blob: d74e08029643d3e5eeb1f8a14d9a78c290549863 [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
Lukas Czerner556615d2014-04-20 23:45:47 -04001720 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001721 return 0;
1722
1723 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1724 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1725
1726 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001727 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001728 return 0;
1729
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001730 /*
1731 * To allow future support for preallocated extents to be added
1732 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001733 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001734 */
Eric Sandeenda0169b2013-11-04 09:58:26 -05001735 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001736 return 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04001737 if (ext4_ext_is_unwritten(ex1) &&
Darrick J. Wonga9b82412014-02-20 21:17:35 -05001738 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1739 atomic_read(&EXT4_I(inode)->i_unwritten) ||
Lukas Czerner556615d2014-04-20 23:45:47 -04001740 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
Darrick J. Wonga9b82412014-02-20 21:17:35 -05001741 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001742#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001743 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001744 return 0;
1745#endif
1746
Theodore Ts'obf89d162010-10-27 21:30:14 -04001747 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001748 return 1;
1749 return 0;
1750}
1751
1752/*
Amit Arora56055d32007-07-17 21:42:38 -04001753 * This function tries to merge the "ex" extent to the next extent in the tree.
1754 * It always tries to merge towards right. If you want to merge towards
1755 * left, pass "ex - 1" as argument instead of "ex".
1756 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1757 * 1 if they got merged.
1758 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001759static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001760 struct ext4_ext_path *path,
1761 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001762{
1763 struct ext4_extent_header *eh;
1764 unsigned int depth, len;
Lukas Czerner556615d2014-04-20 23:45:47 -04001765 int merge_done = 0, unwritten;
Amit Arora56055d32007-07-17 21:42:38 -04001766
1767 depth = ext_depth(inode);
1768 BUG_ON(path[depth].p_hdr == NULL);
1769 eh = path[depth].p_hdr;
1770
1771 while (ex < EXT_LAST_EXTENT(eh)) {
1772 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1773 break;
1774 /* merge with next extent! */
Lukas Czerner556615d2014-04-20 23:45:47 -04001775 unwritten = ext4_ext_is_unwritten(ex);
Amit Arora56055d32007-07-17 21:42:38 -04001776 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1777 + ext4_ext_get_actual_len(ex + 1));
Lukas Czerner556615d2014-04-20 23:45:47 -04001778 if (unwritten)
1779 ext4_ext_mark_unwritten(ex);
Amit Arora56055d32007-07-17 21:42:38 -04001780
1781 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1782 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1783 * sizeof(struct ext4_extent);
1784 memmove(ex + 1, ex + 2, len);
1785 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001786 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001787 merge_done = 1;
1788 WARN_ON(eh->eh_entries == 0);
1789 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001790 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001791 }
1792
1793 return merge_done;
1794}
1795
1796/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001797 * This function does a very simple check to see if we can collapse
1798 * an extent tree with a single extent tree leaf block into the inode.
1799 */
1800static void ext4_ext_try_to_merge_up(handle_t *handle,
1801 struct inode *inode,
1802 struct ext4_ext_path *path)
1803{
1804 size_t s;
1805 unsigned max_root = ext4_ext_space_root(inode, 0);
1806 ext4_fsblk_t blk;
1807
1808 if ((path[0].p_depth != 1) ||
1809 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1810 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1811 return;
1812
1813 /*
1814 * We need to modify the block allocation bitmap and the block
1815 * group descriptor to release the extent tree block. If we
1816 * can't get the journal credits, give up.
1817 */
1818 if (ext4_journal_extend(handle, 2))
1819 return;
1820
1821 /*
1822 * Copy the extent data up to the inode
1823 */
1824 blk = ext4_idx_pblock(path[0].p_idx);
1825 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1826 sizeof(struct ext4_extent_idx);
1827 s += sizeof(struct ext4_extent_header);
1828
Theodore Ts'o10809df82014-09-01 14:40:09 -04001829 path[1].p_maxdepth = path[0].p_maxdepth;
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001830 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1831 path[0].p_depth = 0;
1832 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1833 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1834 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1835
1836 brelse(path[1].p_bh);
1837 ext4_free_blocks(handle, inode, NULL, blk, 1,
Theodore Ts'o71d4f7d2014-07-15 06:02:38 -04001838 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001839}
1840
1841/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001842 * This function tries to merge the @ex extent to neighbours in the tree.
1843 * return 1 if merge left else 0.
1844 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001845static void ext4_ext_try_to_merge(handle_t *handle,
1846 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001847 struct ext4_ext_path *path,
1848 struct ext4_extent *ex) {
1849 struct ext4_extent_header *eh;
1850 unsigned int depth;
1851 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001852
1853 depth = ext_depth(inode);
1854 BUG_ON(path[depth].p_hdr == NULL);
1855 eh = path[depth].p_hdr;
1856
1857 if (ex > EXT_FIRST_EXTENT(eh))
1858 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1859
1860 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001861 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001862
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001863 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001864}
1865
1866/*
Amit Arora25d14f92007-05-24 13:04:13 -04001867 * check if a portion of the "newext" extent overlaps with an
1868 * existing extent.
1869 *
1870 * If there is an overlap discovered, it updates the length of the newext
1871 * such that there will be no overlap, and then returns 1.
1872 * If there is no overlap found, it returns 0.
1873 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001874static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1875 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001876 struct ext4_extent *newext,
1877 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001878{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001879 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001880 unsigned int depth, len1;
1881 unsigned int ret = 0;
1882
1883 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001884 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001885 depth = ext_depth(inode);
1886 if (!path[depth].p_ext)
1887 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001888 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
Amit Arora25d14f92007-05-24 13:04:13 -04001889
1890 /*
1891 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001892 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001893 */
1894 if (b2 < b1) {
1895 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001896 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001897 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001898 b2 = EXT4_LBLK_CMASK(sbi, b2);
Amit Arora25d14f92007-05-24 13:04:13 -04001899 }
1900
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001901 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001902 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001903 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001904 newext->ee_len = cpu_to_le16(len1);
1905 ret = 1;
1906 }
1907
1908 /* check for overlap */
1909 if (b1 + len1 > b2) {
1910 newext->ee_len = cpu_to_le16(b2 - b1);
1911 ret = 1;
1912 }
1913out:
1914 return ret;
1915}
1916
1917/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001918 * ext4_ext_insert_extent:
1919 * tries to merge requsted extent into the existing extent or
1920 * inserts requested extent as new one into the tree,
1921 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001922 */
1923int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -04001924 struct ext4_ext_path **ppath,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001925 struct ext4_extent *newext, int gb_flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07001926{
Theodore Ts'odfe50802014-09-01 14:37:09 -04001927 struct ext4_ext_path *path = *ppath;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001928 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001929 struct ext4_extent *ex, *fex;
1930 struct ext4_extent *nearex; /* nearest extent */
1931 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001932 int depth, len, err;
1933 ext4_lblk_t next;
Lukas Czerner556615d2014-04-20 23:45:47 -04001934 int mb_flags = 0, unwritten;
Alex Tomasa86c6182006-10-11 01:21:03 -07001935
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -04001936 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1937 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
Frank Mayhar273df552010-03-02 11:46:09 -05001938 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1939 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1940 return -EIO;
1941 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001942 depth = ext_depth(inode);
1943 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001944 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001945 if (unlikely(path[depth].p_hdr == NULL)) {
1946 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1947 return -EIO;
1948 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001949
1950 /* try to insert block into found extent and return */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001951 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001952
1953 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001954 * Try to see whether we should rather test the extent on
1955 * right from ex, or from the left of ex. This is because
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04001956 * ext4_find_extent() can return either extent on the
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001957 * left, or on the right from the searched position. This
1958 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001959 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001960 if (ex < EXT_LAST_EXTENT(eh) &&
1961 (le32_to_cpu(ex->ee_block) +
1962 ext4_ext_get_actual_len(ex) <
1963 le32_to_cpu(newext->ee_block))) {
1964 ex += 1;
1965 goto prepend;
1966 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1967 (le32_to_cpu(newext->ee_block) +
1968 ext4_ext_get_actual_len(newext) <
1969 le32_to_cpu(ex->ee_block)))
1970 ex -= 1;
1971
1972 /* Try to append newex to the ex */
1973 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1974 ext_debug("append [%d]%d block to %u:[%d]%d"
1975 "(from %llu)\n",
Lukas Czerner556615d2014-04-20 23:45:47 -04001976 ext4_ext_is_unwritten(newext),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001977 ext4_ext_get_actual_len(newext),
1978 le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04001979 ext4_ext_is_unwritten(ex),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001980 ext4_ext_get_actual_len(ex),
1981 ext4_ext_pblock(ex));
1982 err = ext4_ext_get_access(handle, inode,
1983 path + depth);
1984 if (err)
1985 return err;
Lukas Czerner556615d2014-04-20 23:45:47 -04001986 unwritten = ext4_ext_is_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001987 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001988 + ext4_ext_get_actual_len(newext));
Lukas Czerner556615d2014-04-20 23:45:47 -04001989 if (unwritten)
1990 ext4_ext_mark_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001991 eh = path[depth].p_hdr;
1992 nearex = ex;
1993 goto merge;
1994 }
1995
1996prepend:
1997 /* Try to prepend newex to the ex */
1998 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1999 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2000 "(from %llu)\n",
2001 le32_to_cpu(newext->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04002002 ext4_ext_is_unwritten(newext),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002003 ext4_ext_get_actual_len(newext),
2004 le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04002005 ext4_ext_is_unwritten(ex),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002006 ext4_ext_get_actual_len(ex),
2007 ext4_ext_pblock(ex));
2008 err = ext4_ext_get_access(handle, inode,
2009 path + depth);
2010 if (err)
2011 return err;
2012
Lukas Czerner556615d2014-04-20 23:45:47 -04002013 unwritten = ext4_ext_is_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002014 ex->ee_block = newext->ee_block;
2015 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2016 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2017 + ext4_ext_get_actual_len(newext));
Lukas Czerner556615d2014-04-20 23:45:47 -04002018 if (unwritten)
2019 ext4_ext_mark_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002020 eh = path[depth].p_hdr;
2021 nearex = ex;
2022 goto merge;
2023 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002024 }
2025
Alex Tomasa86c6182006-10-11 01:21:03 -07002026 depth = ext_depth(inode);
2027 eh = path[depth].p_hdr;
2028 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2029 goto has_space;
2030
2031 /* probably next leaf has space for us? */
2032 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04002033 next = EXT_MAX_BLOCKS;
2034 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04002035 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04002036 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002037 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07002038 BUG_ON(npath != NULL);
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04002039 npath = ext4_find_extent(inode, next, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002040 if (IS_ERR(npath))
2041 return PTR_ERR(npath);
2042 BUG_ON(npath->p_depth != path->p_depth);
2043 eh = npath[depth].p_hdr;
2044 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002045 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002046 le16_to_cpu(eh->eh_entries));
2047 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04002048 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07002049 }
2050 ext_debug("next leaf has no free space(%d,%d)\n",
2051 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2052 }
2053
2054 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002055 * There is no free space in the found leaf.
2056 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07002057 */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002058 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -04002059 mb_flags |= EXT4_MB_USE_RESERVED;
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002060 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
Theodore Ts'odfe50802014-09-01 14:37:09 -04002061 ppath, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07002062 if (err)
2063 goto cleanup;
2064 depth = ext_depth(inode);
2065 eh = path[depth].p_hdr;
2066
2067has_space:
2068 nearex = path[depth].p_ext;
2069
Avantika Mathur7e028972006-12-06 20:41:33 -08002070 err = ext4_ext_get_access(handle, inode, path + depth);
2071 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002072 goto cleanup;
2073
2074 if (!nearex) {
2075 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002076 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002077 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002078 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002079 ext4_ext_is_unwritten(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002080 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04002081 nearex = EXT_FIRST_EXTENT(eh);
2082 } else {
2083 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002084 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04002085 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002086 ext_debug("insert %u:%llu:[%d]%d before: "
2087 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002088 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002089 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002090 ext4_ext_is_unwritten(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002091 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002092 nearex);
2093 nearex++;
2094 } else {
2095 /* Insert before */
2096 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04002097 ext_debug("insert %u:%llu:[%d]%d after: "
2098 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04002099 le32_to_cpu(newext->ee_block),
2100 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002101 ext4_ext_is_unwritten(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002102 ext4_ext_get_actual_len(newext),
2103 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002104 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04002105 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2106 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002107 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04002108 "move %d extents from 0x%p to 0x%p\n",
2109 le32_to_cpu(newext->ee_block),
2110 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002111 ext4_ext_is_unwritten(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002112 ext4_ext_get_actual_len(newext),
2113 len, nearex, nearex + 1);
2114 memmove(nearex + 1, nearex,
2115 len * sizeof(struct ext4_extent));
2116 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002117 }
2118
Marcin Slusarze8546d02008-04-17 10:38:59 -04002119 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002120 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002121 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002122 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002123 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002124
2125merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002126 /* try to merge extents */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002127 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002128 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002129
Alex Tomasa86c6182006-10-11 01:21:03 -07002130
2131 /* time to correct all indexes above */
2132 err = ext4_ext_correct_indexes(handle, inode, path);
2133 if (err)
2134 goto cleanup;
2135
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002136 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002137
2138cleanup:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04002139 ext4_ext_drop_refs(npath);
2140 kfree(npath);
Alex Tomasa86c6182006-10-11 01:21:03 -07002141 return err;
2142}
2143
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002144static int ext4_fill_fiemap_extents(struct inode *inode,
2145 ext4_lblk_t block, ext4_lblk_t num,
2146 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002147{
2148 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002149 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002150 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002151 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002152 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002153 int exists, depth = 0, err = 0;
2154 unsigned int flags = 0;
2155 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002156
Lukas Czernerf17722f2011-06-06 00:05:17 -04002157 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002158 num = last - block;
2159 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002160 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002161
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04002162 path = ext4_find_extent(inode, block, &path, 0);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002163 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002164 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002165 err = PTR_ERR(path);
2166 path = NULL;
2167 break;
2168 }
2169
2170 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002171 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002172 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002173 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2174 err = -EIO;
2175 break;
2176 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002177 ex = path[depth].p_ext;
2178 next = ext4_ext_next_allocated_block(path);
2179
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002180 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002181 exists = 0;
2182 if (!ex) {
2183 /* there is no extent yet, so try to allocate
2184 * all requested space */
2185 start = block;
2186 end = block + num;
2187 } else if (le32_to_cpu(ex->ee_block) > block) {
2188 /* need to allocate space before found extent */
2189 start = block;
2190 end = le32_to_cpu(ex->ee_block);
2191 if (block + num < end)
2192 end = block + num;
2193 } else if (block >= le32_to_cpu(ex->ee_block)
2194 + ext4_ext_get_actual_len(ex)) {
2195 /* need to allocate space after found extent */
2196 start = block;
2197 end = block + num;
2198 if (end >= next)
2199 end = next;
2200 } else if (block >= le32_to_cpu(ex->ee_block)) {
2201 /*
2202 * some part of requested space is covered
2203 * by found extent
2204 */
2205 start = block;
2206 end = le32_to_cpu(ex->ee_block)
2207 + ext4_ext_get_actual_len(ex);
2208 if (block + num < end)
2209 end = block + num;
2210 exists = 1;
2211 } else {
2212 BUG();
2213 }
2214 BUG_ON(end <= start);
2215
2216 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002217 es.es_lblk = start;
2218 es.es_len = end - start;
2219 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002220 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002221 es.es_lblk = le32_to_cpu(ex->ee_block);
2222 es.es_len = ext4_ext_get_actual_len(ex);
2223 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner556615d2014-04-20 23:45:47 -04002224 if (ext4_ext_is_unwritten(ex))
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002225 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002226 }
2227
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002228 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002229 * Find delayed extent and update es accordingly. We call
2230 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002231 * last existing extent or not.
2232 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002233 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002234 if (!exists && next_del) {
2235 exists = 1;
Jie Liu72dac952013-06-12 23:13:59 -04002236 flags |= (FIEMAP_EXTENT_DELALLOC |
2237 FIEMAP_EXTENT_UNKNOWN);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002238 }
2239 up_read(&EXT4_I(inode)->i_data_sem);
2240
Zheng Liu69eb33d2013-02-18 00:31:07 -05002241 if (unlikely(es.es_len == 0)) {
2242 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002243 err = -EIO;
2244 break;
2245 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002246
Zheng Liuf7fec032013-02-18 00:28:47 -05002247 /*
2248 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2249 * we need to check next == EXT_MAX_BLOCKS because it is
2250 * possible that an extent is with unwritten and delayed
2251 * status due to when an extent is delayed allocated and
2252 * is allocated by fallocate status tree will track both of
2253 * them in a extent.
