blob: a920c5d29fac0a5b5ef83c9cca18feeb3dcc0fe7 [file] [log] [blame]
Akira Fujita748de672009-06-17 19:24:03 -04001/*
2 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
3 * Written by Takashi Sato <t-sato@yk.jp.nec.com>
4 * Akira Fujita <a-fujita@rs.jp.nec.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2.1 of the GNU Lesser General Public License
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 */
15
16#include <linux/fs.h>
17#include <linux/quotaops.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/slab.h>
Akira Fujita748de672009-06-17 19:24:03 -040019#include "ext4_jbd2.h"
Akira Fujita748de672009-06-17 19:24:03 -040020#include "ext4.h"
Theodore Ts'o4a092d72012-11-28 13:03:30 -050021#include "ext4_extents.h"
Akira Fujita748de672009-06-17 19:24:03 -040022
Akira Fujitae8505972009-09-16 13:46:38 -040023/**
24 * get_ext_path - Find an extent path for designated logical block number.
25 *
26 * @inode: an inode which is searched
27 * @lblock: logical block number to find an extent path
28 * @path: pointer to an extent path pointer (for output)
29 *
Theodore Ts'oed8a1a72014-09-01 14:43:09 -040030 * ext4_find_extent wrapper. Return 0 on success, or a negative error value
Akira Fujitae8505972009-09-16 13:46:38 -040031 * on failure.
32 */
33static inline int
34get_ext_path(struct inode *inode, ext4_lblk_t lblock,
Theodore Ts'o3bdf14b2014-09-01 14:42:09 -040035 struct ext4_ext_path **ppath)
Akira Fujitae8505972009-09-16 13:46:38 -040036{
Dmitry Monakhov0e401102013-03-18 11:40:19 -040037 struct ext4_ext_path *path;
Akira Fujitae8505972009-09-16 13:46:38 -040038
Theodore Ts'oed8a1a72014-09-01 14:43:09 -040039 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
Dmitry Monakhov0e401102013-03-18 11:40:19 -040040 if (IS_ERR(path))
Theodore Ts'o3bdf14b2014-09-01 14:42:09 -040041 return PTR_ERR(path);
42 if (path[ext_depth(inode)].p_ext == NULL) {
43 ext4_ext_drop_refs(path);
44 kfree(path);
45 *ppath = NULL;
46 return -ENODATA;
47 }
48 *ppath = path;
49 return 0;
Akira Fujitae8505972009-09-16 13:46:38 -040050}
Akira Fujita748de672009-06-17 19:24:03 -040051
52/**
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -040053 * ext4_double_down_write_data_sem - Acquire two inodes' write lock
54 * of i_data_sem
Akira Fujita748de672009-06-17 19:24:03 -040055 *
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -040056 * Acquire write lock of i_data_sem of the two inodes
Akira Fujita748de672009-06-17 19:24:03 -040057 */
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -040058void
59ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
Akira Fujita748de672009-06-17 19:24:03 -040060{
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -040061 if (first < second) {
62 down_write(&EXT4_I(first)->i_data_sem);
Theodore Ts'odaf647d2016-04-01 01:31:28 -040063 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -040064 } else {
65 down_write(&EXT4_I(second)->i_data_sem);
Theodore Ts'odaf647d2016-04-01 01:31:28 -040066 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
Akira Fujita748de672009-06-17 19:24:03 -040067
Akira Fujita748de672009-06-17 19:24:03 -040068 }
Akira Fujita748de672009-06-17 19:24:03 -040069}
70
71/**
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -040072 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
Akira Fujita748de672009-06-17 19:24:03 -040073 *
74 * @orig_inode: original inode structure to be released its lock first
75 * @donor_inode: donor inode structure to be released its lock second
Akira Fujitafc04cb42009-11-23 07:24:43 -050076 * Release write lock of i_data_sem of two inodes (orig and donor).
Akira Fujita748de672009-06-17 19:24:03 -040077 */
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -040078void
79ext4_double_up_write_data_sem(struct inode *orig_inode,
80 struct inode *donor_inode)
Akira Fujita748de672009-06-17 19:24:03 -040081{
Akira Fujita748de672009-06-17 19:24:03 -040082 up_write(&EXT4_I(orig_inode)->i_data_sem);
83 up_write(&EXT4_I(donor_inode)->i_data_sem);
84}
85
86/**
Dmitry Monakhov8c854472012-09-26 12:54:52 -040087 * mext_check_coverage - Check that all extents in range has the same type
88 *
89 * @inode: inode in question
90 * @from: block offset of inode
91 * @count: block count to be checked
Lukas Czerner556615d2014-04-20 23:45:47 -040092 * @unwritten: extents expected to be unwritten
Dmitry Monakhov8c854472012-09-26 12:54:52 -040093 * @err: pointer to save error value
94 *
95 * Return 1 if all extents in range has expected type, and zero otherwise.
