blob: 0b16fb4c06d3e176e367bced078d014542a97b2a [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>
Alex Tomasa86c6182006-10-11 01:21:03 -070040#include <asm/uaccess.h>
Eric Sandeen6873fa02008-10-07 00:46:36 -040041#include <linux/fiemap.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040042#include "ext4_jbd2.h"
Theodore Ts'o4a092d72012-11-28 13:03:30 -050043#include "ext4_extents.h"
Tao Maf19d5872012-12-10 14:05:51 -050044#include "xattr.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070045
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040046#include <trace/events/ext4.h>
47
Lukas Czerner5f95d212012-03-19 23:03:19 -040048/*
49 * used by extent splitting.
50 */
51#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
52 due to ENOSPC */
Lukas Czerner556615d2014-04-20 23:45:47 -040053#define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54#define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
Lukas Czerner5f95d212012-03-19 23:03:19 -040055
Dmitry Monakhovdee1f972012-10-10 01:04:58 -040056#define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57#define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
58
Darrick J. Wong7ac59902012-04-29 18:37:10 -040059static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
61{
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
64 __u32 csum;
65
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
69}
70
71static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
73{
74 struct ext4_extent_tail *et;
75
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -040076 if (!ext4_has_metadata_csum(inode->i_sb))
Darrick J. Wong7ac59902012-04-29 18:37:10 -040077 return 1;
78
79 et = find_ext4_extent_tail(eh);
80 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
81 return 0;
82 return 1;
83}
84
85static void ext4_extent_block_csum_set(struct inode *inode,
86 struct ext4_extent_header *eh)
87{
88 struct ext4_extent_tail *et;
89
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -040090 if (!ext4_has_metadata_csum(inode->i_sb))
Darrick J. Wong7ac59902012-04-29 18:37:10 -040091 return;
92
93 et = find_ext4_extent_tail(eh);
94 et->et_checksum = ext4_extent_block_csum(inode, eh);
95}
96
Allison Hendersond583fb82011-05-25 07:41:43 -040097static int ext4_split_extent(handle_t *handle,
98 struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -040099 struct ext4_ext_path **ppath,
Allison Hendersond583fb82011-05-25 07:41:43 -0400100 struct ext4_map_blocks *map,
101 int split_flag,
102 int flags);
103
Lukas Czerner5f95d212012-03-19 23:03:19 -0400104static int ext4_split_extent_at(handle_t *handle,
105 struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -0400106 struct ext4_ext_path **ppath,
Lukas Czerner5f95d212012-03-19 23:03:19 -0400107 ext4_lblk_t split,
108 int split_flag,
109 int flags);
110
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500111static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -0500112 struct extent_status *newes);
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500113
Jan Kara487caee2009-08-17 22:17:20 -0400114static int ext4_ext_truncate_extend_restart(handle_t *handle,
115 struct inode *inode,
116 int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -0700117{
118 int err;
119
Frank Mayhar03901312009-01-07 00:06:22 -0500120 if (!ext4_handle_valid(handle))
121 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700122 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -0400123 return 0;
124 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -0400125 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -0400126 return err;
Jan Kara487caee2009-08-17 22:17:20 -0400127 err = ext4_truncate_restart_trans(handle, inode, needed);
Dmitry Monakhov0617b832010-05-17 01:00:00 -0400128 if (err == 0)
129 err = -EAGAIN;
Jan Kara487caee2009-08-17 22:17:20 -0400130
131 return err;
Alex Tomasa86c6182006-10-11 01:21:03 -0700132}
133
134/*
135 * could return:
136 * - EROFS
137 * - ENOMEM
138 */
139static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
140 struct ext4_ext_path *path)
141{
142 if (path->p_bh) {
143 /* path points to block */
liang xie5d601252014-05-12 22:06:43 -0400144 BUFFER_TRACE(path->p_bh, "get_write_access");
Alex Tomasa86c6182006-10-11 01:21:03 -0700145 return ext4_journal_get_write_access(handle, path->p_bh);
146 }
147 /* path points to leaf/index in inode body */
148 /* we use in-core data, no need to protect them */
149 return 0;
150}
151
152/*
153 * could return:
154 * - EROFS
155 * - ENOMEM
156 * - EIO
157 */
Darrick J. Wong26564972013-04-19 14:04:12 -0400158int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
159 struct inode *inode, struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700160{
161 int err;
Dmitry Monakhov4b1f1662014-07-27 22:28:15 -0400162
163 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
Alex Tomasa86c6182006-10-11 01:21:03 -0700164 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
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400292static inline int
293ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -0400294 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400295 int nofail)
296{
Theodore Ts'odfe50802014-09-01 14:37:09 -0400297 struct ext4_ext_path *path = *ppath;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400298 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
299
Theodore Ts'odfe50802014-09-01 14:37:09 -0400300 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400301 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
302 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
303 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
304}
305
Mingming Caod2a17632008-07-14 17:52:37 -0400306/*
307 * Calculate the number of metadata blocks needed
308 * to allocate @blocks
309 * Worse case is one block per extent
310 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500311int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400312{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500313 struct ext4_inode_info *ei = EXT4_I(inode);
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400314 int idxs;
Mingming Caod2a17632008-07-14 17:52:37 -0400315
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500316 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
317 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400318
319 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500320 * If the new delayed allocation block is contiguous with the
321 * previous da block, it can share index blocks with the
322 * previous block, so we only need to allocate a new index
323 * block every idxs leaf blocks. At ldxs**2 blocks, we need
324 * an additional index block, and at ldxs**3 blocks, yet
325 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400326 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500327 if (ei->i_da_metadata_calc_len &&
328 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400329 int num = 0;
330
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500331 if ((ei->i_da_metadata_calc_len % idxs) == 0)
332 num++;
333 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
334 num++;
335 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
336 num++;
337 ei->i_da_metadata_calc_len = 0;
338 } else
339 ei->i_da_metadata_calc_len++;
340 ei->i_da_metadata_calc_last_lblock++;
341 return num;
342 }
Mingming Caod2a17632008-07-14 17:52:37 -0400343
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500344 /*
345 * In the worst case we need a new set of index blocks at
346 * every level of the inode's extent tree.
347 */
348 ei->i_da_metadata_calc_len = 1;
349 ei->i_da_metadata_calc_last_lblock = lblock;
350 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400351}
352
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400353static int
354ext4_ext_max_entries(struct inode *inode, int depth)
355{
356 int max;
357
358 if (depth == ext_depth(inode)) {
359 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400360 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400361 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400362 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400363 } else {
364 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400365 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400366 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400367 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400368 }
369
370 return max;
371}
372
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400373static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
374{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400375 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400376 int len = ext4_ext_get_actual_len(ext);
Eryu Guan5946d082013-12-03 21:22:21 -0500377 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
378 ext4_lblk_t last = lblock + len - 1;
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400379
Eryu Guan5946d082013-12-03 21:22:21 -0500380 if (lblock > last)
Theodore Ts'o31d4f3a2012-03-11 23:30:16 -0400381 return 0;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400382 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400383}
384
385static int ext4_valid_extent_idx(struct inode *inode,
386 struct ext4_extent_idx *ext_idx)
387{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400388 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400389
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400390 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400391}
392
393static int ext4_valid_extent_entries(struct inode *inode,
394 struct ext4_extent_header *eh,
395 int depth)
396{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400397 unsigned short entries;
398 if (eh->eh_entries == 0)
399 return 1;
400
401 entries = le16_to_cpu(eh->eh_entries);
402
403 if (depth == 0) {
404 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400405 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Eryu Guan5946d082013-12-03 21:22:21 -0500406 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
407 ext4_fsblk_t pblock = 0;
408 ext4_lblk_t lblock = 0;
409 ext4_lblk_t prev = 0;
410 int len = 0;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400411 while (entries) {
412 if (!ext4_valid_extent(inode, ext))
413 return 0;
Eryu Guan5946d082013-12-03 21:22:21 -0500414
415 /* Check for overlapping extents */
416 lblock = le32_to_cpu(ext->ee_block);
417 len = ext4_ext_get_actual_len(ext);
418 if ((lblock <= prev) && prev) {
419 pblock = ext4_ext_pblock(ext);
420 es->s_last_error_block = cpu_to_le64(pblock);
421 return 0;
422 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400423 ext++;
424 entries--;
Eryu Guan5946d082013-12-03 21:22:21 -0500425 prev = lblock + len - 1;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400426 }
427 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400428 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400429 while (entries) {
430 if (!ext4_valid_extent_idx(inode, ext_idx))
431 return 0;
432 ext_idx++;
433 entries--;
434 }
435 }
436 return 1;
437}
438
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400439static int __ext4_ext_check(const char *function, unsigned int line,
440 struct inode *inode, struct ext4_extent_header *eh,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400441 int depth, ext4_fsblk_t pblk)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400442{
443 const char *error_msg;
444 int max = 0;
445
446 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
447 error_msg = "invalid magic";
448 goto corrupted;
449 }
450 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
451 error_msg = "unexpected eh_depth";
452 goto corrupted;
453 }
454 if (unlikely(eh->eh_max == 0)) {
455 error_msg = "invalid eh_max";
456 goto corrupted;
457 }
458 max = ext4_ext_max_entries(inode, depth);
459 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
460 error_msg = "too large eh_max";
461 goto corrupted;
462 }
463 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
464 error_msg = "invalid eh_entries";
465 goto corrupted;
466 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400467 if (!ext4_valid_extent_entries(inode, eh, depth)) {
468 error_msg = "invalid extent entries";
469 goto corrupted;
470 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400471 /* Verify checksum on non-root extent tree nodes */
472 if (ext_depth(inode) != depth &&
473 !ext4_extent_block_csum_verify(inode, eh)) {
474 error_msg = "extent tree corrupted";
475 goto corrupted;
476 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400477 return 0;
478
479corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400480 ext4_error_inode(inode, function, line, 0,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400481 "pblk %llu bad header/extent: %s - magic %x, "
482 "entries %u, max %u(%u), depth %u(%u)",
483 (unsigned long long) pblk, error_msg,
484 le16_to_cpu(eh->eh_magic),
485 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
486 max, le16_to_cpu(eh->eh_depth), depth);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400487 return -EIO;
488}
489
Theodore Ts'oc3491792013-08-16 21:21:41 -0400490#define ext4_ext_check(inode, eh, depth, pblk) \
491 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400492
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400493int ext4_ext_check_inode(struct inode *inode)
494{
Theodore Ts'oc3491792013-08-16 21:21:41 -0400495 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400496}
497
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400498static struct buffer_head *
499__read_extent_tree_block(const char *function, unsigned int line,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400500 struct inode *inode, ext4_fsblk_t pblk, int depth,
501 int flags)
Darrick J. Wongf8489122012-04-29 18:21:10 -0400502{
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400503 struct buffer_head *bh;
504 int err;
Darrick J. Wongf8489122012-04-29 18:21:10 -0400505
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400506 bh = sb_getblk(inode->i_sb, pblk);
507 if (unlikely(!bh))
508 return ERR_PTR(-ENOMEM);
509
510 if (!bh_uptodate_or_lock(bh)) {
511 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
512 err = bh_submit_read(bh);
513 if (err < 0)
514 goto errout;
515 }
Theodore Ts'o7869a4a2013-08-16 22:05:14 -0400516 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400517 return bh;
518 err = __ext4_ext_check(function, line, inode,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400519 ext_block_hdr(bh), depth, pblk);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400520 if (err)
521 goto errout;
Darrick J. Wongf8489122012-04-29 18:21:10 -0400522 set_buffer_verified(bh);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400523 /*
524 * If this is a leaf block, cache all of its entries
525 */
526 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
527 struct ext4_extent_header *eh = ext_block_hdr(bh);
528 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
529 ext4_lblk_t prev = 0;
530 int i;
531
532 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
533 unsigned int status = EXTENT_STATUS_WRITTEN;
534 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
535 int len = ext4_ext_get_actual_len(ex);
536
537 if (prev && (prev != lblk))
538 ext4_es_cache_extent(inode, prev,
539 lblk - prev, ~0,
540 EXTENT_STATUS_HOLE);
541
Lukas Czerner556615d2014-04-20 23:45:47 -0400542 if (ext4_ext_is_unwritten(ex))
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400543 status = EXTENT_STATUS_UNWRITTEN;
544 ext4_es_cache_extent(inode, lblk, len,
545 ext4_ext_pblock(ex), status);
546 prev = lblk + len;
547 }
548 }
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400549 return bh;
550errout:
551 put_bh(bh);
552 return ERR_PTR(err);
553
Darrick J. Wongf8489122012-04-29 18:21:10 -0400554}
555
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400556#define read_extent_tree_block(inode, pblk, depth, flags) \
557 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
558 (depth), (flags))
Darrick J. Wongf8489122012-04-29 18:21:10 -0400559
Theodore Ts'o7869a4a2013-08-16 22:05:14 -0400560/*
561 * This function is called to cache a file's extent information in the
562 * extent status tree
563 */
564int ext4_ext_precache(struct inode *inode)
565{
566 struct ext4_inode_info *ei = EXT4_I(inode);
567 struct ext4_ext_path *path = NULL;
568 struct buffer_head *bh;
569 int i = 0, depth, ret = 0;
570
571 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
572 return 0; /* not an extent-mapped inode */
573
574 down_read(&ei->i_data_sem);
575 depth = ext_depth(inode);
576
577 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
578 GFP_NOFS);
579 if (path == NULL) {
580 up_read(&ei->i_data_sem);
581 return -ENOMEM;
582 }
583
584 /* Don't cache anything if there are no external extent blocks */
585 if (depth == 0)
586 goto out;
587 path[0].p_hdr = ext_inode_hdr(inode);
588 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
589 if (ret)
590 goto out;
591 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
592 while (i >= 0) {
593 /*
594 * If this is a leaf block or we've reached the end of
595 * the index block, go up
596 */
597 if ((i == depth) ||
598 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
599 brelse(path[i].p_bh);
600 path[i].p_bh = NULL;
601 i--;
602 continue;
603 }
604 bh = read_extent_tree_block(inode,
605 ext4_idx_pblock(path[i].p_idx++),
606 depth - i - 1,
607 EXT4_EX_FORCE_CACHE);
608 if (IS_ERR(bh)) {
609 ret = PTR_ERR(bh);
610 break;
611 }
612 i++;
613 path[i].p_bh = bh;
614 path[i].p_hdr = ext_block_hdr(bh);
615 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
616 }
617 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
618out:
619 up_read(&ei->i_data_sem);
620 ext4_ext_drop_refs(path);
621 kfree(path);
622 return ret;
623}
624
Alex Tomasa86c6182006-10-11 01:21:03 -0700625#ifdef EXT_DEBUG
626static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
627{
628 int k, l = path->p_depth;
629
630 ext_debug("path:");
631 for (k = 0; k <= l; k++, path++) {
632 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700633 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400634 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700635 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400636 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700637 le32_to_cpu(path->p_ext->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -0400638 ext4_ext_is_unwritten(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400639 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400640 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700641 } else
642 ext_debug(" []");
643 }
644 ext_debug("\n");
645}
646
647static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
648{
649 int depth = ext_depth(inode);
650 struct ext4_extent_header *eh;
651 struct ext4_extent *ex;
652 int i;
653
654 if (!path)
655 return;
656
657 eh = path[depth].p_hdr;
658 ex = EXT_FIRST_EXTENT(eh);
659
Mingming553f9002009-09-18 13:34:55 -0400660 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
661
Alex Tomasa86c6182006-10-11 01:21:03 -0700662 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400663 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -0400664 ext4_ext_is_unwritten(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400665 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700666 }
667 ext_debug("\n");
668}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400669
670static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
671 ext4_fsblk_t newblock, int level)
672{
673 int depth = ext_depth(inode);
674 struct ext4_extent *ex;
675
676 if (depth != level) {
677 struct ext4_extent_idx *idx;
678 idx = path[level].p_idx;
679 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
680 ext_debug("%d: move %d:%llu in new index %llu\n", level,
681 le32_to_cpu(idx->ei_block),
682 ext4_idx_pblock(idx),
683 newblock);
684 idx++;
685 }
686
687 return;
688 }
689
690 ex = path[depth].p_ext;
691 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
692 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
693 le32_to_cpu(ex->ee_block),
694 ext4_ext_pblock(ex),
Lukas Czerner556615d2014-04-20 23:45:47 -0400695 ext4_ext_is_unwritten(ex),
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400696 ext4_ext_get_actual_len(ex),
697 newblock);
698 ex++;
699 }
700}
701
Alex Tomasa86c6182006-10-11 01:21:03 -0700702#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400703#define ext4_ext_show_path(inode, path)
704#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400705#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700706#endif
707
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500708void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700709{
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -0400710 int depth, i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700711
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -0400712 if (!path)
713 return;
714 depth = path->p_depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700715 for (i = 0; i <= depth; i++, path++)
716 if (path->p_bh) {
717 brelse(path->p_bh);
718 path->p_bh = NULL;
719 }
720}
721
722/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700723 * ext4_ext_binsearch_idx:
724 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400725 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700726 */
727static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500728ext4_ext_binsearch_idx(struct inode *inode,
729 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700730{
731 struct ext4_extent_header *eh = path->p_hdr;
732 struct ext4_extent_idx *r, *l, *m;
733
Alex Tomasa86c6182006-10-11 01:21:03 -0700734
Eric Sandeenbba90742008-01-28 23:58:27 -0500735 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700736
737 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400738 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700739 while (l <= r) {
740 m = l + (r - l) / 2;
741 if (block < le32_to_cpu(m->ei_block))
742 r = m - 1;
743 else
744 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400745 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
746 m, le32_to_cpu(m->ei_block),
747 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700748 }
749
750 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400751 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400752 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700753
754#ifdef CHECK_BINSEARCH
755 {
756 struct ext4_extent_idx *chix, *ix;
757 int k;
758
759 chix = ix = EXT_FIRST_INDEX(eh);
760 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
761 if (k != 0 &&
762 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400763 printk(KERN_DEBUG "k=%d, ix=0x%p, "
764 "first=0x%p\n", k,
765 ix, EXT_FIRST_INDEX(eh));
766 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700767 le32_to_cpu(ix->ei_block),
768 le32_to_cpu(ix[-1].ei_block));
769 }
770 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400771 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700772 if (block < le32_to_cpu(ix->ei_block))
773 break;
774 chix = ix;
775 }
776 BUG_ON(chix != path->p_idx);
777 }
778#endif
779
780}
781
782/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700783 * ext4_ext_binsearch:
784 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400785 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700786 */
787static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500788ext4_ext_binsearch(struct inode *inode,
789 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700790{
791 struct ext4_extent_header *eh = path->p_hdr;
792 struct ext4_extent *r, *l, *m;
793
Alex Tomasa86c6182006-10-11 01:21:03 -0700794 if (eh->eh_entries == 0) {
795 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700796 * this leaf is empty:
797 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700798 */
799 return;
800 }
801
Eric Sandeenbba90742008-01-28 23:58:27 -0500802 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700803
804 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400805 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700806
807 while (l <= r) {
808 m = l + (r - l) / 2;
809 if (block < le32_to_cpu(m->ee_block))
810 r = m - 1;
811 else
812 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400813 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
814 m, le32_to_cpu(m->ee_block),
815 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700816 }
817
818 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400819 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400820 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400821 ext4_ext_pblock(path->p_ext),
Lukas Czerner556615d2014-04-20 23:45:47 -0400822 ext4_ext_is_unwritten(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400823 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700824
825#ifdef CHECK_BINSEARCH
826 {
827 struct ext4_extent *chex, *ex;
828 int k;
829
830 chex = ex = EXT_FIRST_EXTENT(eh);
831 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
832 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400833 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700834 if (block < le32_to_cpu(ex->ee_block))
835 break;
836 chex = ex;
837 }
838 BUG_ON(chex != path->p_ext);
839 }
840#endif
841
842}
843
844int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
845{
846 struct ext4_extent_header *eh;
847
848 eh = ext_inode_hdr(inode);
849 eh->eh_depth = 0;
850 eh->eh_entries = 0;
851 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400852 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700853 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700854 return 0;
855}
856
857struct ext4_ext_path *
Theodore Ts'oed8a1a72014-09-01 14:43:09 -0400858ext4_find_extent(struct inode *inode, ext4_lblk_t block,
859 struct ext4_ext_path **orig_path, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700860{
861 struct ext4_extent_header *eh;
862 struct buffer_head *bh;
Theodore Ts'o705912c2014-09-01 14:34:09 -0400863 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
864 short int depth, i, ppos = 0;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500865 int ret;
Alex Tomasa86c6182006-10-11 01:21:03 -0700866
867 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400868 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700869
Theodore Ts'o10809df82014-09-01 14:40:09 -0400870 if (path) {
Theodore Ts'o523f4312014-09-01 14:38:09 -0400871 ext4_ext_drop_refs(path);
Theodore Ts'o10809df82014-09-01 14:40:09 -0400872 if (depth > path[0].p_maxdepth) {
873 kfree(path);
874 *orig_path = path = NULL;
875 }
876 }
877 if (!path) {
Theodore Ts'o523f4312014-09-01 14:38:09 -0400878 /* account possible depth increase */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800879 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700880 GFP_NOFS);
Theodore Ts'o19008f62014-08-31 15:03:14 -0400881 if (unlikely(!path))
Alex Tomasa86c6182006-10-11 01:21:03 -0700882 return ERR_PTR(-ENOMEM);
Theodore Ts'o10809df82014-09-01 14:40:09 -0400883 path[0].p_maxdepth = depth + 1;
Alex Tomasa86c6182006-10-11 01:21:03 -0700884 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700885 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400886 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700887
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400888 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700889 /* walk through the tree */
890 while (i) {
891 ext_debug("depth %d: num %d, max %d\n",
892 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400893
Alex Tomasa86c6182006-10-11 01:21:03 -0700894 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400895 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700896 path[ppos].p_depth = i;
897 path[ppos].p_ext = NULL;
898
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400899 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
900 flags);
Theodore Ts'o19008f62014-08-31 15:03:14 -0400901 if (unlikely(IS_ERR(bh))) {
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400902 ret = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700903 goto err;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500904 }
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400905
Alex Tomasa86c6182006-10-11 01:21:03 -0700906 eh = ext_block_hdr(bh);
907 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500908 if (unlikely(ppos > depth)) {
909 put_bh(bh);
910 EXT4_ERROR_INODE(inode,
911 "ppos %d > depth %d", ppos, depth);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500912 ret = -EIO;
Frank Mayhar273df552010-03-02 11:46:09 -0500913 goto err;
914 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700915 path[ppos].p_bh = bh;
916 path[ppos].p_hdr = eh;
Alex Tomasa86c6182006-10-11 01:21:03 -0700917 }
918
919 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700920 path[ppos].p_ext = NULL;
921 path[ppos].p_idx = NULL;
922
Alex Tomasa86c6182006-10-11 01:21:03 -0700923 /* find extent */
924 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400925 /* if not an empty leaf */
926 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400927 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700928
929 ext4_ext_show_path(inode, path);
930
931 return path;
932
933err:
934 ext4_ext_drop_refs(path);
Theodore Ts'odfe50802014-09-01 14:37:09 -0400935 kfree(path);
936 if (orig_path)
937 *orig_path = NULL;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500938 return ERR_PTR(ret);
Alex Tomasa86c6182006-10-11 01:21:03 -0700939}
940
941/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700942 * ext4_ext_insert_index:
943 * insert new index [@logical;@ptr] into the block at @curp;
944 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700945 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400946static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
947 struct ext4_ext_path *curp,
948 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700949{
950 struct ext4_extent_idx *ix;
951 int len, err;
952
Avantika Mathur7e028972006-12-06 20:41:33 -0800953 err = ext4_ext_get_access(handle, inode, curp);
954 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700955 return err;
956
Frank Mayhar273df552010-03-02 11:46:09 -0500957 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
958 EXT4_ERROR_INODE(inode,
959 "logical %d == ei_block %d!",
960 logical, le32_to_cpu(curp->p_idx->ei_block));
961 return -EIO;
962 }
Robin Dongd4620312011-07-17 23:43:42 -0400963
964 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
965 >= le16_to_cpu(curp->p_hdr->eh_max))) {
966 EXT4_ERROR_INODE(inode,
967 "eh_entries %d >= eh_max %d!",
968 le16_to_cpu(curp->p_hdr->eh_entries),
969 le16_to_cpu(curp->p_hdr->eh_max));
970 return -EIO;
971 }
972
Alex Tomasa86c6182006-10-11 01:21:03 -0700973 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
974 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400975 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700976 ix = curp->p_idx + 1;
977 } else {
978 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400979 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700980 ix = curp->p_idx;
981 }
982
Eric Gouriou80e675f2011-10-27 11:52:18 -0400983 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
984 BUG_ON(len < 0);
985 if (len > 0) {
986 ext_debug("insert new index %d: "
987 "move %d indices from 0x%p to 0x%p\n",
988 logical, len, ix, ix + 1);
989 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
990 }
991
Tao Maf472e022011-10-17 10:13:46 -0400992 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
993 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
994 return -EIO;
995 }
996
Alex Tomasa86c6182006-10-11 01:21:03 -0700997 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700998 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400999 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001000
Frank Mayhar273df552010-03-02 11:46:09 -05001001 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1002 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1003 return -EIO;
1004 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001005
1006 err = ext4_ext_dirty(handle, inode, curp);
1007 ext4_std_error(inode->i_sb, err);
1008
1009 return err;
1010}
1011
1012/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001013 * ext4_ext_split:
1014 * inserts new subtree into the path, using free index entry
1015 * at depth @at:
1016 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1017 * - makes decision where to split
1018 * - moves remaining extents and index entries (right to the split point)
1019 * into the newly allocated blocks
1020 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -07001021 */
1022static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001023 unsigned int flags,
1024 struct ext4_ext_path *path,
1025 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -07001026{
1027 struct buffer_head *bh = NULL;
1028 int depth = ext_depth(inode);
1029 struct ext4_extent_header *neh;
1030 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -07001031 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001032 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001033 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001034 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -07001035 int err = 0;
1036
1037 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001038 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07001039
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001040 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -07001041 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -05001042 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1043 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1044 return -EIO;
1045 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001046 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1047 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001048 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -07001049 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001050 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -07001051 } else {
1052 border = newext->ee_block;
1053 ext_debug("leaf will be added."