2254 *
2255 * So we could return a unwritten and delayed extent, and
2256 * its block is equal to 'next'.
2257 */
2258 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002259 flags |= FIEMAP_EXTENT_LAST;
2260 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2261 next != EXT_MAX_BLOCKS)) {
2262 EXT4_ERROR_INODE(inode,
2263 "next extent == %u, next "
2264 "delalloc extent = %u",
2265 next, next_del);
2266 err = -EIO;
2267 break;
2268 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002269 }
2270
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002271 if (exists) {
2272 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002273 (__u64)es.es_lblk << blksize_bits,
2274 (__u64)es.es_pblk << blksize_bits,
2275 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002276 flags);
2277 if (err < 0)
2278 break;
2279 if (err == 1) {
2280 err = 0;
2281 break;
2282 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002283 }
2284
Zheng Liu69eb33d2013-02-18 00:31:07 -05002285 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002286 }
2287
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04002288 ext4_ext_drop_refs(path);
2289 kfree(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002290 return err;
2291}
2292
Alex Tomasa86c6182006-10-11 01:21:03 -07002293/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002294 * ext4_ext_put_gap_in_cache:
2295 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002296 * and cache this gap
2297 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002298static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002299ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002300 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002301{
2302 int depth = ext_depth(inode);
Zheng Liu2f8e0a72014-11-25 11:44:37 -05002303 ext4_lblk_t len;
2304 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002305 struct ext4_extent *ex;
Zheng Liu2f8e0a72014-11-25 11:44:37 -05002306 struct extent_status es;
Alex Tomasa86c6182006-10-11 01:21:03 -07002307
2308 ex = path[depth].p_ext;
2309 if (ex == NULL) {
Zheng Liu2f8e0a72014-11-25 11:44:37 -05002310 /* there is no extent yet, so gap is [0;-] */
2311 lblock = 0;
2312 len = EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07002313 ext_debug("cache gap(whole file):");
2314 } else if (block < le32_to_cpu(ex->ee_block)) {
2315 lblock = block;
2316 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002317 ext_debug("cache gap(before): %u [%u:%u]",
2318 block,
2319 le32_to_cpu(ex->ee_block),
2320 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002321 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002322 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002323 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002324 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002325 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002326
2327 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002328 ext_debug("cache gap(after): [%u:%u] %u",
2329 le32_to_cpu(ex->ee_block),
2330 ext4_ext_get_actual_len(ex),
2331 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002332 BUG_ON(next == lblock);
2333 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002334 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07002335 BUG();
2336 }
2337
Zheng Liu2f8e0a72014-11-25 11:44:37 -05002338 ext4_es_find_delayed_extent_range(inode, lblock, lblock + len - 1, &es);
2339 if (es.es_len) {
2340 /* There's delayed extent containing lblock? */
2341 if (es.es_lblk <= lblock)
2342 return;
2343 len = min(es.es_lblk - lblock, len);
2344 }
2345 ext_debug(" -> %u:%u\n", lblock, len);
2346 ext4_es_insert_extent(inode, lblock, len, ~0, EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002347}
2348
2349/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002350 * ext4_ext_rm_idx:
2351 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002352 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002353static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002354 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002355{
Alex Tomasa86c6182006-10-11 01:21:03 -07002356 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002357 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002358
2359 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002360 depth--;
2361 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002362 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002363 if (unlikely(path->p_hdr->eh_entries == 0)) {
2364 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2365 return -EIO;
2366 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002367 err = ext4_ext_get_access(handle, inode, path);
2368 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002369 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002370
2371 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2372 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2373 len *= sizeof(struct ext4_extent_idx);
2374 memmove(path->p_idx, path->p_idx + 1, len);
2375 }
2376
Marcin Slusarze8546d02008-04-17 10:38:59 -04002377 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002378 err = ext4_ext_dirty(handle, inode, path);
2379 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002380 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002381 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002382 trace_ext4_ext_rm_idx(inode, leaf);
2383
Peter Huewe7dc57612011-02-21 21:01:42 -05002384 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002385 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002386
2387 while (--depth >= 0) {
2388 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2389 break;
2390 path--;
2391 err = ext4_ext_get_access(handle, inode, path);
2392 if (err)
2393 break;
2394 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2395 err = ext4_ext_dirty(handle, inode, path);
2396 if (err)
2397 break;
2398 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002399 return err;
2400}
2401
2402/*
Mingming Caoee12b632008-08-19 22:16:05 -04002403 * ext4_ext_calc_credits_for_single_extent:
2404 * This routine returns max. credits that needed to insert an extent
2405 * to the extent tree.
2406 * When pass the actual path, the caller should calculate credits
2407 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002408 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002409int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002410 struct ext4_ext_path *path)
2411{
Alex Tomasa86c6182006-10-11 01:21:03 -07002412 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002413 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002414 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002415
Alex Tomasa86c6182006-10-11 01:21:03 -07002416 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002417 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002418 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2419
2420 /*
2421 * There are some space in the leaf tree, no
2422 * need to account for leaf block credit
2423 *
2424 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002425 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002426 * accounted.
2427 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002428 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002429 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002430 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002431 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002432 }
2433
Mingming Cao525f4ed2008-08-19 22:15:58 -04002434 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002435}
Alex Tomasa86c6182006-10-11 01:21:03 -07002436
Mingming Caoee12b632008-08-19 22:16:05 -04002437/*
Jan Karafffb2732013-06-04 13:01:11 -04002438 * How many index/leaf blocks need to change/allocate to add @extents extents?
Mingming Caoee12b632008-08-19 22:16:05 -04002439 *
Jan Karafffb2732013-06-04 13:01:11 -04002440 * If we add a single extent, then in the worse case, each tree level
2441 * index/leaf need to be changed in case of the tree split.
Mingming Caoee12b632008-08-19 22:16:05 -04002442 *
Jan Karafffb2732013-06-04 13:01:11 -04002443 * If more extents are inserted, they could cause the whole tree split more
2444 * than once, but this is really rare.
Mingming Caoee12b632008-08-19 22:16:05 -04002445 */
Jan Karafffb2732013-06-04 13:01:11 -04002446int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
Mingming Caoee12b632008-08-19 22:16:05 -04002447{
2448 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002449 int depth;
2450
2451 /* If we are converting the inline data, only one is needed here. */
2452 if (ext4_has_inline_data(inode))
2453 return 1;
2454
2455 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002456
Jan Karafffb2732013-06-04 13:01:11 -04002457 if (extents <= 1)
Mingming Caoee12b632008-08-19 22:16:05 -04002458 index = depth * 2;
2459 else
2460 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002461
Mingming Caoee12b632008-08-19 22:16:05 -04002462 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002463}
2464
Theodore Ts'o981250c2013-06-12 11:48:29 -04002465static inline int get_default_free_blocks_flags(struct inode *inode)
2466{
2467 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2468 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2469 else if (ext4_should_journal_data(inode))
2470 return EXT4_FREE_BLOCKS_FORGET;
2471 return 0;
2472}
2473
Alex Tomasa86c6182006-10-11 01:21:03 -07002474static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002475 struct ext4_extent *ex,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002476 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002477 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002478{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002479 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Eric Whitney345ee942014-11-23 00:59:39 -05002480 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002481 ext4_fsblk_t pblk;
Theodore Ts'o981250c2013-06-12 11:48:29 -04002482 int flags = get_default_free_blocks_flags(inode);
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002483
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002484 /*
2485 * For bigalloc file systems, we never free a partial cluster
2486 * at the beginning of the extent. Instead, we make a note
2487 * that we tried freeing the cluster, and check to see if we
2488 * need to free it on a subsequent call to ext4_remove_blocks,
Eric Whitney345ee942014-11-23 00:59:39 -05002489 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002490 */
2491 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2492
Aditya Kalid8990242011-09-09 19:18:51 -04002493 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002494 /*
2495 * If we have a partial cluster, and it's different from the
2496 * cluster of the last block, we need to explicitly free the
2497 * partial cluster here.
2498 */
2499 pblk = ext4_ext_pblock(ex) + ee_len - 1;
Eric Whitney345ee942014-11-23 00:59:39 -05002500 if (*partial_cluster > 0 &&
2501 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002502 ext4_free_blocks(handle, inode, NULL,
2503 EXT4_C2B(sbi, *partial_cluster),
2504 sbi->s_cluster_ratio, flags);
2505 *partial_cluster = 0;
2506 }
2507
Alex Tomasa86c6182006-10-11 01:21:03 -07002508#ifdef EXTENTS_STATS
2509 {
2510 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002511 spin_lock(&sbi->s_ext_stats_lock);
2512 sbi->s_ext_blocks += ee_len;
2513 sbi->s_ext_extents++;
2514 if (ee_len < sbi->s_ext_min)
2515 sbi->s_ext_min = ee_len;
2516 if (ee_len > sbi->s_ext_max)
2517 sbi->s_ext_max = ee_len;
2518 if (ext_depth(inode) > sbi->s_depth_max)
2519 sbi->s_depth_max = ext_depth(inode);
2520 spin_unlock(&sbi->s_ext_stats_lock);
2521 }
2522#endif
2523 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002524 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002525 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002526 ext4_lblk_t num;
Eric Whitney345ee942014-11-23 00:59:39 -05002527 long long first_cluster;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002528
Amit Aroraa2df2a62007-07-17 21:42:41 -04002529 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002530 pblk = ext4_ext_pblock(ex) + ee_len - num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002531 /*
2532 * Usually we want to free partial cluster at the end of the
2533 * extent, except for the situation when the cluster is still
2534 * used by any other extent (partial_cluster is negative).
2535 */
2536 if (*partial_cluster < 0 &&
Eric Whitney345ee942014-11-23 00:59:39 -05002537 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
Lukas Czernerd23142c2013-05-27 23:33:35 -04002538 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2539
2540 ext_debug("free last %u blocks starting %llu partial %lld\n",
2541 num, pblk, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002542 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2543 /*
2544 * If the block range to be freed didn't start at the
2545 * beginning of a cluster, and we removed the entire
Lukas Czernerd23142c2013-05-27 23:33:35 -04002546 * extent and the cluster is not used by any other extent,
2547 * save the partial cluster here, since we might need to
Eric Whitney345ee942014-11-23 00:59:39 -05002548 * delete if we determine that the truncate or punch hole
2549 * operation has removed all of the blocks in the cluster.
2550 * If that cluster is used by another extent, preserve its
2551 * negative value so it isn't freed later on.
Lukas Czernerd23142c2013-05-27 23:33:35 -04002552 *
Eric Whitney345ee942014-11-23 00:59:39 -05002553 * If the whole extent wasn't freed, we've reached the
2554 * start of the truncated/punched region and have finished
2555 * removing blocks. If there's a partial cluster here it's
2556 * shared with the remainder of the extent and is no longer
2557 * a candidate for removal.
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002558 */
Eric Whitney345ee942014-11-23 00:59:39 -05002559 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2560 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2561 if (first_cluster != -*partial_cluster)
2562 *partial_cluster = first_cluster;
2563 } else {
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002564 *partial_cluster = 0;
Eric Whitney345ee942014-11-23 00:59:39 -05002565 }
Lukas Czerner78fb9cd2013-05-27 23:32:35 -04002566 } else
2567 ext4_error(sbi->s_sb, "strange request: removal(2) "
2568 "%u-%u from %u:%u\n",
2569 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002570 return 0;
2571}
2572
Allison Hendersond583fb82011-05-25 07:41:43 -04002573
2574/*
2575 * ext4_ext_rm_leaf() Removes the extents associated with the
Eric Whitney5bf43762014-11-23 00:58:11 -05002576 * blocks appearing between "start" and "end". Both "start"
2577 * and "end" must appear in the same extent or EIO is returned.
Allison Hendersond583fb82011-05-25 07:41:43 -04002578 *
2579 * @handle: The journal handle
2580 * @inode: The files inode
2581 * @path: The path to the leaf
Lukas Czernerd23142c2013-05-27 23:33:35 -04002582 * @partial_cluster: The cluster which we'll have to free if all extents
Eric Whitney5bf43762014-11-23 00:58:11 -05002583 * has been released from it. However, if this value is
2584 * negative, it's a cluster just to the right of the
2585 * punched region and it must not be freed.
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
Aditya Kalid8990242011-09-09 19:18:51 -04002624 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2625
Alex Tomasa86c6182006-10-11 01:21:03 -07002626 while (ex >= EXT_FIRST_EXTENT(eh) &&
2627 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002628
Lukas Czerner556615d2014-04-20 23:45:47 -04002629 if (ext4_ext_is_unwritten(ex))
2630 unwritten = 1;
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002631 else
Lukas Czerner556615d2014-04-20 23:45:47 -04002632 unwritten = 0;
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002633
Mingming553f9002009-09-18 13:34:55 -04002634 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
Lukas Czerner556615d2014-04-20 23:45:47 -04002635 unwritten, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002636 path[depth].p_ext = ex;
2637
2638 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002639 b = ex_ee_block+ex_ee_len - 1 < end ?
2640 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002641
2642 ext_debug(" border %u:%u\n", a, b);
2643
Allison Hendersond583fb82011-05-25 07:41:43 -04002644 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002645 if (end < ex_ee_block) {
Lukas Czernerd23142c2013-05-27 23:33:35 -04002646 /*
2647 * We're going to skip this extent and move to another,
Eric Whitneyf4226d92014-11-23 00:55:42 -05002648 * so note that its first cluster is in use to avoid
2649 * freeing it when removing blocks. Eventually, the
2650 * right edge of the truncated/punched region will
2651 * be just to the left.
Lukas Czernerd23142c2013-05-27 23:33:35 -04002652 */
Eric Whitneyf4226d92014-11-23 00:55:42 -05002653 if (sbi->s_cluster_ratio > 1) {
2654 pblk = ext4_ext_pblock(ex);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002655 *partial_cluster =
Eric Whitneyf4226d92014-11-23 00:55:42 -05002656 -(long long) EXT4_B2C(sbi, pblk);
2657 }
Allison Hendersond583fb82011-05-25 07:41:43 -04002658 ex--;
2659 ex_ee_block = le32_to_cpu(ex->ee_block);
2660 ex_ee_len = ext4_ext_get_actual_len(ex);
2661 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002662 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002663 EXT4_ERROR_INODE(inode,
2664 "can not handle truncate %u:%u "
2665 "on extent %u:%u",
2666 start, end, ex_ee_block,
2667 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002668 err = -EIO;
2669 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002670 } else if (a != ex_ee_block) {
2671 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002672 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002673 } else {
2674 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002675 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002676 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002677 /*
2678 * 3 for leaf, sb, and inode plus 2 (bmap and group
2679 * descriptor) for each block group; assume two block
2680 * groups plus ex_ee_len/blocks_per_block_group for
2681 * the worst case
2682 */
2683 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002684 if (ex == EXT_FIRST_EXTENT(eh)) {
2685 correct_index = 1;
2686 credits += (ext_depth(inode)) + 1;
2687 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002688 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002689
Jan Kara487caee2009-08-17 22:17:20 -04002690 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002691 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002692 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002693
2694 err = ext4_ext_get_access(handle, inode, path + depth);
2695 if (err)
2696 goto out;
2697
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002698 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2699 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002700 if (err)
2701 goto out;
2702
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002703 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002704 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002705 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002706
Alex Tomasa86c6182006-10-11 01:21:03 -07002707 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002708 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04002709 * Do not mark unwritten if all the blocks in the
Amit Arora749269f2007-07-18 09:02:56 -04002710 * extent have been removed.
2711 */
Lukas Czerner556615d2014-04-20 23:45:47 -04002712 if (unwritten && num)
2713 ext4_ext_mark_unwritten(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002714 /*
2715 * If the extent was completely released,
2716 * we need to remove it from the leaf
2717 */
2718 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002719 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002720 /*
2721 * For hole punching, we need to scoot all the
2722 * extents up when an extent is removed so that
2723 * we dont have blank extents in the middle
2724 */
2725 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2726 sizeof(struct ext4_extent));
2727
2728 /* Now get rid of the one at the end */
2729 memset(EXT_LAST_EXTENT(eh), 0,
2730 sizeof(struct ext4_extent));
2731 }
2732 le16_add_cpu(&eh->eh_entries, -1);
Eric Whitney5bf43762014-11-23 00:58:11 -05002733 }
Allison Hendersond583fb82011-05-25 07:41:43 -04002734
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002735 err = ext4_ext_dirty(handle, inode, path + depth);
2736 if (err)
2737 goto out;
2738
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002739 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002740 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002741 ex--;
2742 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002743 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002744 }
2745
2746 if (correct_index && eh->eh_entries)
2747 err = ext4_ext_correct_indexes(handle, inode, path);
2748
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002749 /*
Eric Whitneyad6599a2014-04-01 19:49:30 -04002750 * If there's a partial cluster and at least one extent remains in
2751 * the leaf, free the partial cluster if it isn't shared with the
Eric Whitney5bf43762014-11-23 00:58:11 -05002752 * current extent. If it is shared with the current extent
2753 * we zero partial_cluster because we've reached the start of the
2754 * truncated/punched region and we're done removing blocks.