96 */
97static int
98mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
Lukas Czerner556615d2014-04-20 23:45:47 -040099 int unwritten, int *err)
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400100{
101 struct ext4_ext_path *path = NULL;
102 struct ext4_extent *ext;
Dmitry Monakhov0e401102013-03-18 11:40:19 -0400103 int ret = 0;
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400104 ext4_lblk_t last = from + count;
105 while (from < last) {
106 *err = get_ext_path(inode, from, &path);
107 if (*err)
Dmitry Monakhov0e401102013-03-18 11:40:19 -0400108 goto out;
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400109 ext = path[ext_depth(inode)].p_ext;
Lukas Czerner556615d2014-04-20 23:45:47 -0400110 if (unwritten != ext4_ext_is_unwritten(ext))
Dmitry Monakhov0e401102013-03-18 11:40:19 -0400111 goto out;
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400112 from += ext4_ext_get_actual_len(ext);
113 ext4_ext_drop_refs(path);
114 }
Dmitry Monakhov0e401102013-03-18 11:40:19 -0400115 ret = 1;
116out:
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -0400117 ext4_ext_drop_refs(path);
118 kfree(path);
Dmitry Monakhov0e401102013-03-18 11:40:19 -0400119 return ret;
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400120}
121
122/**
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400123 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
124 *
125 * @inode1: the inode structure
126 * @inode2: the inode structure
Xiaoguang Wang65dd8322014-10-11 19:56:34 -0400127 * @index1: page index
128 * @index2: page index
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400129 * @page: result page vector
130 *
131 * Grab two locked pages for inode's by inode order
132 */
133static int
134mext_page_double_lock(struct inode *inode1, struct inode *inode2,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400135 pgoff_t index1, pgoff_t index2, struct page *page[2])
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400136{
137 struct address_space *mapping[2];
138 unsigned fl = AOP_FLAG_NOFS;
139
140 BUG_ON(!inode1 || !inode2);
141 if (inode1 < inode2) {
142 mapping[0] = inode1->i_mapping;
143 mapping[1] = inode2->i_mapping;
144 } else {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400145 pgoff_t tmp = index1;
146 index1 = index2;
147 index2 = tmp;
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400148 mapping[0] = inode2->i_mapping;
149 mapping[1] = inode1->i_mapping;
150 }
151
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400152 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400153 if (!page[0])
154 return -ENOMEM;
155
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400156 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400157 if (!page[1]) {
158 unlock_page(page[0]);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300159 put_page(page[0]);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400160 return -ENOMEM;
161 }
Dmitry Monakhov7e8b12c2013-04-11 23:24:58 -0400162 /*
163 * grab_cache_page_write_begin() may not wait on page's writeback if
164 * BDI not demand that. But it is reasonable to be very conservative
165 * here and explicitly wait on page's writeback
166 */
167 wait_on_page_writeback(page[0]);
168 wait_on_page_writeback(page[1]);
Fabian Frederickbf865462015-06-12 23:47:33 -0400169 if (inode1 > inode2)
170 swap(page[0], page[1]);
171
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400172 return 0;
173}
174
175/* Force page buffers uptodate w/o dropping page's lock */
176static int
177mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
178{
179 struct inode *inode = page->mapping->host;
180 sector_t block;
181 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
182 unsigned int blocksize, block_start, block_end;
183 int i, err, nr = 0, partial = 0;
184 BUG_ON(!PageLocked(page));
185 BUG_ON(PageWriteback(page));
186
187 if (PageUptodate(page))
188 return 0;
189
190 blocksize = 1 << inode->i_blkbits;
191 if (!page_has_buffers(page))