1054 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001055 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -07001056 }
1057
1058 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001059 * If error occurs, then we break processing
1060 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -07001061 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001062 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -07001063 */
1064
1065 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001066 * Get array to track all allocated blocks.
1067 * We need this to handle errors and free blocks
1068 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -07001069 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -08001070 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07001071 if (!ablocks)
1072 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001073
1074 /* allocate all needed blocks */
1075 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1076 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -04001077 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -04001078 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001079 if (newblock == 0)
1080 goto cleanup;
1081 ablocks[a] = newblock;
1082 }
1083
1084 /* initialize new leaf */
1085 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -05001086 if (unlikely(newblock == 0)) {
1087 EXT4_ERROR_INODE(inode, "newblock == 0!");
1088 err = -EIO;
1089 goto cleanup;
1090 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001091 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001092 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001093 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001094 goto cleanup;
1095 }
1096 lock_buffer(bh);
1097
Avantika Mathur7e028972006-12-06 20:41:33 -08001098 err = ext4_journal_get_create_access(handle, bh);
1099 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001100 goto cleanup;
1101
1102 neh = ext_block_hdr(bh);
1103 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001104 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001105 neh->eh_magic = EXT4_EXT_MAGIC;
1106 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001107
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001108 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -05001109 if (unlikely(path[depth].p_hdr->eh_entries !=
1110 path[depth].p_hdr->eh_max)) {
1111 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1112 path[depth].p_hdr->eh_entries,
1113 path[depth].p_hdr->eh_max);
1114 err = -EIO;
1115 goto cleanup;
1116 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001117 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001118 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1119 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07001120 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001121 struct ext4_extent *ex;
1122 ex = EXT_FIRST_EXTENT(neh);
1123 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001124 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001125 }
1126
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001127 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001128 set_buffer_uptodate(bh);
1129 unlock_buffer(bh);
1130
Frank Mayhar03901312009-01-07 00:06:22 -05001131 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001132 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001133 goto cleanup;
1134 brelse(bh);
1135 bh = NULL;
1136
1137 /* correct old leaf */
1138 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001139 err = ext4_ext_get_access(handle, inode, path + depth);
1140 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001141 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001142 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001143 err = ext4_ext_dirty(handle, inode, path + depth);
1144 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001145 goto cleanup;
1146
1147 }
1148
1149 /* create intermediate indexes */
1150 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001151 if (unlikely(k < 0)) {
1152 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1153 err = -EIO;
1154 goto cleanup;
1155 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001156 if (k)
1157 ext_debug("create %d intermediate indices\n", k);
1158 /* insert new index into current index block */
1159 /* current depth stored in i var */
1160 i = depth - 1;
1161 while (k--) {
1162 oldblock = newblock;
1163 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001164 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001165 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001166 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001167 goto cleanup;
1168 }
1169 lock_buffer(bh);
1170
Avantika Mathur7e028972006-12-06 20:41:33 -08001171 err = ext4_journal_get_create_access(handle, bh);
1172 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001173 goto cleanup;
1174
1175 neh = ext_block_hdr(bh);
1176 neh->eh_entries = cpu_to_le16(1);
1177 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001178 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001179 neh->eh_depth = cpu_to_le16(depth - i);
1180 fidx = EXT_FIRST_INDEX(neh);
1181 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001182 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001183
Eric Sandeenbba90742008-01-28 23:58:27 -05001184 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1185 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001186
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001187 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001188 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1189 EXT_LAST_INDEX(path[i].p_hdr))) {
1190 EXT4_ERROR_INODE(inode,
1191 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1192 le32_to_cpu(path[i].p_ext->ee_block));
1193 err = -EIO;
1194 goto cleanup;
1195 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001196 /* start copy indexes */
1197 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1198 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1199 EXT_MAX_INDEX(path[i].p_hdr));
1200 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001201 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001202 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001203 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001204 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001205 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001206 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001207 set_buffer_uptodate(bh);
1208 unlock_buffer(bh);
1209
Frank Mayhar03901312009-01-07 00:06:22 -05001210 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001211 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001212 goto cleanup;
1213 brelse(bh);
1214 bh = NULL;
1215
1216 /* correct old index */
1217 if (m) {
1218 err = ext4_ext_get_access(handle, inode, path + i);
1219 if (err)
1220 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001221 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001222 err = ext4_ext_dirty(handle, inode, path + i);
1223 if (err)
1224 goto cleanup;
1225 }
1226
1227 i--;
1228 }
1229
1230 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001231 err = ext4_ext_insert_index(handle, inode, path + at,
1232 le32_to_cpu(border), newblock);
1233
1234cleanup:
1235 if (bh) {
1236 if (buffer_locked(bh))
1237 unlock_buffer(bh);
1238 brelse(bh);
1239 }
1240
1241 if (err) {
1242 /* free all allocated blocks in error case */
1243 for (i = 0; i < depth; i++) {
1244 if (!ablocks[i])
1245 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001246 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001247 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001248 }
1249 }
1250 kfree(ablocks);
1251
1252 return err;
1253}
1254
1255/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001256 * ext4_ext_grow_indepth:
1257 * implements tree growing procedure:
1258 * - allocates new block
1259 * - moves top-level data (index block or leaf) into the new block
1260 * - initializes new top-level, creating index that points to the
1261 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001262 */
1263static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Dmitry Monakhovbe5cd902014-10-01 22:57:09 -04001264 unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07001265{
Alex Tomasa86c6182006-10-11 01:21:03 -07001266 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001267 struct buffer_head *bh;
Dmitry Monakhovbe5cd902014-10-01 22:57:09 -04001268 ext4_fsblk_t newblock, goal = 0;
1269 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
Alex Tomasa86c6182006-10-11 01:21:03 -07001270 int err = 0;
1271
Dmitry Monakhovbe5cd902014-10-01 22:57:09 -04001272 /* Try to prepend new index to old one */
1273 if (ext_depth(inode))
1274 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1275 if (goal > le32_to_cpu(es->s_first_data_block)) {
1276 flags |= EXT4_MB_HINT_TRY_GOAL;
1277 goal--;
1278 } else
1279 goal = ext4_inode_to_goal_block(inode);
1280 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1281 NULL, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07001282 if (newblock == 0)
1283 return err;
1284
1285 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001286 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001287 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001288 lock_buffer(bh);
1289
Avantika Mathur7e028972006-12-06 20:41:33 -08001290 err = ext4_journal_get_create_access(handle, bh);
1291 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001292 unlock_buffer(bh);
1293 goto out;
1294 }
1295
1296 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001297 memmove(bh->b_data, EXT4_I(inode)->i_data,
1298 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001299
1300 /* set size of new block */
1301 neh = ext_block_hdr(bh);
1302 /* old root could have indexes or leaves
1303 * so calculate e_max right way */
1304 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001305 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001306 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001307 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001308 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001309 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001310 set_buffer_uptodate(bh);
1311 unlock_buffer(bh);
1312
Frank Mayhar03901312009-01-07 00:06:22 -05001313 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001314 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001315 goto out;
1316
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001317 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001318 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001319 neh->eh_entries = cpu_to_le16(1);
1320 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1321 if (neh->eh_depth == 0) {
1322 /* Root extent block becomes index block */
1323 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1324 EXT_FIRST_INDEX(neh)->ei_block =
1325 EXT_FIRST_EXTENT(neh)->ee_block;
1326 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001327 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001328 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001329 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001330 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001331
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001332 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001333 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001334out:
1335 brelse(bh);
1336
1337 return err;
1338}
1339
1340/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001341 * ext4_ext_create_new_leaf:
1342 * finds empty index and adds new leaf.
1343 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001344 */
1345static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001346 unsigned int mb_flags,
1347 unsigned int gb_flags,
Theodore Ts'odfe50802014-09-01 14:37:09 -04001348 struct ext4_ext_path **ppath,
Allison Henderson55f020d2011-05-25 07:41:26 -04001349 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001350{
Theodore Ts'odfe50802014-09-01 14:37:09 -04001351 struct ext4_ext_path *path = *ppath;
Alex Tomasa86c6182006-10-11 01:21:03 -07001352 struct ext4_ext_path *curp;
1353 int depth, i, err = 0;
1354
1355repeat:
1356 i = depth = ext_depth(inode);
1357
1358 /* walk up to the tree and look for free index entry */
1359 curp = path + depth;
1360 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1361 i--;
1362 curp--;
1363 }
1364
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001365 /* we use already allocated block for index block,
1366 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001367 if (EXT_HAS_FREE_INDEX(curp)) {
1368 /* if we found index with free entry, then use that
1369 * entry: create all needed subtree and add new leaf */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001370 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001371 if (err)
1372 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001373
1374 /* refill path */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04001375 path = ext4_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001376 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
Theodore Ts'odfe50802014-09-01 14:37:09 -04001377 ppath, gb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001378 if (IS_ERR(path))
1379 err = PTR_ERR(path);
1380 } else {
1381 /* tree is full, time to grow in depth */
Dmitry Monakhovbe5cd902014-10-01 22:57:09 -04001382 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001383 if (err)
1384 goto out;
1385
1386 /* refill path */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04001387 path = ext4_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001388 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
Theodore Ts'odfe50802014-09-01 14:37:09 -04001389 ppath, gb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001390 if (IS_ERR(path)) {
1391 err = PTR_ERR(path);
1392 goto out;
1393 }
1394
1395 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001396 * only first (depth 0 -> 1) produces free space;
1397 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001398 */
1399 depth = ext_depth(inode);
1400 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001401 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001402 goto repeat;
1403 }
1404 }
1405
1406out:
1407 return err;
1408}
1409
1410/*
Alex Tomas1988b512008-01-28 23:58:27 -05001411 * search the closest allocated block to the left for *logical
1412 * and returns it at @logical + it's physical address at @phys
1413 * if *logical is the smallest allocated block, the function
1414 * returns 0 at @phys
1415 * return value contains 0 (success) or error code
1416 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001417static int ext4_ext_search_left(struct inode *inode,
1418 struct ext4_ext_path *path,
1419 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001420{
1421 struct ext4_extent_idx *ix;
1422 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001423 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001424
Frank Mayhar273df552010-03-02 11:46:09 -05001425 if (unlikely(path == NULL)) {
1426 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1427 return -EIO;
1428 }
Alex Tomas1988b512008-01-28 23:58:27 -05001429 depth = path->p_depth;
1430 *phys = 0;
1431
1432 if (depth == 0 && path->p_ext == NULL)
1433 return 0;
1434
1435 /* usually extent in the path covers blocks smaller
1436 * then *logical, but it can be that extent is the
1437 * first one in the file */
1438
1439 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001440 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001441 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001442 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1443 EXT4_ERROR_INODE(inode,
1444 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1445 *logical, le32_to_cpu(ex->ee_block));
1446 return -EIO;
1447 }
Alex Tomas1988b512008-01-28 23:58:27 -05001448 while (--depth >= 0) {
1449 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001450 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1451 EXT4_ERROR_INODE(inode,
1452 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001453 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001454 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001455 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001456 depth);
1457 return -EIO;
1458 }
Alex Tomas1988b512008-01-28 23:58:27 -05001459 }
1460 return 0;
1461 }
1462
Frank Mayhar273df552010-03-02 11:46:09 -05001463 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1464 EXT4_ERROR_INODE(inode,
1465 "logical %d < ee_block %d + ee_len %d!",
1466 *logical, le32_to_cpu(ex->ee_block), ee_len);
1467 return -EIO;
1468 }
Alex Tomas1988b512008-01-28 23:58:27 -05001469
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001470 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001471 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001472 return 0;
1473}
1474
1475/*
1476 * search the closest allocated block to the right for *logical
1477 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001478 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001479 * returns 0 at @phys
1480 * return value contains 0 (success) or error code
1481 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001482static int ext4_ext_search_right(struct inode *inode,
1483 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001484 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1485 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001486{
1487 struct buffer_head *bh = NULL;
1488 struct ext4_extent_header *eh;
1489 struct ext4_extent_idx *ix;
1490 struct ext4_extent *ex;
1491 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001492 int depth; /* Note, NOT eh_depth; depth from top of tree */
1493 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001494
Frank Mayhar273df552010-03-02 11:46:09 -05001495 if (unlikely(path == NULL)) {
1496 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1497 return -EIO;
1498 }
Alex Tomas1988b512008-01-28 23:58:27 -05001499 depth = path->p_depth;
1500 *phys = 0;
1501
1502 if (depth == 0 && path->p_ext == NULL)
1503 return 0;
1504
1505 /* usually extent in the path covers blocks smaller
1506 * then *logical, but it can be that extent is the
1507 * first one in the file */
1508
1509 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001510 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001511 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001512 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1513 EXT4_ERROR_INODE(inode,
1514 "first_extent(path[%d].p_hdr) != ex",
1515 depth);
1516 return -EIO;
1517 }
Alex Tomas1988b512008-01-28 23:58:27 -05001518 while (--depth >= 0) {
1519 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001520 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1521 EXT4_ERROR_INODE(inode,
1522 "ix != EXT_FIRST_INDEX *logical %d!",
1523 *logical);
1524 return -EIO;
1525 }
Alex Tomas1988b512008-01-28 23:58:27 -05001526 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001527 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001528 }
1529
Frank Mayhar273df552010-03-02 11:46:09 -05001530 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1531 EXT4_ERROR_INODE(inode,
1532 "logical %d < ee_block %d + ee_len %d!",
1533 *logical, le32_to_cpu(ex->ee_block), ee_len);
1534 return -EIO;
1535 }
Alex Tomas1988b512008-01-28 23:58:27 -05001536
1537 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1538 /* next allocated block in this leaf */
1539 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001540 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001541 }
1542
1543 /* go up and search for index to the right */
1544 while (--depth >= 0) {
1545 ix = path[depth].p_idx;
1546 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001547 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001548 }
1549
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001550 /* we've gone up to the root and found no index to the right */
1551 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001552
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001553got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001554 /* we've found index to the right, let's
1555 * follow it and find the closest allocated
1556 * block to the right */
1557 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001558 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001559 while (++depth < path->p_depth) {
Eric Sandeen395a87b2009-03-10 18:18:47 -04001560 /* subtract from p_depth to get proper eh_depth */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001561 bh = read_extent_tree_block(inode, block,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001562 path->p_depth - depth, 0);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001563 if (IS_ERR(bh))
1564 return PTR_ERR(bh);
1565 eh = ext_block_hdr(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001566 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001567 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001568 put_bh(bh);
1569 }
1570
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001571 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001572 if (IS_ERR(bh))
1573 return PTR_ERR(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001574 eh = ext_block_hdr(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001575 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001576found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001577 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001578 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001579 *ret_ex = ex;
1580 if (bh)
1581 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001582 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001583}
1584
1585/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001586 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001587 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001588 * NOTE: it considers block number from index entry as
1589 * allocated block. Thus, index entries have to be consistent
1590 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001591 */
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04001592ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001593ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1594{
1595 int depth;
1596
1597 BUG_ON(path == NULL);
1598 depth = path->p_depth;
1599
1600 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001601 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001602
1603 while (depth >= 0) {
1604 if (depth == path->p_depth) {
1605 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001606 if (path[depth].p_ext &&
1607 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001608 EXT_LAST_EXTENT(path[depth].p_hdr))
1609 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1610 } else {
1611 /* index */
1612 if (path[depth].p_idx !=
1613 EXT_LAST_INDEX(path[depth].p_hdr))
1614 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1615 }
1616 depth--;
1617 }
1618
Lukas Czernerf17722f2011-06-06 00:05:17 -04001619 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001620}
1621
1622/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001623 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001624 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001625 */
Robin Dong57187892011-07-23 21:49:07 -04001626static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001627{
1628 int depth;
1629
1630 BUG_ON(path == NULL);
1631 depth = path->p_depth;
1632
1633 /* zero-tree has no leaf blocks at all */
1634 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001635 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001636
1637 /* go to index block */
1638 depth--;
1639
1640 while (depth >= 0) {
1641 if (path[depth].p_idx !=
1642 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001643 return (ext4_lblk_t)
1644 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001645 depth--;
1646 }
1647
Lukas Czernerf17722f2011-06-06 00:05:17 -04001648 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001649}
1650
1651/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001652 * ext4_ext_correct_indexes:
1653 * if leaf gets modified and modified extent is first in the leaf,
1654 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001655 * TODO: do we need to correct tree in all cases?