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002755 */
Eric Whitney5bf43762014-11-23 00:58:11 -05002756 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2757 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2758 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2759 ext4_free_blocks(handle, inode, NULL,
2760 EXT4_C2B(sbi, *partial_cluster),
2761 sbi->s_cluster_ratio,
2762 get_default_free_blocks_flags(inode));
2763 }
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002764 *partial_cluster = 0;
2765 }
2766
Alex Tomasa86c6182006-10-11 01:21:03 -07002767 /* if this leaf is free, then we should
2768 * remove it from index block above */
2769 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002770 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002771
2772out:
2773 return err;
2774}
2775
2776/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002777 * ext4_ext_more_to_rm:
2778 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002779 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002780static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002781ext4_ext_more_to_rm(struct ext4_ext_path *path)
2782{
2783 BUG_ON(path->p_idx == NULL);
2784
2785 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2786 return 0;
2787
2788 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002789 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002790 * so we have to consider current index for truncation
2791 */
2792 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2793 return 0;
2794 return 1;
2795}
2796
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002797int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2798 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002799{
Eric Whitneyf4226d92014-11-23 00:55:42 -05002800 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002801 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002802 struct ext4_ext_path *path = NULL;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002803 long long partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002804 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002805 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002806
Lukas Czerner5f95d212012-03-19 23:03:19 -04002807 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002808
2809 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002810 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002811 if (IS_ERR(handle))
2812 return PTR_ERR(handle);
2813
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002814again:
Lukas Czerner61801322013-05-27 23:32:35 -04002815 trace_ext4_ext_remove_space(inode, start, end, depth);
Aditya Kalid8990242011-09-09 19:18:51 -04002816
Alex Tomasa86c6182006-10-11 01:21:03 -07002817 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002818 * Check if we are removing extents inside the extent tree. If that
2819 * is the case, we are going to punch a hole inside the extent tree
2820 * so we have to check whether we need to split the extent covering
2821 * the last block to remove so we can easily remove the part of it
2822 * in ext4_ext_rm_leaf().
2823 */
2824 if (end < EXT_MAX_BLOCKS - 1) {
2825 struct ext4_extent *ex;
Eric Whitneyf4226d92014-11-23 00:55:42 -05002826 ext4_lblk_t ee_block, ex_end, lblk;
2827 ext4_fsblk_t pblk;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002828
Eric Whitneyf4226d92014-11-23 00:55:42 -05002829 /* find extent for or closest extent to this block */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04002830 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002831 if (IS_ERR(path)) {
2832 ext4_journal_stop(handle);
2833 return PTR_ERR(path);
2834 }
2835 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002836 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002837 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002838 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002839 if (depth) {
2840 EXT4_ERROR_INODE(inode,
2841 "path[%d].p_hdr == NULL",
2842 depth);
2843 err = -EIO;
2844 }
2845 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002846 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002847
2848 ee_block = le32_to_cpu(ex->ee_block);
Eric Whitneyf4226d92014-11-23 00:55:42 -05002849 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002850
2851 /*
2852 * See if the last block is inside the extent, if so split
2853 * the extent at 'end' block so we can easily remove the
2854 * tail of the first part of the split extent in
2855 * ext4_ext_rm_leaf().
2856 */
Eric Whitneyf4226d92014-11-23 00:55:42 -05002857 if (end >= ee_block && end < ex_end) {
2858
2859 /*
2860 * If we're going to split the extent, note that
2861 * the cluster containing the block after 'end' is
2862 * in use to avoid freeing it when removing blocks.
2863 */
2864 if (sbi->s_cluster_ratio > 1) {
2865 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2866 partial_cluster =
2867 -(long long) EXT4_B2C(sbi, pblk);
2868 }
2869
Lukas Czerner5f95d212012-03-19 23:03:19 -04002870 /*
2871 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002872 * block in the first new extent. Also we should not
2873 * fail removing space due to ENOSPC so try to use
2874 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002875 */
Theodore Ts'odfe50802014-09-01 14:37:09 -04002876 err = ext4_force_split_extent_at(handle, inode, &path,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04002877 end + 1, 1);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002878 if (err < 0)
2879 goto out;
Eric Whitneyf4226d92014-11-23 00:55:42 -05002880
2881 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2882 /*
2883 * If there's an extent to the right its first cluster
2884 * contains the immediate right boundary of the
2885 * truncated/punched region. Set partial_cluster to
2886 * its negative value so it won't be freed if shared
2887 * with the current extent. The end < ee_block case
2888 * is handled in ext4_ext_rm_leaf().
2889 */
2890 lblk = ex_end + 1;
2891 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2892 &ex);
2893 if (err)
2894 goto out;
2895 if (pblk)
2896 partial_cluster =
2897 -(long long) EXT4_B2C(sbi, pblk);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002898 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002899 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002900 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002901 * We start scanning from right side, freeing all the blocks
2902 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002903 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002904 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002905 if (path) {
2906 int k = i = depth;
2907 while (--k > 0)
2908 path[k].p_block =
2909 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2910 } else {
2911 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2912 GFP_NOFS);
2913 if (path == NULL) {
2914 ext4_journal_stop(handle);
2915 return -ENOMEM;
2916 }
Theodore Ts'o10809df82014-09-01 14:40:09 -04002917 path[0].p_maxdepth = path[0].p_depth = depth;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002918 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002919 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002920
Theodore Ts'oc3491792013-08-16 21:21:41 -04002921 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
Ashish Sangwan968dee72012-07-22 22:49:08 -04002922 err = -EIO;
2923 goto out;
2924 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002925 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002926 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002927
2928 while (i >= 0 && err == 0) {
2929 if (i == depth) {
2930 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002931 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002932 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002933 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002934 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002935 brelse(path[i].p_bh);
2936 path[i].p_bh = NULL;
2937 i--;
2938 continue;
2939 }
2940
2941 /* this is index block */
2942 if (!path[i].p_hdr) {
2943 ext_debug("initialize header\n");
2944 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002945 }
2946
Alex Tomasa86c6182006-10-11 01:21:03 -07002947 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002948 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002949 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2950 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2951 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2952 path[i].p_hdr,
2953 le16_to_cpu(path[i].p_hdr->eh_entries));
2954 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002955 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002956 path[i].p_idx--;
2957 }
2958
2959 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2960 i, EXT_FIRST_INDEX(path[i].p_hdr),
2961 path[i].p_idx);
2962 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002963 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002964 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002965 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002966 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002967 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002968 bh = read_extent_tree_block(inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002969 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2970 EXT4_EX_NOCACHE);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002971 if (IS_ERR(bh)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002972 /* should we reset i_size? */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002973 err = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002974 break;
2975 }
Theodore Ts'o76828c82013-07-15 12:27:47 -04002976 /* Yield here to deal with large extent trees.
2977 * Should be a no-op if we did IO above. */
2978 cond_resched();
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002979 if (WARN_ON(i + 1 > depth)) {
2980 err = -EIO;
2981 break;
2982 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002983 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002984
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002985 /* save actual number of indexes since this
2986 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002987 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2988 i++;
2989 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002990 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002991 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002992 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002993 * handle must be already prepared by the
2994 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002995 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002996 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002997 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002998 brelse(path[i].p_bh);
2999 path[i].p_bh = NULL;
3000 i--;
3001 ext_debug("return to level %d\n", i);
3002 }
3003 }
3004
Lukas Czerner61801322013-05-27 23:32:35 -04003005 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3006 partial_cluster, path->p_hdr->eh_entries);
Aditya Kalid8990242011-09-09 19:18:51 -04003007
Eric Whitney0756b902014-11-23 00:59:39 -05003008 /*
3009 * If we still have something in the partial cluster and we have removed
Aditya Kali7b415bf2011-09-09 19:04:51 -04003010 * even the first extent, then we should free the blocks in the partial
Eric Whitney0756b902014-11-23 00:59:39 -05003011 * cluster as well. (This code will only run when there are no leaves
3012 * to the immediate left of the truncated/punched region.)
3013 */
3014 if (partial_cluster > 0 && err == 0) {
3015 /* don't zero partial_cluster since it's not used afterwards */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003016 ext4_free_blocks(handle, inode, NULL,
Eric Whitneyf4226d92014-11-23 00:55:42 -05003017 EXT4_C2B(sbi, partial_cluster),
Eric Whitney0756b902014-11-23 00:59:39 -05003018 sbi->s_cluster_ratio,
3019 get_default_free_blocks_flags(inode));
Aditya Kali7b415bf2011-09-09 19:04:51 -04003020 }
3021
Alex Tomasa86c6182006-10-11 01:21:03 -07003022 /* TODO: flexible tree reduction should be here */
3023 if (path->p_hdr->eh_entries == 0) {
3024 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003025 * truncate to zero freed all the tree,
3026 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07003027 */
3028 err = ext4_ext_get_access(handle, inode, path);
3029 if (err == 0) {
3030 ext_inode_hdr(inode)->eh_depth = 0;
3031 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04003032 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07003033 err = ext4_ext_dirty(handle, inode, path);
3034 }
3035 }
3036out:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04003037 ext4_ext_drop_refs(path);
3038 kfree(path);
3039 path = NULL;
Theodore Ts'odfe50802014-09-01 14:37:09 -04003040 if (err == -EAGAIN)
3041 goto again;
Alex Tomasa86c6182006-10-11 01:21:03 -07003042 ext4_journal_stop(handle);
3043
3044 return err;
3045}
3046
3047/*
3048 * called at mount time
3049 */
3050void ext4_ext_init(struct super_block *sb)
3051{
3052 /*
3053 * possible initialization would be here
3054 */
3055
Theodore Ts'o83982b62009-01-06 14:53:16 -05003056 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04003057#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04003058 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01003059#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04003060 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07003061#endif
3062#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04003063 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07003064#endif
3065#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04003066 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07003067#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04003068 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04003069#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07003070#ifdef EXTENTS_STATS
3071 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3072 EXT4_SB(sb)->s_ext_min = 1 << 30;
3073 EXT4_SB(sb)->s_ext_max = 0;
3074#endif
3075 }
3076}
3077
3078/*
3079 * called at umount time
3080 */
3081void ext4_ext_release(struct super_block *sb)
3082{
Theodore Ts'o83982b62009-01-06 14:53:16 -05003083 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07003084 return;
3085
3086#ifdef EXTENTS_STATS
3087 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3088 struct ext4_sb_info *sbi = EXT4_SB(sb);
3089 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3090 sbi->s_ext_blocks, sbi->s_ext_extents,
3091 sbi->s_ext_blocks / sbi->s_ext_extents);
3092 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3093 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3094 }
3095#endif
3096}
3097
Zheng Liud7b2a002013-08-28 14:47:06 -04003098static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3099{
3100 ext4_lblk_t ee_block;
3101 ext4_fsblk_t ee_pblock;
3102 unsigned int ee_len;
3103
3104 ee_block = le32_to_cpu(ex->ee_block);
3105 ee_len = ext4_ext_get_actual_len(ex);
3106 ee_pblock = ext4_ext_pblock(ex);
3107
3108 if (ee_len == 0)
3109 return 0;
3110
3111 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3112 EXTENT_STATUS_WRITTEN);
3113}
3114
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003115/* FIXME!! we need to try to merge to left or right after zero-out */
3116static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3117{
Lukas Czerner24075182010-10-27 21:30:06 -04003118 ext4_fsblk_t ee_pblock;
3119 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04003120 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003121
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003122 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003123 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003124
Michael Halcrow2058f832015-04-12 00:55:10 -04003125 if (ext4_encrypted_inode(inode))
3126 return ext4_encrypted_zeroout(inode, ex);
3127
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04003128 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04003129 if (ret > 0)
3130 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003131
Lukas Czerner24075182010-10-27 21:30:06 -04003132 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003133}
3134
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003135/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003136 * ext4_split_extent_at() splits an extent at given block.
3137 *
3138 * @handle: the journal handle
3139 * @inode: the file inode
3140 * @path: the path to the extent
3141 * @split: the logical block where the extent is splitted.
3142 * @split_flags: indicates if the extent could be zeroout if split fails, and
Lukas Czerner556615d2014-04-20 23:45:47 -04003143 * the states(init or unwritten) of new extents.
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003144 * @flags: flags used to insert new extent to extent tree.
3145 *
3146 *
3147 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3148 * of which are deterimined by split_flag.
3149 *
3150 * There are two cases:
3151 * a> the extent are splitted into two extent.
3152 * b> split is not needed, and just mark the extent.
3153 *
3154 * return 0 on success.
3155 */
3156static int ext4_split_extent_at(handle_t *handle,
3157 struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003158 struct ext4_ext_path **ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003159 ext4_lblk_t split,
3160 int split_flag,
3161 int flags)
3162{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003163 struct ext4_ext_path *path = *ppath;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003164 ext4_fsblk_t newblock;
3165 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04003166 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003167 struct ext4_extent *ex2 = NULL;
3168 unsigned int ee_len, depth;
3169 int err = 0;
3170
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003171 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3172 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3173
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003174 ext_debug("ext4_split_extents_at: inode %lu, logical"
3175 "block %llu\n", inode->i_ino, (unsigned long long)split);
3176
3177 ext4_ext_show_leaf(inode, path);
3178
3179 depth = ext_depth(inode);
3180 ex = path[depth].p_ext;
3181 ee_block = le32_to_cpu(ex->ee_block);
3182 ee_len = ext4_ext_get_actual_len(ex);
3183 newblock = split - ee_block + ext4_ext_pblock(ex);
3184
3185 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Lukas Czerner556615d2014-04-20 23:45:47 -04003186 BUG_ON(!ext4_ext_is_unwritten(ex) &&
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003187 split_flag & (EXT4_EXT_MAY_ZEROOUT |
Lukas Czerner556615d2014-04-20 23:45:47 -04003188 EXT4_EXT_MARK_UNWRIT1 |
3189 EXT4_EXT_MARK_UNWRIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003190
3191 err = ext4_ext_get_access(handle, inode, path + depth);
3192 if (err)
3193 goto out;
3194
3195 if (split == ee_block) {
3196 /*
3197 * case b: block @split is the block that the extent begins with
3198 * then we just change the state of the extent, and splitting
3199 * is not needed.
3200 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003201 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3202 ext4_ext_mark_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003203 else
3204 ext4_ext_mark_initialized(ex);
3205
3206 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003207 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003208
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003209 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003210 goto out;
3211 }
3212
3213 /* case a */
3214 memcpy(&orig_ex, ex, sizeof(orig_ex));
3215 ex->ee_len = cpu_to_le16(split - ee_block);
Lukas Czerner556615d2014-04-20 23:45:47 -04003216 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3217 ext4_ext_mark_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003218
3219 /*
3220 * path may lead to new leaf, not to original leaf any more
3221 * after ext4_ext_insert_extent() returns,
3222 */
3223 err = ext4_ext_dirty(handle, inode, path + depth);
3224 if (err)
3225 goto fix_extent_len;
3226
3227 ex2 = &newex;
3228 ex2->ee_block = cpu_to_le32(split);
3229 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3230 ext4_ext_store_pblock(ex2, newblock);
Lukas Czerner556615d2014-04-20 23:45:47 -04003231 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3232 ext4_ext_mark_unwritten(ex2);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003233
Theodore Ts'odfe50802014-09-01 14:37:09 -04003234 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003235 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003236 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003237 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003238 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003239 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003240 zero_ex.ee_len = cpu_to_le16(
3241 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003242 ext4_ext_store_pblock(&zero_ex,
3243 ext4_ext_pblock(ex2));
3244 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003245 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003246 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003247 zero_ex.ee_len = cpu_to_le16(
3248 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003249 ext4_ext_store_pblock(&zero_ex,
3250 ext4_ext_pblock(ex));
3251 }
3252 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003253 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003254 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003255 zero_ex.ee_len = cpu_to_le16(
3256 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003257 ext4_ext_store_pblock(&zero_ex,
3258 ext4_ext_pblock(&orig_ex));
3259 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003260
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003261 if (err)
3262 goto fix_extent_len;
3263 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003264 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003265 ext4_ext_try_to_merge(handle, inode, path, ex);
3266 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003267 if (err)
3268 goto fix_extent_len;
3269
3270 /* update extent status tree */
Zheng Liud7b2a002013-08-28 14:47:06 -04003271 err = ext4_zeroout_es(inode, &zero_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003272
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003273 goto out;
3274 } else if (err)
3275 goto fix_extent_len;
3276
3277out:
3278 ext4_ext_show_leaf(inode, path);
3279 return err;
3280
3281fix_extent_len:
3282 ex->ee_len = orig_ex.ee_len;
Dmitry Monakhov29faed12014-07-27 22:30:29 -04003283 ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003284 return err;
3285}
3286
3287/*
3288 * ext4_split_extents() splits an extent and mark extent which is covered
3289 * by @map as split_flags indicates
3290 *
Anatol Pomozov70261f52013-08-28 14:40:12 -04003291 * It may result in splitting the extent into multiple extents (up to three)
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003292 * There are three possibilities:
3293 * a> There is no split required
3294 * b> Splits in two extents: Split is happening at either end of the extent
3295 * c> Splits in three extents: Somone is splitting in middle of the extent
3296 *
3297 */
3298static int ext4_split_extent(handle_t *handle,
3299 struct inode *inode,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003300 struct ext4_ext_path **ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003301 struct ext4_map_blocks *map,
3302 int split_flag,
3303 int flags)
3304{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003305 struct ext4_ext_path *path = *ppath;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003306 ext4_lblk_t ee_block;
3307 struct ext4_extent *ex;
3308 unsigned int ee_len, depth;
3309 int err = 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04003310 int unwritten;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003311 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003312 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003313
3314 depth = ext_depth(inode);
3315 ex = path[depth].p_ext;
3316 ee_block = le32_to_cpu(ex->ee_block);
3317 ee_len = ext4_ext_get_actual_len(ex);
Lukas Czerner556615d2014-04-20 23:45:47 -04003318 unwritten = ext4_ext_is_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003319
3320 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003321 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003322 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
Lukas Czerner556615d2014-04-20 23:45:47 -04003323 if (unwritten)
3324 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3325 EXT4_EXT_MARK_UNWRIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003326 if (split_flag & EXT4_EXT_DATA_VALID2)
3327 split_flag1 |= EXT4_EXT_DATA_VALID1;
Theodore Ts'odfe50802014-09-01 14:37:09 -04003328 err = ext4_split_extent_at(handle, inode, ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003329 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003330 if (err)
3331 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003332 } else {
3333 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003334 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003335 /*
3336 * Update path is required because previous ext4_split_extent_at() may
3337 * result in split of original leaf or extent zeroout.