192 create_empty_buffers(page, blocksize, 0);
193
194 head = page_buffers(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300195 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400196 for (bh = head, block_start = 0; bh != head || !block_start;
197 block++, block_start = block_end, bh = bh->b_this_page) {
198 block_end = block_start + blocksize;
199 if (block_end <= from || block_start >= to) {
200 if (!buffer_uptodate(bh))
201 partial = 1;
202 continue;
203 }
204 if (buffer_uptodate(bh))
205 continue;
206 if (!buffer_mapped(bh)) {
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400207 err = ext4_get_block(inode, block, bh, 0);
208 if (err) {
209 SetPageError(page);
210 return err;
211 }
212 if (!buffer_mapped(bh)) {
213 zero_user(page, block_start, blocksize);
Dan Carpenterdf3a98b2014-02-17 20:46:40 -0500214 set_buffer_uptodate(bh);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400215 continue;
216 }
217 }
218 BUG_ON(nr >= MAX_BUF_PER_PAGE);
219 arr[nr++] = bh;
220 }
221 /* No io required */
222 if (!nr)
223 goto out;
224
225 for (i = 0; i < nr; i++) {
226 bh = arr[i];
227 if (!bh_uptodate_or_lock(bh)) {
228 err = bh_submit_read(bh);
229 if (err)
230 return err;
231 }
232 }
233out:
234 if (!partial)
235 SetPageUptodate(page);
236 return 0;
237}
238
239/**
Akira Fujita748de672009-06-17 19:24:03 -0400240 * move_extent_per_page - Move extent data per page
241 *
242 * @o_filp: file structure of original file
243 * @donor_inode: donor inode
244 * @orig_page_offset: page index on original file
Xiaoguang Wang65dd8322014-10-11 19:56:34 -0400245 * @donor_page_offset: page index on donor file
Akira Fujita748de672009-06-17 19:24:03 -0400246 * @data_offset_in_page: block index where data swapping starts
247 * @block_len_in_page: the number of blocks to be swapped
Lukas Czerner556615d2014-04-20 23:45:47 -0400248 * @unwritten: orig extent is unwritten or not
Akira Fujitaf868a482009-11-23 07:25:48 -0500249 * @err: pointer to save return value
Akira Fujita748de672009-06-17 19:24:03 -0400250 *
251 * Save the data in original inode blocks and replace original inode extents
Xiaoguang Wang65dd8322014-10-11 19:56:34 -0400252 * with donor inode extents by calling ext4_swap_extents().
Akira Fujitaf868a482009-11-23 07:25:48 -0500253 * Finally, write out the saved data in new original inode blocks. Return
254 * replaced block count.
Akira Fujita748de672009-06-17 19:24:03 -0400255 */
256static int
Akira Fujita44fc48f2009-09-05 23:12:41 -0400257move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400258 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
259 int data_offset_in_page,
260 int block_len_in_page, int unwritten, int *err)
Akira Fujita748de672009-06-17 19:24:03 -0400261{
Al Viro496ad9a2013-01-23 17:07:38 -0500262 struct inode *orig_inode = file_inode(o_filp);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400263 struct page *pagep[2] = {NULL, NULL};
Akira Fujita748de672009-06-17 19:24:03 -0400264 handle_t *handle;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400265 ext4_lblk_t orig_blk_offset, donor_blk_offset;
Akira Fujita748de672009-06-17 19:24:03 -0400266 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
Akira Fujitaf868a482009-11-23 07:25:48 -0500267 unsigned int tmp_data_size, data_size, replaced_size;
Eryu Guanbcff2482016-02-12 01:20:43 -0500268 int i, err2, jblocks, retries = 0;
Akira Fujitaf868a482009-11-23 07:25:48 -0500269 int replaced_count = 0;
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400270 int from = data_offset_in_page << orig_inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300271 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
Dmitry Monakhov88c6b612014-11-05 11:52:38 -0500272 struct super_block *sb = orig_inode->i_sb;
Eryu Guanbcff2482016-02-12 01:20:43 -0500273 struct buffer_head *bh = NULL;
Akira Fujita748de672009-06-17 19:24:03 -0400274
275 /*
276 * It needs twice the amount of ordinary journal buffers because
277 * inode and donor_inode may change each different metadata blocks.