1656 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001657static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001658 struct ext4_ext_path *path)
1659{
1660 struct ext4_extent_header *eh;
1661 int depth = ext_depth(inode);
1662 struct ext4_extent *ex;
1663 __le32 border;
1664 int k, err = 0;
1665
1666 eh = path[depth].p_hdr;
1667 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001668
1669 if (unlikely(ex == NULL || eh == NULL)) {
1670 EXT4_ERROR_INODE(inode,
1671 "ex %p == NULL or eh %p == NULL", ex, eh);
1672 return -EIO;
1673 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001674
1675 if (depth == 0) {
1676 /* there is no tree at all */
1677 return 0;
1678 }
1679
1680 if (ex != EXT_FIRST_EXTENT(eh)) {
1681 /* we correct tree if first leaf got modified only */
1682 return 0;
1683 }
1684
1685 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001686 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001687 */
1688 k = depth - 1;
1689 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001690 err = ext4_ext_get_access(handle, inode, path + k);
1691 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001692 return err;
1693 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001694 err = ext4_ext_dirty(handle, inode, path + k);
1695 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001696 return err;
1697
1698 while (k--) {
1699 /* change all left-side indexes */
1700 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1701 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001702 err = ext4_ext_get_access(handle, inode, path + k);
1703 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001704 break;
1705 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001706 err = ext4_ext_dirty(handle, inode, path + k);
1707 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001708 break;
1709 }
1710
1711 return err;
1712}
1713
Akira Fujita748de672009-06-17 19:24:03 -04001714int
Alex Tomasa86c6182006-10-11 01:21:03 -07001715ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1716 struct ext4_extent *ex2)
1717{
Eric Sandeenda0169b2013-11-04 09:58:26 -05001718 unsigned short ext1_ee_len, ext2_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001719
1720 /*
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001721 * Make sure that both extents are initialized. We don't merge
Lukas Czerner556615d2014-04-20 23:45:47 -04001722 * unwritten extents so that we can be sure that end_io code has
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001723 * the extent that was written properly split out and conversion to
1724 * initialized is trivial.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001725 */
Lukas Czerner556615d2014-04-20 23:45:47 -04001726 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001727 return 0;
1728
1729 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1730 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1731
1732 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001733 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001734 return 0;
1735
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001736 /*
1737 * To allow future support for preallocated extents to be added
1738 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001739 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001740 */
Eric Sandeenda0169b2013-11-04 09:58:26 -05001741 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001742 return 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04001743 if (ext4_ext_is_unwritten(ex1) &&
Darrick J. Wonga9b82412014-02-20 21:17:35 -05001744 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1745 atomic_read(&EXT4_I(inode)->i_unwritten) ||
Lukas Czerner556615d2014-04-20 23:45:47 -04001746 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
Darrick J. Wonga9b82412014-02-20 21:17:35 -05001747 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001748#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001749 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001750 return 0;
1751#endif
1752
Theodore Ts'obf89d162010-10-27 21:30:14 -04001753 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001754 return 1;
1755 return 0;
1756}
1757
1758/*
Amit Arora56055d32007-07-17 21:42:38 -04001759 * This function tries to merge the "ex" extent to the next extent in the tree.
1760 * It always tries to merge towards right. If you want to merge towards
1761 * left, pass "ex - 1" as argument instead of "ex".
1762 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1763 * 1 if they got merged.
1764 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001765static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001766 struct ext4_ext_path *path,
1767 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001768{
1769 struct ext4_extent_header *eh;
1770 unsigned int depth, len;
Lukas Czerner556615d2014-04-20 23:45:47 -04001771 int merge_done = 0, unwritten;
Amit Arora56055d32007-07-17 21:42:38 -04001772
1773 depth = ext_depth(inode);
1774 BUG_ON(path[depth].p_hdr == NULL);
1775 eh = path[depth].p_hdr;
1776
1777 while (ex < EXT_LAST_EXTENT(eh)) {
1778 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1779 break;
1780 /* merge with next extent! */
Lukas Czerner556615d2014-04-20 23:45:47 -04001781 unwritten = ext4_ext_is_unwritten(ex);
Amit Arora56055d32007-07-17 21:42:38 -04001782 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1783 + ext4_ext_get_actual_len(ex + 1));
Lukas Czerner556615d2014-04-20 23:45:47 -04001784 if (unwritten)
1785 ext4_ext_mark_unwritten(ex);
Amit Arora56055d32007-07-17 21:42:38 -04001786
1787 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1788 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1789 * sizeof(struct ext4_extent);
1790 memmove(ex + 1, ex + 2, len);
1791 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001792 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001793 merge_done = 1;
1794 WARN_ON(eh->eh_entries == 0);
1795 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001796 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001797 }
1798
1799 return merge_done;
1800}
1801
1802/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001803 * This function does a very simple check to see if we can collapse
1804 * an extent tree with a single extent tree leaf block into the inode.
1805 */
1806static void ext4_ext_try_to_merge_up(handle_t *handle,
1807 struct inode *inode,
1808 struct ext4_ext_path *path)
1809{
1810 size_t s;
1811 unsigned max_root = ext4_ext_space_root(inode, 0);
1812 ext4_fsblk_t blk;
1813
1814 if ((path[0].p_depth != 1) ||
1815 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1816 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1817 return;
1818
1819 /*
1820 * We need to modify the block allocation bitmap and the block
1821 * group descriptor to release the extent tree block. If we
1822 * can't get the journal credits, give up.
1823 */
1824 if (ext4_journal_extend(handle, 2))
1825 return;
1826
1827 /*
1828 * Copy the extent data up to the inode
1829 */
1830 blk = ext4_idx_pblock(path[0].p_idx);
1831 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1832 sizeof(struct ext4_extent_idx);
1833 s += sizeof(struct ext4_extent_header);
1834
Theodore Ts'o10809df82014-09-01 14:40:09 -04001835 path[1].p_maxdepth = path[0].p_maxdepth;
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001836 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1837 path[0].p_depth = 0;
1838 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1839 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1840 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1841
1842 brelse(path[1].p_bh);
1843 ext4_free_blocks(handle, inode, NULL, blk, 1,
Theodore Ts'o71d4f7d2014-07-15 06:02:38 -04001844 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001845}
1846
1847/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001848 * This function tries to merge the @ex extent to neighbours in the tree.
1849 * return 1 if merge left else 0.
1850 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001851static void ext4_ext_try_to_merge(handle_t *handle,
1852 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001853 struct ext4_ext_path *path,
1854 struct ext4_extent *ex) {
1855 struct ext4_extent_header *eh;
1856 unsigned int depth;
1857 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001858
1859 depth = ext_depth(inode);
1860 BUG_ON(path[depth].p_hdr == NULL);
1861 eh = path[depth].p_hdr;
1862
1863 if (ex > EXT_FIRST_EXTENT(eh))
1864 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1865
1866 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001867 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001868
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001869 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001870}
1871
1872/*
Amit Arora25d14f92007-05-24 13:04:13 -04001873 * check if a portion of the "newext" extent overlaps with an
1874 * existing extent.
1875 *
1876 * If there is an overlap discovered, it updates the length of the newext
1877 * such that there will be no overlap, and then returns 1.
1878 * If there is no overlap found, it returns 0.
1879 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001880static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1881 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001882 struct ext4_extent *newext,
1883 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001884{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001885 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001886 unsigned int depth, len1;
1887 unsigned int ret = 0;
1888
1889 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001890 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001891 depth = ext_depth(inode);
1892 if (!path[depth].p_ext)
1893 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001894 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
Amit Arora25d14f92007-05-24 13:04:13 -04001895
1896 /*
1897 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001898 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001899 */
1900 if (b2 < b1) {
1901 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001902 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001903 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001904 b2 = EXT4_LBLK_CMASK(sbi, b2);
Amit Arora25d14f92007-05-24 13:04:13 -04001905 }
1906
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001907 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001908 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001909 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001910 newext->ee_len = cpu_to_le16(len1);
1911 ret = 1;
1912 }
1913
1914 /* check for overlap */
1915 if (b1 + len1 > b2) {
1916 newext->ee_len = cpu_to_le16(b2 - b1);
1917 ret = 1;
1918 }
1919out:
1920 return ret;
1921}
1922
1923/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001924 * ext4_ext_insert_extent:
1925 * tries to merge requsted extent into the existing extent or
1926 * inserts requested extent as new one into the tree,
1927 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001928 */
1929int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -04001930 struct ext4_ext_path **ppath,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001931 struct ext4_extent *newext, int gb_flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07001932{
Theodore Ts'odfe50802014-09-01 14:37:09 -04001933 struct ext4_ext_path *path = *ppath;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001934 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001935 struct ext4_extent *ex, *fex;
1936 struct ext4_extent *nearex; /* nearest extent */
1937 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001938 int depth, len, err;
1939 ext4_lblk_t next;
Lukas Czerner556615d2014-04-20 23:45:47 -04001940 int mb_flags = 0, unwritten;
Alex Tomasa86c6182006-10-11 01:21:03 -07001941
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -04001942 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1943 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
Frank Mayhar273df552010-03-02 11:46:09 -05001944 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1945 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1946 return -EIO;
1947 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001948 depth = ext_depth(inode);
1949 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001950 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001951 if (unlikely(path[depth].p_hdr == NULL)) {
1952 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1953 return -EIO;
1954 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001955
1956 /* try to insert block into found extent and return */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001957 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001958
1959 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001960 * Try to see whether we should rather test the extent on
1961 * right from ex, or from the left of ex. This is because
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04001962 * ext4_find_extent() can return either extent on the
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001963 * left, or on the right from the searched position. This
1964 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001965 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001966 if (ex < EXT_LAST_EXTENT(eh) &&
1967 (le32_to_cpu(ex->ee_block) +
1968 ext4_ext_get_actual_len(ex) <
1969 le32_to_cpu(newext->ee_block))) {
1970 ex += 1;
1971 goto prepend;
1972 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1973 (le32_to_cpu(newext->ee_block) +
1974 ext4_ext_get_actual_len(newext) <
1975 le32_to_cpu(ex->ee_block)))
1976 ex -= 1;
1977
1978 /* Try to append newex to the ex */
1979 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1980 ext_debug("append [%d]%d block to %u:[%d]%d"
1981 "(from %llu)\n",
Lukas Czerner556615d2014-04-20 23:45:47 -04001982 ext4_ext_is_unwritten(newext),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001983 ext4_ext_get_actual_len(newext),
1984 le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04001985 ext4_ext_is_unwritten(ex),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001986 ext4_ext_get_actual_len(ex),
1987 ext4_ext_pblock(ex));
1988 err = ext4_ext_get_access(handle, inode,
1989 path + depth);
1990 if (err)
1991 return err;
Lukas Czerner556615d2014-04-20 23:45:47 -04001992 unwritten = ext4_ext_is_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001993 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001994 + ext4_ext_get_actual_len(newext));
Lukas Czerner556615d2014-04-20 23:45:47 -04001995 if (unwritten)
1996 ext4_ext_mark_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001997 eh = path[depth].p_hdr;
1998 nearex = ex;
1999 goto merge;
2000 }
2001
2002prepend:
2003 /* Try to prepend newex to the ex */
2004 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2005 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2006 "(from %llu)\n",
2007 le32_to_cpu(newext->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04002008 ext4_ext_is_unwritten(newext),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002009 ext4_ext_get_actual_len(newext),
2010 le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04002011 ext4_ext_is_unwritten(ex),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002012 ext4_ext_get_actual_len(ex),
2013 ext4_ext_pblock(ex));
2014 err = ext4_ext_get_access(handle, inode,
2015 path + depth);
2016 if (err)
2017 return err;
2018
Lukas Czerner556615d2014-04-20 23:45:47 -04002019 unwritten = ext4_ext_is_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002020 ex->ee_block = newext->ee_block;
2021 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2022 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2023 + ext4_ext_get_actual_len(newext));
Lukas Czerner556615d2014-04-20 23:45:47 -04002024 if (unwritten)
2025 ext4_ext_mark_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002026 eh = path[depth].p_hdr;
2027 nearex = ex;
2028 goto merge;
2029 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002030 }
2031
Alex Tomasa86c6182006-10-11 01:21:03 -07002032 depth = ext_depth(inode);
2033 eh = path[depth].p_hdr;
2034 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2035 goto has_space;
2036
2037 /* probably next leaf has space for us? */
2038 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04002039 next = EXT_MAX_BLOCKS;
2040 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04002041 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04002042 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002043 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07002044 BUG_ON(npath != NULL);
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04002045 npath = ext4_find_extent(inode, next, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002046 if (IS_ERR(npath))
2047 return PTR_ERR(npath);
2048 BUG_ON(npath->p_depth != path->p_depth);
2049 eh = npath[depth].p_hdr;
2050 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002051 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002052 le16_to_cpu(eh->eh_entries));
2053 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04002054 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07002055 }
2056 ext_debug("next leaf has no free space(%d,%d)\n",
2057 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2058 }
2059
2060 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002061 * There is no free space in the found leaf.
2062 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07002063 */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002064 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -04002065 mb_flags |= EXT4_MB_USE_RESERVED;
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002066 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
Theodore Ts'odfe50802014-09-01 14:37:09 -04002067 ppath, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07002068 if (err)
2069 goto cleanup;
2070 depth = ext_depth(inode);
2071 eh = path[depth].p_hdr;
2072
2073has_space:
2074 nearex = path[depth].p_ext;
2075
Avantika Mathur7e028972006-12-06 20:41:33 -08002076 err = ext4_ext_get_access(handle, inode, path + depth);
2077 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002078 goto cleanup;
2079
2080 if (!nearex) {
2081 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002082 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002083 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002084 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002085 ext4_ext_is_unwritten(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002086 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04002087 nearex = EXT_FIRST_EXTENT(eh);
2088 } else {
2089 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002090 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04002091 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002092 ext_debug("insert %u:%llu:[%d]%d before: "
2093 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002094 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002095 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002096 ext4_ext_is_unwritten(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002097 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002098 nearex);
2099 nearex++;
2100 } else {
2101 /* Insert before */
2102 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04002103 ext_debug("insert %u:%llu:[%d]%d after: "
2104 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04002105 le32_to_cpu(newext->ee_block),
2106 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002107 ext4_ext_is_unwritten(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002108 ext4_ext_get_actual_len(newext),
2109 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002110 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04002111 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2112 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002113 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04002114 "move %d extents from 0x%p to 0x%p\n",
2115 le32_to_cpu(newext->ee_block),
2116 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002117 ext4_ext_is_unwritten(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002118 ext4_ext_get_actual_len(newext),
2119 len, nearex, nearex + 1);
2120 memmove(nearex + 1, nearex,
2121 len * sizeof(struct ext4_extent));
2122 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002123 }
2124
Marcin Slusarze8546d02008-04-17 10:38:59 -04002125 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002126 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002127 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002128 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002129 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002130
2131merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002132 /* try to merge extents */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002133 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002134 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002135
Alex Tomasa86c6182006-10-11 01:21:03 -07002136
2137 /* time to correct all indexes above */
2138 err = ext4_ext_correct_indexes(handle, inode, path);
2139 if (err)
2140 goto cleanup;
2141
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002142 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002143
2144cleanup:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04002145 ext4_ext_drop_refs(npath);
2146 kfree(npath);
Alex Tomasa86c6182006-10-11 01:21:03 -07002147 return err;
2148}
2149
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002150static int ext4_fill_fiemap_extents(struct inode *inode,
2151 ext4_lblk_t block, ext4_lblk_t num,
2152 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002153{
2154 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002155 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002156 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002157 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002158 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002159 int exists, depth = 0, err = 0;
2160 unsigned int flags = 0;
2161 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002162
Lukas Czernerf17722f2011-06-06 00:05:17 -04002163 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002164 num = last - block;
2165 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002166 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002167
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04002168 path = ext4_find_extent(inode, block, &path, 0);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002169 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002170 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002171 err = PTR_ERR(path);
2172 path = NULL;
2173 break;
2174 }
2175
2176 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002177 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002178 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002179 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2180 err = -EIO;
2181 break;
2182 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002183 ex = path[depth].p_ext;
2184 next = ext4_ext_next_allocated_block(path);
2185
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002186 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002187 exists = 0;
2188 if (!ex) {
2189 /* there is no extent yet, so try to allocate
2190 * all requested space */
2191 start = block;
2192 end = block + num;
2193 } else if (le32_to_cpu(ex->ee_block) > block) {
2194 /* need to allocate space before found extent */
2195 start = block;
2196 end = le32_to_cpu(ex->ee_block);
2197 if (block + num < end)
2198 end = block + num;
2199 } else if (block >= le32_to_cpu(ex->ee_block)
2200 + ext4_ext_get_actual_len(ex)) {
2201 /* need to allocate space after found extent */
2202 start = block;
2203 end = block + num;
2204 if (end >= next)
2205 end = next;
2206 } else if (block >= le32_to_cpu(ex->ee_block)) {
2207 /*
2208 * some part of requested space is covered
2209 * by found extent
2210 */
2211 start = block;
2212 end = le32_to_cpu(ex->ee_block)
2213 + ext4_ext_get_actual_len(ex);
2214 if (block + num < end)
2215 end = block + num;
2216 exists = 1;
2217 } else {
2218 BUG();
2219 }
2220 BUG_ON(end <= start);
2221
2222 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002223 es.es_lblk = start;
2224 es.es_len = end - start;
2225 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002226 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002227 es.es_lblk = le32_to_cpu(ex->ee_block);
2228 es.es_len = ext4_ext_get_actual_len(ex);
2229 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner556615d2014-04-20 23:45:47 -04002230 if (ext4_ext_is_unwritten(ex))
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002231 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002232 }
2233
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002234 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002235 * Find delayed extent and update es accordingly. We call
2236 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002237 * last existing extent or not.
2238 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002239 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002240 if (!exists && next_del) {
2241 exists = 1;
Jie Liu72dac952013-06-12 23:13:59 -04002242 flags |= (FIEMAP_EXTENT_DELALLOC |
2243 FIEMAP_EXTENT_UNKNOWN);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002244 }
2245 up_read(&EXT4_I(inode)->i_data_sem);
2246
Zheng Liu69eb33d2013-02-18 00:31:07 -05002247 if (unlikely(es.es_len == 0)) {
2248 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002249 err = -EIO;
2250 break;
2251 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002252
Zheng Liuf7fec032013-02-18 00:28:47 -05002253 /*
2254 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2255 * we need to check next == EXT_MAX_BLOCKS because it is
2256 * possible that an extent is with unwritten and delayed
2257 * status due to when an extent is delayed allocated and
2258 * is allocated by fallocate status tree will track both of
2259 * them in a extent.
2260 *
2261 * So we could return a unwritten and delayed extent, and
2262 * its block is equal to 'next'.
2263 */
2264 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002265 flags |= FIEMAP_EXTENT_LAST;
2266 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2267 next != EXT_MAX_BLOCKS)) {
2268 EXT4_ERROR_INODE(inode,
2269 "next extent == %u, next "
2270 "delalloc extent = %u",
2271 next, next_del);
2272 err = -EIO;
2273 break;
2274 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002275 }
2276
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002277 if (exists) {
2278 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002279 (__u64)es.es_lblk << blksize_bits,
2280 (__u64)es.es_pblk << blksize_bits,
2281 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002282 flags);
2283 if (err < 0)
2284 break;
2285 if (err == 1) {
2286 err = 0;
2287 break;
2288 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002289 }
2290
Zheng Liu69eb33d2013-02-18 00:31:07 -05002291 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002292 }
2293
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04002294 ext4_ext_drop_refs(path);
2295 kfree(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002296 return err;
2297}
2298
Alex Tomasa86c6182006-10-11 01:21:03 -07002299/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002300 * ext4_ext_put_gap_in_cache:
2301 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002302 * and cache this gap
2303 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002304static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002305ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002306 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002307{
2308 int depth = ext_depth(inode);
Andi Shyti27b1b222013-08-28 14:00:00 -04002309 unsigned long len = 0;
2310 ext4_lblk_t lblock = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002311 struct ext4_extent *ex;
2312
2313 ex = path[depth].p_ext;
2314 if (ex == NULL) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002315 /*
2316 * there is no extent yet, so gap is [0;-] and we
2317 * don't cache it
2318 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002319 ext_debug("cache gap(whole file):");
2320 } else if (block < le32_to_cpu(ex->ee_block)) {
2321 lblock = block;
2322 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002323 ext_debug("cache gap(before): %u [%u:%u]",
2324 block,
2325 le32_to_cpu(ex->ee_block),
2326 ext4_ext_get_actual_len(ex));
Zheng Liud100eef2013-02-18 00:29:59 -05002327 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2328 ext4_es_insert_extent(inode, lblock, len, ~0,
2329 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002330 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002331 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002332 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002333 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002334 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002335
2336 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002337 ext_debug("cache gap(after): [%u:%u] %u",
2338 le32_to_cpu(ex->ee_block),
2339 ext4_ext_get_actual_len(ex),
2340 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002341 BUG_ON(next == lblock);
2342 len = next - lblock;
Zheng Liud100eef2013-02-18 00:29:59 -05002343 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2344 ext4_es_insert_extent(inode, lblock, len, ~0,
2345 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002346 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07002347 BUG();
2348 }
2349
Eric Sandeenbba90742008-01-28 23:58:27 -05002350 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002351}
2352
2353/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002354 * ext4_ext_rm_idx:
2355 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002356 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002357static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002358 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002359{
Alex Tomasa86c6182006-10-11 01:21:03 -07002360 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002361 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002362
2363 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002364 depth--;
2365 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002366 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002367 if (unlikely(path->p_hdr->eh_entries == 0)) {
2368 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2369 return -EIO;
2370 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002371 err = ext4_ext_get_access(handle, inode, path);
2372 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002373 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002374
2375 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2376 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2377 len *= sizeof(struct ext4_extent_idx);
2378 memmove(path->p_idx, path->p_idx + 1, len);
2379 }
2380
Marcin Slusarze8546d02008-04-17 10:38:59 -04002381 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002382 err = ext4_ext_dirty(handle, inode, path);
2383 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002384 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002385 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002386 trace_ext4_ext_rm_idx(inode, leaf);
2387
Peter Huewe7dc57612011-02-21 21:01:42 -05002388 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002389 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002390
2391 while (--depth >= 0) {
2392 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2393 break;
2394 path--;
2395 err = ext4_ext_get_access(handle, inode, path);
2396 if (err)
2397 break;
2398 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2399 err = ext4_ext_dirty(handle, inode, path);
2400 if (err)
2401 break;
2402 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002403 return err;
2404}
2405
2406/*
Mingming Caoee12b632008-08-19 22:16:05 -04002407 * ext4_ext_calc_credits_for_single_extent:
2408 * This routine returns max. credits that needed to insert an extent
2409 * to the extent tree.
2410 * When pass the actual path, the caller should calculate credits
2411 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002412 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002413int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002414 struct ext4_ext_path *path)
2415{
Alex Tomasa86c6182006-10-11 01:21:03 -07002416 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002417 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002418 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002419
Alex Tomasa86c6182006-10-11 01:21:03 -07002420 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002421 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002422 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2423
2424 /*
2425 * There are some space in the leaf tree, no
2426 * need to account for leaf block credit
2427 *
2428 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002429 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002430 * accounted.