3338 */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04003339 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003340 if (IS_ERR(path))
3341 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003342 depth = ext_depth(inode);
3343 ex = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04003344 if (!ex) {
3345 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3346 (unsigned long) map->m_lblk);
3347 return -EIO;
3348 }
Lukas Czerner556615d2014-04-20 23:45:47 -04003349 unwritten = ext4_ext_is_unwritten(ex);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003350 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003351
3352 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003353 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
Lukas Czerner556615d2014-04-20 23:45:47 -04003354 if (unwritten) {
3355 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003356 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
Lukas Czerner556615d2014-04-20 23:45:47 -04003357 EXT4_EXT_MARK_UNWRIT2);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003358 }
Theodore Ts'odfe50802014-09-01 14:37:09 -04003359 err = ext4_split_extent_at(handle, inode, ppath,
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003360 map->m_lblk, split_flag1, flags);
3361 if (err)
3362 goto out;
3363 }
3364
3365 ext4_ext_show_leaf(inode, path);
3366out:
Zheng Liu3a225672013-03-10 21:20:23 -04003367 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003368}
3369
Amit Arora56055d32007-07-17 21:42:38 -04003370/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003371 * This function is called by ext4_ext_map_blocks() if someone tries to write
Lukas Czerner556615d2014-04-20 23:45:47 -04003372 * to an unwritten extent. It may result in splitting the unwritten
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003373 * extent into multiple extents (up to three - one initialized and two
Lukas Czerner556615d2014-04-20 23:45:47 -04003374 * unwritten).
Amit Arora56055d32007-07-17 21:42:38 -04003375 * There are three possibilities:
3376 * a> There is no split required: Entire extent should be initialized
3377 * b> Splits in two extents: Write is happening at either end of the extent
3378 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003379 *
3380 * Pre-conditions:
Lukas Czerner556615d2014-04-20 23:45:47 -04003381 * - The extent pointed to by 'path' is unwritten.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003382 * - The extent pointed to by 'path' contains a superset
3383 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3384 *
3385 * Post-conditions on success:
3386 * - the returned value is the number of blocks beyond map->l_lblk
3387 * that are allocated and initialized.
3388 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003389 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003390static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003391 struct inode *inode,
3392 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003393 struct ext4_ext_path **ppath,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003394 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003395{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003396 struct ext4_ext_path *path = *ppath;
Zheng Liu67a5da52012-08-17 09:54:17 -04003397 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003398 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003399 struct ext4_map_blocks split_map;
3400 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003401 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003402 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003403 unsigned int ee_len, depth, map_len = map->m_len;
3404 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003405 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003406 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003407
3408 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3409 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003410 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003411
Zheng Liu67a5da52012-08-17 09:54:17 -04003412 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003413 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3414 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003415 if (eof_block < map->m_lblk + map_len)
3416 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003417
3418 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003419 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003420 ex = path[depth].p_ext;
3421 ee_block = le32_to_cpu(ex->ee_block);
3422 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003423 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003424
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003425 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3426
3427 /* Pre-conditions */
Lukas Czerner556615d2014-04-20 23:45:47 -04003428 BUG_ON(!ext4_ext_is_unwritten(ex));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003429 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003430
3431 /*
3432 * Attempt to transfer newly initialized blocks from the currently
Lukas Czerner556615d2014-04-20 23:45:47 -04003433 * unwritten extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003434 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003435 * memmove() calls. Transferring to the left is the common case in
3436 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3437 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003438 *
3439 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003440 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003441 * This would require removing the extent if the transfer
3442 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003443 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003444 * same extent tree node.
3445 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003446 if ((map->m_lblk == ee_block) &&
3447 /* See if we can merge left */
3448 (map_len < ee_len) && /*L1*/
3449 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003450 ext4_lblk_t prev_lblk;
3451 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003452 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003453
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003454 abut_ex = ex - 1;
3455 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3456 prev_len = ext4_ext_get_actual_len(abut_ex);
3457 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003458 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003459
3460 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003461 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003462 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003463 * - C1: abut_ex is initialized,
3464 * - C2: abut_ex is logically abutting ex,
3465 * - C3: abut_ex is physically abutting ex,
3466 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003467 * overflowing the (initialized) length limit.
3468 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003469 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003470 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3471 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003472 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003473 err = ext4_ext_get_access(handle, inode, path + depth);
3474 if (err)
3475 goto out;
3476
3477 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003478 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003479
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003480 /* Shift the start of ex by 'map_len' blocks */
3481 ex->ee_block = cpu_to_le32(ee_block + map_len);
3482 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3483 ex->ee_len = cpu_to_le16(ee_len - map_len);
Lukas Czerner556615d2014-04-20 23:45:47 -04003484 ext4_ext_mark_unwritten(ex); /* Restore the flag */
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003485
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003486 /* Extend abut_ex by 'map_len' blocks */
3487 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003488
3489 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003490 allocated = map_len;
3491 }
3492 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3493 (map_len < ee_len) && /*L1*/
3494 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3495 /* See if we can merge right */
3496 ext4_lblk_t next_lblk;
3497 ext4_fsblk_t next_pblk, ee_pblk;
3498 unsigned int next_len;
3499
3500 abut_ex = ex + 1;
3501 next_lblk = le32_to_cpu(abut_ex->ee_block);
3502 next_len = ext4_ext_get_actual_len(abut_ex);
3503 next_pblk = ext4_ext_pblock(abut_ex);
3504 ee_pblk = ext4_ext_pblock(ex);
3505
3506 /*
3507 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3508 * upon those conditions:
3509 * - C1: abut_ex is initialized,
3510 * - C2: abut_ex is logically abutting ex,
3511 * - C3: abut_ex is physically abutting ex,
3512 * - C4: abut_ex can receive the additional blocks without
3513 * overflowing the (initialized) length limit.
3514 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003515 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003516 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3517 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3518 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3519 err = ext4_ext_get_access(handle, inode, path + depth);
3520 if (err)
3521 goto out;
3522
3523 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3524 map, ex, abut_ex);
3525
3526 /* Shift the start of abut_ex by 'map_len' blocks */
3527 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3528 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3529 ex->ee_len = cpu_to_le16(ee_len - map_len);
Lukas Czerner556615d2014-04-20 23:45:47 -04003530 ext4_ext_mark_unwritten(ex); /* Restore the flag */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003531
3532 /* Extend abut_ex by 'map_len' blocks */
3533 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3534
3535 /* Result: number of initialized blocks past m_lblk */
3536 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003537 }
3538 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003539 if (allocated) {
3540 /* Mark the block containing both extents as dirty */
3541 ext4_ext_dirty(handle, inode, path + depth);
3542
3543 /* Update path to point to the right extent */
3544 path[depth].p_ext = abut_ex;
3545 goto out;
3546 } else
3547 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003548
Yongqiang Yang667eff32011-05-03 12:25:07 -04003549 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003550 /*
3551 * It is safe to convert extent to initialized via explicit
Yongqiang Yang9e740562014-01-06 14:05:23 -05003552 * zeroout only if extent is fully inside i_size or new_size.
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003553 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003554 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003555
Zheng Liu67a5da52012-08-17 09:54:17 -04003556 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3557 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003558 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003559
3560 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3561 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003562 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003563 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003564 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003565 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003566 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003567 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003568
3569 err = ext4_ext_get_access(handle, inode, path + depth);
3570 if (err)
3571 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003572 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003573 ext4_ext_try_to_merge(handle, inode, path, ex);
3574 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003575 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003576 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003577
Amit Arora56055d32007-07-17 21:42:38 -04003578 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003579 * four cases:
3580 * 1. split the extent into three extents.
3581 * 2. split the extent into two extents, zeroout the first half.
3582 * 3. split the extent into two extents, zeroout the second half.
3583 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003584 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003585 split_map.m_lblk = map->m_lblk;
3586 split_map.m_len = map->m_len;
3587
Zheng Liu67a5da52012-08-17 09:54:17 -04003588 if (max_zeroout && (allocated > map->m_len)) {
3589 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003590 /* case 3 */
3591 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003592 cpu_to_le32(map->m_lblk);
3593 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003594 ext4_ext_store_pblock(&zero_ex,
3595 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3596 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003597 if (err)
3598 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003599 split_map.m_lblk = map->m_lblk;
3600 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003601 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003602 /* case 2 */
3603 if (map->m_lblk != ee_block) {
3604 zero_ex.ee_block = ex->ee_block;
3605 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3606 ee_block);
3607 ext4_ext_store_pblock(&zero_ex,
3608 ext4_ext_pblock(ex));
3609 err = ext4_ext_zeroout(inode, &zero_ex);
3610 if (err)
3611 goto out;
3612 }
3613
Yongqiang Yang667eff32011-05-03 12:25:07 -04003614 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003615 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3616 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003617 }
3618 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003619
Jan Karaae9e9c6a2014-10-30 10:53:17 -04003620 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3621 flags);
3622 if (err > 0)
3623 err = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003624out:
Zheng Liuadb23552013-03-10 21:13:05 -04003625 /* If we have gotten a failure, don't zero out status tree */
3626 if (!err)
Zheng Liud7b2a002013-08-28 14:47:06 -04003627 err = ext4_zeroout_es(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003628 return err ? err : allocated;
3629}
3630
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003631/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003632 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003633 * ext4_get_blocks_dio_write() when DIO to write
Lukas Czerner556615d2014-04-20 23:45:47 -04003634 * to an unwritten extent.
Mingming Cao00314622009-09-28 15:49:08 -04003635 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003636 * Writing to an unwritten extent may result in splitting the unwritten
3637 * extent into multiple initialized/unwritten extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003638 * There are three possibilities:
Lukas Czerner556615d2014-04-20 23:45:47 -04003639 * a> There is no split required: Entire extent should be unwritten
Mingming Cao00314622009-09-28 15:49:08 -04003640 * b> Splits in two extents: Write is happening at either end of the extent
3641 * c> Splits in three extents: Somone is writing in middle of the extent
3642 *
Lukas Czernerb8a86842014-03-18 18:05:35 -04003643 * This works the same way in the case of initialized -> unwritten conversion.
3644 *
Mingming Cao00314622009-09-28 15:49:08 -04003645 * One of more index blocks maybe needed if the extent tree grow after
Lukas Czerner556615d2014-04-20 23:45:47 -04003646 * the unwritten extent split. To prevent ENOSPC occur at the IO
3647 * complete, we need to split the unwritten extent before DIO submit
3648 * the IO. The unwritten extent called at this time will be split
3649 * into three unwritten extent(at most). After IO complete, the part
Mingming Cao00314622009-09-28 15:49:08 -04003650 * being filled will be convert to initialized by the end_io callback function
3651 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003652 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003653 * Returns the size of unwritten extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003654 */
Lukas Czernerb8a86842014-03-18 18:05:35 -04003655static int ext4_split_convert_extents(handle_t *handle,
Mingming Cao00314622009-09-28 15:49:08 -04003656 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003657 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003658 struct ext4_ext_path **ppath,
Mingming Cao00314622009-09-28 15:49:08 -04003659 int flags)
3660{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003661 struct ext4_ext_path *path = *ppath;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003662 ext4_lblk_t eof_block;
3663 ext4_lblk_t ee_block;
3664 struct ext4_extent *ex;
3665 unsigned int ee_len;
3666 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003667
Lukas Czernerb8a86842014-03-18 18:05:35 -04003668 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3669 __func__, inode->i_ino,
3670 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003671
3672 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3673 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003674 if (eof_block < map->m_lblk + map->m_len)
3675 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003676 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003677 * It is safe to convert extent to initialized via explicit
3678 * zeroout only if extent is fully insde i_size or new_size.
3679 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003680 depth = ext_depth(inode);
3681 ex = path[depth].p_ext;
3682 ee_block = le32_to_cpu(ex->ee_block);
3683 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003684
Lukas Czernerb8a86842014-03-18 18:05:35 -04003685 /* Convert to unwritten */
3686 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3687 split_flag |= EXT4_EXT_DATA_VALID1;
3688 /* Convert to initialized */
3689 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3690 split_flag |= ee_block + ee_len <= eof_block ?
3691 EXT4_EXT_MAY_ZEROOUT : 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04003692 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
Lukas Czernerb8a86842014-03-18 18:05:35 -04003693 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003694 flags |= EXT4_GET_BLOCKS_PRE_IO;
Theodore Ts'odfe50802014-09-01 14:37:09 -04003695 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003696}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003697
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003698static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003699 struct inode *inode,
3700 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04003701 struct ext4_ext_path **ppath)
Mingming Cao00314622009-09-28 15:49:08 -04003702{
Theodore Ts'odfe50802014-09-01 14:37:09 -04003703 struct ext4_ext_path *path = *ppath;
Mingming Cao00314622009-09-28 15:49:08 -04003704 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003705 ext4_lblk_t ee_block;
3706 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003707 int depth;
3708 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003709
3710 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003711 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003712 ee_block = le32_to_cpu(ex->ee_block);
3713 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003714
Yongqiang Yang197217a2011-05-03 11:45:29 -04003715 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3716 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003717 (unsigned long long)ee_block, ee_len);
3718
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003719 /* If extent is larger than requested it is a clear sign that we still
3720 * have some extent state machine issues left. So extent_split is still
3721 * required.
3722 * TODO: Once all related issues will be fixed this situation should be
3723 * illegal.
3724 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003725 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003726#ifdef EXT4_DEBUG
3727 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3728 " len %u; IO logical block %llu, len %u\n",
3729 inode->i_ino, (unsigned long long)ee_block, ee_len,
3730 (unsigned long long)map->m_lblk, map->m_len);
3731#endif
Theodore Ts'odfe50802014-09-01 14:37:09 -04003732 err = ext4_split_convert_extents(handle, inode, map, ppath,
Lukas Czernerb8a86842014-03-18 18:05:35 -04003733 EXT4_GET_BLOCKS_CONVERT);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003734 if (err < 0)
Theodore Ts'odfe50802014-09-01 14:37:09 -04003735 return err;
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04003736 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
Theodore Ts'odfe50802014-09-01 14:37:09 -04003737 if (IS_ERR(path))
3738 return PTR_ERR(path);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003739 depth = ext_depth(inode);
3740 ex = path[depth].p_ext;
3741 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003742
Mingming Cao00314622009-09-28 15:49:08 -04003743 err = ext4_ext_get_access(handle, inode, path + depth);
3744 if (err)
3745 goto out;
3746 /* first mark the extent as initialized */
3747 ext4_ext_mark_initialized(ex);
3748
Yongqiang Yang197217a2011-05-03 11:45:29 -04003749 /* note: ext4_ext_correct_indexes() isn't needed here because
3750 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003751 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003752 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003753
Mingming Cao00314622009-09-28 15:49:08 -04003754 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003755 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003756out:
3757 ext4_ext_show_leaf(inode, path);
3758 return err;
3759}
3760
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003761static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3762 sector_t block, int count)
3763{
3764 int i;
3765 for (i = 0; i < count; i++)
3766 unmap_underlying_metadata(bdev, block + i);
3767}
3768
Theodore Ts'o58590b02010-10-27 21:23:12 -04003769/*
3770 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3771 */
3772static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003773 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003774 struct ext4_ext_path *path,
3775 unsigned int len)
3776{
3777 int i, depth;
3778 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003779 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003780
3781 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3782 return 0;
3783
3784 depth = ext_depth(inode);
3785 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003786
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003787 /*
3788 * We're going to remove EOFBLOCKS_FL entirely in future so we
3789 * do not care for this case anymore. Simply remove the flag
3790 * if there are no extents.