278 */
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400279again:
280 *err = 0;
Akira Fujita748de672009-06-17 19:24:03 -0400281 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
Theodore Ts'o9924a922013-02-08 21:59:22 -0500282 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
Akira Fujita748de672009-06-17 19:24:03 -0400283 if (IS_ERR(handle)) {
Akira Fujitaf868a482009-11-23 07:25:48 -0500284 *err = PTR_ERR(handle);
285 return 0;
Akira Fujita748de672009-06-17 19:24:03 -0400286 }
287
Akira Fujita748de672009-06-17 19:24:03 -0400288 orig_blk_offset = orig_page_offset * blocks_per_page +
289 data_offset_in_page;
290
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400291 donor_blk_offset = donor_page_offset * blocks_per_page +
292 data_offset_in_page;
293
Akira Fujitaf868a482009-11-23 07:25:48 -0500294 /* Calculate data_size */
Akira Fujita748de672009-06-17 19:24:03 -0400295 if ((orig_blk_offset + block_len_in_page - 1) ==
296 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
297 /* Replace the last block */
Akira Fujitaf868a482009-11-23 07:25:48 -0500298 tmp_data_size = orig_inode->i_size & (blocksize - 1);
Akira Fujita748de672009-06-17 19:24:03 -0400299 /*
Akira Fujitaf868a482009-11-23 07:25:48 -0500300 * If data_size equal zero, it shows data_size is multiples of
Akira Fujita748de672009-06-17 19:24:03 -0400301 * blocksize. So we set appropriate value.
302 */
Akira Fujitaf868a482009-11-23 07:25:48 -0500303 if (tmp_data_size == 0)
304 tmp_data_size = blocksize;
Akira Fujita748de672009-06-17 19:24:03 -0400305
Akira Fujitaf868a482009-11-23 07:25:48 -0500306 data_size = tmp_data_size +
Akira Fujita748de672009-06-17 19:24:03 -0400307 ((block_len_in_page - 1) << orig_inode->i_blkbits);
Akira Fujitaf868a482009-11-23 07:25:48 -0500308 } else
309 data_size = block_len_in_page << orig_inode->i_blkbits;
Akira Fujita748de672009-06-17 19:24:03 -0400310
Akira Fujitaf868a482009-11-23 07:25:48 -0500311 replaced_size = data_size;
312
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400313 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400314 donor_page_offset, pagep);
Akira Fujitaf868a482009-11-23 07:25:48 -0500315 if (unlikely(*err < 0))
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400316 goto stop_journal;
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400317 /*
Lukas Czerner556615d2014-04-20 23:45:47 -0400318 * If orig extent was unwritten it can become initialized
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400319 * at any time after i_data_sem was dropped, in order to
320 * serialize with delalloc we have recheck extent while we
321 * hold page's lock, if it is still the case data copy is not
322 * necessary, just swap data blocks between orig and donor.
323 */
Lukas Czerner556615d2014-04-20 23:45:47 -0400324 if (unwritten) {
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400325 ext4_double_down_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400326 /* If any of extents in range became initialized we have to
327 * fallback to data copying */
Lukas Czerner556615d2014-04-20 23:45:47 -0400328 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
329 block_len_in_page, 1, err);
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400330 if (*err)
331 goto drop_data_sem;
Akira Fujita748de672009-06-17 19:24:03 -0400332
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400333 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
Lukas Czerner556615d2014-04-20 23:45:47 -0400334 block_len_in_page, 1, err);
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400335 if (*err)
336 goto drop_data_sem;
337
Lukas Czerner556615d2014-04-20 23:45:47 -0400338 if (!unwritten) {
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400339 ext4_double_up_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400340 goto data_copy;
341 }
342 if ((page_has_private(pagep[0]) &&
343 !try_to_release_page(pagep[0], 0)) ||
344 (page_has_private(pagep[1]) &&
345 !try_to_release_page(pagep[1], 0))) {
346 *err = -EBUSY;
347 goto drop_data_sem;
348 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400349 replaced_count = ext4_swap_extents(handle, orig_inode,
350 donor_inode, orig_blk_offset,
351 donor_blk_offset,
352 block_len_in_page, 1, err);
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400353 drop_data_sem:
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400354 ext4_double_up_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhov8c854472012-09-26 12:54:52 -0400355 goto unlock_pages;
356 }
357data_copy:
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400358 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
359 if (*err)
360 goto unlock_pages;
361
362 /* At this point all buffers in range are uptodate, old mapping layout
363 * is no longer required, try to drop it now. */
364 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