2431 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002432 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002433 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002434 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002435 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002436 }
2437
Mingming Cao525f4ed2008-08-19 22:15:58 -04002438 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002439}
Alex Tomasa86c6182006-10-11 01:21:03 -07002440
Mingming Caoee12b632008-08-19 22:16:05 -04002441/*
Jan Karafffb2732013-06-04 13:01:11 -04002442 * How many index/leaf blocks need to change/allocate to add @extents extents?
Mingming Caoee12b632008-08-19 22:16:05 -04002443 *
Jan Karafffb2732013-06-04 13:01:11 -04002444 * If we add a single extent, then in the worse case, each tree level
2445 * index/leaf need to be changed in case of the tree split.
Mingming Caoee12b632008-08-19 22:16:05 -04002446 *
Jan Karafffb2732013-06-04 13:01:11 -04002447 * If more extents are inserted, they could cause the whole tree split more
2448 * than once, but this is really rare.
Mingming Caoee12b632008-08-19 22:16:05 -04002449 */
Jan Karafffb2732013-06-04 13:01:11 -04002450int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
Mingming Caoee12b632008-08-19 22:16:05 -04002451{
2452 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002453 int depth;
2454
2455 /* If we are converting the inline data, only one is needed here. */
2456 if (ext4_has_inline_data(inode))
2457 return 1;
2458
2459 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002460
Jan Karafffb2732013-06-04 13:01:11 -04002461 if (extents <= 1)
Mingming Caoee12b632008-08-19 22:16:05 -04002462 index = depth * 2;
2463 else
2464 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002465
Mingming Caoee12b632008-08-19 22:16:05 -04002466 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002467}
2468
Theodore Ts'o981250c2013-06-12 11:48:29 -04002469static inline int get_default_free_blocks_flags(struct inode *inode)
2470{
2471 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2472 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2473 else if (ext4_should_journal_data(inode))
2474 return EXT4_FREE_BLOCKS_FORGET;
2475 return 0;
2476}
2477
Alex Tomasa86c6182006-10-11 01:21:03 -07002478static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002479 struct ext4_extent *ex,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002480 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002481 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002482{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002483 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002484 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002485 ext4_fsblk_t pblk;
Theodore Ts'o981250c2013-06-12 11:48:29 -04002486 int flags = get_default_free_blocks_flags(inode);
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002487
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002488 /*
2489 * For bigalloc file systems, we never free a partial cluster
2490 * at the beginning of the extent. Instead, we make a note
2491 * that we tried freeing the cluster, and check to see if we
2492 * need to free it on a subsequent call to ext4_remove_blocks,
2493 * or at the end of the ext4_truncate() operation.
2494 */
2495 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2496
Aditya Kalid8990242011-09-09 19:18:51 -04002497 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002498 /*
2499 * If we have a partial cluster, and it's different from the
2500 * cluster of the last block, we need to explicitly free the
2501 * partial cluster here.
2502 */
2503 pblk = ext4_ext_pblock(ex) + ee_len - 1;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002504 if ((*partial_cluster > 0) &&
2505 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002506 ext4_free_blocks(handle, inode, NULL,
2507 EXT4_C2B(sbi, *partial_cluster),
2508 sbi->s_cluster_ratio, flags);
2509 *partial_cluster = 0;
2510 }
2511
Alex Tomasa86c6182006-10-11 01:21:03 -07002512#ifdef EXTENTS_STATS
2513 {
2514 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002515 spin_lock(&sbi->s_ext_stats_lock);
2516 sbi->s_ext_blocks += ee_len;
2517 sbi->s_ext_extents++;
2518 if (ee_len < sbi->s_ext_min)
2519 sbi->s_ext_min = ee_len;
2520 if (ee_len > sbi->s_ext_max)
2521 sbi->s_ext_max = ee_len;
2522 if (ext_depth(inode) > sbi->s_depth_max)
2523 sbi->s_depth_max = ext_depth(inode);
2524 spin_unlock(&sbi->s_ext_stats_lock);
2525 }
2526#endif
2527 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002528 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002529 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002530 ext4_lblk_t num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002531 unsigned int unaligned;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002532
Amit Aroraa2df2a62007-07-17 21:42:41 -04002533 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002534 pblk = ext4_ext_pblock(ex) + ee_len - num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002535 /*
2536 * Usually we want to free partial cluster at the end of the
2537 * extent, except for the situation when the cluster is still
2538 * used by any other extent (partial_cluster is negative).
2539 */
2540 if (*partial_cluster < 0 &&
2541 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2542 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2543
2544 ext_debug("free last %u blocks starting %llu partial %lld\n",
2545 num, pblk, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002546 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2547 /*
2548 * If the block range to be freed didn't start at the
2549 * beginning of a cluster, and we removed the entire
Lukas Czernerd23142c2013-05-27 23:33:35 -04002550 * extent and the cluster is not used by any other extent,
2551 * save the partial cluster here, since we might need to
2552 * delete if we determine that the truncate operation has
2553 * removed all of the blocks in the cluster.
2554 *
2555 * On the other hand, if we did not manage to free the whole
2556 * extent, we have to mark the cluster as used (store negative
2557 * cluster number in partial_cluster).
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002558 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05002559 unaligned = EXT4_PBLK_COFF(sbi, pblk);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002560 if (unaligned && (ee_len == num) &&
2561 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002562 *partial_cluster = EXT4_B2C(sbi, pblk);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002563 else if (unaligned)
2564 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2565 else if (*partial_cluster > 0)
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002566 *partial_cluster = 0;
Lukas Czerner78fb9cd2013-05-27 23:32:35 -04002567 } else
2568 ext4_error(sbi->s_sb, "strange request: removal(2) "
2569 "%u-%u from %u:%u\n",
2570 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002571 return 0;
2572}
2573
Allison Hendersond583fb82011-05-25 07:41:43 -04002574
2575/*
2576 * ext4_ext_rm_leaf() Removes the extents associated with the
2577 * blocks appearing between "start" and "end", and splits the extents
2578 * if "start" and "end" appear in the same extent
2579 *
2580 * @handle: The journal handle
2581 * @inode: The files inode
2582 * @path: The path to the leaf
Lukas Czernerd23142c2013-05-27 23:33:35 -04002583 * @partial_cluster: The cluster which we'll have to free if all extents
2584 * has been released from it. It gets negative in case
2585 * that the cluster is still used.
Allison Hendersond583fb82011-05-25 07:41:43 -04002586 * @start: The first block to remove
2587 * @end: The last block to remove
2588 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002589static int
2590ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002591 struct ext4_ext_path *path,
2592 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002593 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002594{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002595 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002596 int err = 0, correct_index = 0;
2597 int depth = ext_depth(inode), credits;
2598 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002599 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002600 unsigned num;
2601 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002602 unsigned short ex_ee_len;
Lukas Czerner556615d2014-04-20 23:45:47 -04002603 unsigned unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002604 struct ext4_extent *ex;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002605 ext4_fsblk_t pblk;
Alex Tomasa86c6182006-10-11 01:21:03 -07002606
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002607 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002608 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002609 if (!path[depth].p_hdr)
2610 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2611 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002612 if (unlikely(path[depth].p_hdr == NULL)) {
2613 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2614 return -EIO;
2615 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002616 /* find where to start removing */
Ashish Sangwan6ae06ff2013-07-01 08:12:41 -04002617 ex = path[depth].p_ext;
2618 if (!ex)
2619 ex = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002620
2621 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002622 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002623
Eric Whitneyc0634492014-03-13 23:34:16 -04002624 /*
2625 * If we're starting with an extent other than the last one in the
2626 * node, we need to see if it shares a cluster with the extent to
2627 * the right (towards the end of the file). If its leftmost cluster
2628 * is this extent's rightmost cluster and it is not cluster aligned,
2629 * we'll mark it as a partial that is not to be deallocated.
2630 */
2631
2632 if (ex != EXT_LAST_EXTENT(eh)) {
2633 ext4_fsblk_t current_pblk, right_pblk;
2634 long long current_cluster, right_cluster;
2635
2636 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2637 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2638 right_pblk = ext4_ext_pblock(ex + 1);
2639 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2640 if (current_cluster == right_cluster &&
2641 EXT4_PBLK_COFF(sbi, right_pblk))
2642 *partial_cluster = -right_cluster;
2643 }
2644
Aditya Kalid8990242011-09-09 19:18:51 -04002645 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2646
Alex Tomasa86c6182006-10-11 01:21:03 -07002647 while (ex >= EXT_FIRST_EXTENT(eh) &&
2648 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002649
Lukas Czerner556615d2014-04-20 23:45:47 -04002650 if (ext4_ext_is_unwritten(ex))
2651 unwritten = 1;
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002652 else
Lukas Czerner556615d2014-04-20 23:45:47 -04002653 unwritten = 0;
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002654
Mingming553f9002009-09-18 13:34:55 -04002655 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
Lukas Czerner556615d2014-04-20 23:45:47 -04002656 unwritten, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002657 path[depth].p_ext = ex;
2658
2659 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002660 b = ex_ee_block+ex_ee_len - 1 < end ?
2661 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002662
2663 ext_debug(" border %u:%u\n", a, b);
2664
Allison Hendersond583fb82011-05-25 07:41:43 -04002665 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002666 if (end < ex_ee_block) {
Lukas Czernerd23142c2013-05-27 23:33:35 -04002667 /*
2668 * We're going to skip this extent and move to another,
2669 * so if this extent is not cluster aligned we have
2670 * to mark the current cluster as used to avoid
2671 * accidentally freeing it later on
2672 */
2673 pblk = ext4_ext_pblock(ex);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05002674 if (EXT4_PBLK_COFF(sbi, pblk))
Lukas Czernerd23142c2013-05-27 23:33:35 -04002675 *partial_cluster =
2676 -((long long)EXT4_B2C(sbi, pblk));
Allison Hendersond583fb82011-05-25 07:41:43 -04002677 ex--;
2678 ex_ee_block = le32_to_cpu(ex->ee_block);
2679 ex_ee_len = ext4_ext_get_actual_len(ex);
2680 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002681 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002682 EXT4_ERROR_INODE(inode,
2683 "can not handle truncate %u:%u "
2684 "on extent %u:%u",
2685 start, end, ex_ee_block,
2686 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002687 err = -EIO;
2688 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002689 } else if (a != ex_ee_block) {
2690 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002691 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002692 } else {
2693 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002694 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002695 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002696 /*
2697 * 3 for leaf, sb, and inode plus 2 (bmap and group
2698 * descriptor) for each block group; assume two block
2699 * groups plus ex_ee_len/blocks_per_block_group for
2700 * the worst case
2701 */
2702 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002703 if (ex == EXT_FIRST_EXTENT(eh)) {
2704 correct_index = 1;
2705 credits += (ext_depth(inode)) + 1;
2706 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002707 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002708
Jan Kara487caee2009-08-17 22:17:20 -04002709 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002710 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002711 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002712
2713 err = ext4_ext_get_access(handle, inode, path + depth);
2714 if (err)
2715 goto out;
2716
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002717 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2718 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002719 if (err)
2720 goto out;
2721
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002722 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002723 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002724 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002725
Alex Tomasa86c6182006-10-11 01:21:03 -07002726 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002727 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04002728 * Do not mark unwritten if all the blocks in the
Amit Arora749269f2007-07-18 09:02:56 -04002729 * extent have been removed.
2730 */
Lukas Czerner556615d2014-04-20 23:45:47 -04002731 if (unwritten && num)
2732 ext4_ext_mark_unwritten(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002733 /*
2734 * If the extent was completely released,
2735 * we need to remove it from the leaf
2736 */
2737 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002738 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002739 /*
2740 * For hole punching, we need to scoot all the
2741 * extents up when an extent is removed so that
2742 * we dont have blank extents in the middle
2743 */
2744 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2745 sizeof(struct ext4_extent));
2746
2747 /* Now get rid of the one at the end */
2748 memset(EXT_LAST_EXTENT(eh), 0,
2749 sizeof(struct ext4_extent));
2750 }
2751 le16_add_cpu(&eh->eh_entries, -1);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002752 } else if (*partial_cluster > 0)
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002753 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002754
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002755 err = ext4_ext_dirty(handle, inode, path + depth);
2756 if (err)
2757 goto out;
2758
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002759 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002760 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002761 ex--;
2762 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002763 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002764 }
2765
2766 if (correct_index && eh->eh_entries)
2767 err = ext4_ext_correct_indexes(handle, inode, path);
2768
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002769 /*
Eric Whitneyad6599a2014-04-01 19:49:30 -04002770 * If there's a partial cluster and at least one extent remains in
2771 * the leaf, free the partial cluster if it isn't shared with the
2772 * current extent. If there's a partial cluster and no extents
2773 * remain in the leaf, it can't be freed here. It can only be
2774 * freed when it's possible to determine if it's not shared with
2775 * any other extent - when the next leaf is processed or when space
2776 * removal is complete.
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002777 */
Eric Whitneyad6599a2014-04-01 19:49:30 -04002778 if (*partial_cluster > 0 && eh->eh_entries &&
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002779 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2780 *partial_cluster)) {
Theodore Ts'o981250c2013-06-12 11:48:29 -04002781 int flags = get_default_free_blocks_flags(inode);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002782
2783 ext4_free_blocks(handle, inode, NULL,
2784 EXT4_C2B(sbi, *partial_cluster),
2785 sbi->s_cluster_ratio, flags);
2786 *partial_cluster = 0;
2787 }
2788
Alex Tomasa86c6182006-10-11 01:21:03 -07002789 /* if this leaf is free, then we should
2790 * remove it from index block above */
2791 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002792 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002793
2794out:
2795 return err;
2796}
2797
2798/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002799 * ext4_ext_more_to_rm:
2800 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002801 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002802static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002803ext4_ext_more_to_rm(struct ext4_ext_path *path)
2804{
2805 BUG_ON(path->p_idx == NULL);
2806
2807 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2808 return 0;
2809
2810 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002811 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002812 * so we have to consider current index for truncation
2813 */
2814 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2815 return 0;
2816 return 1;
2817}
2818
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002819int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2820 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002821{
2822 struct super_block *sb = inode->i_sb;
2823 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002824 struct ext4_ext_path *path = NULL;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002825 long long partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002826 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002827 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002828
Lukas Czerner5f95d212012-03-19 23:03:19 -04002829 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002830
2831 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002832 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002833 if (IS_ERR(handle))
2834 return PTR_ERR(handle);
2835
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002836again:
Lukas Czerner61801322013-05-27 23:32:35 -04002837 trace_ext4_ext_remove_space(inode, start, end, depth);
Aditya Kalid8990242011-09-09 19:18:51 -04002838
Alex Tomasa86c6182006-10-11 01:21:03 -07002839 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002840 * Check if we are removing extents inside the extent tree. If that
2841 * is the case, we are going to punch a hole inside the extent tree
2842 * so we have to check whether we need to split the extent covering
2843 * the last block to remove so we can easily remove the part of it
2844 * in ext4_ext_rm_leaf().
2845 */
2846 if (end < EXT_MAX_BLOCKS - 1) {
2847 struct ext4_extent *ex;
2848 ext4_lblk_t ee_block;
2849
2850 /* find extent for this block */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04002851 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002852 if (IS_ERR(path)) {
2853 ext4_journal_stop(handle);
2854 return PTR_ERR(path);
2855 }
2856 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002857 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002858 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002859 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002860 if (depth) {
2861 EXT4_ERROR_INODE(inode,
2862 "path[%d].p_hdr == NULL",
2863 depth);
2864 err = -EIO;
2865 }
2866 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002867 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002868
2869 ee_block = le32_to_cpu(ex->ee_block);
2870
2871 /*
2872 * See if the last block is inside the extent, if so split
2873 * the extent at 'end' block so we can easily remove the
2874 * tail of the first part of the split extent in
2875 * ext4_ext_rm_leaf().
2876 */
2877 if (end >= ee_block &&
2878 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
Lukas Czerner5f95d212012-03-19 23:03:19 -04002879 /*
2880 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002881 * block in the first new extent. Also we should not
2882 * fail removing space due to ENOSPC so try to use
2883 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002884 */
Theodore Ts'odfe50802014-09-01 14:37:09 -04002885 err = ext4_force_split_extent_at(handle, inode, &path,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04002886 end + 1, 1);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002887 if (err < 0)
2888 goto out;
2889 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002890 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002891 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002892 * We start scanning from right side, freeing all the blocks
2893 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002894 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002895 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002896 if (path) {
2897 int k = i = depth;
2898 while (--k > 0)
2899 path[k].p_block =
2900 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2901 } else {
2902 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2903 GFP_NOFS);
2904 if (path == NULL) {
2905 ext4_journal_stop(handle);
2906 return -ENOMEM;
2907 }
Theodore Ts'o10809df82014-09-01 14:40:09 -04002908 path[0].p_maxdepth = path[0].p_depth = depth;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002909 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002910 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002911
Theodore Ts'oc3491792013-08-16 21:21:41 -04002912 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
Ashish Sangwan968dee72012-07-22 22:49:08 -04002913 err = -EIO;
2914 goto out;
2915 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002916 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002917 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002918
2919 while (i >= 0 && err == 0) {
2920 if (i == depth) {
2921 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002922 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002923 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002924 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002925 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002926 brelse(path[i].p_bh);
2927 path[i].p_bh = NULL;
2928 i--;
2929 continue;
2930 }
2931
2932 /* this is index block */
2933 if (!path[i].p_hdr) {
2934 ext_debug("initialize header\n");
2935 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002936 }
2937
Alex Tomasa86c6182006-10-11 01:21:03 -07002938 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002939 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002940 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2941 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2942 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2943 path[i].p_hdr,
2944 le16_to_cpu(path[i].p_hdr->eh_entries));
2945 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002946 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002947 path[i].p_idx--;
2948 }
2949
2950 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2951 i, EXT_FIRST_INDEX(path[i].p_hdr),
2952 path[i].p_idx);
2953 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002954 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002955 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002956 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002957 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002958 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002959 bh = read_extent_tree_block(inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002960 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2961 EXT4_EX_NOCACHE);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002962 if (IS_ERR(bh)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002963 /* should we reset i_size? */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002964 err = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002965 break;
2966 }
Theodore Ts'o76828c82013-07-15 12:27:47 -04002967 /* Yield here to deal with large extent trees.
2968 * Should be a no-op if we did IO above. */
2969 cond_resched();
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002970 if (WARN_ON(i + 1 > depth)) {
2971 err = -EIO;
2972 break;
2973 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002974 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002975
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002976 /* save actual number of indexes since this
2977 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002978 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2979 i++;
2980 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002981 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002982 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002983 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002984 * handle must be already prepared by the
2985 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002986 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002987 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002988 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002989 brelse(path[i].p_bh);
2990 path[i].p_bh = NULL;
2991 i--;
2992 ext_debug("return to level %d\n", i);
2993 }
2994 }
2995
Lukas Czerner61801322013-05-27 23:32:35 -04002996 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2997 partial_cluster, path->p_hdr->eh_entries);
Aditya Kalid8990242011-09-09 19:18:51 -04002998
Aditya Kali7b415bf2011-09-09 19:04:51 -04002999 /* If we still have something in the partial cluster and we have removed
3000 * even the first extent, then we should free the blocks in the partial
3001 * cluster as well. */
Lukas Czernerd23142c2013-05-27 23:33:35 -04003002 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
Theodore Ts'o981250c2013-06-12 11:48:29 -04003003 int flags = get_default_free_blocks_flags(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003004
3005 ext4_free_blocks(handle, inode, NULL,
3006 EXT4_C2B(EXT4_SB(sb), partial_cluster),
3007 EXT4_SB(sb)->s_cluster_ratio, flags);
3008 partial_cluster = 0;
3009 }
3010
Alex Tomasa86c6182006-10-11 01:21:03 -07003011 /* TODO: flexible tree reduction should be here */
3012 if (path->p_hdr->eh_entries == 0) {
3013 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003014 * truncate to zero freed all the tree,
3015 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07003016 */
3017 err = ext4_ext_get_access(handle, inode, path);
3018 if (err == 0) {
3019 ext_inode_hdr(inode)->eh_depth = 0;
3020 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04003021 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07003022 err = ext4_ext_dirty(handle, inode, path);
3023 }
3024 }
3025out:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04003026 ext4_ext_drop_refs(path);
3027 kfree(path);
3028 path = NULL;
Theodore Ts'odfe50802014-09-01 14:37:09 -04003029 if (err == -EAGAIN)
3030 goto again;
Alex Tomasa86c6182006-10-11 01:21:03 -07003031 ext4_journal_stop(handle);
3032
3033 return err;
3034}
3035
3036/*
3037 * called at mount time
3038 */
3039void ext4_ext_init(struct super_block *sb)
3040{
3041 /*
3042 * possible initialization would be here
3043 */
3044
Theodore Ts'o83982b62009-01-06 14:53:16 -05003045 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04003046#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04003047 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01003048#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04003049 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07003050#endif
3051#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04003052 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07003053#endif
3054#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04003055 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07003056#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04003057 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04003058#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07003059#ifdef EXTENTS_STATS
3060 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3061 EXT4_SB(sb)->s_ext_min = 1 << 30;
3062 EXT4_SB(sb)->s_ext_max = 0;
3063#endif
3064 }
3065}
3066
3067/*
3068 * called at umount time
3069 */
3070void ext4_ext_release(struct super_block *sb)
3071{
Theodore Ts'o83982b62009-01-06 14:53:16 -05003072 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07003073 return;
3074
3075#ifdef EXTENTS_STATS
3076 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3077 struct ext4_sb_info *sbi = EXT4_SB(sb);
3078 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3079 sbi->s_ext_blocks, sbi->s_ext_extents,
3080 sbi->s_ext_blocks / sbi->s_ext_extents);
3081 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3082 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3083 }
3084#endif
3085}
3086
Zheng Liud7b2a002013-08-28 14:47:06 -04003087static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3088{
3089 ext4_lblk_t ee_block;
3090 ext4_fsblk_t ee_pblock;
3091 unsigned int ee_len;
3092
3093 ee_block = le32_to_cpu(ex->ee_block);
3094 ee_len = ext4_ext_get_actual_len(ex);
3095 ee_pblock = ext4_ext_pblock(ex);
3096
3097 if (ee_len == 0)
3098 return 0;
3099
3100 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3101 EXTENT_STATUS_WRITTEN);
3102}
3103
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003104/* FIXME!! we need to try to merge to left or right after zero-out */
3105static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3106{
Lukas Czerner24075182010-10-27 21:30:06 -04003107 ext4_fsblk_t ee_pblock;
3108 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04003109 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003110
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003111 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003112 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003113
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04003114 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04003115 if (ret > 0)
3116 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003117
Lukas Czerner24075182010-10-27 21:30:06 -04003118 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003119}
3120
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003121/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003122 * ext4_split_extent_at() splits an extent at given block.
