 return PTR_ERR(npath);
1926 BUG_ON(npath->p_depth != path->p_depth);
1927 eh = npath[depth].p_hdr;
1928 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001929 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001930 le16_to_cpu(eh->eh_entries));
1931 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04001932 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07001933 }
1934 ext_debug("next leaf has no free space(%d,%d)\n",
1935 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1936 }
1937
1938 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001939 * There is no free space in the found leaf.
1940 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001941 */
Lukas Czerner27dd4382013-04-09 22:11:22 -04001942 if (flag & EXT4_GET_BLOCKS_METADATA_NOFAIL)
1943 flags = EXT4_MB_USE_RESERVED;
Allison Henderson55f020d2011-05-25 07:41:26 -04001944 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001945 if (err)
1946 goto cleanup;
1947 depth = ext_depth(inode);
1948 eh = path[depth].p_hdr;
1949
1950has_space:
1951 nearex = path[depth].p_ext;
1952
Avantika Mathur7e028972006-12-06 20:41:33 -08001953 err = ext4_ext_get_access(handle, inode, path + depth);
1954 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001955 goto cleanup;
1956
1957 if (!nearex) {
1958 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001959 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001960 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001961 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001962 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001963 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04001964 nearex = EXT_FIRST_EXTENT(eh);
1965 } else {
1966 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001967 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04001968 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001969 ext_debug("insert %u:%llu:[%d]%d before: "
1970 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001971 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001972 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001973 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001974 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04001975 nearex);
1976 nearex++;
1977 } else {
1978 /* Insert before */
1979 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04001980 ext_debug("insert %u:%llu:[%d]%d after: "
1981 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04001982 le32_to_cpu(newext->ee_block),
1983 ext4_ext_pblock(newext),
1984 ext4_ext_is_uninitialized(newext),
1985 ext4_ext_get_actual_len(newext),
1986 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001987 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04001988 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1989 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001990 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04001991 "move %d extents from 0x%p to 0x%p\n",
1992 le32_to_cpu(newext->ee_block),
1993 ext4_ext_pblock(newext),
1994 ext4_ext_is_uninitialized(newext),
1995 ext4_ext_get_actual_len(newext),
1996 len, nearex, nearex + 1);
1997 memmove(nearex + 1, nearex,
1998 len * sizeof(struct ext4_extent));
1999 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002000 }
2001
Marcin Slusarze8546d02008-04-17 10:38:59 -04002002 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002003 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002004 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002005 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002006 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002007
2008merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002009 /* try to merge extents */
Jiaying Zhang744692d2010-03-04 16:14:02 -05002010 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002011 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002012
Alex Tomasa86c6182006-10-11 01:21:03 -07002013
2014 /* time to correct all indexes above */
2015 err = ext4_ext_correct_indexes(handle, inode, path);
2016 if (err)
2017 goto cleanup;
2018
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002019 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002020
2021cleanup:
2022 if (npath) {
2023 ext4_ext_drop_refs(npath);
2024 kfree(npath);
2025 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002026 return err;
2027}
2028
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002029static int ext4_fill_fiemap_extents(struct inode *inode,
2030 ext4_lblk_t block, ext4_lblk_t num,
2031 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002032{
2033 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002034 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002035 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002036 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002037 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002038 int exists, depth = 0, err = 0;
2039 unsigned int flags = 0;
2040 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002041
Lukas Czernerf17722f2011-06-06 00:05:17 -04002042 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002043 num = last - block;
2044 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002045 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002046
2047 if (path && ext_depth(inode) != depth) {
2048 /* depth was changed. we have to realloc path */
2049 kfree(path);
2050 path = NULL;
2051 }
2052
Eric Sandeen6873fa02008-10-07 00:46:36 -04002053 path = ext4_ext_find_extent(inode, block, path);
2054 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002055 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002056 err = PTR_ERR(path);
2057 path = NULL;
2058 break;
2059 }
2060
2061 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002062 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002063 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002064 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2065 err = -EIO;
2066 break;
2067 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002068 ex = path[depth].p_ext;
2069 next = ext4_ext_next_allocated_block(path);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002070 ext4_ext_drop_refs(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002071
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002072 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002073 exists = 0;
2074 if (!ex) {
2075 /* there is no extent yet, so try to allocate
2076 * all requested space */
2077 start = block;
2078 end = block + num;
2079 } else if (le32_to_cpu(ex->ee_block) > block) {
2080 /* need to allocate space before found extent */
2081 start = block;
2082 end = le32_to_cpu(ex->ee_block);
2083 if (block + num < end)
2084 end = block + num;
2085 } else if (block >= le32_to_cpu(ex->ee_block)
2086 + ext4_ext_get_actual_len(ex)) {
2087 /* need to allocate space after found extent */
2088 start = block;
2089 end = block + num;
2090 if (end >= next)
2091 end = next;
2092 } else if (block >= le32_to_cpu(ex->ee_block)) {
2093 /*
2094 * some part of requested space is covered
2095 * by found extent
2096 */
2097 start = block;
2098 end = le32_to_cpu(ex->ee_block)
2099 + ext4_ext_get_actual_len(ex);
2100 if (block + num < end)
2101 end = block + num;
2102 exists = 1;
2103 } else {
2104 BUG();
2105 }
2106 BUG_ON(end <= start);
2107
2108 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002109 es.es_lblk = start;
2110 es.es_len = end - start;
2111 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002112 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002113 es.es_lblk = le32_to_cpu(ex->ee_block);
2114 es.es_len = ext4_ext_get_actual_len(ex);
2115 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002116 if (ext4_ext_is_uninitialized(ex))
2117 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002118 }
2119
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002120 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002121 * Find delayed extent and update es accordingly. We call
2122 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002123 * last existing extent or not.
2124 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002125 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002126 if (!exists && next_del) {
2127 exists = 1;
2128 flags |= FIEMAP_EXTENT_DELALLOC;
2129 }
2130 up_read(&EXT4_I(inode)->i_data_sem);
2131
Zheng Liu69eb33d2013-02-18 00:31:07 -05002132 if (unlikely(es.es_len == 0)) {
2133 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002134 err = -EIO;
2135 break;
2136 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002137
Zheng Liuf7fec032013-02-18 00:28:47 -05002138 /*
2139 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2140 * we need to check next == EXT_MAX_BLOCKS because it is
2141 * possible that an extent is with unwritten and delayed
2142 * status due to when an extent is delayed allocated and
2143 * is allocated by fallocate status tree will track both of
2144 * them in a extent.
2145 *
2146 * So we could return a unwritten and delayed extent, and
2147 * its block is equal to 'next'.
2148 */
2149 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002150 flags |= FIEMAP_EXTENT_LAST;
2151 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2152 next != EXT_MAX_BLOCKS)) {
2153 EXT4_ERROR_INODE(inode,
2154 "next extent == %u, next "
2155 "delalloc extent = %u",
2156 next, next_del);
2157 err = -EIO;
2158 break;
2159 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002160 }
2161
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002162 if (exists) {
2163 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002164 (__u64)es.es_lblk << blksize_bits,
2165 (__u64)es.es_pblk << blksize_bits,
2166 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002167 flags);
2168 if (err < 0)
2169 break;
2170 if (err == 1) {
2171 err = 0;
2172 break;
2173 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002174 }
2175
Zheng Liu69eb33d2013-02-18 00:31:07 -05002176 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002177 }
2178
2179 if (path) {
2180 ext4_ext_drop_refs(path);
2181 kfree(path);
2182 }
2183
2184 return err;
2185}
2186
Alex Tomasa86c6182006-10-11 01:21:03 -07002187/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002188 * ext4_ext_put_gap_in_cache:
2189 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002190 * and cache this gap
2191 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002192static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002193ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002194 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002195{
2196 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002197 unsigned long len;
2198 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002199 struct ext4_extent *ex;
2200
2201 ex = path[depth].p_ext;
2202 if (ex == NULL) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002203 /*
2204 * there is no extent yet, so gap is [0;-] and we
2205 * don't cache it
2206 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002207 ext_debug("cache gap(whole file):");
2208 } else if (block < le32_to_cpu(ex->ee_block)) {
2209 lblock = block;
2210 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002211 ext_debug("cache gap(before): %u [%u:%u]",
2212 block,
2213 le32_to_cpu(ex->ee_block),
2214 ext4_ext_get_actual_len(ex));
Zheng Liud100eef2013-02-18 00:29:59 -05002215 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2216 ext4_es_insert_extent(inode, lblock, len, ~0,
2217 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002218 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002219 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002220 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002221 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002222 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002223
2224 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002225 ext_debug("cache gap(after): [%u:%u] %u",
2226 le32_to_cpu(ex->ee_block),
2227 ext4_ext_get_actual_len(ex),
2228 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002229 BUG_ON(next == lblock);
2230 len = next - lblock;
Zheng Liud100eef2013-02-18 00:29:59 -05002231 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2232 ext4_es_insert_extent(inode, lblock, len, ~0,
2233 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002234 } else {
2235 lblock = len = 0;
2236 BUG();
2237 }
2238
Eric Sandeenbba90742008-01-28 23:58:27 -05002239 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002240}
2241
2242/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002243 * ext4_ext_rm_idx:
2244 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002245 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002246static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002247 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002248{
Alex Tomasa86c6182006-10-11 01:21:03 -07002249 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002250 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002251
2252 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002253 depth--;
2254 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002255 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002256 if (unlikely(path->p_hdr->eh_entries == 0)) {
2257 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2258 return -EIO;
2259 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002260 err = ext4_ext_get_access(handle, inode, path);
2261 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002262 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002263
2264 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2265 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2266 len *= sizeof(struct ext4_extent_idx);
2267 memmove(path->p_idx, path->p_idx + 1, len);
2268 }
2269
Marcin Slusarze8546d02008-04-17 10:38:59 -04002270 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002271 err = ext4_ext_dirty(handle, inode, path);
2272 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002273 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002274 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002275 trace_ext4_ext_rm_idx(inode, leaf);
2276
Peter Huewe7dc57612011-02-21 21:01:42 -05002277 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002278 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002279
2280 while (--depth >= 0) {
2281 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2282 break;
2283 path--;
2284 err = ext4_ext_get_access(handle, inode, path);
2285 if (err)
2286 break;
2287 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2288 err = ext4_ext_dirty(handle, inode, path);
2289 if (err)
2290 break;
2291 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002292 return err;
2293}
2294
2295/*
Mingming Caoee12b632008-08-19 22:16:05 -04002296 * ext4_ext_calc_credits_for_single_extent:
2297 * This routine returns max. credits that needed to insert an extent
2298 * to the extent tree.