3791 */
3792 if (unlikely(!eh->eh_entries))
3793 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003794 last_ex = EXT_LAST_EXTENT(eh);
3795 /*
3796 * We should clear the EOFBLOCKS_FL flag if we are writing the
3797 * last block in the last extent in the file. We test this by
3798 * first checking to see if the caller to
3799 * ext4_ext_get_blocks() was interested in the last block (or
3800 * a block beyond the last block) in the current extent. If
3801 * this turns out to be false, we can bail out from this
3802 * function immediately.
3803 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003804 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003805 ext4_ext_get_actual_len(last_ex))
3806 return 0;
3807 /*
3808 * If the caller does appear to be planning to write at or
3809 * beyond the end of the current extent, we then test to see
3810 * if the current extent is the last extent in the file, by
3811 * checking to make sure it was reached via the rightmost node
3812 * at each level of the tree.
3813 */
3814 for (i = depth-1; i >= 0; i--)
3815 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3816 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003817out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003818 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3819 return ext4_mark_inode_dirty(handle, inode);
3820}
3821
Aditya Kali7b415bf2011-09-09 19:04:51 -04003822/**
3823 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3824 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003825 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003826 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003827int ext4_find_delalloc_range(struct inode *inode,
3828 ext4_lblk_t lblk_start,
3829 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003830{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003831 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003832
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003833 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003834 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003835 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003836 else if (es.es_lblk <= lblk_start &&
3837 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003838 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003839 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003840 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003841 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003842 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003843}
3844
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003845int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003846{
3847 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3848 ext4_lblk_t lblk_start, lblk_end;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003849 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003850 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3851
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003852 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003853}
3854
3855/**
3856 * Determines how many complete clusters (out of those specified by the 'map')
3857 * are under delalloc and were reserved quota for.
3858 * This function is called when we are writing out the blocks that were
3859 * originally written with their allocation delayed, but then the space was
3860 * allocated using fallocate() before the delayed allocation could be resolved.
3861 * The cases to look for are:
3862 * ('=' indicated delayed allocated blocks
3863 * '-' indicates non-delayed allocated blocks)
3864 * (a) partial clusters towards beginning and/or end outside of allocated range
3865 * are not delalloc'ed.
3866 * Ex:
3867 * |----c---=|====c====|====c====|===-c----|
3868 * |++++++ allocated ++++++|
3869 * ==> 4 complete clusters in above example
3870 *
3871 * (b) partial cluster (outside of allocated range) towards either end is
3872 * marked for delayed allocation. In this case, we will exclude that
3873 * cluster.
3874 * Ex:
3875 * |----====c========|========c========|
3876 * |++++++ allocated ++++++|
3877 * ==> 1 complete clusters in above example
3878 *
3879 * Ex:
3880 * |================c================|
3881 * |++++++ allocated ++++++|
3882 * ==> 0 complete clusters in above example
3883 *
3884 * The ext4_da_update_reserve_space will be called only if we
3885 * determine here that there were some "entire" clusters that span
3886 * this 'allocated' range.
3887 * In the non-bigalloc case, this function will just end up returning num_blks
3888 * without ever calling ext4_find_delalloc_range.
3889 */
3890static unsigned int
3891get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3892 unsigned int num_blks)
3893{
3894 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3895 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3896 ext4_lblk_t lblk_from, lblk_to, c_offset;
3897 unsigned int allocated_clusters = 0;
3898
3899 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3900 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3901
3902 /* max possible clusters for this allocation */
3903 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3904
Aditya Kalid8990242011-09-09 19:18:51 -04003905 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3906
Aditya Kali7b415bf2011-09-09 19:04:51 -04003907 /* Check towards left side */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003908 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003909 if (c_offset) {
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003910 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003911 lblk_to = lblk_from + c_offset - 1;
3912
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003913 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003914 allocated_clusters--;
3915 }
3916
3917 /* Now check towards right. */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003918 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003919 if (allocated_clusters && c_offset) {
3920 lblk_from = lblk_start + num_blks;
3921 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3922
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003923 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003924 allocated_clusters--;
3925 }
3926
3927 return allocated_clusters;
3928}
3929
Mingming Cao00314622009-09-28 15:49:08 -04003930static int
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003931convert_initialized_extent(handle_t *handle, struct inode *inode,
3932 struct ext4_map_blocks *map,
Theodore Ts'o4f224b82014-09-01 14:36:09 -04003933 struct ext4_ext_path **ppath, int flags,
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003934 unsigned int allocated, ext4_fsblk_t newblock)
Lukas Czernerb8a86842014-03-18 18:05:35 -04003935{
Theodore Ts'o4f224b82014-09-01 14:36:09 -04003936 struct ext4_ext_path *path = *ppath;
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003937 struct ext4_extent *ex;
3938 ext4_lblk_t ee_block;
3939 unsigned int ee_len;
3940 int depth;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003941 int err = 0;
3942
3943 /*
3944 * Make sure that the extent is no bigger than we support with
Lukas Czerner556615d2014-04-20 23:45:47 -04003945 * unwritten extent
Lukas Czernerb8a86842014-03-18 18:05:35 -04003946 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003947 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3948 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003949
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003950 depth = ext_depth(inode);
3951 ex = path[depth].p_ext;
3952 ee_block = le32_to_cpu(ex->ee_block);
3953 ee_len = ext4_ext_get_actual_len(ex);
3954
3955 ext_debug("%s: inode %lu, logical"
3956 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3957 (unsigned long long)ee_block, ee_len);
3958
3959 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Theodore Ts'odfe50802014-09-01 14:37:09 -04003960 err = ext4_split_convert_extents(handle, inode, map, ppath,
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003961 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3962 if (err < 0)
3963 return err;
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04003964 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04003965 if (IS_ERR(path))
3966 return PTR_ERR(path);
3967 depth = ext_depth(inode);
3968 ex = path[depth].p_ext;
3969 if (!ex) {
3970 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3971 (unsigned long) map->m_lblk);
3972 return -EIO;
3973 }
3974 }
3975
3976 err = ext4_ext_get_access(handle, inode, path + depth);
3977 if (err)
3978 return err;
3979 /* first mark the extent as unwritten */
3980 ext4_ext_mark_unwritten(ex);
3981
3982 /* note: ext4_ext_correct_indexes() isn't needed here because
3983 * borders are not changed
3984 */
3985 ext4_ext_try_to_merge(handle, inode, path, ex);
3986
3987 /* Mark modified extent as dirty */
3988 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3989 if (err)
3990 return err;
3991 ext4_ext_show_leaf(inode, path);
3992
3993 ext4_update_inode_fsync_trans(handle, inode, 1);
3994 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3995 if (err)
3996 return err;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003997 map->m_flags |= EXT4_MAP_UNWRITTEN;
3998 if (allocated > map->m_len)
3999 allocated = map->m_len;
4000 map->m_len = allocated;
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04004001 return allocated;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004002}
4003
4004static int
Lukas Czerner556615d2014-04-20 23:45:47 -04004005ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004006 struct ext4_map_blocks *map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04004007 struct ext4_ext_path **ppath, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004008 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04004009{
Theodore Ts'odfe50802014-09-01 14:37:09 -04004010 struct ext4_ext_path *path = *ppath;
Mingming Cao00314622009-09-28 15:49:08 -04004011 int ret = 0;
4012 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004013 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04004014
Lukas Czerner556615d2014-04-20 23:45:47 -04004015 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
Zheng Liu88635ca2011-12-28 19:00:25 -05004016 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004017 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04004018 flags, allocated);
4019 ext4_ext_show_leaf(inode, path);
4020
Lukas Czerner27dd4382013-04-09 22:11:22 -04004021 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04004022 * When writing into unwritten space, we should not fail to
Lukas Czerner27dd4382013-04-09 22:11:22 -04004023 * allocate metadata blocks for the new extent block if needed.
4024 */
4025 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4026
Lukas Czerner556615d2014-04-20 23:45:47 -04004027 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
Zheng Liub5645532012-11-08 14:33:43 -05004028 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04004029
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004030 /* get_block() before submit the IO, split the extent */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004031 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
Theodore Ts'odfe50802014-09-01 14:37:09 -04004032 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4033 flags | EXT4_GET_BLOCKS_CONVERT);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004034 if (ret <= 0)
4035 goto out;
Mingming5f524952009-11-10 10:48:04 -05004036 /*
4037 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004038 * that this IO needs to conversion to written when IO is
Mingming5f524952009-11-10 10:48:04 -05004039 * completed
4040 */
Tao Ma0edeb712011-10-31 17:30:44 -04004041 if (io)
4042 ext4_set_io_unwritten_flag(inode, io);
4043 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004044 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05004045 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004046 goto out;
4047 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004048 /* IO end_io complete, convert the filled extent to written */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004049 if (flags & EXT4_GET_BLOCKS_CONVERT) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04004050 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Theodore Ts'odfe50802014-09-01 14:37:09 -04004051 ppath);
Theodore Ts'o58590b02010-10-27 21:23:12 -04004052 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05004053 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05004054 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4055 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04004056 } else
4057 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04004058 map->m_flags |= EXT4_MAP_MAPPED;
Eric Whitney15cc1762014-02-12 10:42:45 -05004059 map->m_pblk = newblock;
Zheng Liucdee7842013-03-10 21:08:52 -04004060 if (allocated > map->m_len)
4061 allocated = map->m_len;
4062 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04004063 goto out2;
4064 }
4065 /* buffered IO case */
4066 /*
4067 * repeat fallocate creation request
4068 * we already have an unwritten extent
4069 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004070 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
Zheng Liua25a4e12013-02-18 00:28:04 -05004071 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004072 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05004073 }
Mingming Cao00314622009-09-28 15:49:08 -04004074
4075 /* buffered READ or buffered write_begin() lookup */
4076 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4077 /*
4078 * We have blocks reserved already. We
4079 * return allocated blocks so that delalloc
4080 * won't do block reservation for us. But
4081 * the buffer head will be unmapped so that
4082 * a read from the block returns 0s.
4083 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004084 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004085 goto out1;
4086 }
4087
4088 /* buffered write, writepage time, convert*/
Theodore Ts'odfe50802014-09-01 14:37:09 -04004089 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004090 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004091 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04004092out:
4093 if (ret <= 0) {
4094 err = ret;
4095 goto out2;
4096 } else
4097 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004098 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004099 /*
4100 * if we allocated more blocks than requested
4101 * we need to make sure we unmap the extra block
4102 * allocated. The actual needed block will get
4103 * unmapped later when we find the buffer_head marked
4104 * new.
4105 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004106 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004107 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004108 newblock + map->m_len,
4109 allocated - map->m_len);
4110 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004111 }
Zheng Liu3a225672013-03-10 21:20:23 -04004112 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004113
4114 /*
4115 * If we have done fallocate with the offset that is already
4116 * delayed allocated, we would have block reservation
4117 * and quota reservation done in the delayed write path.
4118 * But fallocate would have already updated quota and block
4119 * count for this offset. So cancel these reservation
4120 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004121 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4122 unsigned int reserved_clusters;
4123 reserved_clusters = get_reserved_cluster_alloc(inode,
4124 map->m_lblk, map->m_len);
4125 if (reserved_clusters)
4126 ext4_da_update_reserve_space(inode,
4127 reserved_clusters,
4128 0);
4129 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004130
Mingming Cao00314622009-09-28 15:49:08 -04004131map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004132 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004133 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4134 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4135 map->m_len);
4136 if (err < 0)
4137 goto out2;
4138 }
Mingming Cao00314622009-09-28 15:49:08 -04004139out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004140 if (allocated > map->m_len)
4141 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04004142 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004143 map->m_pblk = newblock;
4144 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04004145out2:
Mingming Cao00314622009-09-28 15:49:08 -04004146 return err ? err : allocated;
4147}
Theodore Ts'o58590b02010-10-27 21:23:12 -04004148
Mingming Cao00314622009-09-28 15:49:08 -04004149/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004150 * get_implied_cluster_alloc - check to see if the requested
4151 * allocation (in the map structure) overlaps with a cluster already
4152 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04004153 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004154 * @map The requested lblk->pblk mapping
4155 * @ex The extent structure which might contain an implied
4156 * cluster allocation
4157 *
4158 * This function is called by ext4_ext_map_blocks() after we failed to
4159 * find blocks that were already in the inode's extent tree. Hence,
4160 * we know that the beginning of the requested region cannot overlap
4161 * the extent from the inode's extent tree. There are three cases we
4162 * want to catch. The first is this case:
4163 *
4164 * |--- cluster # N--|
4165 * |--- extent ---| |---- requested region ---|
4166 * |==========|
4167 *
4168 * The second case that we need to test for is this one:
4169 *
4170 * |--------- cluster # N ----------------|
4171 * |--- requested region --| |------- extent ----|
4172 * |=======================|
4173 *
4174 * The third case is when the requested region lies between two extents
4175 * within the same cluster:
4176 * |------------- cluster # N-------------|
4177 * |----- ex -----| |---- ex_right ----|
4178 * |------ requested region ------|
4179 * |================|
4180 *
4181 * In each of the above cases, we need to set the map->m_pblk and
4182 * map->m_len so it corresponds to the return the extent labelled as
4183 * "|====|" from cluster #N, since it is already in use for data in
4184 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4185 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4186 * as a new "allocated" block region. Otherwise, we will return 0 and
4187 * ext4_ext_map_blocks() will then allocate one or more new clusters
4188 * by calling ext4_mb_new_blocks().
4189 */
Aditya Kalid8990242011-09-09 19:18:51 -04004190static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004191 struct ext4_map_blocks *map,
4192 struct ext4_extent *ex,
4193 struct ext4_ext_path *path)
4194{
Aditya Kalid8990242011-09-09 19:18:51 -04004195 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004196 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004197 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05004198 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004199 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4200 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4201 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4202
4203 /* The extent passed in that we are trying to match */
4204 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4205 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4206
4207 /* The requested region passed into ext4_map_blocks() */
4208 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004209
4210 if ((rr_cluster_start == ex_cluster_end) ||
4211 (rr_cluster_start == ex_cluster_start)) {
4212 if (rr_cluster_start == ex_cluster_end)
4213 ee_start += ee_len - 1;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004214 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004215 map->m_len = min(map->m_len,
4216 (unsigned) sbi->s_cluster_ratio - c_offset);
4217 /*
4218 * Check for and handle this case:
4219 *
4220 * |--------- cluster # N-------------|
4221 * |------- extent ----|
4222 * |--- requested region ---|
4223 * |===========|
4224 */
4225
4226 if (map->m_lblk < ee_block)
4227 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4228
4229 /*
4230 * Check for the case where there is already another allocated
4231 * block to the right of 'ex' but before the end of the cluster.