3123 *
3124 * @handle: the journal handle
3125 * @inode: the file inode
3126 * @path: the path to the extent
3127 * @split: the logical block where the extent is splitted.
3128 * @split_flags: indicates if the extent could be zeroout if split fails, and
Lukas Czerner556615d2014-04-20 23:45:47 -04003129 * the states(init or unwritten) of new extents.
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003130 * @flags: flags used to insert new extent to extent tree.
3131 *
3132 *
3133 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3134 * of which are deterimined by split_flag.
3135 *
3136 * There are two cases:
3137 * a> the extent are splitted into two extent.
3138 * b> split is not needed, and just mark the extent.
3139 *
3140 * return 0 on success.
3141 */
3142static int ext4_split_extent_at(handle_t *handle,
3143 struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003144 struct ext4_ext_path **ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003145 ext4_lblk_t split,
3146 int split_flag,
3147 int flags)
3148{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003149 struct ext4_ext_path *path = *ppath;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003150 ext4_fsblk_t newblock;
3151 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04003152 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003153 struct ext4_extent *ex2 = NULL;
3154 unsigned int ee_len, depth;
3155 int err = 0;
3156
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003157 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3158 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3159
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003160 ext_debug("ext4_split_extents_at: inode %lu, logical"
3161 "block %llu\n", inode->i_ino, (unsigned long long)split);
3162
3163 ext4_ext_show_leaf(inode, path);
3164
3165 depth = ext_depth(inode);
3166 ex = path[depth].p_ext;
3167 ee_block = le32_to_cpu(ex->ee_block);
3168 ee_len = ext4_ext_get_actual_len(ex);
3169 newblock = split - ee_block + ext4_ext_pblock(ex);
3170
3171 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Lukas Czerner556615d2014-04-20 23:45:47 -04003172 BUG_ON(!ext4_ext_is_unwritten(ex) &&
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003173 split_flag & (EXT4_EXT_MAY_ZEROOUT |
Lukas Czerner556615d2014-04-20 23:45:47 -04003174 EXT4_EXT_MARK_UNWRIT1 |
3175 EXT4_EXT_MARK_UNWRIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003176
3177 err = ext4_ext_get_access(handle, inode, path + depth);
3178 if (err)
3179 goto out;
3180
3181 if (split == ee_block) {
3182 /*
3183 * case b: block @split is the block that the extent begins with
3184 * then we just change the state of the extent, and splitting
3185 * is not needed.
3186 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003187 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3188 ext4_ext_mark_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003189 else
3190 ext4_ext_mark_initialized(ex);
3191
3192 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003193 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003194
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003195 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003196 goto out;
3197 }
3198
3199 /* case a */
3200 memcpy(&orig_ex, ex, sizeof(orig_ex));
3201 ex->ee_len = cpu_to_le16(split - ee_block);
Lukas Czerner556615d2014-04-20 23:45:47 -04003202 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3203 ext4_ext_mark_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003204
3205 /*
3206 * path may lead to new leaf, not to original leaf any more
3207 * after ext4_ext_insert_extent() returns,
3208 */
3209 err = ext4_ext_dirty(handle, inode, path + depth);
3210 if (err)
3211 goto fix_extent_len;
3212
3213 ex2 = &newex;
3214 ex2->ee_block = cpu_to_le32(split);
3215 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3216 ext4_ext_store_pblock(ex2, newblock);
Lukas Czerner556615d2014-04-20 23:45:47 -04003217 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3218 ext4_ext_mark_unwritten(ex2);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003219
Theodore Ts'odfe50802014-09-01 14:37:09 -04003220 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003221 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003222 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003223 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003224 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003225 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003226 zero_ex.ee_len = cpu_to_le16(
3227 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003228 ext4_ext_store_pblock(&zero_ex,
3229 ext4_ext_pblock(ex2));
3230 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003231 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003232 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003233 zero_ex.ee_len = cpu_to_le16(
3234 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003235 ext4_ext_store_pblock(&zero_ex,
3236 ext4_ext_pblock(ex));
3237 }
3238 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003239 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003240 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003241 zero_ex.ee_len = cpu_to_le16(
3242 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003243 ext4_ext_store_pblock(&zero_ex,
3244 ext4_ext_pblock(&orig_ex));
3245 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003246
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003247 if (err)
3248 goto fix_extent_len;
3249 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003250 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003251 ext4_ext_try_to_merge(handle, inode, path, ex);
3252 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003253 if (err)
3254 goto fix_extent_len;
3255
3256 /* update extent status tree */
Zheng Liud7b2a002013-08-28 14:47:06 -04003257 err = ext4_zeroout_es(inode, &zero_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003258
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003259 goto out;
3260 } else if (err)
3261 goto fix_extent_len;
3262
3263out:
3264 ext4_ext_show_leaf(inode, path);
3265 return err;
3266
3267fix_extent_len:
3268 ex->ee_len = orig_ex.ee_len;
Dmitry Monakhov29faed12014-07-27 22:30:29 -04003269 ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003270 return err;
3271}
3272
3273/*
3274 * ext4_split_extents() splits an extent and mark extent which is covered
3275 * by @map as split_flags indicates
3276 *
Anatol Pomozov70261f52013-08-28 14:40:12 -04003277 * It may result in splitting the extent into multiple extents (up to three)
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003278 * There are three possibilities:
3279 * a> There is no split required
3280 * b> Splits in two extents: Split is happening at either end of the extent
3281 * c> Splits in three extents: Somone is splitting in middle of the extent
3282 *
3283 */
3284static int ext4_split_extent(handle_t *handle,
3285 struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003286 struct ext4_ext_path **ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003287 struct ext4_map_blocks *map,
3288 int split_flag,
3289 int flags)
3290{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003291 struct ext4_ext_path *path = *ppath;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003292 ext4_lblk_t ee_block;
3293 struct ext4_extent *ex;
3294 unsigned int ee_len, depth;
3295 int err = 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04003296 int unwritten;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003297 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003298 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003299
3300 depth = ext_depth(inode);
3301 ex = path[depth].p_ext;
3302 ee_block = le32_to_cpu(ex->ee_block);
3303 ee_len = ext4_ext_get_actual_len(ex);
Lukas Czerner556615d2014-04-20 23:45:47 -04003304 unwritten = ext4_ext_is_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003305
3306 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003307 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003308 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
Lukas Czerner556615d2014-04-20 23:45:47 -04003309 if (unwritten)
3310 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3311 EXT4_EXT_MARK_UNWRIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003312 if (split_flag & EXT4_EXT_DATA_VALID2)
3313 split_flag1 |= EXT4_EXT_DATA_VALID1;
Theodore Ts'odfe50802014-09-01 14:37:09 -04003314 err = ext4_split_extent_at(handle, inode, ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003315 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003316 if (err)
3317 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003318 } else {
3319 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003320 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003321 /*
3322 * Update path is required because previous ext4_split_extent_at() may
3323 * result in split of original leaf or extent zeroout.
3324 */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04003325 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003326 if (IS_ERR(path))
3327 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003328 depth = ext_depth(inode);
3329 ex = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04003330 if (!ex) {
3331 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3332 (unsigned long) map->m_lblk);
3333 return -EIO;
3334 }
Lukas Czerner556615d2014-04-20 23:45:47 -04003335 unwritten = ext4_ext_is_unwritten(ex);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003336 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003337
3338 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003339 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
Lukas Czerner556615d2014-04-20 23:45:47 -04003340 if (unwritten) {
3341 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003342 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
Lukas Czerner556615d2014-04-20 23:45:47 -04003343 EXT4_EXT_MARK_UNWRIT2);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003344 }
Theodore Ts'odfe50802014-09-01 14:37:09 -04003345 err = ext4_split_extent_at(handle, inode, ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003346 map->m_lblk, split_flag1, flags);
3347 if (err)
3348 goto out;
3349 }
3350
3351 ext4_ext_show_leaf(inode, path);
3352out:
Zheng Liu3a225672013-03-10 21:20:23 -04003353 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003354}
3355
Amit Arora56055d32007-07-17 21:42:38 -04003356/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003357 * This function is called by ext4_ext_map_blocks() if someone tries to write
Lukas Czerner556615d2014-04-20 23:45:47 -04003358 * to an unwritten extent. It may result in splitting the unwritten
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003359 * extent into multiple extents (up to three - one initialized and two
Lukas Czerner556615d2014-04-20 23:45:47 -04003360 * unwritten).
Amit Arora56055d32007-07-17 21:42:38 -04003361 * There are three possibilities:
3362 * a> There is no split required: Entire extent should be initialized
3363 * b> Splits in two extents: Write is happening at either end of the extent
3364 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003365 *
3366 * Pre-conditions:
Lukas Czerner556615d2014-04-20 23:45:47 -04003367 * - The extent pointed to by 'path' is unwritten.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003368 * - The extent pointed to by 'path' contains a superset
3369 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3370 *
3371 * Post-conditions on success:
3372 * - the returned value is the number of blocks beyond map->l_lblk
3373 * that are allocated and initialized.
3374 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003375 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003376static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003377 struct inode *inode,
3378 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003379 struct ext4_ext_path **ppath,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003380 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003381{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003382 struct ext4_ext_path *path = *ppath;
Zheng Liu67a5da52012-08-17 09:54:17 -04003383 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003384 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003385 struct ext4_map_blocks split_map;
3386 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003387 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003388 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003389 unsigned int ee_len, depth, map_len = map->m_len;
3390 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003391 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003392 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003393
3394 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3395 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003396 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003397
Zheng Liu67a5da52012-08-17 09:54:17 -04003398 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003399 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3400 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003401 if (eof_block < map->m_lblk + map_len)
3402 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003403
3404 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003405 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003406 ex = path[depth].p_ext;
3407 ee_block = le32_to_cpu(ex->ee_block);
3408 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003409 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003410
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003411 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3412
3413 /* Pre-conditions */
Lukas Czerner556615d2014-04-20 23:45:47 -04003414 BUG_ON(!ext4_ext_is_unwritten(ex));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003415 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003416
3417 /*
3418 * Attempt to transfer newly initialized blocks from the currently
Lukas Czerner556615d2014-04-20 23:45:47 -04003419 * unwritten extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003420 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003421 * memmove() calls. Transferring to the left is the common case in
3422 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3423 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003424 *
3425 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003426 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003427 * This would require removing the extent if the transfer
3428 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003429 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003430 * same extent tree node.
3431 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003432 if ((map->m_lblk == ee_block) &&
3433 /* See if we can merge left */
3434 (map_len < ee_len) && /*L1*/
3435 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003436 ext4_lblk_t prev_lblk;
3437 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003438 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003439
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003440 abut_ex = ex - 1;
3441 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3442 prev_len = ext4_ext_get_actual_len(abut_ex);
3443 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003444 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003445
3446 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003447 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003448 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003449 * - C1: abut_ex is initialized,
3450 * - C2: abut_ex is logically abutting ex,
3451 * - C3: abut_ex is physically abutting ex,
3452 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003453 * overflowing the (initialized) length limit.
3454 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003455 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003456 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3457 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003458 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003459 err = ext4_ext_get_access(handle, inode, path + depth);
3460 if (err)
3461 goto out;
3462
3463 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003464 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003465
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003466 /* Shift the start of ex by 'map_len' blocks */
3467 ex->ee_block = cpu_to_le32(ee_block + map_len);
3468 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3469 ex->ee_len = cpu_to_le16(ee_len - map_len);
Lukas Czerner556615d2014-04-20 23:45:47 -04003470 ext4_ext_mark_unwritten(ex); /* Restore the flag */
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003471
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003472 /* Extend abut_ex by 'map_len' blocks */
3473 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003474
3475 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003476 allocated = map_len;
3477 }
3478 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3479 (map_len < ee_len) && /*L1*/
3480 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3481 /* See if we can merge right */
3482 ext4_lblk_t next_lblk;
3483 ext4_fsblk_t next_pblk, ee_pblk;
3484 unsigned int next_len;
3485
3486 abut_ex = ex + 1;
3487 next_lblk = le32_to_cpu(abut_ex->ee_block);
3488 next_len = ext4_ext_get_actual_len(abut_ex);
3489 next_pblk = ext4_ext_pblock(abut_ex);
3490 ee_pblk = ext4_ext_pblock(ex);
3491
3492 /*
3493 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3494 * upon those conditions:
3495 * - C1: abut_ex is initialized,
3496 * - C2: abut_ex is logically abutting ex,
3497 * - C3: abut_ex is physically abutting ex,
3498 * - C4: abut_ex can receive the additional blocks without
3499 * overflowing the (initialized) length limit.
3500 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003501 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003502 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3503 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3504 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3505 err = ext4_ext_get_access(handle, inode, path + depth);
3506 if (err)
3507 goto out;
3508
3509 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3510 map, ex, abut_ex);
3511
3512 /* Shift the start of abut_ex by 'map_len' blocks */
3513 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3514 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3515 ex->ee_len = cpu_to_le16(ee_len - map_len);
Lukas Czerner556615d2014-04-20 23:45:47 -04003516 ext4_ext_mark_unwritten(ex); /* Restore the flag */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003517
3518 /* Extend abut_ex by 'map_len' blocks */
3519 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3520
3521 /* Result: number of initialized blocks past m_lblk */
3522 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003523 }
3524 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003525 if (allocated) {
3526 /* Mark the block containing both extents as dirty */
3527 ext4_ext_dirty(handle, inode, path + depth);
3528
3529 /* Update path to point to the right extent */
3530 path[depth].p_ext = abut_ex;
3531 goto out;
3532 } else
3533 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003534
Yongqiang Yang667eff32011-05-03 12:25:07 -04003535 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003536 /*
3537 * It is safe to convert extent to initialized via explicit
Yongqiang Yang9e740562014-01-06 14:05:23 -05003538 * zeroout only if extent is fully inside i_size or new_size.
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003539 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003540 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003541
Zheng Liu67a5da52012-08-17 09:54:17 -04003542 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3543 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003544 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003545
3546 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3547 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003548 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003549 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003550 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003551 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003552 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003553 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003554
3555 err = ext4_ext_get_access(handle, inode, path + depth);
3556 if (err)
3557 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003558 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003559 ext4_ext_try_to_merge(handle, inode, path, ex);
3560 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003561 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003562 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003563
Amit Arora56055d32007-07-17 21:42:38 -04003564 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003565 * four cases:
3566 * 1. split the extent into three extents.
3567 * 2. split the extent into two extents, zeroout the first half.
3568 * 3. split the extent into two extents, zeroout the second half.
3569 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003570 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003571 split_map.m_lblk = map->m_lblk;
3572 split_map.m_len = map->m_len;
3573
Zheng Liu67a5da52012-08-17 09:54:17 -04003574 if (max_zeroout && (allocated > map->m_len)) {
3575 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003576 /* case 3 */
3577 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003578 cpu_to_le32(map->m_lblk);
3579 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003580 ext4_ext_store_pblock(&zero_ex,
3581 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3582 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003583 if (err)
3584 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003585 split_map.m_lblk = map->m_lblk;
3586 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003587 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003588 /* case 2 */
3589 if (map->m_lblk != ee_block) {
3590 zero_ex.ee_block = ex->ee_block;
3591 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3592 ee_block);
3593 ext4_ext_store_pblock(&zero_ex,
3594 ext4_ext_pblock(ex));
3595 err = ext4_ext_zeroout(inode, &zero_ex);
3596 if (err)
3597 goto out;
3598 }
3599
Yongqiang Yang667eff32011-05-03 12:25:07 -04003600 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003601 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3602 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003603 }
3604 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003605
Jan Karaae9e9c6a2014-10-30 10:53:17 -04003606 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3607 flags);
3608 if (err > 0)
3609 err = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003610out:
Zheng Liuadb23552013-03-10 21:13:05 -04003611 /* If we have gotten a failure, don't zero out status tree */
3612 if (!err)
Zheng Liud7b2a002013-08-28 14:47:06 -04003613 err = ext4_zeroout_es(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003614 return err ? err : allocated;
3615}
3616
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003617/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003618 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003619 * ext4_get_blocks_dio_write() when DIO to write
Lukas Czerner556615d2014-04-20 23:45:47 -04003620 * to an unwritten extent.
Mingming Cao00314622009-09-28 15:49:08 -04003621 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003622 * Writing to an unwritten extent may result in splitting the unwritten
3623 * extent into multiple initialized/unwritten extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003624 * There are three possibilities:
Lukas Czerner556615d2014-04-20 23:45:47 -04003625 * a> There is no split required: Entire extent should be unwritten
Mingming Cao00314622009-09-28 15:49:08 -04003626 * b> Splits in two extents: Write is happening at either end of the extent
3627 * c> Splits in three extents: Somone is writing in middle of the extent
3628 *
Lukas Czernerb8a86842014-03-18 18:05:35 -04003629 * This works the same way in the case of initialized -> unwritten conversion.
3630 *
Mingming Cao00314622009-09-28 15:49:08 -04003631 * One of more index blocks maybe needed if the extent tree grow after
Lukas Czerner556615d2014-04-20 23:45:47 -04003632 * the unwritten extent split. To prevent ENOSPC occur at the IO
3633 * complete, we need to split the unwritten extent before DIO submit
3634 * the IO. The unwritten extent called at this time will be split
3635 * into three unwritten extent(at most). After IO complete, the part
Mingming Cao00314622009-09-28 15:49:08 -04003636 * being filled will be convert to initialized by the end_io callback function
3637 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003638 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003639 * Returns the size of unwritten extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003640 */
Lukas Czernerb8a86842014-03-18 18:05:35 -04003641static int ext4_split_convert_extents(handle_t *handle,
Mingming Cao00314622009-09-28 15:49:08 -04003642 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003643 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003644 struct ext4_ext_path **ppath,
Mingming Cao00314622009-09-28 15:49:08 -04003645 int flags)
3646{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003647 struct ext4_ext_path *path = *ppath;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003648 ext4_lblk_t eof_block;
3649 ext4_lblk_t ee_block;
3650 struct ext4_extent *ex;
3651 unsigned int ee_len;
3652 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003653
Lukas Czernerb8a86842014-03-18 18:05:35 -04003654 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3655 __func__, inode->i_ino,
3656 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003657
3658 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3659 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003660 if (eof_block < map->m_lblk + map->m_len)
3661 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003662 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003663 * It is safe to convert extent to initialized via explicit
3664 * zeroout only if extent is fully insde i_size or new_size.
3665 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003666 depth = ext_depth(inode);
3667 ex = path[depth].p_ext;
3668 ee_block = le32_to_cpu(ex->ee_block);
3669 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003670
Lukas Czernerb8a86842014-03-18 18:05:35 -04003671 /* Convert to unwritten */
3672 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3673 split_flag |= EXT4_EXT_DATA_VALID1;
3674 /* Convert to initialized */
3675 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3676 split_flag |= ee_block + ee_len <= eof_block ?
3677 EXT4_EXT_MAY_ZEROOUT : 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04003678 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
Lukas Czernerb8a86842014-03-18 18:05:35 -04003679 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003680 flags |= EXT4_GET_BLOCKS_PRE_IO;
Theodore Ts'odfe50802014-09-01 14:37:09 -04003681 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003682}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003683
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003684static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003685 struct inode *inode,
3686 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003687 struct ext4_ext_path **ppath)
Mingming Cao00314622009-09-28 15:49:08 -04003688{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003689 struct ext4_ext_path *path = *ppath;
Mingming Cao00314622009-09-28 15:49:08 -04003690 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003691 ext4_lblk_t ee_block;
3692 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003693 int depth;
3694 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003695
3696 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003697 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003698 ee_block = le32_to_cpu(ex->ee_block);
3699 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003700
Yongqiang Yang197217a2011-05-03 11:45:29 -04003701 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3702 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003703 (unsigned long long)ee_block, ee_len);
3704
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003705 /* If extent is larger than requested it is a clear sign that we still
3706 * have some extent state machine issues left. So extent_split is still
3707 * required.
3708 * TODO: Once all related issues will be fixed this situation should be
3709 * illegal.
3710 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003711 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003712#ifdef EXT4_DEBUG
3713 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3714 " len %u; IO logical block %llu, len %u\n",
3715 inode->i_ino, (unsigned long long)ee_block, ee_len,
3716 (unsigned long long)map->m_lblk, map->m_len);
3717#endif
Theodore Ts'odfe50802014-09-01 14:37:09 -04003718 err = ext4_split_convert_extents(handle, inode, map, ppath,
Lukas Czernerb8a86842014-03-18 18:05:35 -04003719 EXT4_GET_BLOCKS_CONVERT);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003720 if (err < 0)
Theodore Ts'odfe50802014-09-01 14:37:09 -04003721 return err;
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04003722 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
Theodore Ts'odfe50802014-09-01 14:37:09 -04003723 if (IS_ERR(path))
3724 return PTR_ERR(path);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003725 depth = ext_depth(inode);
3726 ex = path[depth].p_ext;
3727 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003728
Mingming Cao00314622009-09-28 15:49:08 -04003729 err = ext4_ext_get_access(handle, inode, path + depth);
3730 if (err)
3731 goto out;
3732 /* first mark the extent as initialized */
3733 ext4_ext_mark_initialized(ex);
3734
Yongqiang Yang197217a2011-05-03 11:45:29 -04003735 /* note: ext4_ext_correct_indexes() isn't needed here because
3736 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003737 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003738 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003739
Mingming Cao00314622009-09-28 15:49:08 -04003740 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003741 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003742out:
3743 ext4_ext_show_leaf(inode, path);
3744 return err;
3745}
3746
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003747static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3748 sector_t block, int count)
3749{
3750 int i;
3751 for (i = 0; i < count; i++)
3752 unmap_underlying_metadata(bdev, block + i);
3753}
3754
Theodore Ts'o58590b02010-10-27 21:23:12 -04003755/*
3756 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3757 */
3758static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003759 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003760 struct ext4_ext_path *path,
3761 unsigned int len)
3762{
3763 int i, depth;
3764 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003765 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003766
3767 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3768 return 0;
3769
3770 depth = ext_depth(inode);
3771 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003772
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003773 /*
3774 * We're going to remove EOFBLOCKS_FL entirely in future so we
3775 * do not care for this case anymore. Simply remove the flag
3776 * if there are no extents.