2299 * When pass the actual path, the caller should calculate credits
2300 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002301 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002302int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002303 struct ext4_ext_path *path)
2304{
Alex Tomasa86c6182006-10-11 01:21:03 -07002305 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002306 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002307 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002308
Alex Tomasa86c6182006-10-11 01:21:03 -07002309 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002310 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002311 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2312
2313 /*
2314 * There are some space in the leaf tree, no
2315 * need to account for leaf block credit
2316 *
2317 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002318 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002319 * accounted.
2320 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002321 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002322 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002323 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002324 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002325 }
2326
Mingming Cao525f4ed2008-08-19 22:15:58 -04002327 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002328}
Alex Tomasa86c6182006-10-11 01:21:03 -07002329
Mingming Caoee12b632008-08-19 22:16:05 -04002330/*
2331 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2332 *
2333 * if nrblocks are fit in a single extent (chunk flag is 1), then
2334 * in the worse case, each tree level index/leaf need to be changed
2335 * if the tree split due to insert a new extent, then the old tree
2336 * index/leaf need to be updated too
2337 *
2338 * If the nrblocks are discontiguous, they could cause
2339 * the whole tree split more than once, but this is really rare.
2340 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002341int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04002342{
2343 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002344 int depth;
2345
2346 /* If we are converting the inline data, only one is needed here. */
2347 if (ext4_has_inline_data(inode))
2348 return 1;
2349
2350 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002351
Mingming Caoee12b632008-08-19 22:16:05 -04002352 if (chunk)
2353 index = depth * 2;
2354 else
2355 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002356
Mingming Caoee12b632008-08-19 22:16:05 -04002357 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002358}
2359
2360static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002361 struct ext4_extent *ex,
2362 ext4_fsblk_t *partial_cluster,
2363 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002364{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002365 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002366 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002367 ext4_fsblk_t pblk;
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002368 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002369
Alex Tomasc9de5602008-01-29 00:19:52 -05002370 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002371 flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2372 else if (ext4_should_journal_data(inode))
2373 flags |= EXT4_FREE_BLOCKS_FORGET;
2374
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002375 /*
2376 * For bigalloc file systems, we never free a partial cluster
2377 * at the beginning of the extent. Instead, we make a note
2378 * that we tried freeing the cluster, and check to see if we
2379 * need to free it on a subsequent call to ext4_remove_blocks,
2380 * or at the end of the ext4_truncate() operation.
2381 */
2382 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2383
Aditya Kalid8990242011-09-09 19:18:51 -04002384 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002385 /*
2386 * If we have a partial cluster, and it's different from the
2387 * cluster of the last block, we need to explicitly free the
2388 * partial cluster here.
2389 */
2390 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2391 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2392 ext4_free_blocks(handle, inode, NULL,
2393 EXT4_C2B(sbi, *partial_cluster),
2394 sbi->s_cluster_ratio, flags);
2395 *partial_cluster = 0;
2396 }
2397
Alex Tomasa86c6182006-10-11 01:21:03 -07002398#ifdef EXTENTS_STATS
2399 {
2400 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002401 spin_lock(&sbi->s_ext_stats_lock);
2402 sbi->s_ext_blocks += ee_len;
2403 sbi->s_ext_extents++;
2404 if (ee_len < sbi->s_ext_min)
2405 sbi->s_ext_min = ee_len;
2406 if (ee_len > sbi->s_ext_max)
2407 sbi->s_ext_max = ee_len;
2408 if (ext_depth(inode) > sbi->s_depth_max)
2409 sbi->s_depth_max = ext_depth(inode);
2410 spin_unlock(&sbi->s_ext_stats_lock);
2411 }
2412#endif
2413 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002414 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002415 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002416 ext4_lblk_t num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002417
Amit Aroraa2df2a62007-07-17 21:42:41 -04002418 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002419 pblk = ext4_ext_pblock(ex) + ee_len - num;
2420 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2421 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2422 /*
2423 * If the block range to be freed didn't start at the
2424 * beginning of a cluster, and we removed the entire
2425 * extent, save the partial cluster here, since we
2426 * might need to delete if we determine that the
2427 * truncate operation has removed all of the blocks in
2428 * the cluster.
2429 */
2430 if (pblk & (sbi->s_cluster_ratio - 1) &&
2431 (ee_len == num))
2432 *partial_cluster = EXT4_B2C(sbi, pblk);
2433 else
2434 *partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002435 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002436 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002437 /* head removal */
2438 ext4_lblk_t num;
2439 ext4_fsblk_t start;
2440
2441 num = to - from;
2442 start = ext4_ext_pblock(ex);
2443
2444 ext_debug("free first %u blocks starting %llu\n", num, start);
H Hartley Sweetenee90d572011-10-18 11:01:51 -04002445 ext4_free_blocks(handle, inode, NULL, start, num, flags);
Allison Hendersond583fb82011-05-25 07:41:43 -04002446
Alex Tomasa86c6182006-10-11 01:21:03 -07002447 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002448 printk(KERN_INFO "strange request: removal(2) "
2449 "%u-%u from %u:%u\n",
2450 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002451 }
2452 return 0;
2453}
2454
Allison Hendersond583fb82011-05-25 07:41:43 -04002455
2456/*
2457 * ext4_ext_rm_leaf() Removes the extents associated with the
2458 * blocks appearing between "start" and "end", and splits the extents
2459 * if "start" and "end" appear in the same extent
2460 *
2461 * @handle: The journal handle
2462 * @inode: The files inode
2463 * @path: The path to the leaf
2464 * @start: The first block to remove
2465 * @end: The last block to remove
2466 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002467static int
2468ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002469 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2470 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002471{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002472 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002473 int err = 0, correct_index = 0;
2474 int depth = ext_depth(inode), credits;
2475 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002476 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002477 unsigned num;
2478 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002479 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002480 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002481 struct ext4_extent *ex;
2482
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002483 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002484 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002485 if (!path[depth].p_hdr)
2486 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2487 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002488 if (unlikely(path[depth].p_hdr == NULL)) {
2489 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2490 return -EIO;
2491 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002492 /* find where to start removing */
2493 ex = EXT_LAST_EXTENT(eh);
2494
2495 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002496 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002497
Aditya Kalid8990242011-09-09 19:18:51 -04002498 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2499
Alex Tomasa86c6182006-10-11 01:21:03 -07002500 while (ex >= EXT_FIRST_EXTENT(eh) &&
2501 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002502
2503 if (ext4_ext_is_uninitialized(ex))
2504 uninitialized = 1;
2505 else
2506 uninitialized = 0;
2507
Mingming553f9002009-09-18 13:34:55 -04002508 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2509 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002510 path[depth].p_ext = ex;
2511
2512 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002513 b = ex_ee_block+ex_ee_len - 1 < end ?
2514 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002515
2516 ext_debug(" border %u:%u\n", a, b);
2517
Allison Hendersond583fb82011-05-25 07:41:43 -04002518 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002519 if (end < ex_ee_block) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002520 ex--;
2521 ex_ee_block = le32_to_cpu(ex->ee_block);
2522 ex_ee_len = ext4_ext_get_actual_len(ex);
2523 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002524 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002525 EXT4_ERROR_INODE(inode,
2526 "can not handle truncate %u:%u "
2527 "on extent %u:%u",
2528 start, end, ex_ee_block,
2529 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002530 err = -EIO;
2531 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002532 } else if (a != ex_ee_block) {
2533 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002534 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002535 } else {
2536 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002537 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002538 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002539 /*
2540 * 3 for leaf, sb, and inode plus 2 (bmap and group
2541 * descriptor) for each block group; assume two block
2542 * groups plus ex_ee_len/blocks_per_block_group for
2543 * the worst case
2544 */
2545 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002546 if (ex == EXT_FIRST_EXTENT(eh)) {
2547 correct_index = 1;
2548 credits += (ext_depth(inode)) + 1;
2549 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002550 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002551
Jan Kara487caee2009-08-17 22:17:20 -04002552 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002553 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002554 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002555
2556 err = ext4_ext_get_access(handle, inode, path + depth);
2557 if (err)
2558 goto out;
2559
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002560 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2561 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002562 if (err)
2563 goto out;
2564
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002565 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002566 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002567 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002568
Alex Tomasa86c6182006-10-11 01:21:03 -07002569 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002570 /*
2571 * Do not mark uninitialized if all the blocks in the
2572 * extent have been removed.
2573 */
2574 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002575 ext4_ext_mark_uninitialized(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002576 /*
2577 * If the extent was completely released,
2578 * we need to remove it from the leaf
2579 */
2580 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002581 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002582 /*
2583 * For hole punching, we need to scoot all the
2584 * extents up when an extent is removed so that
2585 * we dont have blank extents in the middle
2586 */
2587 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2588 sizeof(struct ext4_extent));
2589
2590 /* Now get rid of the one at the end */
2591 memset(EXT_LAST_EXTENT(eh), 0,
2592 sizeof(struct ext4_extent));
2593 }
2594 le16_add_cpu(&eh->eh_entries, -1);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002595 } else
2596 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002597
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002598 err = ext4_ext_dirty(handle, inode, path + depth);
2599 if (err)
2600 goto out;
2601
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002602 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002603 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002604 ex--;
2605 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002606 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002607 }
2608
2609 if (correct_index && eh->eh_entries)
2610 err = ext4_ext_correct_indexes(handle, inode, path);
2611
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002612 /*
2613 * If there is still a entry in the leaf node, check to see if
2614 * it references the partial cluster. This is the only place
2615 * where it could; if it doesn't, we can free the cluster.