365 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
366 *err = -EBUSY;
367 goto unlock_pages;
Akira Fujita748de672009-06-17 19:24:03 -0400368 }
Dmitry Monakhov6e263142014-07-27 22:32:27 -0400369 ext4_double_down_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400370 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
371 orig_blk_offset, donor_blk_offset,
372 block_len_in_page, 1, err);
Dmitry Monakhov6e263142014-07-27 22:32:27 -0400373 ext4_double_up_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400374 if (*err) {
Akira Fujitaf868a482009-11-23 07:25:48 -0500375 if (replaced_count) {
376 block_len_in_page = replaced_count;
377 replaced_size =
378 block_len_in_page << orig_inode->i_blkbits;
Akira Fujitaac48b0a2009-11-24 10:31:56 -0500379 } else
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400380 goto unlock_pages;
Akira Fujitaf868a482009-11-23 07:25:48 -0500381 }
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400382 /* Perform all necessary steps similar write_begin()/write_end()
383 * but keeping in mind that i_size will not change */
Eryu Guanbcff2482016-02-12 01:20:43 -0500384 if (!page_has_buffers(pagep[0]))
385 create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
386 bh = page_buffers(pagep[0]);
387 for (i = 0; i < data_offset_in_page; i++)
388 bh = bh->b_this_page;
389 for (i = 0; i < block_len_in_page; i++) {
390 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
391 if (*err < 0)
392 break;
Eryu Guan6ffe77b2016-02-21 18:38:44 -0500393 bh = bh->b_this_page;
Eryu Guanbcff2482016-02-12 01:20:43 -0500394 }
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400395 if (!*err)
396 *err = block_commit_write(pagep[0], from, from + replaced_size);
Akira Fujita748de672009-06-17 19:24:03 -0400397
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400398 if (unlikely(*err < 0))
399 goto repair_branches;
Akira Fujita748de672009-06-17 19:24:03 -0400400
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400401 /* Even in case of data=writeback it is reasonable to pin
402 * inode to transaction, to prevent unexpected data loss */
Jan Karaee0876b2016-04-24 00:56:08 -0400403 *err = ext4_jbd2_inode_add_write(handle, orig_inode);
Akira Fujita748de672009-06-17 19:24:03 -0400404
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400405unlock_pages:
406 unlock_page(pagep[0]);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300407 put_page(pagep[0]);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400408 unlock_page(pagep[1]);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300409 put_page(pagep[1]);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400410stop_journal:
Akira Fujita748de672009-06-17 19:24:03 -0400411 ext4_journal_stop(handle);
Dmitry Monakhov88c6b612014-11-05 11:52:38 -0500412 if (*err == -ENOSPC &&
413 ext4_should_retry_alloc(sb, &retries))
414 goto again;
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400415 /* Buffer was busy because probably is pinned to journal transaction,
416 * force transaction commit may help to free it. */
Dmitry Monakhov88c6b612014-11-05 11:52:38 -0500417 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
418 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400419 goto again;
Akira Fujitaf868a482009-11-23 07:25:48 -0500420 return replaced_count;
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400421
422repair_branches:
423 /*
424 * This should never ever happen!
425 * Extents are swapped already, but we are not able to copy data.
426 * Try to swap extents to it's original places
427 */
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400428 ext4_double_down_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400429 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
430 orig_blk_offset, donor_blk_offset,
431 block_len_in_page, 0, &err2);
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400432 ext4_double_up_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhovbb557482012-09-26 12:52:07 -0400433 if (replaced_count != block_len_in_page) {
434 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
435 "Unable to copy data block,"
436 " data will be lost.");
437 *err = -EIO;
438 }
439 replaced_count = 0;
440 goto unlock_pages;
Akira Fujita748de672009-06-17 19:24:03 -0400441}
442
443/**
Akira Fujitac437b272010-03-04 00:39:24 -0500444 * mext_check_arguments - Check whether move extent can be done
Akira Fujita748de672009-06-17 19:24:03 -0400445 *
446 * @orig_inode: original inode
447 * @donor_inode: donor inode
448 * @orig_start: logical start offset in block for orig
449 * @donor_start: logical start offset in block for donor
450 * @len: the number of blocks to be moved
Akira Fujita748de672009-06-17 19:24:03 -0400451 *
452 * Check the arguments of ext4_move_extents() whether the files can be
453 * exchanged with each other.
454 * Return 0 on success, or a negative error value on failure.