4232 *
4233 * |------------- cluster # N-------------|
4234 * |----- ex -----| |---- ex_right ----|
4235 * |------ requested region ------|
4236 * |================|
4237 */
4238 if (map->m_lblk > ee_block) {
4239 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4240 map->m_len = min(map->m_len, next - map->m_lblk);
4241 }
Aditya Kalid8990242011-09-09 19:18:51 -04004242
4243 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004244 return 1;
4245 }
Aditya Kalid8990242011-09-09 19:18:51 -04004246
4247 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004248 return 0;
4249}
4250
4251
4252/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05004253 * Block allocation/map/preallocation routine for extents based files
4254 *
4255 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004256 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004257 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4258 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004259 *
4260 * return > 0, number of of blocks already mapped/allocated
4261 * if create == 0 and these are pre-allocated blocks
4262 * buffer head is unmapped
4263 * otherwise blocks are mapped
4264 *
4265 * return = 0, if plain look up failed (blocks have not been allocated)
4266 * buffer head is unmapped
4267 *
4268 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004269 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004270int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4271 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004272{
4273 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004274 struct ext4_extent newex, *ex, *ex2;
4275 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004276 ext4_fsblk_t newblock = 0;
Eric Whitneyce37c422014-02-19 18:52:39 -05004277 int free_on_err = 0, err = 0, depth, ret;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004278 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004279 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004280 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004281 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004282 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004283 int set_unwritten = 0;
Jan Karacbd75842014-11-25 11:41:49 -05004284 bool map_from_cluster = false;
Alex Tomasa86c6182006-10-11 01:21:03 -07004285
Mingming84fe3be2009-09-01 08:44:37 -04004286 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004287 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004288 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004289
Alex Tomasa86c6182006-10-11 01:21:03 -07004290 /* find extent for this block */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04004291 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004292 if (IS_ERR(path)) {
4293 err = PTR_ERR(path);
4294 path = NULL;
4295 goto out2;
4296 }
4297
4298 depth = ext_depth(inode);
4299
4300 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004301 * consistent leaf must not be empty;
4302 * this situation is possible, though, _during_ tree modification;
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04004303 * this is why assert can't be put in ext4_find_extent()
Alex Tomasa86c6182006-10-11 01:21:03 -07004304 */
Frank Mayhar273df552010-03-02 11:46:09 -05004305 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4306 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004307 "lblock: %lu, depth: %d pblock %lld",
4308 (unsigned long) map->m_lblk, depth,
4309 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004310 err = -EIO;
4311 goto out2;
4312 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004313
Avantika Mathur7e028972006-12-06 20:41:33 -08004314 ex = path[depth].p_ext;
4315 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004316 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004317 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004318 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004319
Lukas Czernerb8a86842014-03-18 18:05:35 -04004320
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004321 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04004322 * unwritten extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004323 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004324 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004325 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004326
4327 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4328
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004329 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004330 if (in_range(map->m_lblk, ee_block, ee_len)) {
4331 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004332 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004333 allocated = ee_len - (map->m_lblk - ee_block);
4334 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4335 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004336
Lukas Czernerb8a86842014-03-18 18:05:35 -04004337 /*
4338 * If the extent is initialized check whether the
4339 * caller wants to convert it to unwritten.
4340 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004341 if ((!ext4_ext_is_unwritten(ex)) &&
Lukas Czernerb8a86842014-03-18 18:05:35 -04004342 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
Theodore Ts'oe8b83d932014-09-01 14:35:09 -04004343 allocated = convert_initialized_extent(
Theodore Ts'o4f224b82014-09-01 14:36:09 -04004344 handle, inode, map, &path,
4345 flags, allocated, newblock);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004346 goto out2;
Lukas Czerner556615d2014-04-20 23:45:47 -04004347 } else if (!ext4_ext_is_unwritten(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004348 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004349
Lukas Czerner556615d2014-04-20 23:45:47 -04004350 ret = ext4_ext_handle_unwritten_extents(
Theodore Ts'odfe50802014-09-01 14:37:09 -04004351 handle, inode, map, &path, flags,
Lukas Czerner78771912012-03-19 23:05:43 -04004352 allocated, newblock);
Eric Whitneyce37c422014-02-19 18:52:39 -05004353 if (ret < 0)
4354 err = ret;
4355 else
4356 allocated = ret;
Eric Whitney31cf0f22014-03-13 23:14:46 -04004357 goto out2;
Alex Tomasa86c6182006-10-11 01:21:03 -07004358 }
4359 }
4360
4361 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004362 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004363 * we couldn't try to create block if create flag is zero
4364 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004365 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004366 /*
4367 * put just found gap into cache to speed up
4368 * subsequent requests
4369 */
Zheng Liu2f8e0a72014-11-25 11:44:37 -05004370 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004371 goto out2;
4372 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004373
Alex Tomasa86c6182006-10-11 01:21:03 -07004374 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004375 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004376 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004377 newex.ee_block = cpu_to_le32(map->m_lblk);
Eric Whitneyd0abafa2014-01-06 14:00:23 -05004378 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004379
4380 /*
4381 * If we are doing bigalloc, check to see if the extent returned
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04004382 * by ext4_find_extent() implies a cluster we can use.
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004383 */
4384 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004385 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004386 ar.len = allocated = map->m_len;
4387 newblock = map->m_pblk;
Jan Karacbd75842014-11-25 11:41:49 -05004388 map_from_cluster = true;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004389 goto got_allocated_blocks;
4390 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004391
Alex Tomasc9de5602008-01-29 00:19:52 -05004392 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004393 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004394 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4395 if (err)
4396 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004397 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004398 ex2 = NULL;
4399 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004400 if (err)
4401 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004402
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004403 /* Check if the extent after searching to the right implies a
4404 * cluster we can use. */
4405 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004406 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004407 ar.len = allocated = map->m_len;
4408 newblock = map->m_pblk;
Jan Karacbd75842014-11-25 11:41:49 -05004409 map_from_cluster = true;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004410 goto got_allocated_blocks;
4411 }
4412
Amit Arora749269f2007-07-18 09:02:56 -04004413 /*
4414 * See if request is beyond maximum number of blocks we can have in
4415 * a single extent. For an initialized extent this limit is
Lukas Czerner556615d2014-04-20 23:45:47 -04004416 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4417 * EXT_UNWRITTEN_MAX_LEN.
Amit Arora749269f2007-07-18 09:02:56 -04004418 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004419 if (map->m_len > EXT_INIT_MAX_LEN &&
Lukas Czerner556615d2014-04-20 23:45:47 -04004420 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004421 map->m_len = EXT_INIT_MAX_LEN;
Lukas Czerner556615d2014-04-20 23:45:47 -04004422 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4423 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4424 map->m_len = EXT_UNWRITTEN_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004425
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004426 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004427 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004428 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004429 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004430 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004431 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004432 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004433
4434 /* allocate new block */
4435 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004436 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4437 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004438 /*
4439 * We calculate the offset from the beginning of the cluster
4440 * for the logical block number, since when we allocate a
4441 * physical cluster, the physical block should start at the
4442 * same offset from the beginning of the cluster. This is
4443 * needed so that future calls to get_implied_cluster_alloc()
4444 * work correctly.
4445 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004446 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004447 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4448 ar.goal -= offset;
4449 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004450 if (S_ISREG(inode->i_mode))
4451 ar.flags = EXT4_MB_HINT_DATA;
4452 else
4453 /* disable in-core preallocation for non-regular files */
4454 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004455 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4456 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Theodore Ts'oe3cf5d52014-09-04 18:07:25 -04004457 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4458 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
Alex Tomasc9de5602008-01-29 00:19:52 -05004459 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004460 if (!newblock)
4461 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004462 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004463 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004464 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004465 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004466 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4467 if (ar.len > allocated)
4468 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004469
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004470got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004471 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004472 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004473 newex.ee_len = cpu_to_le16(ar.len);
Lukas Czerner556615d2014-04-20 23:45:47 -04004474 /* Mark unwritten */
4475 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4476 ext4_ext_mark_unwritten(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004477 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004478 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004479 * io_end structure was created for every IO write to an
Lukas Czerner556615d2014-04-20 23:45:47 -04004480 * unwritten extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004481 * here we flag the IO that really needs the conversion.
Mingming5f524952009-11-10 10:48:04 -05004482 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004483 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004484 */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004485 if (flags & EXT4_GET_BLOCKS_PRE_IO)
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004486 set_unwritten = 1;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004487 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004488
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004489 err = 0;
4490 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4491 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4492 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004493 if (!err)
Theodore Ts'odfe50802014-09-01 14:37:09 -04004494 err = ext4_ext_insert_extent(handle, inode, &path,
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004495 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004496
4497 if (!err && set_unwritten) {
4498 if (io)
4499 ext4_set_io_unwritten_flag(inode, io);
4500 else
4501 ext4_set_inode_state(inode,
4502 EXT4_STATE_DIO_UNWRITTEN);
4503 }
4504
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004505 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004506 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4507 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004508 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004509 /* not a good idea to call discard here directly,
4510 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004511 ext4_discard_preallocations(inode);
Theodore Ts'oc8e15132013-07-15 00:09:37 -04004512 ext4_free_blocks(handle, inode, NULL, newblock,
4513 EXT4_C2B(sbi, allocated_clusters), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004514 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004515 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004516
Alex Tomasa86c6182006-10-11 01:21:03 -07004517 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004518 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004519 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004520 if (allocated > map->m_len)
4521 allocated = map->m_len;
4522 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004523
Jan Karab436b9b2009-12-08 23:51:10 -05004524 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004525 * Update reserved blocks/metadata blocks after successful
4526 * block allocation which had been deferred till now.
4527 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004528 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004529 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004530 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004531 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004532 */
4533 reserved_clusters = get_reserved_cluster_alloc(inode,
4534 map->m_lblk, allocated);
Eric Whitney9d21c9f2015-04-03 00:17:31 -04004535 if (!map_from_cluster) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04004536 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004537 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f262011-09-09 19:20:51 -04004538 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004539 int reservation = allocated_clusters -
4540 reserved_clusters;
4541 /*
4542 * It seems we claimed few clusters outside of
4543 * the range of this allocation. We should give
4544 * it back to the reservation pool. This can
4545 * happen in the following case:
4546 *
4547 * * Suppose s_cluster_ratio is 4 (i.e., each
4548 * cluster has 4 blocks. Thus, the clusters
4549 * are [0-3],[4-7],[8-11]...
4550 * * First comes delayed allocation write for
4551 * logical blocks 10 & 11. Since there were no
4552 * previous delayed allocated blocks in the
4553 * range [8-11], we would reserve 1 cluster
4554 * for this write.
4555 * * Next comes write for logical blocks 3 to 8.
4556 * In this case, we will reserve 2 clusters
4557 * (for [0-3] and [4-7]; and not for [8-11] as
4558 * that range has a delayed allocated blocks.
4559 * Thus total reserved clusters now becomes 3.
4560 * * Now, during the delayed allocation writeout
4561 * time, we will first write blocks [3-8] and
4562 * allocate 3 clusters for writing these
4563 * blocks. Also, we would claim all these
4564 * three clusters above.
4565 * * Now when we come here to writeout the
4566 * blocks [10-11], we would expect to claim
4567 * the reservation of 1 cluster we had made
4568 * (and we would claim it since there are no
4569 * more delayed allocated blocks in the range
4570 * [8-11]. But our reserved cluster count had
4571 * already gone to 0.
4572 *
4573 * Thus, at the step 4 above when we determine
4574 * that there are still some unwritten delayed
4575 * allocated blocks outside of our current
4576 * block range, we should increment the
4577 * reserved clusters count so that when the
4578 * remaining blocks finally gets written, we
4579 * could claim them.
4580 */
Aditya Kali5356f262011-09-09 19:20:51 -04004581 dquot_reserve_block(inode,
4582 EXT4_C2B(sbi, reservation));
4583 spin_lock(&ei->i_block_reservation_lock);
4584 ei->i_reserved_data_blocks += reservation;
4585 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004586 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004587 /*
4588 * We will claim quota for all newly allocated blocks.
4589 * We're updating the reserved space *after* the
4590 * correction above so we do not accidentally free
4591 * all the metadata reservation because we might
4592 * actually need it later on.
4593 */
4594 ext4_da_update_reserve_space(inode, allocated_clusters,
4595 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004596 }
4597 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004598
4599 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004600 * Cache the extent and update transaction to commit on fdatasync only
Lukas Czerner556615d2014-04-20 23:45:47 -04004601 * when it is _not_ an unwritten extent.
Jan Karab436b9b2009-12-08 23:51:10 -05004602 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004603 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004604 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004605 else
Jan Karab436b9b2009-12-08 23:51:10 -05004606 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004607out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004608 if (allocated > map->m_len)
4609 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004610 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004611 map->m_flags |= EXT4_MAP_MAPPED;
4612 map->m_pblk = newblock;
4613 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004614out2:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04004615 ext4_ext_drop_refs(path);
4616 kfree(path);
Allison Hendersone8613042011-05-25 07:41:46 -04004617
Theodore Ts'o63b99962013-07-16 10:28:47 -04004618 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4619 err ? err : allocated);
Lukas Czerner78771912012-03-19 23:05:43 -04004620 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004621}
4622
Theodore Ts'o819c4922013-04-03 12:47:17 -04004623void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004624{
Alex Tomasa86c6182006-10-11 01:21:03 -07004625 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004626 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004627 int err = 0;
4628
4629 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004630 * TODO: optimization is possible here.
4631 * Probably we need not scan at all,
4632 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004633 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004634
4635 /* we have to know where to truncate from in crash case */
4636 EXT4_I(inode)->i_disksize = inode->i_size;
4637 ext4_mark_inode_dirty(handle, inode);
4638
4639 last_block = (inode->i_size + sb->s_blocksize - 1)
4640 >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004641retry:
Zheng Liu51865fd2012-11-08 21:57:32 -05004642 err = ext4_es_remove_extent(inode, last_block,
4643 EXT_MAX_BLOCKS - last_block);
Theodore Ts'o94eec0f2013-07-29 12:12:56 -04004644 if (err == -ENOMEM) {
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004645 cond_resched();
4646 congestion_wait(BLK_RW_ASYNC, HZ/50);
4647 goto retry;
4648 }
4649 if (err) {
4650 ext4_std_error(inode->i_sb, err);
4651 return;
4652 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04004653 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004654 ext4_std_error(inode->i_sb, err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004655}
4656
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004657static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004658 ext4_lblk_t len, loff_t new_size,
4659 int flags, int mode)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004660{
Al Viro496ad9a2013-01-23 17:07:38 -05004661 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004662 handle_t *handle;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004663 int ret = 0;
4664 int ret2 = 0;
4665 int retries = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004666 struct ext4_map_blocks map;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004667 unsigned int credits;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004668 loff_t epos;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004669
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004670 map.m_lblk = offset;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004671 map.m_len = len;
Greg Harm3c6fe772011-10-31 18:41:47 -04004672 /*
4673 * Don't normalize the request if it can fit in one extent so
4674 * that it doesn't get unnecessarily split into multiple
4675 * extents.
4676 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004677 if (len <= EXT_UNWRITTEN_MAX_LEN)
Greg Harm3c6fe772011-10-31 18:41:47 -04004678 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004679
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004680 /*
4681 * credits to insert 1 extent into extent tree
4682 */
4683 credits = ext4_chunk_trans_blocks(inode, len);
4684
Amit Aroraa2df2a62007-07-17 21:42:41 -04004685retry:
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004686 while (ret >= 0 && len) {
Theodore Ts'o9924a922013-02-08 21:59:22 -05004687 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4688 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004689 if (IS_ERR(handle)) {
4690 ret = PTR_ERR(handle);
4691 break;
4692 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004693 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004694 if (ret <= 0) {
Lukas Czernerf282ac12014-03-18 17:44:35 -04004695 ext4_debug("inode #%lu: block %u: len %u: "
4696 "ext4_ext_map_blocks returned %d",
4697 inode->i_ino, map.m_lblk,
4698 map.m_len, ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004699 ext4_mark_inode_dirty(handle, inode);
4700 ret2 = ext4_journal_stop(handle);
4701 break;
4702 }
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004703 map.m_lblk += ret;
4704 map.m_len = len = len - ret;
4705 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4706 inode->i_ctime = ext4_current_time(inode);
4707 if (new_size) {
4708 if (epos > new_size)
4709 epos = new_size;
4710 if (ext4_update_inode_size(inode, epos) & 0x1)
4711 inode->i_mtime = inode->i_ctime;
4712 } else {
4713 if (epos > inode->i_size)
4714 ext4_set_inode_flag(inode,
4715 EXT4_INODE_EOFBLOCKS);
4716 }
4717 ext4_mark_inode_dirty(handle, inode);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004718 ret2 = ext4_journal_stop(handle);
4719 if (ret2)
4720 break;
4721 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004722 if (ret == -ENOSPC &&
4723 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4724 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004725 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004726 }
Lukas Czernerf282ac12014-03-18 17:44:35 -04004727
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004728 return ret > 0 ? ret2 : ret;
4729}
4730
Lukas Czernerb8a86842014-03-18 18:05:35 -04004731static long ext4_zero_range(struct file *file, loff_t offset,
4732 loff_t len, int mode)
4733{
4734 struct inode *inode = file_inode(file);
4735 handle_t *handle = NULL;
4736 unsigned int max_blocks;
4737 loff_t new_size = 0;
4738 int ret = 0;
4739 int flags;
Dmitry Monakhov69dc9532014-08-27 18:33:49 -04004740 int credits;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004741 int partial_begin, partial_end;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004742 loff_t start, end;
4743 ext4_lblk_t lblk;
4744 struct address_space *mapping = inode->i_mapping;
4745 unsigned int blkbits = inode->i_blkbits;
4746
4747 trace_ext4_zero_range(inode, offset, len, mode);
4748
jon ernst6c5e73d2014-04-18 11:50:35 -04004749 if (!S_ISREG(inode->i_mode))
4750 return -EINVAL;
4751
Namjae Jeone1ee60f2014-05-27 12:48:55 -04004752 /* Call ext4_force_commit to flush all data in case of data=journal. */
4753 if (ext4_should_journal_data(inode)) {
4754 ret = ext4_force_commit(inode->i_sb);
4755 if (ret)
4756 return ret;
4757 }
4758
Lukas Czernerb8a86842014-03-18 18:05:35 -04004759 /*
4760 * Write out all dirty pages to avoid race conditions
4761 * Then release them.