2616 */
2617 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2618 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2619 *partial_cluster)) {
2620 int flags = EXT4_FREE_BLOCKS_FORGET;
2621
2622 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2623 flags |= EXT4_FREE_BLOCKS_METADATA;
2624
2625 ext4_free_blocks(handle, inode, NULL,
2626 EXT4_C2B(sbi, *partial_cluster),
2627 sbi->s_cluster_ratio, flags);
2628 *partial_cluster = 0;
2629 }
2630
Alex Tomasa86c6182006-10-11 01:21:03 -07002631 /* if this leaf is free, then we should
2632 * remove it from index block above */
2633 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002634 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002635
2636out:
2637 return err;
2638}
2639
2640/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002641 * ext4_ext_more_to_rm:
2642 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002643 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002644static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002645ext4_ext_more_to_rm(struct ext4_ext_path *path)
2646{
2647 BUG_ON(path->p_idx == NULL);
2648
2649 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2650 return 0;
2651
2652 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002653 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002654 * so we have to consider current index for truncation
2655 */
2656 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2657 return 0;
2658 return 1;
2659}
2660
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002661int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2662 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002663{
2664 struct super_block *sb = inode->i_sb;
2665 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002666 struct ext4_ext_path *path = NULL;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002667 ext4_fsblk_t partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002668 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002669 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002670
Lukas Czerner5f95d212012-03-19 23:03:19 -04002671 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002672
2673 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002674 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002675 if (IS_ERR(handle))
2676 return PTR_ERR(handle);
2677
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002678again:
Aditya Kalid8990242011-09-09 19:18:51 -04002679 trace_ext4_ext_remove_space(inode, start, depth);
2680
Alex Tomasa86c6182006-10-11 01:21:03 -07002681 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002682 * Check if we are removing extents inside the extent tree. If that
2683 * is the case, we are going to punch a hole inside the extent tree
2684 * so we have to check whether we need to split the extent covering
2685 * the last block to remove so we can easily remove the part of it
2686 * in ext4_ext_rm_leaf().
2687 */
2688 if (end < EXT_MAX_BLOCKS - 1) {
2689 struct ext4_extent *ex;
2690 ext4_lblk_t ee_block;
2691
2692 /* find extent for this block */
2693 path = ext4_ext_find_extent(inode, end, NULL);
2694 if (IS_ERR(path)) {
2695 ext4_journal_stop(handle);
2696 return PTR_ERR(path);
2697 }
2698 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002699 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002700 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002701 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002702 if (depth) {
2703 EXT4_ERROR_INODE(inode,
2704 "path[%d].p_hdr == NULL",
2705 depth);
2706 err = -EIO;
2707 }
2708 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002709 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002710
2711 ee_block = le32_to_cpu(ex->ee_block);
2712
2713 /*
2714 * See if the last block is inside the extent, if so split
2715 * the extent at 'end' block so we can easily remove the
2716 * tail of the first part of the split extent in
2717 * ext4_ext_rm_leaf().
2718 */
2719 if (end >= ee_block &&
2720 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2721 int split_flag = 0;
2722
2723 if (ext4_ext_is_uninitialized(ex))
2724 split_flag = EXT4_EXT_MARK_UNINIT1 |
2725 EXT4_EXT_MARK_UNINIT2;
2726
2727 /*
2728 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002729 * block in the first new extent. Also we should not
2730 * fail removing space due to ENOSPC so try to use
2731 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002732 */
2733 err = ext4_split_extent_at(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04002734 end + 1, split_flag,
2735 EXT4_GET_BLOCKS_PRE_IO |
2736 EXT4_GET_BLOCKS_METADATA_NOFAIL);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002737
2738 if (err < 0)
2739 goto out;
2740 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002741 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002742 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002743 * We start scanning from right side, freeing all the blocks
2744 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002745 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002746 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002747 if (path) {
2748 int k = i = depth;
2749 while (--k > 0)
2750 path[k].p_block =
2751 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2752 } else {
2753 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2754 GFP_NOFS);
2755 if (path == NULL) {
2756 ext4_journal_stop(handle);
2757 return -ENOMEM;
2758 }
2759 path[0].p_depth = depth;
2760 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002761 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002762
Ashish Sangwan968dee72012-07-22 22:49:08 -04002763 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2764 err = -EIO;
2765 goto out;
2766 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002767 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002768 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002769
2770 while (i >= 0 && err == 0) {
2771 if (i == depth) {
2772 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002773 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002774 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002775 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002776 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002777 brelse(path[i].p_bh);
2778 path[i].p_bh = NULL;
2779 i--;
2780 continue;
2781 }
2782
2783 /* this is index block */
2784 if (!path[i].p_hdr) {
2785 ext_debug("initialize header\n");
2786 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002787 }
2788
Alex Tomasa86c6182006-10-11 01:21:03 -07002789 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002790 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002791 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2792 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2793 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2794 path[i].p_hdr,
2795 le16_to_cpu(path[i].p_hdr->eh_entries));
2796 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002797 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002798 path[i].p_idx--;
2799 }
2800
2801 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2802 i, EXT_FIRST_INDEX(path[i].p_hdr),
2803 path[i].p_idx);
2804 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002805 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002806 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002807 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002808 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002809 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'obf89d162010-10-27 21:30:14 -04002810 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002811 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002812 /* should we reset i_size? */
2813 err = -EIO;
2814 break;
2815 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002816 if (WARN_ON(i + 1 > depth)) {
2817 err = -EIO;
2818 break;
2819 }
Darrick J. Wongf8489122012-04-29 18:21:10 -04002820 if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2821 depth - i - 1, bh)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002822 err = -EIO;
2823 break;
2824 }
2825 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002826
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002827 /* save actual number of indexes since this
2828 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002829 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2830 i++;
2831 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002832 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002833 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002834 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002835 * handle must be already prepared by the
2836 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002837 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002838 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002839 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002840 brelse(path[i].p_bh);
2841 path[i].p_bh = NULL;
2842 i--;
2843 ext_debug("return to level %d\n", i);
2844 }
2845 }
2846
Aditya Kalid8990242011-09-09 19:18:51 -04002847 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2848 path->p_hdr->eh_entries);
2849
Aditya Kali7b415bf2011-09-09 19:04:51 -04002850 /* If we still have something in the partial cluster and we have removed
2851 * even the first extent, then we should free the blocks in the partial
2852 * cluster as well. */
2853 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2854 int flags = EXT4_FREE_BLOCKS_FORGET;
2855
2856 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2857 flags |= EXT4_FREE_BLOCKS_METADATA;
2858
2859 ext4_free_blocks(handle, inode, NULL,
2860 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2861 EXT4_SB(sb)->s_cluster_ratio, flags);
2862 partial_cluster = 0;
2863 }
2864
Alex Tomasa86c6182006-10-11 01:21:03 -07002865 /* TODO: flexible tree reduction should be here */
2866 if (path->p_hdr->eh_entries == 0) {
2867 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002868 * truncate to zero freed all the tree,
2869 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002870 */
2871 err = ext4_ext_get_access(handle, inode, path);
2872 if (err == 0) {
2873 ext_inode_hdr(inode)->eh_depth = 0;
2874 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002875 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002876 err = ext4_ext_dirty(handle, inode, path);
2877 }
2878 }
2879out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002880 ext4_ext_drop_refs(path);
2881 kfree(path);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002882 if (err == -EAGAIN) {
2883 path = NULL;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002884 goto again;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002885 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002886 ext4_journal_stop(handle);
2887
2888 return err;
2889}
2890
2891/*
2892 * called at mount time
2893 */
2894void ext4_ext_init(struct super_block *sb)
2895{
2896 /*
2897 * possible initialization would be here
2898 */
2899
Theodore Ts'o83982b62009-01-06 14:53:16 -05002900 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002901#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04002902 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002903#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04002904 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07002905#endif
2906#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04002907 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07002908#endif
2909#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04002910 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07002911#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04002912 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04002913#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07002914#ifdef EXTENTS_STATS
2915 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2916 EXT4_SB(sb)->s_ext_min = 1 << 30;
2917 EXT4_SB(sb)->s_ext_max = 0;
2918#endif
2919 }
2920}
2921
2922/*
2923 * called at umount time
2924 */
2925void ext4_ext_release(struct super_block *sb)
2926{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002927 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002928 return;
2929
2930#ifdef EXTENTS_STATS
2931 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2932 struct ext4_sb_info *sbi = EXT4_SB(sb);
2933 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2934 sbi->s_ext_blocks, sbi->s_ext_extents,
2935 sbi->s_ext_blocks / sbi->s_ext_extents);
2936 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2937 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2938 }
2939#endif
2940}
2941
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002942/* FIXME!! we need to try to merge to left or right after zero-out */
2943static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2944{
Lukas Czerner24075182010-10-27 21:30:06 -04002945 ext4_fsblk_t ee_pblock;
2946 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04002947 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002948
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002949 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04002950 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002951
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002952 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04002953 if (ret > 0)
2954 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002955
Lukas Czerner24075182010-10-27 21:30:06 -04002956 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002957}
2958
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002959/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002960 * ext4_split_extent_at() splits an extent at given block.
2961 *
2962 * @handle: the journal handle
2963 * @inode: the file inode
2964 * @path: the path to the extent
2965 * @split: the logical block where the extent is splitted.
2966 * @split_flags: indicates if the extent could be zeroout if split fails, and
2967 * the states(init or uninit) of new extents.
2968 * @flags: flags used to insert new extent to extent tree.
2969 *
2970 *
2971 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2972 * of which are deterimined by split_flag.
2973 *
2974 * There are two cases:
2975 * a> the extent are splitted into two extent.
2976 * b> split is not needed, and just mark the extent.
2977 *
2978 * return 0 on success.