455 */
456static int
457mext_check_arguments(struct inode *orig_inode,
Kazuya Mio446aaa62009-11-24 10:28:48 -0500458 struct inode *donor_inode, __u64 orig_start,
459 __u64 donor_start, __u64 *len)
Akira Fujita748de672009-06-17 19:24:03 -0400460{
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400461 __u64 orig_eof, donor_eof;
Akira Fujita70d5d3d2009-09-16 14:28:22 -0400462 unsigned int blkbits = orig_inode->i_blkbits;
463 unsigned int blocksize = 1 << blkbits;
464
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400465 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
466 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
467
468
Akira Fujita4a585792009-12-06 23:38:31 -0500469 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
470 ext4_debug("ext4 move extent: suid or sgid is set"
471 " to donor file [ino:orig %lu, donor %lu]\n",
472 orig_inode->i_ino, donor_inode->i_ino);
473 return -EINVAL;
474 }
475
Theodore Ts'o1f5a81e2010-06-02 22:04:39 -0400476 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
477 return -EPERM;
478
Akira Fujita748de672009-06-17 19:24:03 -0400479 /* Ext4 move extent does not support swapfile */
480 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
481 ext4_debug("ext4 move extent: The argument files should "
482 "not be swapfile [ino:orig %lu, donor %lu]\n",
483 orig_inode->i_ino, donor_inode->i_ino);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400484 return -EBUSY;
Akira Fujita748de672009-06-17 19:24:03 -0400485 }
486
Theodore Ts'odaf647d2016-04-01 01:31:28 -0400487 if (IS_NOQUOTA(orig_inode) || IS_NOQUOTA(donor_inode)) {
488 ext4_debug("ext4 move extent: The argument files should "
489 "not be quota files [ino:orig %lu, donor %lu]\n",
490 orig_inode->i_ino, donor_inode->i_ino);
491 return -EBUSY;
492 }
493
Akira Fujita748de672009-06-17 19:24:03 -0400494 /* Ext4 move extent supports only extent based file */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400495 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
Akira Fujita748de672009-06-17 19:24:03 -0400496 ext4_debug("ext4 move extent: orig file is not extents "
497 "based file [ino:orig %lu]\n", orig_inode->i_ino);
498 return -EOPNOTSUPP;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400499 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
Akira Fujita748de672009-06-17 19:24:03 -0400500 ext4_debug("ext4 move extent: donor file is not extents "
501 "based file [ino:donor %lu]\n", donor_inode->i_ino);
502 return -EOPNOTSUPP;
503 }
504
505 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
506 ext4_debug("ext4 move extent: File size is 0 byte\n");
507 return -EINVAL;
508 }
509
510 /* Start offset should be same */
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400511 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
512 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
Akira Fujita748de672009-06-17 19:24:03 -0400513 ext4_debug("ext4 move extent: orig and donor's start "
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400514 "offset are not alligned [ino:orig %lu, donor %lu]\n",
Akira Fujita748de672009-06-17 19:24:03 -0400515 orig_inode->i_ino, donor_inode->i_ino);
516 return -EINVAL;
517 }
518
Lukas Czernerf17722f2011-06-06 00:05:17 -0400519 if ((orig_start >= EXT_MAX_BLOCKS) ||
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400520 (donor_start >= EXT_MAX_BLOCKS) ||
Lukas Czernerf17722f2011-06-06 00:05:17 -0400521 (*len > EXT_MAX_BLOCKS) ||
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400522 (donor_start + *len >= EXT_MAX_BLOCKS) ||
Lukas Czernerf17722f2011-06-06 00:05:17 -0400523 (orig_start + *len >= EXT_MAX_BLOCKS)) {
Eric Sandeen0a80e982009-09-17 11:55:58 -0400524 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
Lukas Czernerf17722f2011-06-06 00:05:17 -0400525 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
Akira Fujita748de672009-06-17 19:24:03 -0400526 orig_inode->i_ino, donor_inode->i_ino);
527 return -EINVAL;
528 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400529 if (orig_eof < orig_start + *len - 1)
530 *len = orig_eof - orig_start;
531 if (donor_eof < donor_start + *len - 1)
532 *len = donor_eof - donor_start;
Akira Fujita748de672009-06-17 19:24:03 -0400533 if (!*len) {
Akira Fujita92c28152009-11-23 07:24:50 -0500534 ext4_debug("ext4 move extent: len should not be 0 "
Akira Fujita748de672009-06-17 19:24:03 -0400535 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
536 donor_inode->i_ino);
537 return -EINVAL;
538 }
539
540 return 0;
541}
542
543/**
Akira Fujita748de672009-06-17 19:24:03 -0400544 * ext4_move_extents - Exchange the specified range of a file
545 *
546 * @o_filp: file structure of the original file
547 * @d_filp: file structure of the donor file
Xiaoguang Wang65dd8322014-10-11 19:56:34 -0400548 * @orig_blk: start offset in block for orig
549 * @donor_blk: start offset in block for donor
Akira Fujita748de672009-06-17 19:24:03 -0400550 * @len: the number of blocks to be moved
551 * @moved_len: moved block length
552 *
553 * This function returns 0 and moved block length is set in moved_len
554 * if succeed, otherwise returns error value.