3777 */
3778 if (unlikely(!eh->eh_entries))
3779 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003780 last_ex = EXT_LAST_EXTENT(eh);
3781 /*
3782 * We should clear the EOFBLOCKS_FL flag if we are writing the
3783 * last block in the last extent in the file. We test this by
3784 * first checking to see if the caller to
3785 * ext4_ext_get_blocks() was interested in the last block (or
3786 * a block beyond the last block) in the current extent. If
3787 * this turns out to be false, we can bail out from this
3788 * function immediately.
3789 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003790 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003791 ext4_ext_get_actual_len(last_ex))
3792 return 0;
3793 /*
3794 * If the caller does appear to be planning to write at or
3795 * beyond the end of the current extent, we then test to see
3796 * if the current extent is the last extent in the file, by
3797 * checking to make sure it was reached via the rightmost node
3798 * at each level of the tree.
3799 */
3800 for (i = depth-1; i >= 0; i--)
3801 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3802 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003803out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003804 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3805 return ext4_mark_inode_dirty(handle, inode);
3806}
3807
Aditya Kali7b415bf2011-09-09 19:04:51 -04003808/**
3809 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3810 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003811 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003812 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003813int ext4_find_delalloc_range(struct inode *inode,
3814 ext4_lblk_t lblk_start,
3815 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003816{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003817 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003818
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003819 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003820 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003821 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003822 else if (es.es_lblk <= lblk_start &&
3823 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003824 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003825 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003826 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003827 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003828 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003829}
3830
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003831int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003832{
3833 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3834 ext4_lblk_t lblk_start, lblk_end;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003835 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003836 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3837
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003838 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003839}
3840
3841/**
3842 * Determines how many complete clusters (out of those specified by the 'map')
3843 * are under delalloc and were reserved quota for.
3844 * This function is called when we are writing out the blocks that were
3845 * originally written with their allocation delayed, but then the space was
3846 * allocated using fallocate() before the delayed allocation could be resolved.
3847 * The cases to look for are:
3848 * ('=' indicated delayed allocated blocks
3849 * '-' indicates non-delayed allocated blocks)
3850 * (a) partial clusters towards beginning and/or end outside of allocated range
3851 * are not delalloc'ed.
3852 * Ex:
3853 * |----c---=|====c====|====c====|===-c----|
3854 * |++++++ allocated ++++++|
3855 * ==> 4 complete clusters in above example
3856 *
3857 * (b) partial cluster (outside of allocated range) towards either end is
3858 * marked for delayed allocation. In this case, we will exclude that
3859 * cluster.
3860 * Ex:
3861 * |----====c========|========c========|
3862 * |++++++ allocated ++++++|
3863 * ==> 1 complete clusters in above example
3864 *
3865 * Ex:
3866 * |================c================|
3867 * |++++++ allocated ++++++|
3868 * ==> 0 complete clusters in above example
3869 *
3870 * The ext4_da_update_reserve_space will be called only if we
3871 * determine here that there were some "entire" clusters that span
3872 * this 'allocated' range.
3873 * In the non-bigalloc case, this function will just end up returning num_blks
3874 * without ever calling ext4_find_delalloc_range.
3875 */
3876static unsigned int
3877get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3878 unsigned int num_blks)
3879{
3880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3881 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3882 ext4_lblk_t lblk_from, lblk_to, c_offset;
3883 unsigned int allocated_clusters = 0;
3884
3885 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3886 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3887
3888 /* max possible clusters for this allocation */
3889 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3890
Aditya Kalid8990242011-09-09 19:18:51 -04003891 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3892
Aditya Kali7b415bf2011-09-09 19:04:51 -04003893 /* Check towards left side */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003894 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003895 if (c_offset) {
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003896 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003897 lblk_to = lblk_from + c_offset - 1;
3898
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003899 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003900 allocated_clusters--;
3901 }
3902
3903 /* Now check towards right. */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003904 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003905 if (allocated_clusters && c_offset) {
3906 lblk_from = lblk_start + num_blks;
3907 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3908
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003909 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003910 allocated_clusters--;
3911 }
3912
3913 return allocated_clusters;
3914}
3915
Mingming Cao00314622009-09-28 15:49:08 -04003916static int
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003917convert_initialized_extent(handle_t *handle, struct inode *inode,
3918 struct ext4_map_blocks *map,
Theodore Ts'o4f224b82014-09-01 14:36:09 -04003919 struct ext4_ext_path **ppath, int flags,
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003920 unsigned int allocated, ext4_fsblk_t newblock)
Lukas Czernerb8a86842014-03-18 18:05:35 -04003921{
Theodore Ts'o4f224b82014-09-01 14:36:09 -04003922 struct ext4_ext_path *path = *ppath;
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003923 struct ext4_extent *ex;
3924 ext4_lblk_t ee_block;
3925 unsigned int ee_len;
3926 int depth;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003927 int err = 0;
3928
3929 /*
3930 * Make sure that the extent is no bigger than we support with
Lukas Czerner556615d2014-04-20 23:45:47 -04003931 * unwritten extent
Lukas Czernerb8a86842014-03-18 18:05:35 -04003932 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003933 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3934 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003935
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003936 depth = ext_depth(inode);
3937 ex = path[depth].p_ext;
3938 ee_block = le32_to_cpu(ex->ee_block);
3939 ee_len = ext4_ext_get_actual_len(ex);
3940
3941 ext_debug("%s: inode %lu, logical"
3942 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3943 (unsigned long long)ee_block, ee_len);
3944
3945 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Theodore Ts'odfe50802014-09-01 14:37:09 -04003946 err = ext4_split_convert_extents(handle, inode, map, ppath,
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003947 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3948 if (err < 0)
3949 return err;
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04003950 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003951 if (IS_ERR(path))
3952 return PTR_ERR(path);
3953 depth = ext_depth(inode);
3954 ex = path[depth].p_ext;
3955 if (!ex) {
3956 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3957 (unsigned long) map->m_lblk);
3958 return -EIO;
3959 }
3960 }
3961
3962 err = ext4_ext_get_access(handle, inode, path + depth);
3963 if (err)
3964 return err;
3965 /* first mark the extent as unwritten */
3966 ext4_ext_mark_unwritten(ex);
3967
3968 /* note: ext4_ext_correct_indexes() isn't needed here because
3969 * borders are not changed
3970 */
3971 ext4_ext_try_to_merge(handle, inode, path, ex);
3972
3973 /* Mark modified extent as dirty */
3974 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3975 if (err)
3976 return err;
3977 ext4_ext_show_leaf(inode, path);
3978
3979 ext4_update_inode_fsync_trans(handle, inode, 1);
3980 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3981 if (err)
3982 return err;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003983 map->m_flags |= EXT4_MAP_UNWRITTEN;
3984 if (allocated > map->m_len)
3985 allocated = map->m_len;
3986 map->m_len = allocated;
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003987 return allocated;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003988}
3989
3990static int
Lukas Czerner556615d2014-04-20 23:45:47 -04003991ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003992 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003993 struct ext4_ext_path **ppath, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003994 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003995{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003996 struct ext4_ext_path *path = *ppath;
Mingming Cao00314622009-09-28 15:49:08 -04003997 int ret = 0;
3998 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003999 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04004000
Lukas Czerner556615d2014-04-20 23:45:47 -04004001 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
Zheng Liu88635ca2011-12-28 19:00:25 -05004002 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004003 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04004004 flags, allocated);
4005 ext4_ext_show_leaf(inode, path);
4006
Lukas Czerner27dd4382013-04-09 22:11:22 -04004007 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04004008 * When writing into unwritten space, we should not fail to
Lukas Czerner27dd4382013-04-09 22:11:22 -04004009 * allocate metadata blocks for the new extent block if needed.
4010 */
4011 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4012
Lukas Czerner556615d2014-04-20 23:45:47 -04004013 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
Zheng Liub5645532012-11-08 14:33:43 -05004014 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04004015
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004016 /* get_block() before submit the IO, split the extent */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004017 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
Theodore Ts'odfe50802014-09-01 14:37:09 -04004018 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4019 flags | EXT4_GET_BLOCKS_CONVERT);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004020 if (ret <= 0)
4021 goto out;
Mingming5f524952009-11-10 10:48:04 -05004022 /*
4023 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004024 * that this IO needs to conversion to written when IO is
Mingming5f524952009-11-10 10:48:04 -05004025 * completed
4026 */
Tao Ma0edeb712011-10-31 17:30:44 -04004027 if (io)
4028 ext4_set_io_unwritten_flag(inode, io);
4029 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004030 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05004031 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004032 goto out;
4033 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004034 /* IO end_io complete, convert the filled extent to written */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004035 if (flags & EXT4_GET_BLOCKS_CONVERT) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04004036 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04004037 ppath);
Theodore Ts'o58590b02010-10-27 21:23:12 -04004038 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05004039 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05004040 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4041 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04004042 } else
4043 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04004044 map->m_flags |= EXT4_MAP_MAPPED;
Eric Whitney15cc1762014-02-12 10:42:45 -05004045 map->m_pblk = newblock;
Zheng Liucdee7842013-03-10 21:08:52 -04004046 if (allocated > map->m_len)
4047 allocated = map->m_len;
4048 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04004049 goto out2;
4050 }
4051 /* buffered IO case */
4052 /*
4053 * repeat fallocate creation request
4054 * we already have an unwritten extent
4055 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004056 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
Zheng Liua25a4e12013-02-18 00:28:04 -05004057 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004058 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05004059 }
Mingming Cao00314622009-09-28 15:49:08 -04004060
4061 /* buffered READ or buffered write_begin() lookup */
4062 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4063 /*
4064 * We have blocks reserved already. We
4065 * return allocated blocks so that delalloc
4066 * won't do block reservation for us. But
4067 * the buffer head will be unmapped so that
4068 * a read from the block returns 0s.
4069 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004070 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004071 goto out1;
4072 }
4073
4074 /* buffered write, writepage time, convert*/
Theodore Ts'odfe50802014-09-01 14:37:09 -04004075 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004076 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004077 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04004078out:
4079 if (ret <= 0) {
4080 err = ret;
4081 goto out2;
4082 } else
4083 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004084 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004085 /*
4086 * if we allocated more blocks than requested
4087 * we need to make sure we unmap the extra block
4088 * allocated. The actual needed block will get
4089 * unmapped later when we find the buffer_head marked
4090 * new.
4091 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004092 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004093 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004094 newblock + map->m_len,
4095 allocated - map->m_len);
4096 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004097 }
Zheng Liu3a225672013-03-10 21:20:23 -04004098 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004099
4100 /*
4101 * If we have done fallocate with the offset that is already
4102 * delayed allocated, we would have block reservation
4103 * and quota reservation done in the delayed write path.
4104 * But fallocate would have already updated quota and block
4105 * count for this offset. So cancel these reservation
4106 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004107 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4108 unsigned int reserved_clusters;
4109 reserved_clusters = get_reserved_cluster_alloc(inode,
4110 map->m_lblk, map->m_len);
4111 if (reserved_clusters)
4112 ext4_da_update_reserve_space(inode,
4113 reserved_clusters,
4114 0);
4115 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004116
Mingming Cao00314622009-09-28 15:49:08 -04004117map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004118 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004119 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4120 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4121 map->m_len);
4122 if (err < 0)
4123 goto out2;
4124 }
Mingming Cao00314622009-09-28 15:49:08 -04004125out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004126 if (allocated > map->m_len)
4127 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04004128 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004129 map->m_pblk = newblock;
4130 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04004131out2:
Mingming Cao00314622009-09-28 15:49:08 -04004132 return err ? err : allocated;
4133}
Theodore Ts'o58590b02010-10-27 21:23:12 -04004134
Mingming Cao00314622009-09-28 15:49:08 -04004135/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004136 * get_implied_cluster_alloc - check to see if the requested
4137 * allocation (in the map structure) overlaps with a cluster already
4138 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04004139 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004140 * @map The requested lblk->pblk mapping
4141 * @ex The extent structure which might contain an implied
4142 * cluster allocation
4143 *
4144 * This function is called by ext4_ext_map_blocks() after we failed to
4145 * find blocks that were already in the inode's extent tree. Hence,
4146 * we know that the beginning of the requested region cannot overlap
4147 * the extent from the inode's extent tree. There are three cases we
4148 * want to catch. The first is this case:
4149 *
4150 * |--- cluster # N--|
4151 * |--- extent ---| |---- requested region ---|
4152 * |==========|
4153 *
4154 * The second case that we need to test for is this one:
4155 *
4156 * |--------- cluster # N ----------------|
4157 * |--- requested region --| |------- extent ----|
4158 * |=======================|
4159 *
4160 * The third case is when the requested region lies between two extents
4161 * within the same cluster:
4162 * |------------- cluster # N-------------|
4163 * |----- ex -----| |---- ex_right ----|
4164 * |------ requested region ------|
4165 * |================|
4166 *
4167 * In each of the above cases, we need to set the map->m_pblk and
4168 * map->m_len so it corresponds to the return the extent labelled as
4169 * "|====|" from cluster #N, since it is already in use for data in
4170 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4171 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4172 * as a new "allocated" block region. Otherwise, we will return 0 and
4173 * ext4_ext_map_blocks() will then allocate one or more new clusters
4174 * by calling ext4_mb_new_blocks().
4175 */
Aditya Kalid8990242011-09-09 19:18:51 -04004176static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004177 struct ext4_map_blocks *map,
4178 struct ext4_extent *ex,
4179 struct ext4_ext_path *path)
4180{
Aditya Kalid8990242011-09-09 19:18:51 -04004181 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004182 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004183 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05004184 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004185 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4186 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4187 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4188
4189 /* The extent passed in that we are trying to match */
4190 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4191 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4192
4193 /* The requested region passed into ext4_map_blocks() */
4194 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004195
4196 if ((rr_cluster_start == ex_cluster_end) ||
4197 (rr_cluster_start == ex_cluster_start)) {
4198 if (rr_cluster_start == ex_cluster_end)
4199 ee_start += ee_len - 1;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004200 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004201 map->m_len = min(map->m_len,
4202 (unsigned) sbi->s_cluster_ratio - c_offset);
4203 /*
4204 * Check for and handle this case:
4205 *
4206 * |--------- cluster # N-------------|
4207 * |------- extent ----|
4208 * |--- requested region ---|
4209 * |===========|
4210 */
4211
4212 if (map->m_lblk < ee_block)
4213 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4214
4215 /*
4216 * Check for the case where there is already another allocated
4217 * block to the right of 'ex' but before the end of the cluster.
4218 *
4219 * |------------- cluster # N-------------|
4220 * |----- ex -----| |---- ex_right ----|
4221 * |------ requested region ------|
4222 * |================|
4223 */
4224 if (map->m_lblk > ee_block) {
4225 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4226 map->m_len = min(map->m_len, next - map->m_lblk);
4227 }
Aditya Kalid8990242011-09-09 19:18:51 -04004228
4229 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004230 return 1;
4231 }
Aditya Kalid8990242011-09-09 19:18:51 -04004232
4233 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004234 return 0;
4235}
4236
4237
4238/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05004239 * Block allocation/map/preallocation routine for extents based files
4240 *
4241 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004242 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004243 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4244 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004245 *
4246 * return > 0, number of of blocks already mapped/allocated
4247 * if create == 0 and these are pre-allocated blocks
4248 * buffer head is unmapped
4249 * otherwise blocks are mapped
4250 *
4251 * return = 0, if plain look up failed (blocks have not been allocated)
4252 * buffer head is unmapped
4253 *
4254 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004255 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004256int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4257 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004258{
4259 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004260 struct ext4_extent newex, *ex, *ex2;
4261 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004262 ext4_fsblk_t newblock = 0;
Eric Whitneyce37c422014-02-19 18:52:39 -05004263 int free_on_err = 0, err = 0, depth, ret;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004264 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004265 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004266 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004267 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004268 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004269 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07004270
Mingming84fe3be2009-09-01 08:44:37 -04004271 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004272 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004273 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004274
Alex Tomasa86c6182006-10-11 01:21:03 -07004275 /* find extent for this block */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04004276 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004277 if (IS_ERR(path)) {
4278 err = PTR_ERR(path);
4279 path = NULL;
4280 goto out2;
4281 }
4282
4283 depth = ext_depth(inode);
4284
4285 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004286 * consistent leaf must not be empty;
4287 * this situation is possible, though, _during_ tree modification;
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04004288 * this is why assert can't be put in ext4_find_extent()
Alex Tomasa86c6182006-10-11 01:21:03 -07004289 */
Frank Mayhar273df552010-03-02 11:46:09 -05004290 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4291 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004292 "lblock: %lu, depth: %d pblock %lld",
4293 (unsigned long) map->m_lblk, depth,
4294 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004295 err = -EIO;
4296 goto out2;
4297 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004298
Avantika Mathur7e028972006-12-06 20:41:33 -08004299 ex = path[depth].p_ext;
4300 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004301 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004302 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004303 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004304
Lukas Czernerb8a86842014-03-18 18:05:35 -04004305
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004306 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04004307 * unwritten extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004308 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004309 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004310 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004311
4312 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4313
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004314 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004315 if (in_range(map->m_lblk, ee_block, ee_len)) {
4316 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004317 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004318 allocated = ee_len - (map->m_lblk - ee_block);
4319 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4320 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004321
Lukas Czernerb8a86842014-03-18 18:05:35 -04004322 /*
4323 * If the extent is initialized check whether the
4324 * caller wants to convert it to unwritten.
4325 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004326 if ((!ext4_ext_is_unwritten(ex)) &&
Lukas Czernerb8a86842014-03-18 18:05:35 -04004327 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04004328 allocated = convert_initialized_extent(
Theodore Ts'o4f224b82014-09-01 14:36:09 -04004329 handle, inode, map, &path,
4330 flags, allocated, newblock);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004331 goto out2;
Lukas Czerner556615d2014-04-20 23:45:47 -04004332 } else if (!ext4_ext_is_unwritten(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004333 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004334
Lukas Czerner556615d2014-04-20 23:45:47 -04004335 ret = ext4_ext_handle_unwritten_extents(
Theodore Ts'odfe50802014-09-01 14:37:09 -04004336 handle, inode, map, &path, flags,
Lukas Czerner78771912012-03-19 23:05:43 -04004337 allocated, newblock);
Eric Whitneyce37c422014-02-19 18:52:39 -05004338 if (ret < 0)
4339 err = ret;
4340 else
4341 allocated = ret;
Eric Whitney31cf0f22014-03-13 23:14:46 -04004342 goto out2;
Alex Tomasa86c6182006-10-11 01:21:03 -07004343 }
4344 }
4345
Aditya Kali7b415bf2011-09-09 19:04:51 -04004346 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05004347 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04004348 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4349
Alex Tomasa86c6182006-10-11 01:21:03 -07004350 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004351 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004352 * we couldn't try to create block if create flag is zero
4353 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004354 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004355 /*
4356 * put just found gap into cache to speed up
4357 * subsequent requests
4358 */
Zheng Liud100eef2013-02-18 00:29:59 -05004359 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4360 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004361 goto out2;
4362 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004363
Alex Tomasa86c6182006-10-11 01:21:03 -07004364 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004365 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004366 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004367 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004368 newex.ee_block = cpu_to_le32(map->m_lblk);
Eric Whitneyd0abafa2014-01-06 14:00:23 -05004369 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004370
4371 /*
4372 * If we are doing bigalloc, check to see if the extent returned
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04004373 * by ext4_find_extent() implies a cluster we can use.
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004374 */
4375 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004376 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004377 ar.len = allocated = map->m_len;
4378 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004379 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004380 goto got_allocated_blocks;
4381 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004382
Alex Tomasc9de5602008-01-29 00:19:52 -05004383 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004384 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004385 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4386 if (err)
4387 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004388 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004389 ex2 = NULL;
4390 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004391 if (err)
4392 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004393
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004394 /* Check if the extent after searching to the right implies a
4395 * cluster we can use. */
4396 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004397 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004398 ar.len = allocated = map->m_len;
4399 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004400 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004401 goto got_allocated_blocks;
4402 }
4403
Amit Arora749269f2007-07-18 09:02:56 -04004404 /*
4405 * See if request is beyond maximum number of blocks we can have in
4406 * a single extent. For an initialized extent this limit is
Lukas Czerner556615d2014-04-20 23:45:47 -04004407 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4408 * EXT_UNWRITTEN_MAX_LEN.
Amit Arora749269f2007-07-18 09:02:56 -04004409 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004410 if (map->m_len > EXT_INIT_MAX_LEN &&
Lukas Czerner556615d2014-04-20 23:45:47 -04004411 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004412 map->m_len = EXT_INIT_MAX_LEN;
Lukas Czerner556615d2014-04-20 23:45:47 -04004413 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4414 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4415 map->m_len = EXT_UNWRITTEN_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004416
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004417 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004418 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004419 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004420 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004421 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004422 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004423 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004424
4425 /* allocate new block */
4426 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004427 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4428 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004429 /*
4430 * We calculate the offset from the beginning of the cluster
4431 * for the logical block number, since when we allocate a
4432 * physical cluster, the physical block should start at the
4433 * same offset from the beginning of the cluster. This is
4434 * needed so that future calls to get_implied_cluster_alloc()
4435 * work correctly.