2979 */
2980static int ext4_split_extent_at(handle_t *handle,
2981 struct inode *inode,
2982 struct ext4_ext_path *path,
2983 ext4_lblk_t split,
2984 int split_flag,
2985 int flags)
2986{
2987 ext4_fsblk_t newblock;
2988 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04002989 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002990 struct ext4_extent *ex2 = NULL;
2991 unsigned int ee_len, depth;
2992 int err = 0;
2993
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04002994 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2995 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2996
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002997 ext_debug("ext4_split_extents_at: inode %lu, logical"
2998 "block %llu\n", inode->i_ino, (unsigned long long)split);
2999
3000 ext4_ext_show_leaf(inode, path);
3001
3002 depth = ext_depth(inode);
3003 ex = path[depth].p_ext;
3004 ee_block = le32_to_cpu(ex->ee_block);
3005 ee_len = ext4_ext_get_actual_len(ex);
3006 newblock = split - ee_block + ext4_ext_pblock(ex);
3007
3008 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003009 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3010 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3011 EXT4_EXT_MARK_UNINIT1 |
3012 EXT4_EXT_MARK_UNINIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003013
3014 err = ext4_ext_get_access(handle, inode, path + depth);
3015 if (err)
3016 goto out;
3017
3018 if (split == ee_block) {
3019 /*
3020 * case b: block @split is the block that the extent begins with
3021 * then we just change the state of the extent, and splitting
3022 * is not needed.
3023 */
3024 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3025 ext4_ext_mark_uninitialized(ex);
3026 else
3027 ext4_ext_mark_initialized(ex);
3028
3029 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003030 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003031
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003032 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003033 goto out;
3034 }
3035
3036 /* case a */
3037 memcpy(&orig_ex, ex, sizeof(orig_ex));
3038 ex->ee_len = cpu_to_le16(split - ee_block);
3039 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3040 ext4_ext_mark_uninitialized(ex);
3041
3042 /*
3043 * path may lead to new leaf, not to original leaf any more
3044 * after ext4_ext_insert_extent() returns,
3045 */
3046 err = ext4_ext_dirty(handle, inode, path + depth);
3047 if (err)
3048 goto fix_extent_len;
3049
3050 ex2 = &newex;
3051 ex2->ee_block = cpu_to_le32(split);
3052 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3053 ext4_ext_store_pblock(ex2, newblock);
3054 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3055 ext4_ext_mark_uninitialized(ex2);
3056
3057 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3058 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003059 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003060 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003061 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003062 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003063 zero_ex.ee_len = cpu_to_le16(
3064 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003065 ext4_ext_store_pblock(&zero_ex,
3066 ext4_ext_pblock(ex2));
3067 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003068 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003069 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003070 zero_ex.ee_len = cpu_to_le16(
3071 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003072 ext4_ext_store_pblock(&zero_ex,
3073 ext4_ext_pblock(ex));
3074 }
3075 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003076 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003077 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003078 zero_ex.ee_len = cpu_to_le16(
3079 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003080 ext4_ext_store_pblock(&zero_ex,
3081 ext4_ext_pblock(&orig_ex));
3082 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003083
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003084 if (err)
3085 goto fix_extent_len;
3086 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003087 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003088 ext4_ext_try_to_merge(handle, inode, path, ex);
3089 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003090 if (err)
3091 goto fix_extent_len;
3092
3093 /* update extent status tree */
3094 err = ext4_es_zeroout(inode, &zero_ex);
3095
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003096 goto out;
3097 } else if (err)
3098 goto fix_extent_len;
3099
3100out:
3101 ext4_ext_show_leaf(inode, path);
3102 return err;
3103
3104fix_extent_len:
3105 ex->ee_len = orig_ex.ee_len;
3106 ext4_ext_dirty(handle, inode, path + depth);
3107 return err;
3108}
3109
3110/*
3111 * ext4_split_extents() splits an extent and mark extent which is covered
3112 * by @map as split_flags indicates
3113 *
3114 * It may result in splitting the extent into multiple extents (upto three)
3115 * There are three possibilities:
3116 * a> There is no split required
3117 * b> Splits in two extents: Split is happening at either end of the extent
3118 * c> Splits in three extents: Somone is splitting in middle of the extent
3119 *
3120 */
3121static int ext4_split_extent(handle_t *handle,
3122 struct inode *inode,
3123 struct ext4_ext_path *path,
3124 struct ext4_map_blocks *map,
3125 int split_flag,
3126 int flags)
3127{
3128 ext4_lblk_t ee_block;
3129 struct ext4_extent *ex;
3130 unsigned int ee_len, depth;
3131 int err = 0;
3132 int uninitialized;
3133 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003134 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003135
3136 depth = ext_depth(inode);
3137 ex = path[depth].p_ext;
3138 ee_block = le32_to_cpu(ex->ee_block);
3139 ee_len = ext4_ext_get_actual_len(ex);
3140 uninitialized = ext4_ext_is_uninitialized(ex);
3141
3142 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003143 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003144 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3145 if (uninitialized)
3146 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3147 EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003148 if (split_flag & EXT4_EXT_DATA_VALID2)
3149 split_flag1 |= EXT4_EXT_DATA_VALID1;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003150 err = ext4_split_extent_at(handle, inode, path,
3151 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003152 if (err)
3153 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003154 } else {
3155 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003156 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003157 /*
3158 * Update path is required because previous ext4_split_extent_at() may
3159 * result in split of original leaf or extent zeroout.
3160 */
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003161 ext4_ext_drop_refs(path);
3162 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3163 if (IS_ERR(path))
3164 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003165 depth = ext_depth(inode);
3166 ex = path[depth].p_ext;
3167 uninitialized = ext4_ext_is_uninitialized(ex);
3168 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003169
3170 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003171 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3172 if (uninitialized) {
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003173 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003174 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3175 EXT4_EXT_MARK_UNINIT2);
3176 }
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003177 err = ext4_split_extent_at(handle, inode, path,
3178 map->m_lblk, split_flag1, flags);
3179 if (err)
3180 goto out;
3181 }
3182
3183 ext4_ext_show_leaf(inode, path);
3184out:
Zheng Liu3a225672013-03-10 21:20:23 -04003185 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003186}
3187
Amit Arora56055d32007-07-17 21:42:38 -04003188/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003189 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04003190 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003191 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04003192 * uninitialized).
3193 * There are three possibilities:
3194 * a> There is no split required: Entire extent should be initialized
3195 * b> Splits in two extents: Write is happening at either end of the extent
3196 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003197 *
3198 * Pre-conditions:
3199 * - The extent pointed to by 'path' is uninitialized.
3200 * - The extent pointed to by 'path' contains a superset
3201 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3202 *
3203 * Post-conditions on success:
3204 * - the returned value is the number of blocks beyond map->l_lblk
3205 * that are allocated and initialized.
3206 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003207 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003208static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003209 struct inode *inode,
3210 struct ext4_map_blocks *map,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003211 struct ext4_ext_path *path,
3212 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003213{
Zheng Liu67a5da52012-08-17 09:54:17 -04003214 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003215 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003216 struct ext4_map_blocks split_map;
3217 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003218 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003219 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003220 unsigned int ee_len, depth, map_len = map->m_len;
3221 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003222 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003223 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003224
3225 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3226 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003227 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003228
Zheng Liu67a5da52012-08-17 09:54:17 -04003229 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003230 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3231 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003232 if (eof_block < map->m_lblk + map_len)
3233 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003234
3235 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003236 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003237 ex = path[depth].p_ext;
3238 ee_block = le32_to_cpu(ex->ee_block);
3239 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003240 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003241
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003242 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3243
3244 /* Pre-conditions */
3245 BUG_ON(!ext4_ext_is_uninitialized(ex));
3246 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003247
3248 /*
3249 * Attempt to transfer newly initialized blocks from the currently
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003250 * uninitialized extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003251 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003252 * memmove() calls. Transferring to the left is the common case in
3253 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3254 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003255 *
3256 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003257 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003258 * This would require removing the extent if the transfer
3259 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003260 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003261 * same extent tree node.
3262 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003263 if ((map->m_lblk == ee_block) &&
3264 /* See if we can merge left */
3265 (map_len < ee_len) && /*L1*/
3266 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003267 ext4_lblk_t prev_lblk;
3268 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003269 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003270
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003271 abut_ex = ex - 1;
3272 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3273 prev_len = ext4_ext_get_actual_len(abut_ex);
3274 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003275 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003276
3277 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003278 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003279 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003280 * - C1: abut_ex is initialized,
3281 * - C2: abut_ex is logically abutting ex,
3282 * - C3: abut_ex is physically abutting ex,
3283 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003284 * overflowing the (initialized) length limit.
3285 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003286 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003287 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3288 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003289 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003290 err = ext4_ext_get_access(handle, inode, path + depth);
3291 if (err)
3292 goto out;
3293
3294 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003295 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003296
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003297 /* Shift the start of ex by 'map_len' blocks */
3298 ex->ee_block = cpu_to_le32(ee_block + map_len);
3299 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3300 ex->ee_len = cpu_to_le16(ee_len - map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003301 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3302
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003303 /* Extend abut_ex by 'map_len' blocks */
3304 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003305
3306 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003307 allocated = map_len;
3308 }
3309 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3310 (map_len < ee_len) && /*L1*/
3311 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3312 /* See if we can merge right */
3313 ext4_lblk_t next_lblk;
3314 ext4_fsblk_t next_pblk, ee_pblk;
3315 unsigned int next_len;
3316
3317 abut_ex = ex + 1;
3318 next_lblk = le32_to_cpu(abut_ex->ee_block);
3319 next_len = ext4_ext_get_actual_len(abut_ex);
3320 next_pblk = ext4_ext_pblock(abut_ex);
3321 ee_pblk = ext4_ext_pblock(ex);
3322
3323 /*
3324 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3325 * upon those conditions:
3326 * - C1: abut_ex is initialized,
3327 * - C2: abut_ex is logically abutting ex,
3328 * - C3: abut_ex is physically abutting ex,
3329 * - C4: abut_ex can receive the additional blocks without
3330 * overflowing the (initialized) length limit.
3331 */
3332 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3333 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3334 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3335 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3336 err = ext4_ext_get_access(handle, inode, path + depth);
3337 if (err)
3338 goto out;
3339
3340 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3341 map, ex, abut_ex);
3342
3343 /* Shift the start of abut_ex by 'map_len' blocks */
3344 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3345 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3346 ex->ee_len = cpu_to_le16(ee_len - map_len);
3347 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3348
3349 /* Extend abut_ex by 'map_len' blocks */
3350 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3351
3352 /* Result: number of initialized blocks past m_lblk */
3353 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003354 }
3355 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003356 if (allocated) {
3357 /* Mark the block containing both extents as dirty */
3358 ext4_ext_dirty(handle, inode, path + depth);
3359
3360 /* Update path to point to the right extent */
3361 path[depth].p_ext = abut_ex;
3362 goto out;
3363 } else
3364 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003365
Yongqiang Yang667eff32011-05-03 12:25:07 -04003366 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003367 /*
3368 * It is safe to convert extent to initialized via explicit
3369 * zeroout only if extent is fully insde i_size or new_size.