555 *
Akira Fujita748de672009-06-17 19:24:03 -0400556 */
557int
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400558ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
559 __u64 donor_blk, __u64 len, __u64 *moved_len)
Akira Fujita748de672009-06-17 19:24:03 -0400560{
Al Viro496ad9a2013-01-23 17:07:38 -0500561 struct inode *orig_inode = file_inode(o_filp);
562 struct inode *donor_inode = file_inode(d_filp);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400563 struct ext4_ext_path *path = NULL;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300564 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400565 ext4_lblk_t o_end, o_start = orig_blk;
566 ext4_lblk_t d_start = donor_blk;
567 int ret;
Akira Fujita748de672009-06-17 19:24:03 -0400568
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -0400569 if (orig_inode->i_sb != donor_inode->i_sb) {
570 ext4_debug("ext4 move extent: The argument files "
571 "should be in same FS [ino:orig %lu, donor %lu]\n",
572 orig_inode->i_ino, donor_inode->i_ino);
573 return -EINVAL;
574 }
575
576 /* orig and donor should be different inodes */
577 if (orig_inode == donor_inode) {
Theodore Ts'of3ce8062009-09-28 15:58:29 -0400578 ext4_debug("ext4 move extent: The argument files should not "
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -0400579 "be same inode [ino:orig %lu, donor %lu]\n",
Theodore Ts'of3ce8062009-09-28 15:58:29 -0400580 orig_inode->i_ino, donor_inode->i_ino);
581 return -EINVAL;
582 }
583
Akira Fujita7247c0c2010-03-04 00:34:58 -0500584 /* Regular file check */
585 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
586 ext4_debug("ext4 move extent: The argument files should be "
587 "regular file [ino:orig %lu, donor %lu]\n",
588 orig_inode->i_ino, donor_inode->i_ino);
589 return -EINVAL;
590 }
Eric Whitney04e22412015-06-21 21:38:03 -0400591
592 /* TODO: it's not obvious how to swap blocks for inodes with full
593 journaling enabled */
Dmitry Monakhovf0660552012-09-26 12:32:54 -0400594 if (ext4_should_journal_data(orig_inode) ||
595 ext4_should_journal_data(donor_inode)) {
Eric Whitney04e22412015-06-21 21:38:03 -0400596 ext4_msg(orig_inode->i_sb, KERN_ERR,
597 "Online defrag not supported with data journaling");
598 return -EOPNOTSUPP;
Dmitry Monakhovf0660552012-09-26 12:32:54 -0400599 }
Eric Whitney04e22412015-06-21 21:38:03 -0400600
Akira Fujitafc04cb42009-11-23 07:24:43 -0500601 /* Protect orig and donor inodes against a truncate */
J. Bruce Fields375e2892012-04-18 15:16:33 -0400602 lock_two_nondirectories(orig_inode, donor_inode);
Akira Fujita748de672009-06-17 19:24:03 -0400603
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400604 /* Wait for all existing dio workers */
605 ext4_inode_block_unlocked_dio(orig_inode);
606 ext4_inode_block_unlocked_dio(donor_inode);
607 inode_dio_wait(orig_inode);
608 inode_dio_wait(donor_inode);
609
Akira Fujitafc04cb42009-11-23 07:24:43 -0500610 /* Protect extent tree against block allocations via delalloc */
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400611 ext4_double_down_write_data_sem(orig_inode, donor_inode);
Akira Fujita748de672009-06-17 19:24:03 -0400612 /* Check the filesystem environment whether move_extent can be done */
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400613 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
614 donor_blk, &len);
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -0400615 if (ret)
Akira Fujita347fa6f2009-09-16 14:25:07 -0400616 goto out;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400617 o_end = o_start + len;
Akira Fujita748de672009-06-17 19:24:03 -0400618
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400619 while (o_start < o_end) {
620 struct ext4_extent *ex;
621 ext4_lblk_t cur_blk, next_blk;
622 pgoff_t orig_page_index, donor_page_index;
623 int offset_in_page;
624 int unwritten, cur_len;