4436 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004437 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004438 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4439 ar.goal -= offset;
4440 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004441 if (S_ISREG(inode->i_mode))
4442 ar.flags = EXT4_MB_HINT_DATA;
4443 else
4444 /* disable in-core preallocation for non-regular files */
4445 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004446 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4447 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -04004448 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4449 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
Alex Tomasc9de5602008-01-29 00:19:52 -05004450 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004451 if (!newblock)
4452 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004453 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004454 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004455 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004456 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004457 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4458 if (ar.len > allocated)
4459 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004460
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004461got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004462 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004463 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004464 newex.ee_len = cpu_to_le16(ar.len);
Lukas Czerner556615d2014-04-20 23:45:47 -04004465 /* Mark unwritten */
4466 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4467 ext4_ext_mark_unwritten(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004468 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004469 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004470 * io_end structure was created for every IO write to an
Lukas Czerner556615d2014-04-20 23:45:47 -04004471 * unwritten extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004472 * here we flag the IO that really needs the conversion.
Mingming5f524952009-11-10 10:48:04 -05004473 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004474 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004475 */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004476 if (flags & EXT4_GET_BLOCKS_PRE_IO)
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004477 set_unwritten = 1;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004478 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004479
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004480 err = 0;
4481 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4482 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4483 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004484 if (!err)
Theodore Ts'odfe50802014-09-01 14:37:09 -04004485 err = ext4_ext_insert_extent(handle, inode, &path,
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004486 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004487
4488 if (!err && set_unwritten) {
4489 if (io)
4490 ext4_set_io_unwritten_flag(inode, io);
4491 else
4492 ext4_set_inode_state(inode,
4493 EXT4_STATE_DIO_UNWRITTEN);
4494 }
4495
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004496 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004497 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4498 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004499 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004500 /* not a good idea to call discard here directly,
4501 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004502 ext4_discard_preallocations(inode);
Theodore Ts'oc8e15132013-07-15 00:09:37 -04004503 ext4_free_blocks(handle, inode, NULL, newblock,
4504 EXT4_C2B(sbi, allocated_clusters), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004505 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004506 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004507
Alex Tomasa86c6182006-10-11 01:21:03 -07004508 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004509 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004510 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004511 if (allocated > map->m_len)
4512 allocated = map->m_len;
4513 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004514
Jan Karab436b9b2009-12-08 23:51:10 -05004515 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004516 * Update reserved blocks/metadata blocks after successful
4517 * block allocation which had been deferred till now.
4518 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004519 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004520 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004521 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004522 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004523 */
4524 reserved_clusters = get_reserved_cluster_alloc(inode,
4525 map->m_lblk, allocated);
4526 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4527 if (reserved_clusters) {
4528 /*
4529 * We have clusters reserved for this range.
4530 * But since we are not doing actual allocation
4531 * and are simply using blocks from previously
4532 * allocated cluster, we should release the
4533 * reservation and not claim quota.
4534 */
4535 ext4_da_update_reserve_space(inode,
4536 reserved_clusters, 0);
4537 }
4538 } else {
4539 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004540 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f262011-09-09 19:20:51 -04004541 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004542 int reservation = allocated_clusters -
4543 reserved_clusters;
4544 /*
4545 * It seems we claimed few clusters outside of
4546 * the range of this allocation. We should give
4547 * it back to the reservation pool. This can
4548 * happen in the following case:
4549 *
4550 * * Suppose s_cluster_ratio is 4 (i.e., each
4551 * cluster has 4 blocks. Thus, the clusters
4552 * are [0-3],[4-7],[8-11]...
4553 * * First comes delayed allocation write for
4554 * logical blocks 10 & 11. Since there were no
4555 * previous delayed allocated blocks in the
4556 * range [8-11], we would reserve 1 cluster
4557 * for this write.
4558 * * Next comes write for logical blocks 3 to 8.
4559 * In this case, we will reserve 2 clusters
4560 * (for [0-3] and [4-7]; and not for [8-11] as
4561 * that range has a delayed allocated blocks.
4562 * Thus total reserved clusters now becomes 3.
4563 * * Now, during the delayed allocation writeout
4564 * time, we will first write blocks [3-8] and
4565 * allocate 3 clusters for writing these
4566 * blocks. Also, we would claim all these
4567 * three clusters above.
4568 * * Now when we come here to writeout the
4569 * blocks [10-11], we would expect to claim
4570 * the reservation of 1 cluster we had made
4571 * (and we would claim it since there are no
4572 * more delayed allocated blocks in the range
4573 * [8-11]. But our reserved cluster count had
4574 * already gone to 0.
4575 *
4576 * Thus, at the step 4 above when we determine
4577 * that there are still some unwritten delayed
4578 * allocated blocks outside of our current
4579 * block range, we should increment the
4580 * reserved clusters count so that when the
4581 * remaining blocks finally gets written, we
4582 * could claim them.
4583 */
Aditya Kali5356f262011-09-09 19:20:51 -04004584 dquot_reserve_block(inode,
4585 EXT4_C2B(sbi, reservation));
4586 spin_lock(&ei->i_block_reservation_lock);
4587 ei->i_reserved_data_blocks += reservation;
4588 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004589 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004590 /*
4591 * We will claim quota for all newly allocated blocks.
4592 * We're updating the reserved space *after* the
4593 * correction above so we do not accidentally free
4594 * all the metadata reservation because we might
4595 * actually need it later on.
4596 */
4597 ext4_da_update_reserve_space(inode, allocated_clusters,
4598 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004599 }
4600 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004601
4602 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004603 * Cache the extent and update transaction to commit on fdatasync only
Lukas Czerner556615d2014-04-20 23:45:47 -04004604 * when it is _not_ an unwritten extent.
Jan Karab436b9b2009-12-08 23:51:10 -05004605 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004606 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004607 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004608 else
Jan Karab436b9b2009-12-08 23:51:10 -05004609 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004610out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004611 if (allocated > map->m_len)
4612 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004613 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004614 map->m_flags |= EXT4_MAP_MAPPED;
4615 map->m_pblk = newblock;
4616 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004617out2:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04004618 ext4_ext_drop_refs(path);
4619 kfree(path);
Allison Hendersone8613042011-05-25 07:41:46 -04004620
Theodore Ts'o63b99962013-07-16 10:28:47 -04004621 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4622 err ? err : allocated);
4623 ext4_es_lru_add(inode);
Lukas Czerner78771912012-03-19 23:05:43 -04004624 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004625}
4626
Theodore Ts'o819c4922013-04-03 12:47:17 -04004627void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004628{
Alex Tomasa86c6182006-10-11 01:21:03 -07004629 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004630 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004631 int err = 0;
4632
4633 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004634 * TODO: optimization is possible here.
4635 * Probably we need not scan at all,
4636 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004637 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004638
4639 /* we have to know where to truncate from in crash case */
4640 EXT4_I(inode)->i_disksize = inode->i_size;
4641 ext4_mark_inode_dirty(handle, inode);
4642
4643 last_block = (inode->i_size + sb->s_blocksize - 1)
4644 >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004645retry:
Zheng Liu51865fd2012-11-08 21:57:32 -05004646 err = ext4_es_remove_extent(inode, last_block,
4647 EXT_MAX_BLOCKS - last_block);
Theodore Ts'o94eec0f2013-07-29 12:12:56 -04004648 if (err == -ENOMEM) {
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004649 cond_resched();
4650 congestion_wait(BLK_RW_ASYNC, HZ/50);
4651 goto retry;
4652 }
4653 if (err) {
4654 ext4_std_error(inode->i_sb, err);
4655 return;
4656 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04004657 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004658 ext4_std_error(inode->i_sb, err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004659}
4660
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004661static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004662 ext4_lblk_t len, loff_t new_size,
4663 int flags, int mode)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004664{
Al Viro496ad9a2013-01-23 17:07:38 -05004665 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004666 handle_t *handle;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004667 int ret = 0;
4668 int ret2 = 0;
4669 int retries = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004670 struct ext4_map_blocks map;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004671 unsigned int credits;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004672 loff_t epos;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004673
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004674 map.m_lblk = offset;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004675 map.m_len = len;
Greg Harm3c6fe772011-10-31 18:41:47 -04004676 /*
4677 * Don't normalize the request if it can fit in one extent so
4678 * that it doesn't get unnecessarily split into multiple
4679 * extents.
4680 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004681 if (len <= EXT_UNWRITTEN_MAX_LEN)
Greg Harm3c6fe772011-10-31 18:41:47 -04004682 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004683
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004684 /*
4685 * credits to insert 1 extent into extent tree
4686 */
4687 credits = ext4_chunk_trans_blocks(inode, len);
4688
Amit Aroraa2df2a62007-07-17 21:42:41 -04004689retry:
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004690 while (ret >= 0 && len) {
Theodore Ts'o9924a922013-02-08 21:59:22 -05004691 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4692 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004693 if (IS_ERR(handle)) {
4694 ret = PTR_ERR(handle);
4695 break;
4696 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004697 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004698 if (ret <= 0) {
Lukas Czernerf282ac12014-03-18 17:44:35 -04004699 ext4_debug("inode #%lu: block %u: len %u: "
4700 "ext4_ext_map_blocks returned %d",
4701 inode->i_ino, map.m_lblk,
4702 map.m_len, ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004703 ext4_mark_inode_dirty(handle, inode);
4704 ret2 = ext4_journal_stop(handle);
4705 break;
4706 }
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004707 map.m_lblk += ret;
4708 map.m_len = len = len - ret;
4709 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4710 inode->i_ctime = ext4_current_time(inode);
4711 if (new_size) {
4712 if (epos > new_size)
4713 epos = new_size;
4714 if (ext4_update_inode_size(inode, epos) & 0x1)
4715 inode->i_mtime = inode->i_ctime;
4716 } else {
4717 if (epos > inode->i_size)
4718 ext4_set_inode_flag(inode,
4719 EXT4_INODE_EOFBLOCKS);
4720 }
4721 ext4_mark_inode_dirty(handle, inode);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004722 ret2 = ext4_journal_stop(handle);
4723 if (ret2)
4724 break;
4725 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004726 if (ret == -ENOSPC &&
4727 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4728 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004729 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004730 }
Lukas Czernerf282ac12014-03-18 17:44:35 -04004731
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004732 return ret > 0 ? ret2 : ret;
4733}
4734
Lukas Czernerb8a86842014-03-18 18:05:35 -04004735static long ext4_zero_range(struct file *file, loff_t offset,
4736 loff_t len, int mode)
4737{
4738 struct inode *inode = file_inode(file);
4739 handle_t *handle = NULL;
4740 unsigned int max_blocks;
4741 loff_t new_size = 0;
4742 int ret = 0;
4743 int flags;
Dmitry Monakhov69dc9532014-08-27 18:33:49 -04004744 int credits;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004745 int partial_begin, partial_end;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004746 loff_t start, end;
4747 ext4_lblk_t lblk;
4748 struct address_space *mapping = inode->i_mapping;
4749 unsigned int blkbits = inode->i_blkbits;
4750
4751 trace_ext4_zero_range(inode, offset, len, mode);
4752
jon ernst6c5e73d2014-04-18 11:50:35 -04004753 if (!S_ISREG(inode->i_mode))
4754 return -EINVAL;
4755
Namjae Jeone1ee60f2014-05-27 12:48:55 -04004756 /* Call ext4_force_commit to flush all data in case of data=journal. */
4757 if (ext4_should_journal_data(inode)) {
4758 ret = ext4_force_commit(inode->i_sb);
4759 if (ret)
4760 return ret;
4761 }
4762
Lukas Czernerb8a86842014-03-18 18:05:35 -04004763 /*
4764 * Write out all dirty pages to avoid race conditions
4765 * Then release them.
4766 */
4767 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4768 ret = filemap_write_and_wait_range(mapping, offset,
4769 offset + len - 1);
4770 if (ret)
4771 return ret;
4772 }
4773
4774 /*
4775 * Round up offset. This is not fallocate, we neet to zero out
4776 * blocks, so convert interior block aligned part of the range to
4777 * unwritten and possibly manually zero out unaligned parts of the
4778 * range.
4779 */
4780 start = round_up(offset, 1 << blkbits);
4781 end = round_down((offset + len), 1 << blkbits);
4782
4783 if (start < offset || end > offset + len)
4784 return -EINVAL;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004785 partial_begin = offset & ((1 << blkbits) - 1);
4786 partial_end = (offset + len) & ((1 << blkbits) - 1);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004787
4788 lblk = start >> blkbits;
4789 max_blocks = (end >> blkbits);
4790 if (max_blocks < lblk)
4791 max_blocks = 0;
4792 else
4793 max_blocks -= lblk;
4794
Lukas Czerner556615d2014-04-20 23:45:47 -04004795 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
Theodore Ts'o713e8dd2014-09-01 14:32:09 -04004796 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4797 EXT4_EX_NOCACHE;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004798 if (mode & FALLOC_FL_KEEP_SIZE)
4799 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4800
4801 mutex_lock(&inode->i_mutex);
4802
4803 /*
4804 * Indirect files do not support unwritten extnets
4805 */
4806 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4807 ret = -EOPNOTSUPP;
4808 goto out_mutex;
4809 }
4810
4811 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4812 offset + len > i_size_read(inode)) {
4813 new_size = offset + len;
4814 ret = inode_newsize_ok(inode, new_size);
4815 if (ret)
4816 goto out_mutex;
4817 /*
4818 * If we have a partial block after EOF we have to allocate
4819 * the entire block.
4820 */
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004821 if (partial_end)
Lukas Czernerb8a86842014-03-18 18:05:35 -04004822 max_blocks += 1;
4823 }
4824
4825 if (max_blocks > 0) {
4826
4827 /* Now release the pages and zero block aligned part of pages*/
4828 truncate_pagecache_range(inode, start, end - 1);
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004829 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004830
4831 /* Wait all existing dio workers, newcomers will block on i_mutex */
4832 ext4_inode_block_unlocked_dio(inode);
4833 inode_dio_wait(inode);
4834
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004835 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4836 flags, mode);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004837 if (ret)
4838 goto out_dio;
Theodore Ts'o713e8dd2014-09-01 14:32:09 -04004839 /*
4840 * Remove entire range from the extent status tree.
4841 *
4842 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4843 * NOT sufficient. I'm not sure why this is the case,
4844 * but let's be conservative and remove the extent
4845 * status tree for the entire inode. There should be
4846 * no outstanding delalloc extents thanks to the
4847 * filemap_write_and_wait_range() call above.
4848 */
4849 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4850 if (ret)
4851 goto out_dio;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004852 }
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004853 if (!partial_begin && !partial_end)
4854 goto out_dio;
4855
Dmitry Monakhov69dc9532014-08-27 18:33:49 -04004856 /*
4857 * In worst case we have to writeout two nonadjacent unwritten
4858 * blocks and update the inode
4859 */
4860 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4861 if (ext4_should_journal_data(inode))
4862 credits += 2;
4863 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004864 if (IS_ERR(handle)) {
4865 ret = PTR_ERR(handle);
4866 ext4_std_error(inode->i_sb, ret);
4867 goto out_dio;
4868 }
4869
4870 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Lukas Czernere5b30412014-04-01 00:59:21 -04004871 if (new_size) {
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04004872 ext4_update_inode_size(inode, new_size);
Lukas Czernere5b30412014-04-01 00:59:21 -04004873 } else {
Lukas Czernerb8a86842014-03-18 18:05:35 -04004874 /*
4875 * Mark that we allocate beyond EOF so the subsequent truncate
4876 * can proceed even if the new size is the same as i_size.
4877 */
4878 if ((offset + len) > i_size_read(inode))
4879 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4880 }
Lukas Czernerb8a86842014-03-18 18:05:35 -04004881 ext4_mark_inode_dirty(handle, inode);
4882
4883 /* Zero out partial block at the edges of the range */
4884 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4885
4886 if (file->f_flags & O_SYNC)
4887 ext4_handle_sync(handle);
4888
4889 ext4_journal_stop(handle);
4890out_dio:
4891 ext4_inode_resume_unlocked_dio(inode);
4892out_mutex:
4893 mutex_unlock(&inode->i_mutex);
4894 return ret;
4895}
4896
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004897/*
4898 * preallocate space for a file. This implements ext4's fallocate file
4899 * operation, which gets called from sys_fallocate system call.
4900 * For block-mapped files, posix_fallocate should fall back to the method
4901 * of writing zeroes to the required new blocks (the same behavior which is
4902 * expected for file systems which do not support fallocate() system call).
4903 */
4904long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4905{
4906 struct inode *inode = file_inode(file);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004907 loff_t new_size = 0;
4908 unsigned int max_blocks;
4909 int ret = 0;
4910 int flags;
4911 ext4_lblk_t lblk;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004912 unsigned int blkbits = inode->i_blkbits;
4913
4914 /* Return error if mode is not supported */
4915 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
Lukas Czernerb8a86842014-03-18 18:05:35 -04004916 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004917 return -EOPNOTSUPP;
4918
4919 if (mode & FALLOC_FL_PUNCH_HOLE)
4920 return ext4_punch_hole(inode, offset, len);
4921
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004922 ret = ext4_convert_inline_data(inode);
4923 if (ret)
4924 return ret;
4925
4926 /*
4927 * currently supporting (pre)allocate mode for extent-based
4928 * files _only_
4929 */
4930 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4931 return -EOPNOTSUPP;
4932
Theodore Ts'o40c406c2014-04-12 22:53:53 -04004933 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4934 return ext4_collapse_range(inode, offset, len);
4935
Lukas Czernerb8a86842014-03-18 18:05:35 -04004936 if (mode & FALLOC_FL_ZERO_RANGE)
4937 return ext4_zero_range(file, offset, len, mode);
4938
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004939 trace_ext4_fallocate_enter(inode, offset, len, mode);
4940 lblk = offset >> blkbits;
4941 /*
4942 * We can't just convert len to max_blocks because
4943 * If blocksize = 4096 offset = 3072 and len = 2048
4944 */
4945 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4946 - lblk;
4947
Lukas Czerner556615d2014-04-20 23:45:47 -04004948 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004949 if (mode & FALLOC_FL_KEEP_SIZE)
4950 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4951
4952 mutex_lock(&inode->i_mutex);
4953
4954 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4955 offset + len > i_size_read(inode)) {
4956 new_size = offset + len;
4957 ret = inode_newsize_ok(inode, new_size);
4958 if (ret)
4959 goto out;
4960 }
4961
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004962 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4963 flags, mode);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004964 if (ret)
4965 goto out;
4966
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004967 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4968 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4969 EXT4_I(inode)->i_sync_tid);
Lukas Czernerf282ac12014-03-18 17:44:35 -04004970 }
Lukas Czernerf282ac12014-03-18 17:44:35 -04004971out:
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004972 mutex_unlock(&inode->i_mutex);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004973 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4974 return ret;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004975}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004976
4977/*
Mingming Cao00314622009-09-28 15:49:08 -04004978 * This function convert a range of blocks to written extents
4979 * The caller of this function will pass the start offset and the size.
4980 * all unwritten extents within this range will be converted to
4981 * written extents.
4982 *
4983 * This function is called from the direct IO end io call back
4984 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004985 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004986 */
Jan Kara6b523df2013-06-04 13:21:11 -04004987int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4988 loff_t offset, ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004989{
Mingming Cao00314622009-09-28 15:49:08 -04004990 unsigned int max_blocks;
4991 int ret = 0;
4992 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004993 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004994 unsigned int credits, blkbits = inode->i_blkbits;
4995
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004996 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004997 /*
4998 * We can't just convert len to max_blocks because
4999 * If blocksize = 4096 offset = 3072 and len = 2048
5000 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005001 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5002 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04005003 /*
Jan Kara6b523df2013-06-04 13:21:11 -04005004 * This is somewhat ugly but the idea is clear: When transaction is
5005 * reserved, everything goes into it. Otherwise we rather start several
5006 * smaller transactions for conversion of each extent separately.
Mingming Cao00314622009-09-28 15:49:08 -04005007 */
Jan Kara6b523df2013-06-04 13:21:11 -04005008 if (handle) {
5009 handle = ext4_journal_start_reserved(handle,
5010 EXT4_HT_EXT_CONVERT);
5011 if (IS_ERR(handle))
5012 return PTR_ERR(handle);
5013 credits = 0;
5014 } else {
5015 /*
5016 * credits to insert 1 extent into extent tree
5017 */
5018 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5019 }
Mingming Cao00314622009-09-28 15:49:08 -04005020 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005021 map.m_lblk += ret;
5022 map.m_len = (max_blocks -= ret);
Jan Kara6b523df2013-06-04 13:21:11 -04005023 if (credits) {
5024 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5025 credits);
5026 if (IS_ERR(handle)) {
5027 ret = PTR_ERR(handle);
5028 break;
5029 }
Mingming Cao00314622009-09-28 15:49:08 -04005030 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005031 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05005032 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05005033 if (ret <= 0)
5034 ext4_warning(inode->i_sb,
5035 "inode #%lu: block %u: len %u: "
5036 "ext4_ext_map_blocks returned %d",
5037 inode->i_ino, map.m_lblk,
5038 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04005039 ext4_mark_inode_dirty(handle, inode);
Jan Kara6b523df2013-06-04 13:21:11 -04005040 if (credits)
5041 ret2 = ext4_journal_stop(handle);
5042 if (ret <= 0 || ret2)
Mingming Cao00314622009-09-28 15:49:08 -04005043 break;
5044 }
Jan Kara6b523df2013-06-04 13:21:11 -04005045 if (!credits)
5046 ret2 = ext4_journal_stop(handle);
Mingming Cao00314622009-09-28 15:49:08 -04005047 return ret > 0 ? ret2 : ret;
5048}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005049
Mingming Cao00314622009-09-28 15:49:08 -04005050/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05005051 * If newes is not existing extent (newes->ec_pblk equals zero) find
5052 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005053 * return start of the next delayed extent.