3370 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003371 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003372
Zheng Liu67a5da52012-08-17 09:54:17 -04003373 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3374 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003375 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003376
3377 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3378 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003379 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003380 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003381 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003382 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003383 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003384 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003385
3386 err = ext4_ext_get_access(handle, inode, path + depth);
3387 if (err)
3388 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003389 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003390 ext4_ext_try_to_merge(handle, inode, path, ex);
3391 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003392 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003393 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003394
Amit Arora56055d32007-07-17 21:42:38 -04003395 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003396 * four cases:
3397 * 1. split the extent into three extents.
3398 * 2. split the extent into two extents, zeroout the first half.
3399 * 3. split the extent into two extents, zeroout the second half.
3400 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003401 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003402 split_map.m_lblk = map->m_lblk;
3403 split_map.m_len = map->m_len;
3404
Zheng Liu67a5da52012-08-17 09:54:17 -04003405 if (max_zeroout && (allocated > map->m_len)) {
3406 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003407 /* case 3 */
3408 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003409 cpu_to_le32(map->m_lblk);
3410 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003411 ext4_ext_store_pblock(&zero_ex,
3412 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3413 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003414 if (err)
3415 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003416 split_map.m_lblk = map->m_lblk;
3417 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003418 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003419 /* case 2 */
3420 if (map->m_lblk != ee_block) {
3421 zero_ex.ee_block = ex->ee_block;
3422 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3423 ee_block);
3424 ext4_ext_store_pblock(&zero_ex,
3425 ext4_ext_pblock(ex));
3426 err = ext4_ext_zeroout(inode, &zero_ex);
3427 if (err)
3428 goto out;
3429 }
3430
Yongqiang Yang667eff32011-05-03 12:25:07 -04003431 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003432 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3433 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003434 }
3435 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003436
3437 allocated = ext4_split_extent(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003438 &split_map, split_flag, flags);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003439 if (allocated < 0)
3440 err = allocated;
3441
Amit Arora56055d32007-07-17 21:42:38 -04003442out:
Zheng Liuadb23552013-03-10 21:13:05 -04003443 /* If we have gotten a failure, don't zero out status tree */
3444 if (!err)
3445 err = ext4_es_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003446 return err ? err : allocated;
3447}
3448
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003449/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003450 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003451 * ext4_get_blocks_dio_write() when DIO to write
3452 * to an uninitialized extent.
3453 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003454 * Writing to an uninitialized extent may result in splitting the uninitialized
Wang Sheng-Hui30cb27d2012-08-18 22:38:07 -04003455 * extent into multiple initialized/uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003456 * There are three possibilities:
3457 * a> There is no split required: Entire extent should be uninitialized
3458 * b> Splits in two extents: Write is happening at either end of the extent
3459 * c> Splits in three extents: Somone is writing in middle of the extent
3460 *
3461 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003462 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003463 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003464 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003465 * into three uninitialized extent(at most). After IO complete, the part
3466 * being filled will be convert to initialized by the end_io callback function
3467 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003468 *
3469 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003470 */
3471static int ext4_split_unwritten_extents(handle_t *handle,
3472 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003473 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003474 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003475 int flags)
3476{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003477 ext4_lblk_t eof_block;
3478 ext4_lblk_t ee_block;
3479 struct ext4_extent *ex;
3480 unsigned int ee_len;
3481 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003482
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003483 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3484 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003485 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003486
3487 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3488 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003489 if (eof_block < map->m_lblk + map->m_len)
3490 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003491 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003492 * It is safe to convert extent to initialized via explicit
3493 * zeroout only if extent is fully insde i_size or new_size.
3494 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003495 depth = ext_depth(inode);
3496 ex = path[depth].p_ext;
3497 ee_block = le32_to_cpu(ex->ee_block);
3498 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003499
Yongqiang Yang667eff32011-05-03 12:25:07 -04003500 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3501 split_flag |= EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003502 if (flags & EXT4_GET_BLOCKS_CONVERT)
3503 split_flag |= EXT4_EXT_DATA_VALID2;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003504 flags |= EXT4_GET_BLOCKS_PRE_IO;
3505 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003506}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003507
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003508static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003509 struct inode *inode,
3510 struct ext4_map_blocks *map,
3511 struct ext4_ext_path *path)
Mingming Cao00314622009-09-28 15:49:08 -04003512{
3513 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003514 ext4_lblk_t ee_block;
3515 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003516 int depth;
3517 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003518
3519 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003520 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003521 ee_block = le32_to_cpu(ex->ee_block);
3522 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003523
Yongqiang Yang197217a2011-05-03 11:45:29 -04003524 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3525 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003526 (unsigned long long)ee_block, ee_len);
3527
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003528 /* If extent is larger than requested it is a clear sign that we still
3529 * have some extent state machine issues left. So extent_split is still
3530 * required.
3531 * TODO: Once all related issues will be fixed this situation should be
3532 * illegal.
3533 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003534 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003535#ifdef EXT4_DEBUG
3536 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3537 " len %u; IO logical block %llu, len %u\n",
3538 inode->i_ino, (unsigned long long)ee_block, ee_len,
3539 (unsigned long long)map->m_lblk, map->m_len);
3540#endif
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003541 err = ext4_split_unwritten_extents(handle, inode, map, path,
3542 EXT4_GET_BLOCKS_CONVERT);
3543 if (err < 0)
3544 goto out;
3545 ext4_ext_drop_refs(path);
3546 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3547 if (IS_ERR(path)) {
3548 err = PTR_ERR(path);
3549 goto out;
3550 }
3551 depth = ext_depth(inode);
3552 ex = path[depth].p_ext;
3553 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003554
Mingming Cao00314622009-09-28 15:49:08 -04003555 err = ext4_ext_get_access(handle, inode, path + depth);
3556 if (err)
3557 goto out;
3558 /* first mark the extent as initialized */
3559 ext4_ext_mark_initialized(ex);
3560
Yongqiang Yang197217a2011-05-03 11:45:29 -04003561 /* note: ext4_ext_correct_indexes() isn't needed here because
3562 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003563 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003564 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003565
Mingming Cao00314622009-09-28 15:49:08 -04003566 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003567 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003568out:
3569 ext4_ext_show_leaf(inode, path);
3570 return err;
3571}
3572
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003573static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3574 sector_t block, int count)
3575{
3576 int i;
3577 for (i = 0; i < count; i++)
3578 unmap_underlying_metadata(bdev, block + i);
3579}
3580
Theodore Ts'o58590b02010-10-27 21:23:12 -04003581/*
3582 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3583 */
3584static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003585 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003586 struct ext4_ext_path *path,
3587 unsigned int len)
3588{
3589 int i, depth;
3590 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003591 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003592
3593 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3594 return 0;
3595
3596 depth = ext_depth(inode);
3597 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003598
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003599 /*
3600 * We're going to remove EOFBLOCKS_FL entirely in future so we
3601 * do not care for this case anymore. Simply remove the flag
3602 * if there are no extents.
3603 */
3604 if (unlikely(!eh->eh_entries))
3605 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003606 last_ex = EXT_LAST_EXTENT(eh);
3607 /*
3608 * We should clear the EOFBLOCKS_FL flag if we are writing the
3609 * last block in the last extent in the file. We test this by
3610 * first checking to see if the caller to
3611 * ext4_ext_get_blocks() was interested in the last block (or
3612 * a block beyond the last block) in the current extent. If
3613 * this turns out to be false, we can bail out from this
3614 * function immediately.
3615 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003616 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003617 ext4_ext_get_actual_len(last_ex))
3618 return 0;
3619 /*
3620 * If the caller does appear to be planning to write at or
3621 * beyond the end of the current extent, we then test to see
3622 * if the current extent is the last extent in the file, by
3623 * checking to make sure it was reached via the rightmost node
3624 * at each level of the tree.
3625 */
3626 for (i = depth-1; i >= 0; i--)
3627 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3628 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003629out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003630 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3631 return ext4_mark_inode_dirty(handle, inode);
3632}
3633
Aditya Kali7b415bf2011-09-09 19:04:51 -04003634/**
3635 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3636 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003637 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003638 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003639int ext4_find_delalloc_range(struct inode *inode,
3640 ext4_lblk_t lblk_start,
3641 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003642{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003643 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003644
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003645 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003646 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003647 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003648 else if (es.es_lblk <= lblk_start &&
3649 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003650 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003651 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003652 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003653 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003654 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003655}
3656
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003657int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003658{
3659 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3660 ext4_lblk_t lblk_start, lblk_end;
3661 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3662 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3663
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003664 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003665}
3666
3667/**
3668 * Determines how many complete clusters (out of those specified by the 'map')
3669 * are under delalloc and were reserved quota for.
3670 * This function is called when we are writing out the blocks that were
3671 * originally written with their allocation delayed, but then the space was
3672 * allocated using fallocate() before the delayed allocation could be resolved.
3673 * The cases to look for are:
3674 * ('=' indicated delayed allocated blocks
3675 * '-' indicates non-delayed allocated blocks)
3676 * (a) partial clusters towards beginning and/or end outside of allocated range
3677 * are not delalloc'ed.
3678 * Ex:
3679 * |----c---=|====c====|====c====|===-c----|
3680 * |++++++ allocated ++++++|
3681 * ==> 4 complete clusters in above example
3682 *
3683 * (b) partial cluster (outside of allocated range) towards either end is
3684 * marked for delayed allocation. In this case, we will exclude that
3685 * cluster.
3686 * Ex:
3687 * |----====c========|========c========|
3688 * |++++++ allocated ++++++|
3689 * ==> 1 complete clusters in above example
3690 *
3691 * Ex:
3692 * |================c================|
3693 * |++++++ allocated ++++++|
3694 * ==> 0 complete clusters in above example
3695 *
3696 * The ext4_da_update_reserve_space will be called only if we
3697 * determine here that there were some "entire" clusters that span
3698 * this 'allocated' range.