4762 */
4763 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4764 ret = filemap_write_and_wait_range(mapping, offset,
4765 offset + len - 1);
4766 if (ret)
4767 return ret;
4768 }
4769
4770 /*
4771 * Round up offset. This is not fallocate, we neet to zero out
4772 * blocks, so convert interior block aligned part of the range to
4773 * unwritten and possibly manually zero out unaligned parts of the
4774 * range.
4775 */
4776 start = round_up(offset, 1 << blkbits);
4777 end = round_down((offset + len), 1 << blkbits);
4778
4779 if (start < offset || end > offset + len)
4780 return -EINVAL;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004781 partial_begin = offset & ((1 << blkbits) - 1);
4782 partial_end = (offset + len) & ((1 << blkbits) - 1);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004783
4784 lblk = start >> blkbits;
4785 max_blocks = (end >> blkbits);
4786 if (max_blocks < lblk)
4787 max_blocks = 0;
4788 else
4789 max_blocks -= lblk;
4790
Lukas Czernerb8a86842014-03-18 18:05:35 -04004791 mutex_lock(&inode->i_mutex);
4792
4793 /*
4794 * Indirect files do not support unwritten extnets
4795 */
4796 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4797 ret = -EOPNOTSUPP;
4798 goto out_mutex;
4799 }
4800
4801 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4802 offset + len > i_size_read(inode)) {
4803 new_size = offset + len;
4804 ret = inode_newsize_ok(inode, new_size);
4805 if (ret)
4806 goto out_mutex;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004807 }
4808
Lukas Czerner0f2af212015-04-03 00:09:13 -04004809 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4810 if (mode & FALLOC_FL_KEEP_SIZE)
4811 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4812
4813 /* Preallocate the range including the unaligned edges */
4814 if (partial_begin || partial_end) {
4815 ret = ext4_alloc_file_blocks(file,
4816 round_down(offset, 1 << blkbits) >> blkbits,
4817 (round_up((offset + len), 1 << blkbits) -
4818 round_down(offset, 1 << blkbits)) >> blkbits,
4819 new_size, flags, mode);
4820 if (ret)
4821 goto out_mutex;
4822
4823 }
4824
4825 /* Zero range excluding the unaligned edges */
Lukas Czernerb8a86842014-03-18 18:05:35 -04004826 if (max_blocks > 0) {
Lukas Czerner0f2af212015-04-03 00:09:13 -04004827 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4828 EXT4_EX_NOCACHE);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004829
4830 /* Now release the pages and zero block aligned part of pages*/
4831 truncate_pagecache_range(inode, start, end - 1);
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004832 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004833
4834 /* Wait all existing dio workers, newcomers will block on i_mutex */
4835 ext4_inode_block_unlocked_dio(inode);
4836 inode_dio_wait(inode);
4837
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004838 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4839 flags, mode);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004840 if (ret)
4841 goto out_dio;
4842 }
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004843 if (!partial_begin && !partial_end)
4844 goto out_dio;
4845
Dmitry Monakhov69dc9532014-08-27 18:33:49 -04004846 /*
4847 * In worst case we have to writeout two nonadjacent unwritten
4848 * blocks and update the inode
4849 */
4850 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4851 if (ext4_should_journal_data(inode))
4852 credits += 2;
4853 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004854 if (IS_ERR(handle)) {
4855 ret = PTR_ERR(handle);
4856 ext4_std_error(inode->i_sb, ret);
4857 goto out_dio;
4858 }
4859
4860 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Lukas Czernere5b30412014-04-01 00:59:21 -04004861 if (new_size) {
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04004862 ext4_update_inode_size(inode, new_size);
Lukas Czernere5b30412014-04-01 00:59:21 -04004863 } else {
Lukas Czernerb8a86842014-03-18 18:05:35 -04004864 /*
4865 * Mark that we allocate beyond EOF so the subsequent truncate
4866 * can proceed even if the new size is the same as i_size.
4867 */
4868 if ((offset + len) > i_size_read(inode))
4869 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4870 }
Lukas Czernerb8a86842014-03-18 18:05:35 -04004871 ext4_mark_inode_dirty(handle, inode);
4872
4873 /* Zero out partial block at the edges of the range */
4874 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4875
4876 if (file->f_flags & O_SYNC)
4877 ext4_handle_sync(handle);
4878
4879 ext4_journal_stop(handle);
4880out_dio:
4881 ext4_inode_resume_unlocked_dio(inode);
4882out_mutex:
4883 mutex_unlock(&inode->i_mutex);
4884 return ret;
4885}
4886
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004887/*
4888 * preallocate space for a file. This implements ext4's fallocate file
4889 * operation, which gets called from sys_fallocate system call.
4890 * For block-mapped files, posix_fallocate should fall back to the method
4891 * of writing zeroes to the required new blocks (the same behavior which is
4892 * expected for file systems which do not support fallocate() system call).
4893 */
4894long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4895{
4896 struct inode *inode = file_inode(file);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004897 loff_t new_size = 0;
4898 unsigned int max_blocks;
4899 int ret = 0;
4900 int flags;
4901 ext4_lblk_t lblk;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004902 unsigned int blkbits = inode->i_blkbits;
4903
Michael Halcrow2058f832015-04-12 00:55:10 -04004904 /*
4905 * Encrypted inodes can't handle collapse range or insert
4906 * range since we would need to re-encrypt blocks with a
4907 * different IV or XTS tweak (which are based on the logical
4908 * block number).
4909 *
4910 * XXX It's not clear why zero range isn't working, but we'll
4911 * leave it disabled for encrypted inodes for now. This is a
4912 * bug we should fix....
4913 */
4914 if (ext4_encrypted_inode(inode) &&
4915 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)))
4916 return -EOPNOTSUPP;
4917
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004918 /* Return error if mode is not supported */
4919 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
Lukas Czernerb8a86842014-03-18 18:05:35 -04004920 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004921 return -EOPNOTSUPP;
4922
4923 if (mode & FALLOC_FL_PUNCH_HOLE)
4924 return ext4_punch_hole(inode, offset, len);
4925
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004926 ret = ext4_convert_inline_data(inode);
4927 if (ret)
4928 return ret;
4929
Theodore Ts'o40c406c2014-04-12 22:53:53 -04004930 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4931 return ext4_collapse_range(inode, offset, len);
4932
Lukas Czernerb8a86842014-03-18 18:05:35 -04004933 if (mode & FALLOC_FL_ZERO_RANGE)
4934 return ext4_zero_range(file, offset, len, mode);
4935
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004936 trace_ext4_fallocate_enter(inode, offset, len, mode);
4937 lblk = offset >> blkbits;
4938 /*
4939 * We can't just convert len to max_blocks because
4940 * If blocksize = 4096 offset = 3072 and len = 2048
4941 */
4942 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4943 - lblk;
4944
Lukas Czerner556615d2014-04-20 23:45:47 -04004945 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004946 if (mode & FALLOC_FL_KEEP_SIZE)
4947 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4948
4949 mutex_lock(&inode->i_mutex);
4950
Davide Italiano280227a2015-05-02 23:21:15 -04004951 /*
4952 * We only support preallocation for extent-based files only
4953 */
4954 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4955 ret = -EOPNOTSUPP;
4956 goto out;
4957 }
4958
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004959 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4960 offset + len > i_size_read(inode)) {
4961 new_size = offset + len;
4962 ret = inode_newsize_ok(inode, new_size);
4963 if (ret)
4964 goto out;
4965 }
4966
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004967 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4968 flags, mode);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004969 if (ret)
4970 goto out;
4971
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004972 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4973 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4974 EXT4_I(inode)->i_sync_tid);
Lukas Czernerf282ac12014-03-18 17:44:35 -04004975 }
Lukas Czernerf282ac12014-03-18 17:44:35 -04004976out:
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004977 mutex_unlock(&inode->i_mutex);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004978 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4979 return ret;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004980}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004981
4982/*
Mingming Cao00314622009-09-28 15:49:08 -04004983 * This function convert a range of blocks to written extents
4984 * The caller of this function will pass the start offset and the size.
4985 * all unwritten extents within this range will be converted to
4986 * written extents.
4987 *
4988 * This function is called from the direct IO end io call back
4989 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004990 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004991 */
Jan Kara6b523df2013-06-04 13:21:11 -04004992int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4993 loff_t offset, ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004994{
Mingming Cao00314622009-09-28 15:49:08 -04004995 unsigned int max_blocks;
4996 int ret = 0;
4997 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004998 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004999 unsigned int credits, blkbits = inode->i_blkbits;
5000
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005001 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04005002 /*
5003 * We can't just convert len to max_blocks because
5004 * If blocksize = 4096 offset = 3072 and len = 2048
5005 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005006 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5007 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04005008 /*
Jan Kara6b523df2013-06-04 13:21:11 -04005009 * This is somewhat ugly but the idea is clear: When transaction is
5010 * reserved, everything goes into it. Otherwise we rather start several
5011 * smaller transactions for conversion of each extent separately.
Mingming Cao00314622009-09-28 15:49:08 -04005012 */
Jan Kara6b523df2013-06-04 13:21:11 -04005013 if (handle) {
5014 handle = ext4_journal_start_reserved(handle,
5015 EXT4_HT_EXT_CONVERT);
5016 if (IS_ERR(handle))
5017 return PTR_ERR(handle);
5018 credits = 0;
5019 } else {
5020 /*
5021 * credits to insert 1 extent into extent tree
5022 */
5023 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5024 }
Mingming Cao00314622009-09-28 15:49:08 -04005025 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005026 map.m_lblk += ret;
5027 map.m_len = (max_blocks -= ret);
Jan Kara6b523df2013-06-04 13:21:11 -04005028 if (credits) {
5029 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5030 credits);
5031 if (IS_ERR(handle)) {
5032 ret = PTR_ERR(handle);
5033 break;
5034 }
Mingming Cao00314622009-09-28 15:49:08 -04005035 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005036 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05005037 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05005038 if (ret <= 0)
5039 ext4_warning(inode->i_sb,
5040 "inode #%lu: block %u: len %u: "
5041 "ext4_ext_map_blocks returned %d",
5042 inode->i_ino, map.m_lblk,
5043 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04005044 ext4_mark_inode_dirty(handle, inode);
Jan Kara6b523df2013-06-04 13:21:11 -04005045 if (credits)
5046 ret2 = ext4_journal_stop(handle);
5047 if (ret <= 0 || ret2)
Mingming Cao00314622009-09-28 15:49:08 -04005048 break;
5049 }
Jan Kara6b523df2013-06-04 13:21:11 -04005050 if (!credits)
5051 ret2 = ext4_journal_stop(handle);
Mingming Cao00314622009-09-28 15:49:08 -04005052 return ret > 0 ? ret2 : ret;
5053}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005054
Mingming Cao00314622009-09-28 15:49:08 -04005055/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05005056 * If newes is not existing extent (newes->ec_pblk equals zero) find
5057 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005058 * return start of the next delayed extent.
5059 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05005060 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005061 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05005062 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04005063 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005064static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05005065 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04005066{
Zheng Liub3aff3e2012-11-08 21:57:37 -05005067 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05005068 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005069
Zheng Liu69eb33d2013-02-18 00:31:07 -05005070 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04005071 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5072 newes->es_lblk + newes->es_len - 1, &es);
5073
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005074 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05005075 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005076 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005077 */
Zheng Liu06b0c882013-02-18 00:26:51 -05005078 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05005079 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005080 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005081
Zheng Liu69eb33d2013-02-18 00:31:07 -05005082 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05005083 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05005084 newes->es_len = min(es.es_lblk - newes->es_lblk,
5085 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005086 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005087 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005088
Zheng Liu69eb33d2013-02-18 00:31:07 -05005089 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005090 }
5091
Zheng Liu69eb33d2013-02-18 00:31:07 -05005092 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04005093 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05005094 if (es.es_len == 0)
5095 next_del = EXT_MAX_BLOCKS;
5096 else
5097 next_del = es.es_lblk;
5098
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005099 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005100}
Eric Sandeen6873fa02008-10-07 00:46:36 -04005101/* fiemap flags we can handle specified here */
5102#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5103
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05005104static int ext4_xattr_fiemap(struct inode *inode,
5105 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04005106{
5107 __u64 physical = 0;
5108 __u64 length;
5109 __u32 flags = FIEMAP_EXTENT_LAST;
5110 int blockbits = inode->i_sb->s_blocksize_bits;
5111 int error = 0;
5112
5113 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005114 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04005115 struct ext4_iloc iloc;
5116 int offset; /* offset of xattr in inode */
5117
5118 error = ext4_get_inode_loc(inode, &iloc);
5119 if (error)
5120 return error;
Jan Karaa60697f2013-05-31 19:38:56 -04005121 physical = (__u64)iloc.bh->b_blocknr << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005122 offset = EXT4_GOOD_OLD_INODE_SIZE +
5123 EXT4_I(inode)->i_extra_isize;
5124 physical += offset;
5125 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5126 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04005127 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005128 } else { /* external block */
Jan Karaa60697f2013-05-31 19:38:56 -04005129 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005130 length = inode->i_sb->s_blocksize;
5131 }
5132
5133 if (physical)
5134 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5135 length, flags);
5136 return (error < 0 ? error : 0);
5137}
5138
5139int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5140 __u64 start, __u64 len)
5141{
5142 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005143 int error = 0;
5144
Tao Ma94191982012-12-10 14:06:02 -05005145 if (ext4_has_inline_data(inode)) {
5146 int has_inline = 1;
5147
Dmitry Monakhovd952d692014-12-02 16:11:20 -05005148 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5149 start, len);
Tao Ma94191982012-12-10 14:06:02 -05005150
5151 if (has_inline)
5152 return error;
5153 }
5154
Theodore Ts'o7869a4a2013-08-16 22:05:14 -04005155 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5156 error = ext4_ext_precache(inode);
5157 if (error)
5158 return error;
5159 }
5160
Eric Sandeen6873fa02008-10-07 00:46:36 -04005161 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005162 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Theodore Ts'oad7fefb2015-01-02 15:16:00 -05005163 return generic_block_fiemap(inode, fieinfo, start, len,
5164 ext4_get_block);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005165
5166 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5167 return -EBADR;
5168
5169 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5170 error = ext4_xattr_fiemap(inode, fieinfo);
5171 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005172 ext4_lblk_t len_blks;
5173 __u64 last_blk;
5174
Eric Sandeen6873fa02008-10-07 00:46:36 -04005175 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005176 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04005177 if (last_blk >= EXT_MAX_BLOCKS)
5178 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005179 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005180
5181 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005182 * Walk the extent tree gathering extent information
5183 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04005184 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005185 error = ext4_fill_fiemap_extents(inode, start_blk,
5186 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005187 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04005188 return error;
5189}
Namjae Jeon9eb79482014-02-23 15:18:59 -05005190
5191/*
5192 * ext4_access_path:
5193 * Function to access the path buffer for marking it dirty.
5194 * It also checks if there are sufficient credits left in the journal handle
5195 * to update path.
5196 */
5197static int
5198ext4_access_path(handle_t *handle, struct inode *inode,
5199 struct ext4_ext_path *path)
5200{
5201 int credits, err;
5202
5203 if (!ext4_handle_valid(handle))
5204 return 0;
5205
5206 /*
5207 * Check if need to extend journal credits
5208 * 3 for leaf, sb, and inode plus 2 (bmap and group
5209 * descriptor) for each block group; assume two block
5210 * groups
5211 */
5212 if (handle->h_buffer_credits < 7) {
5213 credits = ext4_writepage_trans_blocks(inode);
5214 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5215 /* EAGAIN is success */
5216 if (err && err != -EAGAIN)
5217 return err;
5218 }
5219
5220 err = ext4_ext_get_access(handle, inode, path);
5221 return err;
5222}
5223
5224/*
5225 * ext4_ext_shift_path_extents:
5226 * Shift the extents of a path structure lying between path[depth].p_ext
5227 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5228 * from starting block for each extent.