5054 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05005055 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005056 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05005057 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04005058 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005059static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05005060 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04005061{
Zheng Liub3aff3e2012-11-08 21:57:37 -05005062 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05005063 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005064
Zheng Liu69eb33d2013-02-18 00:31:07 -05005065 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04005066 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5067 newes->es_lblk + newes->es_len - 1, &es);
5068
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005069 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05005070 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005071 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005072 */
Zheng Liu06b0c882013-02-18 00:26:51 -05005073 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05005074 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005075 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005076
Zheng Liu69eb33d2013-02-18 00:31:07 -05005077 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05005078 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05005079 newes->es_len = min(es.es_lblk - newes->es_lblk,
5080 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005081 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005082 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005083
Zheng Liu69eb33d2013-02-18 00:31:07 -05005084 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005085 }
5086
Zheng Liu69eb33d2013-02-18 00:31:07 -05005087 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04005088 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05005089 if (es.es_len == 0)
5090 next_del = EXT_MAX_BLOCKS;
5091 else
5092 next_del = es.es_lblk;
5093
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005094 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005095}
Eric Sandeen6873fa02008-10-07 00:46:36 -04005096/* fiemap flags we can handle specified here */
5097#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5098
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05005099static int ext4_xattr_fiemap(struct inode *inode,
5100 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04005101{
5102 __u64 physical = 0;
5103 __u64 length;
5104 __u32 flags = FIEMAP_EXTENT_LAST;
5105 int blockbits = inode->i_sb->s_blocksize_bits;
5106 int error = 0;
5107
5108 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005109 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04005110 struct ext4_iloc iloc;
5111 int offset; /* offset of xattr in inode */
5112
5113 error = ext4_get_inode_loc(inode, &iloc);
5114 if (error)
5115 return error;
Jan Karaa60697f2013-05-31 19:38:56 -04005116 physical = (__u64)iloc.bh->b_blocknr << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005117 offset = EXT4_GOOD_OLD_INODE_SIZE +
5118 EXT4_I(inode)->i_extra_isize;
5119 physical += offset;
5120 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5121 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04005122 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005123 } else { /* external block */
Jan Karaa60697f2013-05-31 19:38:56 -04005124 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005125 length = inode->i_sb->s_blocksize;
5126 }
5127
5128 if (physical)
5129 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5130 length, flags);
5131 return (error < 0 ? error : 0);
5132}
5133
5134int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5135 __u64 start, __u64 len)
5136{
5137 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005138 int error = 0;
5139
Tao Ma94191982012-12-10 14:06:02 -05005140 if (ext4_has_inline_data(inode)) {
5141 int has_inline = 1;
5142
5143 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5144
5145 if (has_inline)
5146 return error;
5147 }
5148
Theodore Ts'o7869a4a2013-08-16 22:05:14 -04005149 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5150 error = ext4_ext_precache(inode);
5151 if (error)
5152 return error;
5153 }
5154
Eric Sandeen6873fa02008-10-07 00:46:36 -04005155 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005156 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04005157 return generic_block_fiemap(inode, fieinfo, start, len,
5158 ext4_get_block);
5159
5160 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5161 return -EBADR;
5162
5163 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5164 error = ext4_xattr_fiemap(inode, fieinfo);
5165 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005166 ext4_lblk_t len_blks;
5167 __u64 last_blk;
5168
Eric Sandeen6873fa02008-10-07 00:46:36 -04005169 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005170 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04005171 if (last_blk >= EXT_MAX_BLOCKS)
5172 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005173 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005174
5175 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005176 * Walk the extent tree gathering extent information
5177 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04005178 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005179 error = ext4_fill_fiemap_extents(inode, start_blk,
5180 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005181 }
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04005182 ext4_es_lru_add(inode);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005183 return error;
5184}
Namjae Jeon9eb79482014-02-23 15:18:59 -05005185
5186/*
5187 * ext4_access_path:
5188 * Function to access the path buffer for marking it dirty.
5189 * It also checks if there are sufficient credits left in the journal handle
5190 * to update path.
5191 */
5192static int
5193ext4_access_path(handle_t *handle, struct inode *inode,
5194 struct ext4_ext_path *path)
5195{
5196 int credits, err;
5197
5198 if (!ext4_handle_valid(handle))
5199 return 0;
5200
5201 /*
5202 * Check if need to extend journal credits
5203 * 3 for leaf, sb, and inode plus 2 (bmap and group
5204 * descriptor) for each block group; assume two block
5205 * groups
5206 */
5207 if (handle->h_buffer_credits < 7) {
5208 credits = ext4_writepage_trans_blocks(inode);
5209 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5210 /* EAGAIN is success */
5211 if (err && err != -EAGAIN)
5212 return err;
5213 }
5214
5215 err = ext4_ext_get_access(handle, inode, path);
5216 return err;
5217}
5218
5219/*
5220 * ext4_ext_shift_path_extents:
5221 * Shift the extents of a path structure lying between path[depth].p_ext
5222 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5223 * from starting block for each extent.
5224 */
5225static int
5226ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5227 struct inode *inode, handle_t *handle,
5228 ext4_lblk_t *start)
5229{
5230 int depth, err = 0;
5231 struct ext4_extent *ex_start, *ex_last;
5232 bool update = 0;
5233 depth = path->p_depth;
5234
5235 while (depth >= 0) {
5236 if (depth == path->p_depth) {
5237 ex_start = path[depth].p_ext;
5238 if (!ex_start)
5239 return -EIO;
5240
5241 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5242 if (!ex_last)
5243 return -EIO;
5244
5245 err = ext4_access_path(handle, inode, path + depth);
5246 if (err)
5247 goto out;
5248
5249 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5250 update = 1;
5251
Zheng Liu847c6c42014-04-12 12:45:55 -04005252 *start = le32_to_cpu(ex_last->ee_block) +
Namjae Jeon9eb79482014-02-23 15:18:59 -05005253 ext4_ext_get_actual_len(ex_last);
5254
5255 while (ex_start <= ex_last) {
Zheng Liu847c6c42014-04-12 12:45:55 -04005256 le32_add_cpu(&ex_start->ee_block, -shift);
Lukas Czerner6dd834e2014-04-18 10:55:24 -04005257 /* Try to merge to the left. */
5258 if ((ex_start >
5259 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5260 ext4_ext_try_to_merge_right(inode,
5261 path, ex_start - 1))
5262 ex_last--;
5263 else
5264 ex_start++;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005265 }
5266 err = ext4_ext_dirty(handle, inode, path + depth);
5267 if (err)
5268 goto out;
5269
5270 if (--depth < 0 || !update)
5271 break;
5272 }
5273
5274 /* Update index too */
5275 err = ext4_access_path(handle, inode, path + depth);
5276 if (err)
5277 goto out;
5278
Zheng Liu847c6c42014-04-12 12:45:55 -04005279 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005280 err = ext4_ext_dirty(handle, inode, path + depth);
5281 if (err)
5282 goto out;
5283
5284 /* we are done if current index is not a starting index */
5285 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5286 break;
5287
5288 depth--;
5289 }
5290
5291out:
5292 return err;
5293}
5294
5295/*
5296 * ext4_ext_shift_extents:
5297 * All the extents which lies in the range from start to the last allocated
5298 * block for the file are shifted downwards by shift blocks.
5299 * On success, 0 is returned, error otherwise.
5300 */
5301static int
5302ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5303 ext4_lblk_t start, ext4_lblk_t shift)
5304{
5305 struct ext4_ext_path *path;
5306 int ret = 0, depth;
5307 struct ext4_extent *extent;
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005308 ext4_lblk_t stop_block;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005309 ext4_lblk_t ex_start, ex_end;
5310
5311 /* Let path point to the last extent */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005312 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005313 if (IS_ERR(path))
5314 return PTR_ERR(path);
5315
5316 depth = path->p_depth;
5317 extent = path[depth].p_ext;
Theodore Ts'oee4bd0d92014-09-01 14:41:09 -04005318 if (!extent)
5319 goto out;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005320
Zheng Liu847c6c42014-04-12 12:45:55 -04005321 stop_block = le32_to_cpu(extent->ee_block) +
5322 ext4_ext_get_actual_len(extent);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005323
5324 /* Nothing to shift, if hole is at the end of file */
5325 if (start >= stop_block)
Theodore Ts'oee4bd0d92014-09-01 14:41:09 -04005326 goto out;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005327
5328 /*
5329 * Don't start shifting extents until we make sure the hole is big
5330 * enough to accomodate the shift.
5331 */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005332 path = ext4_find_extent(inode, start - 1, &path, 0);
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005333 if (IS_ERR(path))
5334 return PTR_ERR(path);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005335 depth = path->p_depth;
5336 extent = path[depth].p_ext;
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005337 if (extent) {
5338 ex_start = le32_to_cpu(extent->ee_block);
5339 ex_end = le32_to_cpu(extent->ee_block) +
Zheng Liu847c6c42014-04-12 12:45:55 -04005340 ext4_ext_get_actual_len(extent);
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005341 } else {
5342 ex_start = 0;
5343 ex_end = 0;
5344 }
Namjae Jeon9eb79482014-02-23 15:18:59 -05005345
5346 if ((start == ex_start && shift > ex_start) ||
5347 (shift > start - ex_end))
5348 return -EINVAL;
5349
5350 /* Its safe to start updating extents */
5351 while (start < stop_block) {
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005352 path = ext4_find_extent(inode, start, &path, 0);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005353 if (IS_ERR(path))
5354 return PTR_ERR(path);
5355 depth = path->p_depth;
5356 extent = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04005357 if (!extent) {
5358 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5359 (unsigned long) start);
5360 return -EIO;
5361 }
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005362 if (start > le32_to_cpu(extent->ee_block)) {
Namjae Jeon9eb79482014-02-23 15:18:59 -05005363 /* Hole, move to the next extent */
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005364 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5365 path[depth].p_ext++;
5366 } else {
5367 start = ext4_ext_next_allocated_block(path);
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005368 continue;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005369 }
5370 }
5371 ret = ext4_ext_shift_path_extents(path, shift, inode,
5372 handle, &start);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005373 if (ret)
5374 break;
5375 }
Theodore Ts'oee4bd0d92014-09-01 14:41:09 -04005376out:
5377 ext4_ext_drop_refs(path);
5378 kfree(path);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005379 return ret;
5380}
5381
5382/*
5383 * ext4_collapse_range:
5384 * This implements the fallocate's collapse range functionality for ext4
5385 * Returns: 0 and non-zero on error.
5386 */
5387int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5388{
5389 struct super_block *sb = inode->i_sb;
5390 ext4_lblk_t punch_start, punch_stop;
5391 handle_t *handle;
5392 unsigned int credits;
Namjae Jeona8680e02014-04-19 16:37:31 -04005393 loff_t new_size, ioffset;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005394 int ret;
5395
Namjae Jeon9eb79482014-02-23 15:18:59 -05005396 /* Collapse range works only on fs block size aligned offsets. */
Namjae Jeonee98fa32014-07-29 10:45:31 -04005397 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5398 len & (EXT4_CLUSTER_SIZE(sb) - 1))
Namjae Jeon9eb79482014-02-23 15:18:59 -05005399 return -EINVAL;
5400
5401 if (!S_ISREG(inode->i_mode))
Theodore Ts'o86f1ca32014-04-18 11:52:11 -04005402 return -EINVAL;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005403
5404 trace_ext4_collapse_range(inode, offset, len);
5405
5406 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5407 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5408
Namjae Jeon1ce01c42014-04-10 22:58:20 -04005409 /* Call ext4_force_commit to flush all data in case of data=journal. */
5410 if (ext4_should_journal_data(inode)) {
5411 ret = ext4_force_commit(inode->i_sb);
5412 if (ret)
5413 return ret;
5414 }
5415
Namjae Jeona8680e02014-04-19 16:37:31 -04005416 /*
5417 * Need to round down offset to be aligned with page size boundary
5418 * for page size > block size.
5419 */
5420 ioffset = round_down(offset, PAGE_SIZE);
5421
Namjae Jeon9eb79482014-02-23 15:18:59 -05005422 /* Write out all dirty pages */
Namjae Jeona8680e02014-04-19 16:37:31 -04005423 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5424 LLONG_MAX);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005425 if (ret)
5426 return ret;
5427
5428 /* Take mutex lock */
5429 mutex_lock(&inode->i_mutex);
5430
Lukas Czerner23fffa92014-04-12 09:56:41 -04005431 /*
5432 * There is no need to overlap collapse range with EOF, in which case
5433 * it is effectively a truncate operation
5434 */
5435 if (offset + len >= i_size_read(inode)) {
5436 ret = -EINVAL;
5437 goto out_mutex;
5438 }
5439
Namjae Jeon9eb79482014-02-23 15:18:59 -05005440 /* Currently just for extent based files */
5441 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5442 ret = -EOPNOTSUPP;
5443 goto out_mutex;
5444 }
5445
Namjae Jeona8680e02014-04-19 16:37:31 -04005446 truncate_pagecache(inode, ioffset);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005447
5448 /* Wait for existing dio to complete */
5449 ext4_inode_block_unlocked_dio(inode);
5450 inode_dio_wait(inode);
5451
5452 credits = ext4_writepage_trans_blocks(inode);
5453 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5454 if (IS_ERR(handle)) {
5455 ret = PTR_ERR(handle);
5456 goto out_dio;
5457 }
5458
5459 down_write(&EXT4_I(inode)->i_data_sem);
5460 ext4_discard_preallocations(inode);
5461
5462 ret = ext4_es_remove_extent(inode, punch_start,
Lukas Czerner2c1d2322014-04-18 10:43:21 -04005463 EXT_MAX_BLOCKS - punch_start);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005464 if (ret) {
5465 up_write(&EXT4_I(inode)->i_data_sem);
5466 goto out_stop;
5467 }
5468
5469 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5470 if (ret) {
5471 up_write(&EXT4_I(inode)->i_data_sem);
5472 goto out_stop;
5473 }
Lukas Czerneref24f6c2014-04-18 10:50:23 -04005474 ext4_discard_preallocations(inode);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005475
5476 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5477 punch_stop - punch_start);
5478 if (ret) {
5479 up_write(&EXT4_I(inode)->i_data_sem);
5480 goto out_stop;
5481 }
5482
5483 new_size = i_size_read(inode) - len;
Lukas Czerner9337d5d2014-04-18 10:48:25 -04005484 i_size_write(inode, new_size);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005485 EXT4_I(inode)->i_disksize = new_size;
5486
Namjae Jeon9eb79482014-02-23 15:18:59 -05005487 up_write(&EXT4_I(inode)->i_data_sem);
5488 if (IS_SYNC(inode))
5489 ext4_handle_sync(handle);
5490 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5491 ext4_mark_inode_dirty(handle, inode);
5492
5493out_stop:
5494 ext4_journal_stop(handle);
5495out_dio:
5496 ext4_inode_resume_unlocked_dio(inode);
5497out_mutex:
5498 mutex_unlock(&inode->i_mutex);
5499 return ret;
5500}
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005501
5502/**
5503 * ext4_swap_extents - Swap extents between two inodes
5504 *
5505 * @inode1: First inode
5506 * @inode2: Second inode
5507 * @lblk1: Start block for first inode
5508 * @lblk2: Start block for second inode
5509 * @count: Number of blocks to swap
5510 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5511 * @erp: Pointer to save error value
5512 *
5513 * This helper routine does exactly what is promise "swap extents". All other
5514 * stuff such as page-cache locking consistency, bh mapping consistency or
5515 * extent's data copying must be performed by caller.
5516 * Locking:
5517 * i_mutex is held for both inodes
5518 * i_data_sem is locked for write for both inodes
5519 * Assumptions:
5520 * All pages from requested range are locked for both inodes
5521 */
5522int
5523ext4_swap_extents(handle_t *handle, struct inode *inode1,
5524 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5525 ext4_lblk_t count, int unwritten, int *erp)
5526{
5527 struct ext4_ext_path *path1 = NULL;
5528 struct ext4_ext_path *path2 = NULL;
5529 int replaced_count = 0;
5530
5531 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5532 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5533 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5534 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5535
5536 *erp = ext4_es_remove_extent(inode1, lblk1, count);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005537 if (unlikely(*erp))
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005538 return 0;
5539 *erp = ext4_es_remove_extent(inode2, lblk2, count);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005540 if (unlikely(*erp))
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005541 return 0;
5542
5543 while (count) {
5544 struct ext4_extent *ex1, *ex2, tmp_ex;
5545 ext4_lblk_t e1_blk, e2_blk;
5546 int e1_len, e2_len, len;
5547 int split = 0;
5548
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005549 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005550 if (unlikely(IS_ERR(path1))) {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005551 *erp = PTR_ERR(path1);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005552 path1 = NULL;
5553 finish:
5554 count = 0;
5555 goto repeat;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005556 }
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005557 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005558 if (unlikely(IS_ERR(path2))) {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005559 *erp = PTR_ERR(path2);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005560 path2 = NULL;
5561 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005562 }
5563 ex1 = path1[path1->p_depth].p_ext;
5564 ex2 = path2[path2->p_depth].p_ext;
5565 /* Do we have somthing to swap ? */
5566 if (unlikely(!ex2 || !ex1))
Theodore Ts'o19008f62014-08-31 15:03:14 -04005567 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005568
5569 e1_blk = le32_to_cpu(ex1->ee_block);
5570 e2_blk = le32_to_cpu(ex2->ee_block);
5571 e1_len = ext4_ext_get_actual_len(ex1);
5572 e2_len = ext4_ext_get_actual_len(ex2);
5573
5574 /* Hole handling */
5575 if (!in_range(lblk1, e1_blk, e1_len) ||
5576 !in_range(lblk2, e2_blk, e2_len)) {
5577 ext4_lblk_t next1, next2;
5578
5579 /* if hole after extent, then go to next extent */
5580 next1 = ext4_ext_next_allocated_block(path1);
5581 next2 = ext4_ext_next_allocated_block(path2);
5582 /* If hole before extent, then shift to that extent */
5583 if (e1_blk > lblk1)
5584 next1 = e1_blk;
5585 if (e2_blk > lblk2)
5586 next2 = e1_blk;
5587 /* Do we have something to swap */
5588 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
Theodore Ts'o19008f62014-08-31 15:03:14 -04005589 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005590 /* Move to the rightest boundary */
5591 len = next1 - lblk1;
5592 if (len < next2 - lblk2)
5593 len = next2 - lblk2;
5594 if (len > count)
5595 len = count;
5596 lblk1 += len;
5597 lblk2 += len;
5598 count -= len;
5599 goto repeat;
5600 }
5601
5602 /* Prepare left boundary */
5603 if (e1_blk < lblk1) {
5604 split = 1;
5605 *erp = ext4_force_split_extent_at(handle, inode1,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005606 &path1, lblk1, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005607 if (unlikely(*erp))
5608 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005609 }
5610 if (e2_blk < lblk2) {
5611 split = 1;
5612 *erp = ext4_force_split_extent_at(handle, inode2,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005613 &path2, lblk2, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005614 if (unlikely(*erp))
5615 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005616 }
Theodore Ts'odfe50802014-09-01 14:37:09 -04005617 /* ext4_split_extent_at() may result in leaf extent split,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005618 * path must to be revalidated. */
5619 if (split)
5620 goto repeat;
5621
5622 /* Prepare right boundary */
5623 len = count;
5624 if (len > e1_blk + e1_len - lblk1)
5625 len = e1_blk + e1_len - lblk1;
5626 if (len > e2_blk + e2_len - lblk2)
5627 len = e2_blk + e2_len - lblk2;
5628
5629 if (len != e1_len) {
5630 split = 1;
5631 *erp = ext4_force_split_extent_at(handle, inode1,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005632 &path1, lblk1 + len, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005633 if (unlikely(*erp))
5634 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005635 }
5636 if (len != e2_len) {
5637 split = 1;
5638 *erp = ext4_force_split_extent_at(handle, inode2,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005639 &path2, lblk2 + len, 0);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005640 if (*erp)
Theodore Ts'o19008f62014-08-31 15:03:14 -04005641 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005642 }
Theodore Ts'odfe50802014-09-01 14:37:09 -04005643 /* ext4_split_extent_at() may result in leaf extent split,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005644 * path must to be revalidated. */
5645 if (split)
5646 goto repeat;
5647
5648 BUG_ON(e2_len != e1_len);
5649 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005650 if (unlikely(*erp))
5651 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005652 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005653 if (unlikely(*erp))
5654 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005655
5656 /* Both extents are fully inside boundaries. Swap it now */
5657 tmp_ex = *ex1;
5658 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5659 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5660 ex1->ee_len = cpu_to_le16(e2_len);
5661 ex2->ee_len = cpu_to_le16(e1_len);
5662 if (unwritten)
5663 ext4_ext_mark_unwritten(ex2);
5664 if (ext4_ext_is_unwritten(&tmp_ex))
5665 ext4_ext_mark_unwritten(ex1);
5666
5667 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5668 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5669 *erp = ext4_ext_dirty(handle, inode2, path2 +
5670 path2->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005671 if (unlikely(*erp))
5672 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005673 *erp = ext4_ext_dirty(handle, inode1, path1 +
5674 path1->p_depth);
5675 /*
5676 * Looks scarry ah..? second inode already points to new blocks,
5677 * and it was successfully dirtied. But luckily error may happen
5678 * only due to journal error, so full transaction will be
5679 * aborted anyway.
5680 */
Theodore Ts'o19008f62014-08-31 15:03:14 -04005681 if (unlikely(*erp))
5682 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005683 lblk1 += len;
5684 lblk2 += len;
5685 replaced_count += len;
5686 count -= len;
5687
5688 repeat:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04005689 ext4_ext_drop_refs(path1);
5690 kfree(path1);
5691 ext4_ext_drop_refs(path2);
5692 kfree(path2);
5693 path1 = path2 = NULL;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005694 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005695 return replaced_count;
5696}