3699 * In the non-bigalloc case, this function will just end up returning num_blks
3700 * without ever calling ext4_find_delalloc_range.
3701 */
3702static unsigned int
3703get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3704 unsigned int num_blks)
3705{
3706 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3707 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3708 ext4_lblk_t lblk_from, lblk_to, c_offset;
3709 unsigned int allocated_clusters = 0;
3710
3711 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3712 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3713
3714 /* max possible clusters for this allocation */
3715 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3716
Aditya Kalid8990242011-09-09 19:18:51 -04003717 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3718
Aditya Kali7b415bf2011-09-09 19:04:51 -04003719 /* Check towards left side */
3720 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3721 if (c_offset) {
3722 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3723 lblk_to = lblk_from + c_offset - 1;
3724
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003725 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003726 allocated_clusters--;
3727 }
3728
3729 /* Now check towards right. */
3730 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3731 if (allocated_clusters && c_offset) {
3732 lblk_from = lblk_start + num_blks;
3733 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3734
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003735 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003736 allocated_clusters--;
3737 }
3738
3739 return allocated_clusters;
3740}
3741
Mingming Cao00314622009-09-28 15:49:08 -04003742static int
3743ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003744 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003745 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003746 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003747{
3748 int ret = 0;
3749 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003750 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003751
Zheng Liu88635ca2011-12-28 19:00:25 -05003752 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3753 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003754 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003755 flags, allocated);
3756 ext4_ext_show_leaf(inode, path);
3757
Lukas Czerner27dd4382013-04-09 22:11:22 -04003758 /*
3759 * When writing into uninitialized space, we should not fail to
3760 * allocate metadata blocks for the new extent block if needed.
3761 */
3762 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3763
Zheng Liub5645532012-11-08 14:33:43 -05003764 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3765 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04003766
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003767 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003768 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003769 ret = ext4_split_unwritten_extents(handle, inode, map,
3770 path, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04003771 if (ret <= 0)
3772 goto out;
Mingming5f524952009-11-10 10:48:04 -05003773 /*
3774 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003775 * that this IO needs to conversion to written when IO is
Mingming5f524952009-11-10 10:48:04 -05003776 * completed
3777 */
Tao Ma0edeb712011-10-31 17:30:44 -04003778 if (io)
3779 ext4_set_io_unwritten_flag(inode, io);
3780 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003781 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05003782 map->m_flags |= EXT4_MAP_UNWRITTEN;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003783 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003784 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003785 goto out;
3786 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003787 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003788 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003789 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Mingming Cao00314622009-09-28 15:49:08 -04003790 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003791 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003792 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003793 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3794 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003795 } else
3796 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04003797 map->m_flags |= EXT4_MAP_MAPPED;
3798 if (allocated > map->m_len)
3799 allocated = map->m_len;
3800 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003801 goto out2;
3802 }
3803 /* buffered IO case */
3804 /*
3805 * repeat fallocate creation request
3806 * we already have an unwritten extent
3807 */
Zheng Liua25a4e12013-02-18 00:28:04 -05003808 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3809 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003810 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05003811 }
Mingming Cao00314622009-09-28 15:49:08 -04003812
3813 /* buffered READ or buffered write_begin() lookup */
3814 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3815 /*
3816 * We have blocks reserved already. We
3817 * return allocated blocks so that delalloc
3818 * won't do block reservation for us. But
3819 * the buffer head will be unmapped so that
3820 * a read from the block returns 0s.
3821 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003822 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003823 goto out1;
3824 }
3825
3826 /* buffered write, writepage time, convert*/
Lukas Czerner27dd4382013-04-09 22:11:22 -04003827 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003828 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05003829 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04003830out:
3831 if (ret <= 0) {
3832 err = ret;
3833 goto out2;
3834 } else
3835 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003836 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003837 /*
3838 * if we allocated more blocks than requested
3839 * we need to make sure we unmap the extra block
3840 * allocated. The actual needed block will get
3841 * unmapped later when we find the buffer_head marked
3842 * new.
3843 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003844 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003845 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003846 newblock + map->m_len,
3847 allocated - map->m_len);
3848 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003849 }
Zheng Liu3a225672013-03-10 21:20:23 -04003850 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003851
3852 /*
3853 * If we have done fallocate with the offset that is already
3854 * delayed allocated, we would have block reservation
3855 * and quota reservation done in the delayed write path.
3856 * But fallocate would have already updated quota and block
3857 * count for this offset. So cancel these reservation
3858 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003859 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3860 unsigned int reserved_clusters;
3861 reserved_clusters = get_reserved_cluster_alloc(inode,
3862 map->m_lblk, map->m_len);
3863 if (reserved_clusters)
3864 ext4_da_update_reserve_space(inode,
3865 reserved_clusters,
3866 0);
3867 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003868
Mingming Cao00314622009-09-28 15:49:08 -04003869map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003870 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003871 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3872 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3873 map->m_len);
3874 if (err < 0)
3875 goto out2;
3876 }
Mingming Cao00314622009-09-28 15:49:08 -04003877out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003878 if (allocated > map->m_len)
3879 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003880 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003881 map->m_pblk = newblock;
3882 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003883out2:
3884 if (path) {
3885 ext4_ext_drop_refs(path);
3886 kfree(path);
3887 }
3888 return err ? err : allocated;
3889}
Theodore Ts'o58590b02010-10-27 21:23:12 -04003890
Mingming Cao00314622009-09-28 15:49:08 -04003891/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003892 * get_implied_cluster_alloc - check to see if the requested
3893 * allocation (in the map structure) overlaps with a cluster already
3894 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04003895 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003896 * @map The requested lblk->pblk mapping
3897 * @ex The extent structure which might contain an implied
3898 * cluster allocation
3899 *
3900 * This function is called by ext4_ext_map_blocks() after we failed to
3901 * find blocks that were already in the inode's extent tree. Hence,
3902 * we know that the beginning of the requested region cannot overlap
3903 * the extent from the inode's extent tree. There are three cases we
3904 * want to catch. The first is this case:
3905 *
3906 * |--- cluster # N--|
3907 * |--- extent ---| |---- requested region ---|
3908 * |==========|
3909 *
3910 * The second case that we need to test for is this one:
3911 *
3912 * |--------- cluster # N ----------------|
3913 * |--- requested region --| |------- extent ----|
3914 * |=======================|
3915 *
3916 * The third case is when the requested region lies between two extents
3917 * within the same cluster:
3918 * |------------- cluster # N-------------|
3919 * |----- ex -----| |---- ex_right ----|
3920 * |------ requested region ------|
3921 * |================|
3922 *
3923 * In each of the above cases, we need to set the map->m_pblk and
3924 * map->m_len so it corresponds to the return the extent labelled as
3925 * "|====|" from cluster #N, since it is already in use for data in
3926 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3927 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3928 * as a new "allocated" block region. Otherwise, we will return 0 and
3929 * ext4_ext_map_blocks() will then allocate one or more new clusters
3930 * by calling ext4_mb_new_blocks().
3931 */
Aditya Kalid8990242011-09-09 19:18:51 -04003932static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003933 struct ext4_map_blocks *map,
3934 struct ext4_extent *ex,
3935 struct ext4_ext_path *path)
3936{
Aditya Kalid8990242011-09-09 19:18:51 -04003937 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003938 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3939 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05003940 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003941 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3942 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3943 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3944
3945 /* The extent passed in that we are trying to match */
3946 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3947 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3948
3949 /* The requested region passed into ext4_map_blocks() */
3950 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003951
3952 if ((rr_cluster_start == ex_cluster_end) ||
3953 (rr_cluster_start == ex_cluster_start)) {
3954 if (rr_cluster_start == ex_cluster_end)
3955 ee_start += ee_len - 1;
3956 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3957 c_offset;
3958 map->m_len = min(map->m_len,
3959 (unsigned) sbi->s_cluster_ratio - c_offset);
3960 /*
3961 * Check for and handle this case:
3962 *
3963 * |--------- cluster # N-------------|
3964 * |------- extent ----|
3965 * |--- requested region ---|
3966 * |===========|
3967 */
3968
3969 if (map->m_lblk < ee_block)
3970 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3971
3972 /*
3973 * Check for the case where there is already another allocated
3974 * block to the right of 'ex' but before the end of the cluster.