5229 */
5230static int
5231ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5232 struct inode *inode, handle_t *handle,
5233 ext4_lblk_t *start)
5234{
5235 int depth, err = 0;
5236 struct ext4_extent *ex_start, *ex_last;
5237 bool update = 0;
5238 depth = path->p_depth;
5239
5240 while (depth >= 0) {
5241 if (depth == path->p_depth) {
5242 ex_start = path[depth].p_ext;
5243 if (!ex_start)
5244 return -EIO;
5245
5246 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005247
5248 err = ext4_access_path(handle, inode, path + depth);
5249 if (err)
5250 goto out;
5251
5252 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5253 update = 1;
5254
Zheng Liu847c6c42014-04-12 12:45:55 -04005255 *start = le32_to_cpu(ex_last->ee_block) +
Namjae Jeon9eb79482014-02-23 15:18:59 -05005256 ext4_ext_get_actual_len(ex_last);
5257
5258 while (ex_start <= ex_last) {
Zheng Liu847c6c42014-04-12 12:45:55 -04005259 le32_add_cpu(&ex_start->ee_block, -shift);
Lukas Czerner6dd834e2014-04-18 10:55:24 -04005260 /* Try to merge to the left. */
5261 if ((ex_start >
5262 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5263 ext4_ext_try_to_merge_right(inode,
5264 path, ex_start - 1))
5265 ex_last--;
5266 else
5267 ex_start++;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005268 }
5269 err = ext4_ext_dirty(handle, inode, path + depth);
5270 if (err)
5271 goto out;
5272
5273 if (--depth < 0 || !update)
5274 break;
5275 }
5276
5277 /* Update index too */
5278 err = ext4_access_path(handle, inode, path + depth);
5279 if (err)
5280 goto out;
5281
Zheng Liu847c6c42014-04-12 12:45:55 -04005282 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005283 err = ext4_ext_dirty(handle, inode, path + depth);
5284 if (err)
5285 goto out;
5286
5287 /* we are done if current index is not a starting index */
5288 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5289 break;
5290
5291 depth--;
5292 }
5293
5294out:
5295 return err;
5296}
5297
5298/*
5299 * ext4_ext_shift_extents:
5300 * All the extents which lies in the range from start to the last allocated
5301 * block for the file are shifted downwards by shift blocks.
5302 * On success, 0 is returned, error otherwise.
5303 */
5304static int
5305ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5306 ext4_lblk_t start, ext4_lblk_t shift)
5307{
5308 struct ext4_ext_path *path;
5309 int ret = 0, depth;
5310 struct ext4_extent *extent;
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005311 ext4_lblk_t stop_block;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005312 ext4_lblk_t ex_start, ex_end;
5313
5314 /* Let path point to the last extent */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005315 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005316 if (IS_ERR(path))
5317 return PTR_ERR(path);
5318
5319 depth = path->p_depth;
5320 extent = path[depth].p_ext;
Theodore Ts'oee4bd0d92014-09-01 14:41:09 -04005321 if (!extent)
5322 goto out;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005323
Zheng Liu847c6c42014-04-12 12:45:55 -04005324 stop_block = le32_to_cpu(extent->ee_block) +
5325 ext4_ext_get_actual_len(extent);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005326
5327 /* Nothing to shift, if hole is at the end of file */
5328 if (start >= stop_block)
Theodore Ts'oee4bd0d92014-09-01 14:41:09 -04005329 goto out;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005330
5331 /*
5332 * Don't start shifting extents until we make sure the hole is big
5333 * enough to accomodate the shift.
5334 */
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005335 path = ext4_find_extent(inode, start - 1, &path, 0);
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005336 if (IS_ERR(path))
5337 return PTR_ERR(path);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005338 depth = path->p_depth;
5339 extent = path[depth].p_ext;
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005340 if (extent) {
5341 ex_start = le32_to_cpu(extent->ee_block);
5342 ex_end = le32_to_cpu(extent->ee_block) +
Zheng Liu847c6c42014-04-12 12:45:55 -04005343 ext4_ext_get_actual_len(extent);
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005344 } else {
5345 ex_start = 0;
5346 ex_end = 0;
5347 }
Namjae Jeon9eb79482014-02-23 15:18:59 -05005348
5349 if ((start == ex_start && shift > ex_start) ||
5350 (shift > start - ex_end))
5351 return -EINVAL;
5352
5353 /* Its safe to start updating extents */
5354 while (start < stop_block) {
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005355 path = ext4_find_extent(inode, start, &path, 0);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005356 if (IS_ERR(path))
5357 return PTR_ERR(path);
5358 depth = path->p_depth;
5359 extent = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04005360 if (!extent) {
5361 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5362 (unsigned long) start);
5363 return -EIO;
5364 }
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005365 if (start > le32_to_cpu(extent->ee_block)) {
Namjae Jeon9eb79482014-02-23 15:18:59 -05005366 /* Hole, move to the next extent */
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005367 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5368 path[depth].p_ext++;
5369 } else {
5370 start = ext4_ext_next_allocated_block(path);
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005371 continue;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005372 }
5373 }
5374 ret = ext4_ext_shift_path_extents(path, shift, inode,
5375 handle, &start);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005376 if (ret)
5377 break;
5378 }
Theodore Ts'oee4bd0d92014-09-01 14:41:09 -04005379out:
5380 ext4_ext_drop_refs(path);
5381 kfree(path);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005382 return ret;
5383}
5384
5385/*
5386 * ext4_collapse_range:
5387 * This implements the fallocate's collapse range functionality for ext4
5388 * Returns: 0 and non-zero on error.
5389 */
5390int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5391{
5392 struct super_block *sb = inode->i_sb;
5393 ext4_lblk_t punch_start, punch_stop;
5394 handle_t *handle;
5395 unsigned int credits;
Namjae Jeona8680e02014-04-19 16:37:31 -04005396 loff_t new_size, ioffset;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005397 int ret;
5398
Namjae Jeon9eb79482014-02-23 15:18:59 -05005399 /* Collapse range works only on fs block size aligned offsets. */
Namjae Jeonee98fa32014-07-29 10:45:31 -04005400 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5401 len & (EXT4_CLUSTER_SIZE(sb) - 1))
Namjae Jeon9eb79482014-02-23 15:18:59 -05005402 return -EINVAL;
5403
5404 if (!S_ISREG(inode->i_mode))
Theodore Ts'o86f1ca32014-04-18 11:52:11 -04005405 return -EINVAL;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005406
5407 trace_ext4_collapse_range(inode, offset, len);
5408
5409 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5410 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5411
Namjae Jeon1ce01c42014-04-10 22:58:20 -04005412 /* Call ext4_force_commit to flush all data in case of data=journal. */
5413 if (ext4_should_journal_data(inode)) {
5414 ret = ext4_force_commit(inode->i_sb);
5415 if (ret)
5416 return ret;
5417 }
5418
Namjae Jeona8680e02014-04-19 16:37:31 -04005419 /*
5420 * Need to round down offset to be aligned with page size boundary
5421 * for page size > block size.
5422 */
5423 ioffset = round_down(offset, PAGE_SIZE);
5424
Namjae Jeon9eb79482014-02-23 15:18:59 -05005425 /* Write out all dirty pages */
Namjae Jeona8680e02014-04-19 16:37:31 -04005426 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5427 LLONG_MAX);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005428 if (ret)
5429 return ret;
5430
5431 /* Take mutex lock */
5432 mutex_lock(&inode->i_mutex);
5433
Lukas Czerner23fffa92014-04-12 09:56:41 -04005434 /*
5435 * There is no need to overlap collapse range with EOF, in which case
5436 * it is effectively a truncate operation
5437 */
5438 if (offset + len >= i_size_read(inode)) {
5439 ret = -EINVAL;
5440 goto out_mutex;
5441 }
5442
Namjae Jeon9eb79482014-02-23 15:18:59 -05005443 /* Currently just for extent based files */
5444 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5445 ret = -EOPNOTSUPP;
5446 goto out_mutex;
5447 }
5448
Namjae Jeona8680e02014-04-19 16:37:31 -04005449 truncate_pagecache(inode, ioffset);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005450
5451 /* Wait for existing dio to complete */
5452 ext4_inode_block_unlocked_dio(inode);
5453 inode_dio_wait(inode);
5454
5455 credits = ext4_writepage_trans_blocks(inode);
5456 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5457 if (IS_ERR(handle)) {
5458 ret = PTR_ERR(handle);
5459 goto out_dio;
5460 }
5461
5462 down_write(&EXT4_I(inode)->i_data_sem);
5463 ext4_discard_preallocations(inode);
5464
5465 ret = ext4_es_remove_extent(inode, punch_start,
Lukas Czerner2c1d2322014-04-18 10:43:21 -04005466 EXT_MAX_BLOCKS - punch_start);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005467 if (ret) {
5468 up_write(&EXT4_I(inode)->i_data_sem);
5469 goto out_stop;
5470 }
5471
5472 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5473 if (ret) {
5474 up_write(&EXT4_I(inode)->i_data_sem);
5475 goto out_stop;
5476 }
Lukas Czerneref24f6c2014-04-18 10:50:23 -04005477 ext4_discard_preallocations(inode);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005478
5479 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5480 punch_stop - punch_start);
5481 if (ret) {
5482 up_write(&EXT4_I(inode)->i_data_sem);
5483 goto out_stop;
5484 }
5485
5486 new_size = i_size_read(inode) - len;
Lukas Czerner9337d5d2014-04-18 10:48:25 -04005487 i_size_write(inode, new_size);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005488 EXT4_I(inode)->i_disksize = new_size;
5489
Namjae Jeon9eb79482014-02-23 15:18:59 -05005490 up_write(&EXT4_I(inode)->i_data_sem);
5491 if (IS_SYNC(inode))
5492 ext4_handle_sync(handle);
5493 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5494 ext4_mark_inode_dirty(handle, inode);
5495
5496out_stop:
5497 ext4_journal_stop(handle);
5498out_dio:
5499 ext4_inode_resume_unlocked_dio(inode);
5500out_mutex:
5501 mutex_unlock(&inode->i_mutex);
5502 return ret;
5503}
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005504
5505/**
5506 * ext4_swap_extents - Swap extents between two inodes
5507 *
5508 * @inode1: First inode
5509 * @inode2: Second inode
5510 * @lblk1: Start block for first inode
5511 * @lblk2: Start block for second inode
5512 * @count: Number of blocks to swap
5513 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5514 * @erp: Pointer to save error value
5515 *
5516 * This helper routine does exactly what is promise "swap extents". All other
5517 * stuff such as page-cache locking consistency, bh mapping consistency or
5518 * extent's data copying must be performed by caller.
5519 * Locking:
5520 * i_mutex is held for both inodes
5521 * i_data_sem is locked for write for both inodes
5522 * Assumptions:
5523 * All pages from requested range are locked for both inodes
5524 */
5525int
5526ext4_swap_extents(handle_t *handle, struct inode *inode1,
5527 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5528 ext4_lblk_t count, int unwritten, int *erp)
5529{
5530 struct ext4_ext_path *path1 = NULL;
5531 struct ext4_ext_path *path2 = NULL;
5532 int replaced_count = 0;
5533
5534 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5535 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5536 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5537 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5538
5539 *erp = ext4_es_remove_extent(inode1, lblk1, count);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005540 if (unlikely(*erp))
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005541 return 0;
5542 *erp = ext4_es_remove_extent(inode2, lblk2, count);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005543 if (unlikely(*erp))
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005544 return 0;
5545
5546 while (count) {
5547 struct ext4_extent *ex1, *ex2, tmp_ex;
5548 ext4_lblk_t e1_blk, e2_blk;
5549 int e1_len, e2_len, len;
5550 int split = 0;
5551
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005552 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005553 if (unlikely(IS_ERR(path1))) {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005554 *erp = PTR_ERR(path1);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005555 path1 = NULL;
5556 finish:
5557 count = 0;
5558 goto repeat;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005559 }
Theodore Ts'oed8a1a72014-09-01 14:43:09 -04005560 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005561 if (unlikely(IS_ERR(path2))) {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005562 *erp = PTR_ERR(path2);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005563 path2 = NULL;
5564 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005565 }
5566 ex1 = path1[path1->p_depth].p_ext;
5567 ex2 = path2[path2->p_depth].p_ext;
5568 /* Do we have somthing to swap ? */
5569 if (unlikely(!ex2 || !ex1))
Theodore Ts'o19008f62014-08-31 15:03:14 -04005570 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005571
5572 e1_blk = le32_to_cpu(ex1->ee_block);
5573 e2_blk = le32_to_cpu(ex2->ee_block);
5574 e1_len = ext4_ext_get_actual_len(ex1);
5575 e2_len = ext4_ext_get_actual_len(ex2);
5576
5577 /* Hole handling */
5578 if (!in_range(lblk1, e1_blk, e1_len) ||
5579 !in_range(lblk2, e2_blk, e2_len)) {
5580 ext4_lblk_t next1, next2;
5581
5582 /* if hole after extent, then go to next extent */
5583 next1 = ext4_ext_next_allocated_block(path1);
5584 next2 = ext4_ext_next_allocated_block(path2);
5585 /* If hole before extent, then shift to that extent */
5586 if (e1_blk > lblk1)
5587 next1 = e1_blk;
5588 if (e2_blk > lblk2)
5589 next2 = e1_blk;
5590 /* Do we have something to swap */
5591 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
Theodore Ts'o19008f62014-08-31 15:03:14 -04005592 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005593 /* Move to the rightest boundary */
5594 len = next1 - lblk1;
5595 if (len < next2 - lblk2)
5596 len = next2 - lblk2;
5597 if (len > count)
5598 len = count;
5599 lblk1 += len;
5600 lblk2 += len;
5601 count -= len;
5602 goto repeat;
5603 }
5604
5605 /* Prepare left boundary */
5606 if (e1_blk < lblk1) {
5607 split = 1;
5608 *erp = ext4_force_split_extent_at(handle, inode1,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005609 &path1, lblk1, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005610 if (unlikely(*erp))
5611 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005612 }
5613 if (e2_blk < lblk2) {
5614 split = 1;
5615 *erp = ext4_force_split_extent_at(handle, inode2,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005616 &path2, lblk2, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005617 if (unlikely(*erp))
5618 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005619 }
Theodore Ts'odfe50802014-09-01 14:37:09 -04005620 /* ext4_split_extent_at() may result in leaf extent split,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005621 * path must to be revalidated. */
5622 if (split)
5623 goto repeat;
5624
5625 /* Prepare right boundary */
5626 len = count;
5627 if (len > e1_blk + e1_len - lblk1)
5628 len = e1_blk + e1_len - lblk1;
5629 if (len > e2_blk + e2_len - lblk2)
5630 len = e2_blk + e2_len - lblk2;
5631
5632 if (len != e1_len) {
5633 split = 1;
5634 *erp = ext4_force_split_extent_at(handle, inode1,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005635 &path1, lblk1 + len, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005636 if (unlikely(*erp))
5637 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005638 }
5639 if (len != e2_len) {
5640 split = 1;
5641 *erp = ext4_force_split_extent_at(handle, inode2,
Theodore Ts'odfe50802014-09-01 14:37:09 -04005642 &path2, lblk2 + len, 0);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005643 if (*erp)
Theodore Ts'o19008f62014-08-31 15:03:14 -04005644 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005645 }
Theodore Ts'odfe50802014-09-01 14:37:09 -04005646 /* ext4_split_extent_at() may result in leaf extent split,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005647 * path must to be revalidated. */
5648 if (split)
5649 goto repeat;
5650
5651 BUG_ON(e2_len != e1_len);
5652 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->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 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005656 if (unlikely(*erp))
5657 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005658
5659 /* Both extents are fully inside boundaries. Swap it now */
5660 tmp_ex = *ex1;
5661 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5662 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5663 ex1->ee_len = cpu_to_le16(e2_len);
5664 ex2->ee_len = cpu_to_le16(e1_len);
5665 if (unwritten)
5666 ext4_ext_mark_unwritten(ex2);
5667 if (ext4_ext_is_unwritten(&tmp_ex))
5668 ext4_ext_mark_unwritten(ex1);
5669
5670 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5671 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5672 *erp = ext4_ext_dirty(handle, inode2, path2 +
5673 path2->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005674 if (unlikely(*erp))
5675 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005676 *erp = ext4_ext_dirty(handle, inode1, path1 +
5677 path1->p_depth);
5678 /*
5679 * Looks scarry ah..? second inode already points to new blocks,
5680 * and it was successfully dirtied. But luckily error may happen
5681 * only due to journal error, so full transaction will be
5682 * aborted anyway.
5683 */
Theodore Ts'o19008f62014-08-31 15:03:14 -04005684 if (unlikely(*erp))
5685 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005686 lblk1 += len;
5687 lblk2 += len;
5688 replaced_count += len;
5689 count -= len;
5690
5691 repeat:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -04005692 ext4_ext_drop_refs(path1);
5693 kfree(path1);
5694 ext4_ext_drop_refs(path2);
5695 kfree(path2);
5696 path1 = path2 = NULL;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005697 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005698 return replaced_count;
5699}