Akira Fujita748de672009-06-17 19:24:03 -0400625
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400626 ret = get_ext_path(orig_inode, o_start, &path);
627 if (ret)
Akira Fujita748de672009-06-17 19:24:03 -0400628 goto out;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400629 ex = path[path->p_depth].p_ext;
630 next_blk = ext4_ext_next_allocated_block(path);
631 cur_blk = le32_to_cpu(ex->ee_block);
632 cur_len = ext4_ext_get_actual_len(ex);
633 /* Check hole before the start pos */
634 if (cur_blk + cur_len - 1 < o_start) {
635 if (next_blk == EXT_MAX_BLOCKS) {
636 o_start = o_end;
637 ret = -ENODATA;
638 goto out;
639 }
640 d_start += next_blk - o_start;
641 o_start = next_blk;
Theodore Ts'o3bdf14b2014-09-01 14:42:09 -0400642 continue;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400643 /* Check hole after the start pos */
644 } else if (cur_blk > o_start) {
645 /* Skip hole */
646 d_start += cur_blk - o_start;
647 o_start = cur_blk;
648 /* Extent inside requested range ?*/
649 if (cur_blk >= o_end)
650 goto out;
651 } else { /* in_range(o_start, o_blk, o_len) */
652 cur_len += cur_blk - o_start;
Akira Fujita748de672009-06-17 19:24:03 -0400653 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400654 unwritten = ext4_ext_is_unwritten(ex);
655 if (o_end - o_start < cur_len)
656 cur_len = o_end - o_start;
Akira Fujita748de672009-06-17 19:24:03 -0400657
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300658 orig_page_index = o_start >> (PAGE_SHIFT -
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400659 orig_inode->i_blkbits);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300660 donor_page_index = d_start >> (PAGE_SHIFT -
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400661 donor_inode->i_blkbits);
662 offset_in_page = o_start % blocks_per_page;
663 if (cur_len > blocks_per_page- offset_in_page)
664 cur_len = blocks_per_page - offset_in_page;
Akira Fujitafc04cb42009-11-23 07:24:43 -0500665 /*
666 * Up semaphore to avoid following problems:
667 * a. transaction deadlock among ext4_journal_start,
668 * ->write_begin via pagefault, and jbd2_journal_commit
669 * b. racing with ->readpage, ->write_begin, and ext4_get_block
670 * in move_extent_per_page
671 */
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400672 ext4_double_up_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400673 /* Swap original branches with new branches */
674 move_extent_per_page(o_filp, donor_inode,
675 orig_page_index, donor_page_index,
676 offset_in_page, cur_len,
677 unwritten, &ret);
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400678 ext4_double_down_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -0400679 if (ret < 0)
Akira Fujitafc04cb42009-11-23 07:24:43 -0500680 break;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400681 o_start += cur_len;
682 d_start += cur_len;
Akira Fujita748de672009-06-17 19:24:03 -0400683 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400684 *moved_len = o_start - orig_blk;
685 if (*moved_len > len)
686 *moved_len = len;
687
Akira Fujita748de672009-06-17 19:24:03 -0400688out:
Akira Fujita94d7c162009-11-24 10:19:57 -0500689 if (*moved_len) {
690 ext4_discard_preallocations(orig_inode);
691 ext4_discard_preallocations(donor_inode);
692 }
693
Theodore Ts'ob7ea89a2014-09-01 14:39:09 -0400694 ext4_ext_drop_refs(path);
695 kfree(path);
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400696 ext4_double_up_write_data_sem(orig_inode, donor_inode);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -0400697 ext4_inode_resume_unlocked_dio(orig_inode);
698 ext4_inode_resume_unlocked_dio(donor_inode);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400699 unlock_two_nondirectories(orig_inode, donor_inode);
Akira Fujita748de672009-06-17 19:24:03 -0400700
Dmitry Monakhov03bd8b92012-09-26 12:32:19 -0400701 return ret;
Akira Fujita748de672009-06-17 19:24:03 -0400702}