3975 *
3976 * |------------- cluster # N-------------|
3977 * |----- ex -----| |---- ex_right ----|
3978 * |------ requested region ------|
3979 * |================|
3980 */
3981 if (map->m_lblk > ee_block) {
3982 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3983 map->m_len = min(map->m_len, next - map->m_lblk);
3984 }
Aditya Kalid8990242011-09-09 19:18:51 -04003985
3986 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003987 return 1;
3988 }
Aditya Kalid8990242011-09-09 19:18:51 -04003989
3990 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003991 return 0;
3992}
3993
3994
3995/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05003996 * Block allocation/map/preallocation routine for extents based files
3997 *
3998 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003999 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004000 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4001 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004002 *
4003 * return > 0, number of of blocks already mapped/allocated
4004 * if create == 0 and these are pre-allocated blocks
4005 * buffer head is unmapped
4006 * otherwise blocks are mapped
4007 *
4008 * return = 0, if plain look up failed (blocks have not been allocated)
4009 * buffer head is unmapped
4010 *
4011 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004012 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004013int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4014 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004015{
4016 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004017 struct ext4_extent newex, *ex, *ex2;
4018 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004019 ext4_fsblk_t newblock = 0;
Zheng Liu37794732012-11-08 14:47:52 -05004020 int free_on_err = 0, err = 0, depth;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004021 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004022 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004023 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004024 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004025 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004026 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07004027
Mingming84fe3be2009-09-01 08:44:37 -04004028 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004029 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004030 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004031
Alex Tomasa86c6182006-10-11 01:21:03 -07004032 /* find extent for this block */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004033 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07004034 if (IS_ERR(path)) {
4035 err = PTR_ERR(path);
4036 path = NULL;
4037 goto out2;
4038 }
4039
4040 depth = ext_depth(inode);
4041
4042 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004043 * consistent leaf must not be empty;
4044 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07004045 * this is why assert can't be put in ext4_ext_find_extent()
4046 */
Frank Mayhar273df552010-03-02 11:46:09 -05004047 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4048 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004049 "lblock: %lu, depth: %d pblock %lld",
4050 (unsigned long) map->m_lblk, depth,
4051 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004052 err = -EIO;
4053 goto out2;
4054 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004055
Avantika Mathur7e028972006-12-06 20:41:33 -08004056 ex = path[depth].p_ext;
4057 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004058 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004059 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004060 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004061
4062 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004063 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004064 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004065 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004066 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004067
4068 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4069
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004070 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004071 if (in_range(map->m_lblk, ee_block, ee_len)) {
4072 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004073 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004074 allocated = ee_len - (map->m_lblk - ee_block);
4075 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4076 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004077
Zheng Liu69eb33d2013-02-18 00:31:07 -05004078 if (!ext4_ext_is_uninitialized(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004079 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004080
Zheng Liu37794732012-11-08 14:47:52 -05004081 allocated = ext4_ext_handle_uninitialized_extents(
Lukas Czerner78771912012-03-19 23:05:43 -04004082 handle, inode, map, path, flags,
4083 allocated, newblock);
Zheng Liu37794732012-11-08 14:47:52 -05004084 goto out3;
Alex Tomasa86c6182006-10-11 01:21:03 -07004085 }
4086 }
4087
Aditya Kali7b415bf2011-09-09 19:04:51 -04004088 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05004089 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04004090 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4091
Alex Tomasa86c6182006-10-11 01:21:03 -07004092 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004093 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004094 * we couldn't try to create block if create flag is zero
4095 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004096 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004097 /*
4098 * put just found gap into cache to speed up
4099 * subsequent requests
4100 */
Zheng Liud100eef2013-02-18 00:29:59 -05004101 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4102 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004103 goto out2;
4104 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004105
Alex Tomasa86c6182006-10-11 01:21:03 -07004106 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004107 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004108 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004109 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004110 newex.ee_block = cpu_to_le32(map->m_lblk);
4111 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4112
4113 /*
4114 * If we are doing bigalloc, check to see if the extent returned
4115 * by ext4_ext_find_extent() implies a cluster we can use.
4116 */
4117 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004118 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004119 ar.len = allocated = map->m_len;
4120 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004121 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004122 goto got_allocated_blocks;
4123 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004124
Alex Tomasc9de5602008-01-29 00:19:52 -05004125 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004126 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004127 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4128 if (err)
4129 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004130 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004131 ex2 = NULL;
4132 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004133 if (err)
4134 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004135
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004136 /* Check if the extent after searching to the right implies a
4137 * cluster we can use. */
4138 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004139 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004140 ar.len = allocated = map->m_len;
4141 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004142 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004143 goto got_allocated_blocks;
4144 }
4145
Amit Arora749269f2007-07-18 09:02:56 -04004146 /*
4147 * See if request is beyond maximum number of blocks we can have in
4148 * a single extent. For an initialized extent this limit is
4149 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4150 * EXT_UNINIT_MAX_LEN.
4151 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004152 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004153 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004154 map->m_len = EXT_INIT_MAX_LEN;
4155 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004156 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004157 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004158
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004159 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004160 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004161 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004162 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004163 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004164 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004165 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004166
4167 /* allocate new block */
4168 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004169 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4170 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004171 /*
4172 * We calculate the offset from the beginning of the cluster
4173 * for the logical block number, since when we allocate a
4174 * physical cluster, the physical block should start at the
4175 * same offset from the beginning of the cluster. This is
4176 * needed so that future calls to get_implied_cluster_alloc()
4177 * work correctly.
4178 */
4179 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4180 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4181 ar.goal -= offset;
4182 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004183 if (S_ISREG(inode->i_mode))
4184 ar.flags = EXT4_MB_HINT_DATA;
4185 else
4186 /* disable in-core preallocation for non-regular files */
4187 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004188 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4189 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004190 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004191 if (!newblock)
4192 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004193 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004194 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004195 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004196 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004197 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4198 if (ar.len > allocated)
4199 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004200
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004201got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004202 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004203 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004204 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004205 /* Mark uninitialized */
4206 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04004207 ext4_ext_mark_uninitialized(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004208 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004209 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004210 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004211 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004212 * here we flag the IO that really needs the conversion.
Mingming5f524952009-11-10 10:48:04 -05004213 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004214 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004215 */
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004216 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4217 set_unwritten = 1;
Jiaying Zhang744692d2010-03-04 16:14:02 -05004218 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004219 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004220 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004221
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004222 err = 0;
4223 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4224 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4225 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004226 if (!err)
4227 err = ext4_ext_insert_extent(handle, inode, path,
4228 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004229
4230 if (!err && set_unwritten) {
4231 if (io)
4232 ext4_set_io_unwritten_flag(inode, io);
4233 else
4234 ext4_set_inode_state(inode,
4235 EXT4_STATE_DIO_UNWRITTEN);
4236 }
4237
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004238 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004239 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4240 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004241 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004242 /* not a good idea to call discard here directly,
4243 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004244 ext4_discard_preallocations(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004245 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
Maxim Patlasov7132de72011-07-10 19:37:48 -04004246 ext4_ext_get_actual_len(&newex), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004247 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004248 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004249
Alex Tomasa86c6182006-10-11 01:21:03 -07004250 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004251 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004252 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004253 if (allocated > map->m_len)
4254 allocated = map->m_len;
4255 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004256
Jan Karab436b9b2009-12-08 23:51:10 -05004257 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004258 * Update reserved blocks/metadata blocks after successful
4259 * block allocation which had been deferred till now.
4260 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004261 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004262 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004263 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004264 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004265 */
4266 reserved_clusters = get_reserved_cluster_alloc(inode,
4267 map->m_lblk, allocated);
4268 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4269 if (reserved_clusters) {
4270 /*
4271 * We have clusters reserved for this range.
4272 * But since we are not doing actual allocation
4273 * and are simply using blocks from previously
4274 * allocated cluster, we should release the
4275 * reservation and not claim quota.
4276 */
4277 ext4_da_update_reserve_space(inode,
4278 reserved_clusters, 0);
4279 }
4280 } else {
4281 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004282 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f262011-09-09 19:20:51 -04004283 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004284 int reservation = allocated_clusters -
4285 reserved_clusters;
4286 /*
4287 * It seems we claimed few clusters outside of
4288 * the range of this allocation. We should give
4289 * it back to the reservation pool. This can
4290 * happen in the following case:
4291 *
4292 * * Suppose s_cluster_ratio is 4 (i.e., each
4293 * cluster has 4 blocks. Thus, the clusters
4294 * are [0-3],[4-7],[8-11]...
4295 * * First comes delayed allocation write for
4296 * logical blocks 10 & 11. Since there were no
4297 * previous delayed allocated blocks in the
4298 * range [8-11], we would reserve 1 cluster
4299 * for this write.
4300 * * Next comes write for logical blocks 3 to 8.
4301 * In this case, we will reserve 2 clusters
4302 * (for [0-3] and [4-7]; and not for [8-11] as
4303 * that range has a delayed allocated blocks.
4304 * Thus total reserved clusters now becomes 3.
4305 * * Now, during the delayed allocation writeout
4306 * time, we will first write blocks [3-8] and
4307 * allocate 3 clusters for writing these
4308 * blocks. Also, we would claim all these
4309 * three clusters above.
4310 * * Now when we come here to writeout the
4311 * blocks [10-11], we would expect to claim
4312 * the reservation of 1 cluster we had made
4313 * (and we would claim it since there are no
4314 * more delayed allocated blocks in the range
4315 * [8-11]. But our reserved cluster count had
4316 * already gone to 0.
4317 *
4318 * Thus, at the step 4 above when we determine
4319 * that there are still some unwritten delayed
4320 * allocated blocks outside of our current
4321 * block range, we should increment the
4322 * reserved clusters count so that when the
4323 * remaining blocks finally gets written, we
4324 * could claim them.
4325 */
Aditya Kali5356f262011-09-09 19:20:51 -04004326 dquot_reserve_block(inode,
4327 EXT4_C2B(sbi, reservation));
4328 spin_lock(&ei->i_block_reservation_lock);
4329 ei->i_reserved_data_blocks += reservation;
4330 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004331 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004332 /*
4333 * We will claim quota for all newly allocated blocks.
4334 * We're updating the reserved space *after* the
4335 * correction above so we do not accidentally free
4336 * all the metadata reservation because we might
4337 * actually need it later on.
4338 */
4339 ext4_da_update_reserve_space(inode, allocated_clusters,
4340 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004341 }
4342 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004343
4344 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004345 * Cache the extent and update transaction to commit on fdatasync only
4346 * when it is _not_ an uninitialized extent.
4347 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004348 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004349 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004350 else
Jan Karab436b9b2009-12-08 23:51:10 -05004351 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004352out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004353 if (allocated > map->m_len)
4354 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004355 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004356 map->m_flags |= EXT4_MAP_MAPPED;
4357 map->m_pblk = newblock;
4358 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004359out2:
4360 if (path) {
4361 ext4_ext_drop_refs(path);
4362 kfree(path);
4363 }
Allison Hendersone8613042011-05-25 07:41:46 -04004364
Zheng Liu37794732012-11-08 14:47:52 -05004365out3:
Zheng Liu19b303d2012-11-08 14:34:04 -05004366 trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated);
Yongqiang Yange7b319e2011-10-29 09:39:51 -04004367
Lukas Czerner78771912012-03-19 23:05:43 -04004368 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004369}
4370
Theodore Ts'o819c4922013-04-03 12:47:17 -04004371void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004372{
Alex Tomasa86c6182006-10-11 01:21:03 -07004373 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004374 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004375 int err = 0;
4376
4377 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004378 * TODO: optimization is possible here.
4379 * Probably we need not scan at all,
4380 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004381 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004382
4383 /* we have to know where to truncate from in crash case */
4384 EXT4_I(inode)->i_disksize = inode->i_size;
4385 ext4_mark_inode_dirty(handle, inode);
4386
4387 last_block = (inode->i_size + sb->s_blocksize - 1)
4388 >> EXT4_BLOCK_SIZE_BITS(sb);
Zheng Liu51865fd2012-11-08 21:57:32 -05004389 err = ext4_es_remove_extent(inode, last_block,
4390 EXT_MAX_BLOCKS - last_block);
Lukas Czerner5f95d212012-03-19 23:03:19 -04004391 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07004392}
4393
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004394static void ext4_falloc_update_inode(struct inode *inode,
4395 int mode, loff_t new_size, int update_ctime)
4396{
4397 struct timespec now;
4398
4399 if (update_ctime) {
4400 now = current_fs_time(inode->i_sb);
4401 if (!timespec_equal(&inode->i_ctime, &now))
4402 inode->i_ctime = now;
4403 }
4404 /*
4405 * Update only when preallocation was requested beyond
4406 * the file size.
4407 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04004408 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4409 if (new_size > i_size_read(inode))
4410 i_size_write(inode, new_size);
4411 if (new_size > EXT4_I(inode)->i_disksize)
4412 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004413 } else {
4414 /*
4415 * Mark that we allocate beyond EOF so the subsequent truncate
4416 * can proceed even if the new size is the same as i_size.
4417 */
4418 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004419 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004420 }
4421
4422}
4423
Amit Aroraa2df2a62007-07-17 21:42:41 -04004424/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004425 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04004426 * operation, which gets called from sys_fallocate system call.
4427 * For block-mapped files, posix_fallocate should fall back to the method
4428 * of writing zeroes to the required new blocks (the same behavior which is
4429 * expected for file systems which do not support fallocate() system call).
4430 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004431long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004432{
Al Viro496ad9a2013-01-23 17:07:38 -05004433 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004434 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004435 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004436 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004437 int ret = 0;
4438 int ret2 = 0;
4439 int retries = 0;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004440 int flags;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004441 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004442 unsigned int credits, blkbits = inode->i_blkbits;
4443
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004444 /* Return error if mode is not supported */
4445 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4446 return -EOPNOTSUPP;
4447
4448 if (mode & FALLOC_FL_PUNCH_HOLE)
4449 return ext4_punch_hole(file, offset, len);
4450
Tao Ma0c8d4142012-12-10 14:06:03 -05004451 ret = ext4_convert_inline_data(inode);
4452 if (ret)
4453 return ret;
4454
Zheng Liu8bad6fc2013-01-28 09:21:37 -05004455 /*
4456 * currently supporting (pre)allocate mode for extent-based
4457 * files _only_
4458 */
4459 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4460 return -EOPNOTSUPP;
4461
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004462 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004463 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004464 /*
4465 * We can't just convert len to max_blocks because
4466 * If blocksize = 4096 offset = 3072 and len = 2048
4467 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004468 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004469 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004470 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04004471 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04004472 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004473 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004474 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004475 ret = inode_newsize_ok(inode, (len + offset));
4476 if (ret) {
4477 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004478 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004479 return ret;
4480 }
Greg Harm3c6fe772011-10-31 18:41:47 -04004481 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004482 if (mode & FALLOC_FL_KEEP_SIZE)
4483 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
Greg Harm3c6fe772011-10-31 18:41:47 -04004484 /*
4485 * Don't normalize the request if it can fit in one extent so
4486 * that it doesn't get unnecessarily split into multiple
4487 * extents.
4488 */
4489 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4490 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004491
Amit Aroraa2df2a62007-07-17 21:42:41 -04004492retry:
4493 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004494 map.m_lblk = map.m_lblk + ret;
4495 map.m_len = max_blocks = max_blocks - ret;
Theodore Ts'o9924a922013-02-08 21:59:22 -05004496 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4497 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004498 if (IS_ERR(handle)) {
4499 ret = PTR_ERR(handle);
4500 break;
4501 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004502 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004503 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004504#ifdef EXT4FS_DEBUG
Lukas Czernerb06acd32013-01-28 21:21:12 -05004505 ext4_warning(inode->i_sb,
4506 "inode #%lu: block %u: len %u: "
4507 "ext4_ext_map_blocks returned %d",
4508 inode->i_ino, map.m_lblk,
4509 map.m_len, ret);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004510#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04004511 ext4_mark_inode_dirty(handle, inode);
4512 ret2 = ext4_journal_stop(handle);
4513 break;
4514 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004515 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004516 blkbits) >> blkbits))
4517 new_size = offset + len;
4518 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04004519 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004520
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004521 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004522 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04004523 ext4_mark_inode_dirty(handle, inode);
Zheng Liuf4e95b32012-06-30 19:12:57 -04004524 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4525 ext4_handle_sync(handle);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004526 ret2 = ext4_journal_stop(handle);
4527 if (ret2)
4528 break;
4529 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004530 if (ret == -ENOSPC &&
4531 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4532 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004533 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004534 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004535 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004536 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4537 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004538 return ret > 0 ? ret2 : ret;
4539}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004540
4541/*
Mingming Cao00314622009-09-28 15:49:08 -04004542 * This function convert a range of blocks to written extents
4543 * The caller of this function will pass the start offset and the size.
4544 * all unwritten extents within this range will be converted to
4545 * written extents.
4546 *
4547 * This function is called from the direct IO end io call back
4548 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004549 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004550 */
4551int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
Eric Sandeena1de02d2010-02-04 23:58:38 -05004552 ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004553{
4554 handle_t *handle;
Mingming Cao00314622009-09-28 15:49:08 -04004555 unsigned int max_blocks;
4556 int ret = 0;
4557 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004558 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004559 unsigned int credits, blkbits = inode->i_blkbits;
4560
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004561 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004562 /*
4563 * We can't just convert len to max_blocks because
4564 * If blocksize = 4096 offset = 3072 and len = 2048
4565 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004566 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4567 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004568 /*
4569 * credits to insert 1 extent into extent tree
4570 */
4571 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4572 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004573 map.m_lblk += ret;
4574 map.m_len = (max_blocks -= ret);
Theodore Ts'o9924a922013-02-08 21:59:22 -05004575 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
Mingming Cao00314622009-09-28 15:49:08 -04004576 if (IS_ERR(handle)) {
4577 ret = PTR_ERR(handle);
4578 break;
4579 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004580 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004581 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05004582 if (ret <= 0)
4583 ext4_warning(inode->i_sb,
4584 "inode #%lu: block %u: len %u: "
4585 "ext4_ext_map_blocks returned %d",
4586 inode->i_ino, map.m_lblk,
4587 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04004588 ext4_mark_inode_dirty(handle, inode);
4589 ret2 = ext4_journal_stop(handle);
4590 if (ret <= 0 || ret2 )
4591 break;
4592 }
4593 return ret > 0 ? ret2 : ret;
4594}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004595
Mingming Cao00314622009-09-28 15:49:08 -04004596/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004597 * If newes is not existing extent (newes->ec_pblk equals zero) find
4598 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004599 * return start of the next delayed extent.
4600 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05004601 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004602 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05004603 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004604 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004605static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05004606 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004607{
Zheng Liub3aff3e2012-11-08 21:57:37 -05004608 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05004609 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004610
Zheng Liu69eb33d2013-02-18 00:31:07 -05004611 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004612 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4613 newes->es_lblk + newes->es_len - 1, &es);
4614
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004615 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004616 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004617 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004618 */
Zheng Liu06b0c882013-02-18 00:26:51 -05004619 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05004620 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004621 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004622
Zheng Liu69eb33d2013-02-18 00:31:07 -05004623 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05004624 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004625 newes->es_len = min(es.es_lblk - newes->es_lblk,
4626 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004627 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004628 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004629
Zheng Liu69eb33d2013-02-18 00:31:07 -05004630 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004631 }
4632
Zheng Liu69eb33d2013-02-18 00:31:07 -05004633 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004634 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05004635 if (es.es_len == 0)
4636 next_del = EXT_MAX_BLOCKS;
4637 else
4638 next_del = es.es_lblk;
4639
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004640 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004641}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004642/* fiemap flags we can handle specified here */
4643#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4644
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004645static int ext4_xattr_fiemap(struct inode *inode,
4646 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004647{
4648 __u64 physical = 0;
4649 __u64 length;
4650 __u32 flags = FIEMAP_EXTENT_LAST;
4651 int blockbits = inode->i_sb->s_blocksize_bits;
4652 int error = 0;
4653
4654 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004655 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004656 struct ext4_iloc iloc;
4657 int offset; /* offset of xattr in inode */
4658
4659 error = ext4_get_inode_loc(inode, &iloc);
4660 if (error)
4661 return error;
4662 physical = iloc.bh->b_blocknr << blockbits;
4663 offset = EXT4_GOOD_OLD_INODE_SIZE +
4664 EXT4_I(inode)->i_extra_isize;
4665 physical += offset;
4666 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4667 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004668 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004669 } else { /* external block */
4670 physical = EXT4_I(inode)->i_file_acl << blockbits;
4671 length = inode->i_sb->s_blocksize;
4672 }
4673
4674 if (physical)
4675 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4676 length, flags);
4677 return (error < 0 ? error : 0);
4678}
4679
4680int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4681 __u64 start, __u64 len)
4682{
4683 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004684 int error = 0;
4685
Tao Ma94191982012-12-10 14:06:02 -05004686 if (ext4_has_inline_data(inode)) {
4687 int has_inline = 1;
4688
4689 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4690
4691 if (has_inline)
4692 return error;
4693 }
4694
Eric Sandeen6873fa02008-10-07 00:46:36 -04004695 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004696 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004697 return generic_block_fiemap(inode, fieinfo, start, len,
4698 ext4_get_block);
4699
4700 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4701 return -EBADR;
4702
4703 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4704 error = ext4_xattr_fiemap(inode, fieinfo);
4705 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004706 ext4_lblk_t len_blks;
4707 __u64 last_blk;
4708
Eric Sandeen6873fa02008-10-07 00:46:36 -04004709 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004710 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004711 if (last_blk >= EXT_MAX_BLOCKS)
4712 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004713 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004714
4715 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004716 * Walk the extent tree gathering extent information
4717 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004718 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004719 error = ext4_fill_fiemap_extents(inode, start_blk,
4720 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004721 }
4722
4723 return error;
4724}